Study Id50485 Artificial-Intelligence

In-depth: Artificial Intelligence
2019
Statista Digital Market Outlook
February 2019
Management summary (1/2)
Artificial intelligence (AI) essentially refers to computing technologies that are inspired by the ways
people use their brains and nervous systems to reason and make decisions, but typically they
operate quite differently. The concept of AI has been the source of inspiration for many science
fiction writers and futurologists for over a century. Today, advancements in computing and big data
have made it a reality, with machines now being deployed at a large scale across industries. The
application of AI technologies is driving growth at individual, business, and economic levels. In fact, AI
has started to outperform human beings in a range of work activities, including ones requiring
cognitive abilities.
The current AI ecosystem consists of machine learning, robotics, and artificial neural networks
(ANNs). In machine learning, programs learn from existing data and apply this knowledge to new data
or use it to predict data. The field of robotics is concerned with developing and training robots.
Usually, the ability of a robot to interact with people and the world follows general rules and is
predictable. However, current efforts also revolve around using deep learning to train robots to
manipulate situations and act with a certain degree of self-awareness. ANNs are built to mimic the
workings of a human brain. Connected units (artificial neurons) are organized in layers to process
information.
Over the last few decades, the evolution of AI has mostly revolved around the advancement of
linguistic, mathematical, and logical reasoning abilities. However, the next wave of AI advancements is
pushing towards developing emotional intelligence. At the same time, sequential learning, another
feature of Google’s DeepMind, is enabling AIs to learn multiple skills. Over the last few years, deep
learning has made vast improvements in enabling machines to comprehend the physical world to a
certain degree and is used across industries for various tasks. Among the leading economies, China
has invested a lot of research and money into AI in recent years.
One of the major factors driving the current wave of AI growth is the rapid increase in corporate
venture capital (CVC) investment in AI start-ups. On the technology front, rapid advancements in
computing power are driving the industry to the next level. Similarly, open source platforms are
promoting and enabling collaborative learning, which is conducive for the growth of AI. The current
wave of growth in the AI industry is as much about the abundant availability of big data as it is about
the software and hardware. The amount of big data being generated by today’s increasingly digitized
economy is growing at a rate of 40% each year and is expected to reach 44 zettabytes by 2020. This
growth in big data is driving the improvement of AI algorithms.
AI solutions are increasingly being customized to serve the needs of the automotive, healthcare,
education, finance, entertainment, and other industries. In the automotive sector, AI is primarily used
to power autonomous cars, with these systems expected to become standard in new vehicles in the
medium to long term. In the healthcare industry, developments in the field of AI and machine
learning have not only accelerated the pace of innovation in the industry but are also changing entire
operating models. In the education industry, there are attempts to provide customized learning
programs for each student using AI, while in the finance industry, AI wealth management solutions
can offer higher personalization.
2
Management summary (2/2)
With the rise of AI, more and more start-ups venture into the market. Most of them work in the field
of machine learning algorithms, followed by natural language processing. The annual global funding
of AI start-ups experienced an average growth rate of almost 70%, increasing from US$0.6 billion in
2012 to US$4.9 billion in 2016. In the same time frame, the corresponding number of deals grew by
47% from 151 in 2012 to 703 in 2016. In terms of M&A, the number of deals jumped up by 38% in
2017. Looking at the most recent M&A deals, big tech companies like Google, Apple, Amazon,
Microsoft, IBM, or Facebook have often been the acquirer. But Chinese tech giants like Baidu or rising
stars in the start-up world like Twitter, Uber, or Spotify also acquire AI companies.
Companies from various industries are currently developing AI and related applications. Google, IBM
and Microsoft are leading AI innovations in the IT industry, while Amazon and eBay are investing in AI
to improve their eCommerce platform, and ridesharing company Uber is using AI on autonomous
driving, food deliveries, and mapping research. Collaborative development is on the rise, and leading
companies such as Amazon, Apple, Facebook, Google/DeepMind, IBM, and Microsoft are currently
working in partnership towards developing AI applications. The acquisition of small-scale AI
companies by tech giants like Apple, IBM, and Microsoft in relevant fields is on the rise, leading to a
decreasing learning curve. Other leading companies include Baidu, Facebook, and Salesforce.
Questions? U.S.: [email protected] EU: [email protected]
3
Table of contents (1/3)
Management summary 02
Table of contents 04
Introduction 07
▪ Overview 08
▪ Definitions 09
▪ Evolution of artificial intelligence 10
▪ Automation potential 13
▪ Global revenue projection 14
▪ Impact of AI 15
Technologies 20
▪ Machine learning 21
▪ Robotics 22
▪ Artificial neural networks 23
Trends 24
▪ Artificial emotional intelligence 25
▪ Sequential learning 28
▪ Deep learning 29
▪ Rise of China 31
▪ Growth in hardware and software 35
▪ Transfer learning 37
4
Drivers 38
▪ Corporate VC investment 39
▪ Computing power 41
▪ Open platforms 43
▪ Big data 45
Applications 46
▪ Automotive industry 48
▪ Healthcare 56
▪ Education 64
▪ Finance 67
▪ Entertainment 74
Start-ups: funding and M&A 76

▪ Overview 77
▪ AI start-ups by funding 79
▪ Mergers & acquisitions 92
▪ Recent M&A deals 93
Competitive landscape 102

▪ Company comparison 103
▪ Amazon 104
▪ Apple 106
▪ Baidu 107
▪ eBay 109
▪ Facebook 111
▪ Google 113
5
Competitive landscape (continued)

▪ IBM 116
▪ Microsoft 118
▪ Salesforce 120
▪ Uber 121
Appendix 123
▪ Glossary 124
▪ Statista Global Consumer Survey 125
▪ Statista Digital Market Outlook 126
▪ Statista Research & Analysis, Statista Content & Design 127
▪ Authors, imprint, and disclaimer 128
6
Introduction
Artificial intelligence (AI) essentially refers to computing technologies that are
inspired by the ways people use their brains and nervous systems to reason
and make decisions, but they typically operate quite differently.
The concept of AI has been the source of inspiration for many science fiction
writers and futurologists for over a century. Today, advancements in
computing and big data have made it a reality, with machines now being
deployed at a large scale across industries. The application of AI technologies
is driving growth at individual, business, and economic levels. In fact, AI has
started to outperform human beings in a range of work activities, including
ones requiring cognitive abilities.
7
Artificial intelligence mimics the
working of the human brain
Overview
Artificial intelligence (AI) essentially refers to computing technologies that are inspired by the ways
people use their brains and nervous systems to reason and make decisions, but they typically
operate quite differently. The concept of AI has been the source of inspiration for many science
fiction writers and futurologists for over a century. Today, advancements in computing and big data
have made it a reality, with machines now being deployed at a large scale across industries. In fact, AI
has started to outperform human beings in a range of work activities, including ones requiring
cognitive abilities.
Current artificial intelligence ecosystem
Artificial intelligence
Artificial neural
Machine learning Robotics
networks
Supervised
Soft robotics Deep learning
learning
Swarm robotics
Unsupervised Convolutional
learning Touch robotics neural networks
Humanoid robots
Reinforcement Recurrent neural

Serpentine robots
learning networks
8
Source: McKinsey
There are three main types of
artificial intelligence
Definitions
Machine learning
This involves designing new learning algorithms and improving existing ones to enable computers to
act without explicit programming. These algorithms allow computers to analyze large volumes of
complex data to recognize patterns and make predictions and adjustments.
The different types of machine learning are:
▪ Supervised learning
▪ Unsupervised learning
▪ Reinforcement learning
Robotics
This branch of technology is concerned with developing and training robots to interact with people
and the world in general in predictable ways. However, current efforts also revolve around using
deep learning to train robots to manipulate situations and act with a certain degree of self-
awareness. The main fields within robotics are:
▪ Soft robotics
▪ Swarm robotics
▪ Touch robotics
▪ Humanoid robots
▪ Serpentine robots
Artificial neural networks (ANN)

This area is concerned with developing algorithms that mimic the functioning of the neocortex area
of the human brain, where all the thinking occurs. This comparison is not entirely correct because in
a human brain, neurons are not arranged in a linear sequence, as is the case with ANNs.
There are three different types of ANN:
▪ Deep learning
▪ Convolutional neural networks
▪ Recurrent neural network
9
Artificial intelligence has evolved
from being reactive to being aware
Evolution of artificial intelligence (1/3)
Evolution of artificial intelligence
Reactive machines
▪ Most basic AI systems
▪ Cannot use past experiences to influence current decisions
▪ Cannot deviate from assigned tasks

Phase 1 ▪ No interactive participation with the world
▪ Behave exactly the same way each time for a particular situation
▪ Example: IBM’s Deep Blue
Memory-forming machines
▪ Can form memories and draw on past experiences
▪ Cannot make significant improvisations based on memories

Phase 2
▪ Example: self-driving cars
Partially aware machines

▪ Current technologies are trying to reach this stage,
with limited success
▪ Here, the AI system will be influenced by other entities and

will learn contextually
Phase 3
▪ Example: The U.S. Navy is using video simulations to teach
robots human ethics
Fully aware machines

▪ The final phase of AI evolution
▪ Fully self-aware and conscious

Phase 4
▪ Ability to learn from scratch without instructions
▪ Example: Creators of the Deep Q Network claim that it can
play video games without any human instructions
10
Source: Company information
AI is being deployed at scale across
industries
The application of AI technologies is driving growth at individual, business, and economic levels. As
the percentage of the working population declines in many countries around the world, AI provides
the much-needed automation to sustain and accelerate productivity growth at a micro and macro
level.
At the micro level, businesses are now adopting different AI technologies to capture benefits such as
lower labor costs, increased throughput, enhanced quality, and lower downtimes. At the macro level,
automation is expected to result in robust productivity growth; McKinsey estimates that it will be
between 0.8% and 1.4% annually.
Even though this growth has been estimated for all the countries around the world, four economies,
Japan, India, China, and the U.S., which account for the majority of activities that are potentially
automatable, are expected to witness the highest adoption of AI technologies.
Automation potential varies across industries as well. For example, according to a McKinsey report,
the manufacturing industry has an automation potential of 64%, whereas for an industry like
educational services, it is 34%. The factors affecting the pace and extent of automation include
openness to change, cost of developing new technologies, labor market dynamics, the regulatory
framework, and economic benefits. The highest levels of disruption can be found in industries such
as healthcare, financial services, the automotive industry, and education, where advances in AI are
forcing dramatic business model changes for concerned stakeholders.
Timeline: AI evolution
Artificial intelligence
Machine learning
Deep learning
1950 1960 1970 1980 1990 2000 2010
11
Source: nvidia.com
Human-machine relationships are
integral to the progress of AI
However, in spite of these advancements, especially those involving the application of cognitive
thinking, machines are still limited when it comes to improvisation. They mostly follow programmed
algorithms that only allow them to act in a pre-determined manner for each conceived situation and
are therefore subject to a fundamental limitation of data-driven statistical inference. They come up
short when faced with a novel situation since they do not yet have the ‘common sense’ that is the
hallmark of human experience. That said, machines are slowly but steadily starting to beat humans in
a variety of fields, even in those that require superior cognitive abilities. A few examples are:
▪ Libratus: This is an AI developed by Carnegie Mellon University. It was the first computer to defeat
four poker professionals. Owing to the margin of victory, experts were 99.98% sure that the wins
were not due to chance.
▪ AlphaGo: This is an AI system developed by Google’s DeepMind. It beat Lee Sedol, widely regarded
as the world’s best player of Go, an ancient Chinese board game with innumerable permutations
and combinations.
▪ A project by Google’s DeepMind and the University of Oxford, which applied deep learning to a
huge data set of BBC programs to create a lip-reading system, comfortably outperformed a
professional human lip-reader.
All these developments have led to much conjecture as to the future development of AI and the
evolution of the man-machine relationship. Elon Musk, co-founder of Tesla, has called AI “our biggest
existential threat”. Stephen Hawking has strongly warned that a full AI could result in the end of
mankind. Also, philosopher Nick Bostrom, in his widely acclaimed book Superintelligence, speaks
about the possibility of computers overtaking humans in terms of cognitive capabilities. However,
keeping the recent advancements and resulting benefits of AI in mind, along with the need for
human intervention at various levels, the future is expected to see a collaborative approach between
AI and humans.
One of the fascinating things about the search for AI is that it’s been so hard
to predict which parts would be easy or hard. At first, we thought that the
quintessential preoccupations of the officially smart few, like playing chess
or proving theorems—the corridas of nerd machismo—would prove to be
hardest for computers. In fact, they turn out to be easy. Things every dummy
can do, like recognizing objects or picking them up, are much harder. And it
turns out to be much easier to simulate the reasoning of a highly trained
adult expert than to mimic the ordinary learning of every baby.
Alison Gopnik, cognitive scientist
12
Source: Deloitte
Automation potential differs across
industries and countries
Automation potential
Automation potential across industries
64%
44%
42%
38%
34%
Manufacturing Finance and Arts, entertainment, Health care and Educational

insurance and recreation social assistance services
Source: EMSI, Oxford Economic Forecasting, U.S. Bureau of Labor Statistics, McKinsey analysis – data as of January 2017
Automation potential across countries
Japan 56%
India 52%
China 51%
Europe
(France, Germany, 46%
Italy, Spain, UK)
U.S. 46%
Rest of the world 50%
Source: EMSI, Oxford Economic Forecasting, U.S. Bureau of Labor Statistics, McKinsey analysis – data as of January 2017
13
AI revenues are expected to grow at
a CAGR1 of 45%
Global revenue projection
Global revenue projection in billion US$
89.8
71.0
53.2
+45%1
38.0
26.0
17.3
11.3
7.3
4.8
3.2
2016 2017 2018 2019 2020 2021 2022 2023 2024 2025
14
1: CAGR: Compound Annual Growth Rate / average growth rate per year
Source: Tractica
AI is expected to have a big impact
on GVA1 growth rates
Impact of AI (1/5)
A study by Accenture and Frontier Economics expects AI to have a big impact on a country’s gross
value added (GVA). It estimates the annual GVA growth rates in 2035 for a baseline state, based on
current assumptions regarding economic growth, and for an AI steady state, assuming artificial
intelligence is integrated into economic processes.
The extent of the impact depends on the country, ranging from a 0.8 percentage points increase in
potential GVA growth rates in Italy or Spain to 2.0 percentage points in Finland or the U.S.
Potential annual GVA1 growth rates in 2035 in %
Baseline state AI steady state
China 6.3
7.9
India 5.8
7.1
S. Korea 2.9
4.9
U.S. 2.6
4.6
Finland 2.1
4.1
UK 2.5
3.9
Sweden 1.7
3.6
Canada 2.1
3.5
Netherlands 1.6
3.2
Germany 1.4
3.0
Austria 1.4
3.0
France 1.7
2.9
Japan 0.8
2.7
Belgium 1.6
2.7
Spain 1.7
2.5
Italy 1.0
1.8
15
1 Gross value added: a close approximation of a country‘s GDP
Source: Accenture, Frontier Economics
AI has the potential to increase
labor productivity
Impact of AI (2/5)
The study by Accenture and Frontier Economics also estimates that AI has the potential to increase
labor productivity in countries. The impact ranges from an 11 percentage points increase in Spain to
37 percentage points in Sweden.
Impact of AI on labor productivity in developed countries in 2035
Sweden 37%
Finland 36%
U.S. 35%
Japan 34%
Austria 30%
Germany 29%
Netherlands 27%
UK 25%
France 20%
Belgium 17%
Italy 12%
Spain 11%
16
AI has the potential to increase the
growth rate of industries
Impact of AI (3/5)
The study by Accenture and Frontier Economics also estimates that AI has the potential to increase
economic growth rates by a weighted average of 1.7 percentage points by 2035 across 16 industries.
Impact of AI on industry growth in 2035 in %
Information & 3.4

communication 4.8
Manufacturing 2.3
4.4
Financial services 2.4
4.3
Wholesale & retail 2.0
4.0
Transportation 2.1
& storage 4.0
Professional services 2.3
3.8
Healthcare 2.2
3.4
Construction 2.3
3.4
Agriculture, forestry 1.3
& fishing 3.4
Accommodation 1.4
& food services 3.2
Utilities 1.4
3.1
Arts, entertainment 1.9
& recreation 3.1
Social services 1.6
2.8
Public services 0.9
2.3
Other services 0.7
1.7
Education 0.9
1.6
17
AI can boost manufacturing output
by US$4 trillion
Impact of AI (4/5)
The study by Accenture and Frontier Economics also estimates that AI can substantially raise the
economic output of industries. For manufacturing alone, AI can boost GVA1 by almost US$4 trillion in
2035.
Impact of AI on industry output in 2035 in trillion US$
Manufacturing 8.4
12.2
Professional services 7.5
9.3
Wholesale & retail 6.2
8.4
Public services 4.0
4.9
Information & 3.7
communication 4.7
Financial services 3.4
4.6
Construction 2.8
3.3
Transportation 2.1
& storage 2.9
Healthcare 2.3
2.7
Accommodation 1.5
& food services 2.0
Social services 1.1
1.3
Utilities 1.0
1.3
Education 1.1
1.2
Agriculture, forestry 0.6
& fishing 0.8
Other services 0.5
0.6
Arts, entertainment 0.5
& recreation 0.5
18
1 Gross value added – a close approximation of a country‘s GDP
AI has potential incremental value
Impact of AI (5/5)
A study by McKinsey on more than 400 use cases across 19 industries and nine business functions
highlights the use and economic potential of advanced AI techniques. In more than two thirds of use
cases, AI can improve performance as compared to other analytics techniques.
Potential incremental value of AI as compared to other analytics techniques
Travel 128%
Transport
89%
and logistics
Retail 87%
Automotive industry
85%
and assembly
High tech 85%
Oil and gas 79%
Chemicals 67%
Media and
57%
entertainment
Basic materials 56%
Agriculture 55%
Consumer packaged
55%
goods
Banking 50%
Healthcare systems
44%
and services
Public and
44%
social sectors
Telecommunications 44%
Pharmaceuticals
39%
and medical products
Insurance 38%
Advanced electronics 36%

Aerospace
30%
and defense
19
Source: McKinsey
Technologies
The current AI ecosystem consists of machine learning, robotics, and artificial
neural networks. In machine learning, programs learn from existing data and
apply this knowledge to new data or use it to predict data. The field of
robotics is concerned with developing and training robots. Usually, the ability
of a robot to interact with people and the world follows general rules and is
predictable.
However, current efforts also revolve around using deep learning to train
robots to manipulate situations and act with a certain degree of self-
awareness. Artificial neural networks (ANNs) are built to mimic the working of
a human brain. Connected units (artificial neurons) are organized in layers to
process information.
20
Machine learning applies insights
from existing data to new data
Machine learning
In Machine learning, programs learn from existing data and apply this knowledge to new data or use
it to predict data. Machine learning involves designing new learning algorithms and improving
existing ones to enable computers to act without explicit programming. These algorithms allow
computers to analyze large volumes of complex data and are used to complete tasks like
classification, regression, clustering, etc. The different types of machine learning are:
▪ Supervised learning: These techniques train the system to respond appropriately to particular
stimuli. For this, the learning algorithm is fed with a series of inputs as well as with the
corresponding outputs. The algorithm then applies this same set of rules in the future.
▪ Unsupervised learning: Here, the system is not provided with the right answer but is expected to
learn by itself. It does this by exploring the data on its own to find some sort of structure or
patterns. In other words, the AI system uses its experience of solving one problem to solve
another related problem. This type of machine learning can be applied to identify consumers with
similar purchasing behaviors in order to deliver personalized marketing, for example.
▪ Reinforcement learning: Inspired by behaviorist psychology, the algorithm learns through a trial
and error process in which the actions are either virtually ‘rewarded’ or ‘punished’. It then forms a
memory of each experience and uses this learning for subsequent experiences. DeepMind’s (a
Google AI company) win over the world champion in the game of Go is an example of
reinforcement learning.
Illustration of the machine learning process
Machine learning
▪ Supervised learning: specified input

and output
Training data Algorithm ▪ Unsupervised learning: pattern
recognition from past experiences
▪ Reinforcement learning: trial and error
with rewards and punishment
Application of
Output learning to new
data
Source: sas.com, McKinsey
21
AI is used to advance the behavior
potential of robots
Robotics
The field of robotics is concerned with developing and training robots. Usually, the capabilities of a
robot to interact with people and the world follows general rules and is predictable.
However, current efforts also revolve around using deep learning to train robots to manipulate
situations and act with a certain degree of self-awareness. Advances in machine learning, including
computer vision and tactile perception, will continue to be key enablers in advancing the capabilities
of robotics.
Currently, there are the following general types of robots:

▪ Soft robotics: These robots are built out of soft and deformable materials, which gives them the
ability to mimic the movements of living beings. These structures can achieve complex movements
and are more adaptable than traditional rigid robots. For example, Soft Robotics Inc. makes
robotic grippers that are used to handle tender items such as soft foods without damaging them.
▪ Swarm robotics: a field of robotics that deals with the deployment of a large number of mini-
robots that often mimic insects or animals which operate collectively, such as ants or bees.
▪ Touch robotics: Typically used to perform surgeries, these robots deliver a sense of touch, feel,
and vision to the operator. They are usually designed as biologically inspired hands.
▪ Humanoid robots: robots similar in structure to a human being, with a torso, head, arms, and legs.
Some robots might only model a part of the body, for example the upper body. Android robots
resemble a male body, while Gynoids resemble a female body.
▪ Serpentine robots: Robots that are designed to mimic the movement of snakes in order to
navigate through tightly packed spaces.
There is an exponential pace of improvement in hardware and machine

learning algorithms. The computational power required has gone down a
lot. The result is a new class of machines that can operate by themselves in
human space, the advance guard of a new robot industry.
Uriah Baalke, co-founder of Dispatch, 2016
22
Artificial neural networks mimic the
working of a brain
Artificial neural networks
Artificial neural networks (ANNs) are built to mimic the working of a human brain. Connected units
(artificial neurons) are organized in layers to process information. Each unit can transmit a signal to
another unit and thereby simulate a human brain. While neurons in a brain, however, are connected
in a complex and unpredictable manner, artificial neurons are arranged in a linear sequence. The
overall process of converting input into output is based on the programming of each neuron.
There are three types of artificial neural networks:
▪ Deep learning: These algorithms have many layers of neural networks which process information
at many levels. Before the advent of deep learning, ANNs often only had three layers, unlike deep
learning networks, which usually have over 10 layers. This branch of machine learning is especially
important because it is the first family of algorithms that does not require manual intervention.
Instead, it learns from raw data, very much like a human brain does, making use of different types
of sensory inputs. Google, with vast data reserves and advanced computing resources, is the hub
for deep learning across the world. The founding of Google Brain by Andrew Ng marked the
beginning of deep learning at Google. The main difference between deep learning and other
machine learning techniques is that larger neural networks keep improving their performance as
they get access to more and more data, whereas other techniques plateau at an earlier point.
▪ Convolutional neural networks (CNN): These are very similar to ordinary neural networks in their
overall working. The only difference is that the connections between neural layers are similar to
those seen in the animal visual cortex, the part of the brain that processes images. These
architectures are programmed to perceive each input as an image. In the 2012 ImageNet visual
recognition contest, the only entry using a CNN achieved an 84% correct score, compared to 75%
the year before. Since then, they have won all subsequent ImageNet contests, exceeding human
performance in 2015 with a score of 90%.
▪ Recurrent neural network (RNN): These neural networks differ from others in terms of their
architecture. Their neurons are connected to each other, thereby allowing them to send feedback
signals to each other. Here, the information travels in loops from layer to layer so that each bit of
information can be stored as memory and the network can exhibit dynamic behavior. It is due to
this that RNNs have been found to be apt for natural language processing applications.
23
Trends
Over the last few decades, the evolution of AI has mostly revolved around the
advancement of linguistic, mathematical, and logical reasoning abilities.
However, the next wave of AI advancements is pushing towards developing
emotional intelligence. At the same time, sequential learning, another feature
of Google’s DeepMind, is enabling AIs to learn multiple skills.
Over the last few years, deep learning has made vast improvements in
enabling machines to comprehend the physical world to a certain degree and
is used across industries for various tasks. Among the leading economies,
China has been investing a lot of research and money into AI in recent years.
24
AI moves towards artificial
emotional intelligence
Artificial emotional intelligence (1/3)
Over the last few decades, the evolution of AI has mostly revolved around the advancement of
linguistic, mathematical, and logical reasoning abilities. Computers have been able to make massive
calculations and analyze huge amounts of data to recognize patterns much beyond the abilities of
human beings. One area where they have lacked, however, is the ability to recognize and display
human emotions. It has often been thought that the challenge of creating emotionally intelligent
computing systems is just too big to be met anytime soon.
However, with the rising popularity of technologies such as chatbots, it is evident that humans are
drawn to human-like interaction. Therefore, the next wave of AI advancements is aiming to address
just that.
According to Bronwyn van der Merwe, group director at Fjord Australia and New Zealand, Accenture
Interactive’s design and innovation arm, the most successful AI systems in the future will be those
that are able to demonstrate emotional intelligence (EI) very similar to human interaction. EI can be
defined as the ability of individuals to distinguish different feelings and use those emotions to guide
their thinking and behavior.
The growth of EI in machines has been aided by the advancements in our understanding of how the
brain works, also called the neuroscience of EI. There is now overwhelming proof that different
regions of the brain stimulate different responses and emotions. Therefore, if thinking can indeed be
understood as a step-by-step process with a certain degree of sequence, AI machines can also be
programmed to exhibit different emotions and achieve different levels of consciousness. EI improves
an AI system’s ability to not only understand input but also adapt and improvise to provide human-
like responses in real time.
Interestingly, it is the very lack of inherent emotions in machines that make them such good
candidates for EI. According to research, there is a direct correlation between a higher emotional
quotient (EQ) in humans and lower emotional intelligence. According to Tomas Chamorro-Premuzic,
Professor of Business Psychology at University College London and Columbia University, a “higher EQ
is about controlling one’s impulses and inhibiting strong emotions in order to act rationally and minimize
emotional interference”.
As human beings, we have contextual understanding and we have empathy,

and right now there isn’t a lot of that built into AI. We do believe that in the
future the companies that are going to succeed will be those that can build
into their technology that kind of understanding.
Bronwyn van der Merwe, group director at Fjord Australia & New Zealand,
2017
25
Source: Harvard Business Review, MIT
Emotionally intelligent AI systems
have various applications
Nowadays, sensors and other devices are developing advanced abilities to observe and recognize
facial features, body posture, gestures, speech, and physical states. The development of neural
networks and Graphic Processing Units (GPUs) has also played a major role in integrating emotional
recognition features in machines.
Examples of companies integrating varying levels of EI into their AI systems include:
Drive.ai: The autonomous car start-up Drive.ai programs vehicles to not

only recognize people and other things in their environment but also to
interact with the driver through lights, sounds, and movement. In August
2016, the company launched a retrofit kit complete with a sensor array,
computer, and an LED sign to help vehicles communicate with
pedestrians and other drivers.
Cogito Corporation: provides EI for healthcare with AI-driven real-time

voice analysis for insurance call centers and other organizations. Its
voice analytics tools analyze every nanosecond of the conversation and
offer simultaneous recommendations and feedback, such as "you're
speaking too much“, "frequent overlaps," "breathe, pause between
phrases“, or "you're tense.“
Nuralogix: has developed technology that claims to be more accurate

than a polygraph owing to its EI abilities. It uses Transdermal Optical
Imaging and machine learning algorithms to observe blood flow in the
face to reveal hidden emotions. Law enforcement agencies would be
able to use this technology to determine if a subject is lying based on
the blood flow in their faces, an element which cannot be controlled
physically.
Affectiva: a company that spun out of MIT’s media lab, lends its
emotional AI technology to over 1,400 brands worldwide, including CBS,
MARS, and Kellogg’s, some of whom use it to measure the emotional
effects of advertisements. The technology can also be used to assess the
condition of a driver before switching from autonomous to manual
driving.
26
Source: Business Insider, Harvard Business Review, MIT
Huawei aims to build an emotionally
intelligent voice assistance software
More examples of companies integrating varying levels of EI into their AI systems include:
Huawei: is looking to develop an emotionally intelligent voice assistance

software that keeps a conversation going for as long as possible in order
to ensure that the user does not feel alone. According to Felix Zhang,
vice president of software engineering at Huawei, the fictional character
of Samantha, an EI-powered voice assistant in the Hollywood movie
‘Her’, is an inspiration for this technology.
Beyond Verbal: Founded in 2012, the company aims to develop a

technology to decipher emotions from vocal intonations to mainly help
marketing professionals improve customer experience and
communication campaigns. The factors that will be assessed include
valence, arousal, temper, and mood groups.
Receptiviti: Founded in 2014, this Toronto-based company has used NLP

to develop a proprietary technology called Linguistic Inquiry and Word
Count, or LIWC2015, which assesses an individual’s handwriting to
gather insights into their character, emotions, levels of deception, and
decision-making ability.
BRAIQ: While most technologies related to autonomous cars are

centered on how the car interacts with the environment, the technology
being developed by BRAIQ focuses on how the car interacts with its
passengers. This AI system looks to intuitively read a driver’s emotional
signals and to interact with him/her accordingly.
CrowdEmotion: Founded in 2013, this London-based company has

developed a technology to gauge human emotions visually by analyzing
the movements of 43 muscles on an individual’s face. The company
currently has a cloud-based version of the technology and has many
clients, including the BBC.
27
Source: Business Insider, Harvard Business Review, MIT
Sequential learning enables AIs to
learn multiple skills
Sequential learning
For many years, AI scientists have been trying to create ‘general AI’, which is essentially the ability of
the AI to draw on past experiences and analyze them to come up with the best solution to a novel
problem. However, this holy grail has proven to be elusive as scientists haven’t been able to create an
AI system with the ability to build multiple skills on top of each other. In other words, neural
networks, which form the core of deep learning, can achieve excellence in learning a particular task,
such as playing chess or poker, through countless rounds of trial and error. However, once the
network is trained to, say, play chess, it cannot learn another game or task without overwriting its
chess-playing skills. This shortcoming is called ‘catastrophic forgetting’.
However, in March 2017, Google’s DeepMind devised an AI technique called sequential learning that
helped the computers preserve the neural connections that have been most important to learn a
particular task before moving on to something else. This technique draws heavily from neuroscience
studies on animals which demonstrate how they learn continually by preserving brain connections
used for different skills learned in the past.
If we’re going to have computer programs that are more intelligent and
more useful, then they will have to have this ability to learn sequentially.
James Kirkpatrick, research scientist at DeepMind, 2017
The new AI system was put to the test by making it play 10 Atari games in random order. After many
days on each game, the results showed that the AI was as good as a human player on each of those
games. One of the key observations during this experiment was that while playing Enduro, a car
racing game that takes place in different weather conditions and times of the day, the AI treated each
environment setting as a different task.
However, the research team agreed that even though the AI learned to play different games, it did
not master each one as well as a dedicated AI would have done. This was mainly because it
sometimes failed to understand the importance of certain connections for a particular game.
Therefore, even though this was considered to be a major breakthrough in the quest to develop
human-like general AI, there is still some way to go for it to become a reality. Other companies
engaged in similar research include NNaisense, SwiftKey/Microsoft Research, and Facebook AI
Research.
28
Deep learning is used across
industries for various tasks
Deep learning (1/2)
Over the last few years, deep learning has made vast improvements in enabling machines to
comprehend the physical world to a certain degree. Historically, AI-powered computers have been
able to perform only those tasks for which they were explicitly programmed. Even though this has
worked well to make large calculations and achieve such feats as beating professionals at games like
poker, chess, and Go, it has come up short when performing human-like tasks such as recognizing
faces or answering novel questions.
Taking AI a step further, deep learning systems allow machines to make sense of data themselves
and learn as they experience more and more. One of the major breakthroughs in the industry was
made by a team at Stanford led by Andrew Ng, who figured out that graphics processing unit chips
(GPUs), which were originally made to meet the visual processing demands of video games, could be
repurposed for deep learning. The use of these chips to operate neural networks has increased the
capabilities of deep learning manifold. Deep learning is now being used across industries for tasks
such as detecting gene abnormalities, predicting weather patterns, and identifying false insurance
claims.
Deep learning is effectively a computer’s version of the radioactive spider

from Spider Man. If computers can be taught to see, hear and understand
like humans it will become a lot easier to interact with them.
Andrew Ng, founder of Google Brain, 2016
29
One application of deep learning is
to detect fraud
Deep learning (2/2)
Listed below are a few industries where deep learning is making major advances:
Agriculture: Deep learning applications are using data from drones and
satellites to not only predict crop outputs but also to monitor global
water levels to help detect crop diseases before they spread. They are
also being used for less complex tasks, such as sorting good and bad
produce in farms, as demonstrated by Makoto Koike in Japan, who made
use of Google’s open-sourced network TensorFlow to sort cucumbers.
Autonomous cars: Even though there are plenty of cars with driver assist
features, they still require drivers to take over when an unforeseen
event occurs that the car is not programmed to handle. Companies like
Drive.ai and Nvidia are using deep learning to give cars full autonomy.
Nvidia uses CNNs to learn the entire process of steering a car.
Cybersecurity: The application of deep learning to enable a more

comprehensive and sophisticated detection of malware is the latest
trend in cybersecurity solutions. Companies such as Deep Instinct and
Altoros are applying deep learning techniques to not only devise
protection against existing cybersecurity threats but also to recognize
new ones, which have not been detected before. The deep learning
systems collect a very large number of files irrespective of their type and
run tests on them to classify them as either malicious or legitimate. This
data is then fed into the AI engine to enable it to make predictions for
future cases. According to tests conducted by Drebin University and
Siemens CERT, Deep Instinct’s solution was found to be 99.8% accurate
as compared to the 61.5% average score for the top 10 vendors.
Healthcare: By enabling doctors to look at a person’s gene sequence

and molecular makeup, deep learning is facilitating personalized
treatments at a very granular level. A good example is Deep Genomics,
which is using large volumes of data on human cellular composition to
enable machines to predict the outcomes of alterations to the genome.
In drug discovery, companies like Atomwise are making use of the
technique to analyze data on molecular compounds to discover drugs
for new or established diseases.
Insurance: Here, deep learning techniques are used not only to assess
claims at high speed but also to identify anomalies, indicating potential
claims fraud, and determine payouts for legitimate claims. Tractable is
an example of a company that uses images of damaged cars to teach
machines how to estimate future repair costs.
30
China is investing a lot of research
and money in AI
Rise of China: overview
In October 2016, the Obama administration released a strategic plan for AI research, which
mentioned that the U.S. was no longer the world leader in journal articles on deep learning. China
had overtaken them. A similar study by Japan's National Institute of Science and Technology Policy
found China to be a close second to the U.S. in terms of the number of AI studies presented at top
academic conferences in 2015. It is not only academic research where China is making serious
inroads in terms of global AI dominance. Chinese technology companies such as Baidu, Tencent,
Alibaba, and Didi Chuxing have all invested significantly in setting up AI research labs with access to
the huge amount of big data generated by the large Chinese digital economy.
In terms of global AI start-up funding, China ranks well above other countries, including the U.S. In
fact, of the US$15.2 billion invested in AI start-ups globally in 2017, 48% went to Chinese companies,
while only 38% went to companies in the U.S. China is also developing its first ‘national laboratory for
deep learning’. The National Development and Reform Commission (NDRC) has commissioned Baidu
to create the lab in partnership with Tsinghua and Beihang universities and other research institutes.
The lab’s area of focus is expected to include machine-learning-based visual recognition, voice
recognition, new types of human machine interaction, and deep learning intellectual property.
Another visible indication of Chinese dominance is the billions it is investing in U.S. start-ups, many of
them focusing on AI. According to CB Insights, Chinese investment in U.S. technology companies
totaled US$18.2 billion from 2011 to the third quarter of 2016, with US$9.9 billion being spent in
2015 alone. A KPMG study estimated that overall Chinese VC investment across all industries had hit
a record US$40 billion in 2017, with a major focus on AI. Moreover, according to a 2016 study by
Japanese company Astamuse, the number of AI-related patents being submitted by China have
grown at a rate of 186% during the period from 2010 to 2014 as compared to just 26% for the U.S.
China’s highly competitive ecosystem and a quick turnaround time from the conception of an idea to
launching the product are two of the main reasons for this growth, according to Qiang Yang,
computer scientist at Hong Kong University of Science and Technology. According to Andrew Ng, the
founder of Google Brain, everything in China is a lot faster than in most of Silicon Valley, which
means, for example, that the time you have from spotting a business opportunity to responding to it
is a lot shorter.
Over the next year, the world will recognize how much artificial intelligence
is going to transform everything we do. For example, the amount being
invested in artificial intelligence in Asia is growing by the day. 2017 will be
the year investors will look at AI and say, ‘if you’re not investing in it,
you’re missing the boat’.
Egidio Zarrella, clients and innovation partner, KPMG China, 2017
31
Source: CB Insights, scmp.com, technologyreview.com
Baidu adapts its structure to
accommodate AI advancements
Rise of China: Baidu
Baidu is one of the big investors in AI in China. Here are some examples on how Baidu is investing
and what its strategies are:
R&D spending
Apart from leading the AI lab commissioned by NDRC, Baidu is also spending as much as 15% of its
entire revenue on AI-related research, according to Robin Li, the company’s co-founder and CEO.
From 2015 to 2017, this totaled US$2.9 billion – the highest amount in China and one of the highest
amounts in the world.
High-profile appointments
In January 2017, the company appointed former Microsoft executive Qi Lu as its chief operating
officer, with a mandate to focus on technologies such as deep learning, augmented reality, and image
recognition. Even though Lu resigned recently, in May 2018, he was instrumental in making the
company push AI by focusing on autonomous driving and conversational AI. In fact, it was during his
tenure that the company was appointed by the Chinese government to spearhead the country’s
development in autonomous driving. Previously, Baidu had also appointed Andrew Ng as chief
scientist – Andrew had earlier founded Google's first deep learning team1.
Divestments and spin-outs of non-core businesses

In an effort to direct more of its resources towards AI, Baidu sold its food delivery business Baidu
Waimai in August 2017 and span out its utility apps and its mobile ad business in May 2018. The
company is now setting up a new global business unit around its AI-powered services, such as
recommendation engine PopIn and keyboard app Simeji.
Baidu’s current AI projects include:

▪ Deep learning laboratory: This is a collaboration between the Chinese government and Baidu
which was launched in 2017 with an undisclosed budget. Its goal is to integrate resources from the
government, industry leaders, and academic communities to improve the country’s overall R&D
capabilities in deep learning and AI. The main areas of focus include visual perception, speech
recognition, biometrics recognition, and human-machine interaction.
▪ Seattle research center: In October 2017, Baidu opened a research center in Seattle, Washington,
to expand its reach in AI and cloud computing.
▪ Xiaoyu Zaikia (Little Fish): a home robot that is capable of turning its head to listen to whoever is
speaking to it.
▪ Autonomous cars: Baidu is also expected to begin mass production of AI-powered autonomous
cars by 2021. In April 2017, the company launched a platform called Project Apollo that allows
companies to manufacture autonomous vehicles faster by providing them with the tech and open-
source code needed to help their vehicles perceive obstacles and plan their routes. Apart from
this, Baidu is also looking to sell a range of hardware called Apollo Computing Units that can plug
into cars to run its software.
1: Andrew has since left Baidu and has started various AI-related projects, including a US$175
million AI-focused fund in partnership with Greylock Partners, NEA, Sequoia, and SoftBank 32
Source: scmp.com, technologyreview.com, theatlantic.com, Techcrunch, Zhongguancun
Science Park
Most of Tencent‘s AI investments
are health-related
Rise of China: Tencent (1/2)
With the slogan of ‘Make AI Everywhere’, Tencent’s AI efforts mainly focus on three areas: healthcare
AI, robotics, and autonomous cars.
Healthcare
Having launched the AI Medical Innovation System, or AIMIS, an AI-powered diagnostic medical
imaging service, in 2017, the company has subsequently made big investments in AI imaging
technology and medical start-ups around the world. So far, AIMIS labs have been established in over
10 hospitals around the country, and agreements have been signed to increase the count to 100
hospitals in the near future. In 2014, it also launched WeChat Intelligent Healthcare, a platform that
allows people to book appointments, make payments, and more at hospitals and other medical
facilities through WeChat public accounts. Babylon Health, the UK-based digital healthcare start-up,
has recently tied up with Tencent to deploy its AI app on WeChat. This partnership will enable
WeChat’s one billion users to communicate their medical symptoms directly to Babylon’s mobile app,
which will then send back healthcare advice on what to do next.
Share of Tencent’s AI investments in 2014–2017
Others2
10%
Pharmaceuticals
6%
Health crowdfunding Medical O2O1

6% 39%
Medical AI 8%
14%
Genomics
17%
Health monitoring
Source: Technode
1: Medical Online to Offline service category 2: “Others” includes investment in resuscitation 33

training, veterinarian social network, clinic and food safety
Source: China Daily, Computer Business Review, Technode, The Drum
Tencent invested in robotics start-
up UBTech
Rise of China: Tencent (2/2)
Robotics
In March 2018, the company announced the establishment of the Robotics X laboratory in Shenzhen,
China, which, along with the previously set up AI lab, will work in affiliation with the Tencent
Technology Engineering Group (TEG). The lab is expected to focus on various areas ranging from
glasses to humanoid robots, according to the company’s vice president, Yao Xing. In May 2018,
Tencent also led a US$820 million funding round in robotics start-up UBTech and is expected to
collaborate with the company to launch robots that can communicate and interact more naturally
and intuitively.
Autonomous cars
In a bid to develop AI-powered level 3 and level 4 autonomous cars, Tencent has made a string of
partnerships with automotive companies such as BYD, Changan, Dongfeng Liuzhou Motor, GAC
Group, Geely, and FAW China. In October 2017, the company launched an AI system that facilitates
the manufacture of cars capable of calculating the best route, ordering food, and searching for
parking lots with minimum human intervention. These features are expected to be seen in the next
line of autonomous cars launched by companies such as GAC Group, BYD Co, and Zhejiang Geely
Holding Group Co, all backed by Tencent’s proprietary technologies. Tencent has also invested in
Tesla and two other EV start-ups: Nio and Weltmeister.
34
Source: China Daily, Computer Business Review, Technode, The Drum
AI hardware is growing at a similar
pace to AI software
Growth in hardware and software (1/2)
While software seems to be getting all the attention in the AI industry, the market for related
hardware is also witnessing robust growth. According to estimates by IDC, even though software will
account for the highest share of AI revenues, hardware revenues are estimated to grow nearly as fast
as software at an average annual growth rate of over 60% during the period of 2016 to 2020.
Many industry experts consider hardware to be a huge differentiator for AI companies. According to
Sam Altman, co-chairman of OpenAi, the organization has been focusing on building better AI
algorithms until now, but its next phase of growth will come from building better hardware.
Companies like Nvidia, Intel, Qualcomm, AMD and a number of start-ups are also developing chips to
make machine and deep learning faster, cheaper, and more powerful. Even though Nvidia is primarily
associated with the development of video cards for the gaming industry, it has also been focusing on
improving its GPUs to meet advanced AI requirements.
GPUs, which have begun to play a significant role in training large deep neural networks, are
expected to be a major catalyst for AI progress. This is because GPUs offer a huge parallel
architecture that has the ability to handle multiple tasks simultaneously. Since neural networks are
required to process enormous amounts of data, with much of it being high-dimensional, training on
CPUs is simply not adequate.
However, since GPUs were originally not built for neural network training, they experience challenges
related to memory bandwidth and data throughput or end up wasting their high computational
precision, which is not always required. This has resulted in new hardware innovation by companies
such as Google, Graphcore, Intel, and Scortex, which are developing chips with larger memory
bandwidth, higher compute density, efficiency, and performance per watt. These chips can be used
in various areas such as cloud infrastructure as a service, self-driving vehicles, UAVs, and robotics.
35
Source: medium.com, fortune.com
AI software and hardware market to
register robust growth
Growth in hardware and software (2/2)
Worldwide AI for enterprise applications market in million US$
31,237
23,887
+64.3%1
17,284
11,841
7,714
4,806
2,868
841 1,622
358
2016 2017 2018 2019 2020 2021 2022 2023 2024 2025
Source: Tractica
AI-driven hardware sales growth in million US$
115.4
+47.5%1
3.5
2016 2025
Source: Tractica
36
Transfer learning enables AI models
to train on less data
Transfer learning
Deep learning is the branch of AI currently receiving maximum attention globally and has applications
in almost all industries. However, for deep learning models to achieve the desired results, the neural
networks need to be trained with huge amounts of data. For example, in the annual ImageNet Large
Scale Visual Recognition Challenge, which requires teams to challenge their image recognition
models, 1.2 million images across 1,000 object categories were used to train the AI programs. Until
recently, such volumes of data have been imperative in order for deep learning models to perform
well on complex tasks such as speech recognition or machine translation. This has often proven to
be costly and time-consuming and therefore unsuitable for many use cases.
Transfer learning is a new technology developed by AI researchers that enables deep learning
models to train on small data sets by transferring the learnings from a previous task to be reused for
different problems in the same domain. Importantly, this smaller neural network only needs specific
data about the current problem as it has already learned about the overall patterns from the pre-
existing data.
Examples of related start-ups are:
▪ Geometric Intelligence: a start-up acquired by Uber, has developed a machine learning software
called XProp, which requires much less data to learn a new visual task. In a demonstration, CEO
Gary Marcus compared XProp to another deep learning program in a test to recognize
handwritten digits. In this test, XPropr needed only 150 examples to achieve an accuracy of 98% as
compared to the 700 examples required by the other software for a similar level of accuracy.
▪ Owkin: a start-up that uses AI technology to make accurate medical predictions, is building a
platform to enable healthcare professionals to create and share a new transfer-learning-powered
classification algorithm. Interestingly, it requires a data set of medical images that have a minimum
of only two different labels, whether it is for diagnosis, prognosis, or drug response prediction.
Transfer learning can help data scientists mitigate the risks of machine-
learning-driven predictions in any problem domain susceptible to highly
improbable events. For example, cross-fertilization of statistical knowledge
from meteorological models may be useful in predicting “perfect storms” of
congestion in traffic management. Likewise, historical data on “black
swans” in economics, such as stock-market crashes and severe depressions,
may be useful in predicting catastrophic developments in politics and
epidemiology.
James Kobielus, lead analyst for data science, deep learning, and app
development at SiliconANGLE Media, 2017
37
Source: CB Insights, scmp.com, technologyreview.com
Drivers
One of the major factors driving the current wave of AI growth is the rapid
increase in corporate venture capital (CVC) investment in AI start-ups. On the
technology front, rapid advancements in computing power are driving the
industry to the next level. Similarly, open source platforms are promoting and
enabling collaborative learning, which is conducive for the growth in AI.
The current wave of growth in the AI industry is as much about the abundant
availability of big data as it is about software and hardware. The amount of
big data being generated by today’s increasingly digitized economy is growing
at a rate of 40% each year and is expected to reach 44 zettabytes by 2020. This
growth in big data is driving the improvement of AI algorithms.
38
Corporate venture capital funding
for AI increased by 81% in 2017
Corporate VC investment (1/2)
One of the major factors driving the current wave of AI growth is the rapid increase in corporate
venture capital (CVC) investment in AI start-ups. Even as global CVC funding recovered across
industries to increase to US$31.2 billion in 2017 from US$26.5 billion in 2016, the AI sector also
continued its upward trajectory. According to CB Insights, CVC funding for AI companies nearly
doubled in 2017, increasing from US$2.1 billion to US$3.8 billion. The total funding over the period
2013–2017 increased at a huge CAGR of 88.7%, while the number of deals increased from 36 to 198.
In May 2017, Salesforce Ventures announced its fourth investment fund, the Salesforce platform
fund, which will focus on AI start-ups. Notably, the largest CVC funding deal for 2017 was for Nio, a
Chinese electric vehicle start-up which received US$1 billion from Tencent and Lone Pine Capital,
CITIC Capital, and Scottish fund manager Baillie Gifford.
Below is a selected list of corporate investments:
Alphabet: One of the most active AI investors, Alphabet’s largest and

most famous AI investment has been the acquisition of DeepMind
Technologies for US$600 million. The company has made various AI
investments over the last few years, in companies such as BenchSci,
Calico, DNNresearch, Jetpac, Moodstocks, Kaggle, Api.ai, and Granata.
Apple: The company has been on the bleeding edge of AI since the
launch of Siri in 2012. Since then, it has made a number of significant
acquisitions, such as Novauris (speech recognition), Perceptio (image
recognition), VocalIQ (natural language processing), Emotient (face
expression recognition), Turi (wide range of machine learning
applications), and RealFace (facial recognition).
Facebook: AI has become vital for Facebook in its efforts to scale up and
match the growing prowess of Google and Microsoft. Its acquisitions
include Face.com (facial recognition), Jibbigo (speech translation), Wit.ai
(speech recognition), and Zurich Eye (computer vision).
39
Source: CB Insights, GCV Analytics, singularityhub.com, company sources
SoftBank to heavily invest in AI with
its US$100bn vision fund
Corporate VC investment (2/2)
Intel: The most significant of Intel’s AI acquisitions is Nervana Systems, a

deep learning start-up, for US$408 million as it ups the ante in its
competition with Nvidia. Others include Itseez (computer vision),
Movidius (computer vision processors), Indisys (natural language
processing), and Saffron (cognitive computing platform).
Microsoft: Even though Microsoft has been making AI-related

investments for over two decades, its latest initiative, the launch of an AI
fund by Microsoft Ventures, reinforces the company’s commitment to
developing its AI capabilities and to compete with other technology
giants. The first investments of the fund were in the Montréal-based
start-up Element AI, an incubator co-founded by Yoshua Bengio, who is
often called “the godfather of machine”, and in Tact, an AI-powered
assistant for salespeople. The company also acquired AI scheduling tool
Genee and deep learning research lab Maluuba. In addition,
Envisagenics (healthcare), Hazy (data sharing), ZenCity (data analytics)
and Voiceitt (speech recognition) are the first winners of Innovate.AI, the
AI start-up competition announced by the company in October 2017.
SoftBank: is expected to make the biggest investments in AI over the

coming years, mainly with its US$100 billion vision fund. The company’s
founder Masayoshi Son has often spoken about the concept of
Singularity – the time when AI will become smarter than humans – and
has thus been investing in AI-related businesses for the last few years.
His most significant buys include the US$32 billion acquisition of
semiconductor maker ARM Holdings, insurtech start-up Lemonade for
US$120 million, and US$93 million for Petuum, an AI solution
development platform.
40
Source: CB Insights, GCV Analytics, singularityhub.com, company sources
Rapid advancements in computing
power are driving the industry
Computing power (1/2)
The exponential increase in big data and sophistication of analytical capabilities has mandated
significant advancements in computing power. Just to give an idea of the progress chip makers have
made over the years: The current generation of microprocessors delivers over 4 million times the
performance of the first single-chip microprocessor introduced in 1971. It is because of
developments like this that AI scientists are now able to create advanced system designs such as
those supporting multi-core and parallel processing.
According to Dileep George, co-founder of the machine learning start-up Vicarious, at least 80% of
the recent advances in AI are due to advancements in computing power. Moreover, distributed
computing network systems can seamlessly interface with infrastructure platforms and cloud
applications as well as analyze streamed data from sources such as the IoT, sensors, and embedded
intelligent devices.
Below are examples of what some of the key companies are doing in this area:
Google: In May 2016, Google launched its own computer chip for driving
deep neural networks. Called the Tensor Processing Unit (TPU), the chip
requires fewer transistors per operation, which helps it get results more
rapidly. In February 2018, the company made the chip available to other
companies through its cloud computing service.
Intel: recently launched a new range of CPUs called Knights Mill, specially
designed for AI applications. These include the next generation of Intel
Xeon Phi co-processors, whose performance is expected to be up to
four times better than the performance of the previous generation of
deep learning. The company is also considering melding its CPUs with
reprogrammable field programmable gate arrays (FPGA) processors
that, as the name suggests, can be reprogrammed after they are made
to carry out specific tasks. In October 2017, Intel launched a new family
of chips called Nervana Neural Network Processors (NNPs), specially
designed for deep learning. The first version of the chip set, code-named
Lake Crest, is currently made available only to a select set of customers
like Facebook to train complex neural networks to share insights on the
performance of the chips.
IBM: IBM’s new hardware suite, called PowerAI, contains the Power8
server run by the Nvidia Tesla GPUs. In December 2017, the company
launched the Power9 chip, specially designed to improve the
performance of common AI frameworks like Chainer, TensorFlow, and
Caffe. IBM intends to sell the chips to third-party manufacturers and to
cloud vendors, including Google.
41
Source: queue.acm.org, company sources
Microsoft is building the world’s first
AI supercomputer
Computing power (2/2)
Microsoft: In March 2017, Microsoft, in partnership with Nvidia, unveiled

a new hyperscale GPU accelerator called HGX-1, designed for AI
workloads in the cloud. It is an open-source design that enables a CPU
to connect to multiple GPUs, thereby allowing cloud service providers to
offer a number of CPU and GPU machine instance configurations.
In September 2016, the company installed FPGAs across its Azure cloud
computing platform consisting of servers in 15 countries to build what it
called “the world’s first AI supercomputer”. To showcase its power, the
company demonstrated its ability to translate 5 billion words into
another language in less than one-tenth of a second.
Nvidia: Its GPU technology has had the maximum impact on

advancements in global deep learning. Its Volta GPU computing
architecture, launched in May 2017, consists of 21 billion transistors
providing a deep learning performance equivalent to 100 CPUs. In April
2016, Nvidia released the Tesla P100 GPU with 12 times as much
processing power as its previous system. It also launched the ‘world’s
first deep learning supercomputer’: NVIDIA® DGX-1.
42
Source: queue.acm.org, company sources
Open-source platforms promote
and enable collaborative learning
Open platforms (1/2)
Even though AI scientists have made considerable progress in their efforts to make algorithms learn
from patterns, progress has been a bit slow. This is why major technology companies with
considerable proficiency in this domain, including Google, Facebook, and OpenAI, have decided to
move from intellectual property protection to open and free software.
The AI community usually operates in this way and shares its learnings with all its members. In fact,
deep learning came into existence when academics, including University of Toronto professor Geoff
Hinton, openly shared ideas.
Google started this trend in November 2015, when it made its AI engine TensorFlow open source,
freely sharing the underlying code with the world at large. In December 2016, Google again made its
entire DeepMind Lab training environment codebase open source, sharing it on GitHub for anyone
looking to train their own AI systems. The company uses 3D gaming environments to train AI agents
to behave more like human beings. According to team members of DeepMind, this move was
necessary because despite using its AI lab for some time now, DeepMind has “only barely scratched
the surface of what is possible”.
Google is five to seven years ahead of the rest of the world. If they open
source their tools, this can make everybody else better at machine learning.
Chris Nicholson, co-founder at Skymind, 2015
At the same time, Elon Musk’s OpenAI also released its Universe platform for other researchers to
test their agents.
Facebook, which had open sourced its AI hardware design in 2015, did the same with some of its AI
software, such as DeepMask, SharpMask, and MultiPathNet. These tools are essentially used for
segmenting objects within images.
More recently, in April 2017, Chinese internet giant Baidu announced that it would be sharing its
autonomous driving platform, Project Apollo, in an effort to speed up developments in the sector.
According to a company press release by Project Apollo, Baidu will make its software code and
capabilities in various areas, such as obstacle perception, trajectory planning, vehicle control, and
vehicle operating systems, open source as well.
The rise of open-source AI is driving research in the community by providing software developers
and start-ups with tools and resources that were not available to them previously. One benefit of
this can be observed in the personal robots market, where manufacturers now have access to
quality AI software, which had heretofore been a weak point.
43
Source: singularityhub.com, company sources
TensorFlow and Torch are two
popular open AI frameworks
Open platforms (2/2)
▪ TensorFlow: Developed by Google, this open-source software library is made for deep learning or
artificial neural networks and allows users to create neural networks and computation models
using flowgraphs. It is available in C++ and Python.
▪ Theano: Only available in Python, Theano is designed especially for deep learning and supports
platforms like Linux, Mac OS X, and Windows. It allows users to define and evaluate mathematical
calculations, including multi-dimensional arrays.
▪ Torch: an open-source computing framework for machine learning algorithms which offers GPU
support, N-dimensional arrays, numeric optimization routines and linear algebra routines. It is
based on a scripting language called Lua and supports major platforms such as Linux, Android,
Mac OS X, iOS, and Windows.
▪ Caffe: Developed by then UC Berkeley PhD candidate Yangqing Jia, Caffe is a deep learning
framework which can process over 60 million images in a single day using just one NVIDIA K40
GPU. One of its main features is that it allows the user to apply neural networks to the problem
using text without writing code. It supports operating systems such as Ubuntu, Mac OS X, and
Windows.
▪ Microsoft CNTK: a deep learning framework that is billed to be faster than TensorFlow and
supports distributed learning with built-in data readers. The tool supports Windows and Linux.
▪ Deeplearning4j: an open-source deep learning library for the Java Virtual Machine (JVM). It runs in
distributed environments and integrates with both Hadoop and Apache Spark.
▪ Distributed Machine Learning Toolkit (DMTK): Another open-source AI tool developed by

Microsoft, the DMTK consists of three key components: the DMTK framework, the LightLDA topic
model algorithm, and the Distributed (Multisense) Word Embedding algorithm.
▪ Azure ML Studio: used to develop larger machine learning models in the cloud, owing to its wide
range of modelling options and algorithms. It can be used with R and Python programs.
▪ Amazon Machine Learning (AML): a machine learning tool that can be connected to data stored in
Amazon S3, Redshift, or RDS.
▪ MLlib: Another machine learning tool offered by Spark, MLlib integrates with Hadoop and
interoperates with both NumPy and R. It includes various machine learning algorithms for
classification, regression, decision trees, recommendation, clustering, topic modeling, feature
transformations, model evaluation, ML pipeline construction, ML persistence, survival analysis,
frequent itemset, and sequential pattern mining.
44
Source: Datamation, DZone
The growth of big data is driving the
improvement of AI algorithms
Big data
The current wave of growth in the AI industry is as much about the abundant availability of big data
as it is about software and hardware. This is because AI applications, especially deep learning, need
large volumes of data to deliver accurate results.
According to a study conducted by International Data Corporation (IDC), the amount of big data
being generated by today’s increasingly digitized economy is growing at a rate of 40% each year and
is expected to reach 44 zettabytes or 44 trillion gigabytes by 2020.
This huge growth, which is driven mainly by the proliferation of smart devices, the Internet of Things
(IoT) and social media, is absolutely critical to the growth of AI applications across industries. After all,
AI systems get ‘smarter’ in direct proportion to the amount of data they consume.
Worldwide big data revenue in billion US$
Actuals Forecast
103
96
90
84
77
70
64
56
49
42
35
28
23
20 18
12
2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027
Source: Wikibon; SiliconANGLE
45
Source: IDC
Applications
AI solutions are increasingly being customized to serve the needs of the
automotive, healthcare, education, finance, entertainment, and other
industries.
In the automotive sector, AI is primarily used to power autonomous cars, with
these systems expected to become standard in new vehicles in the medium to
long term. In the healthcare industry, developments in the field of AI and
machine learning have not only accelerated the pace of innovation in the
industry but are also changing entire operating models. In the education
industry, there are attempts to provide customized learning programs for
each student using AI, while in the finance industry, AI wealth management
solutions can offer higher personalization.
46
AI applications are used in multiple
industries
Overview
AI solutions are increasingly being customized to serve the needs across the automotive, healthcare,
education, finance, entertainment, and other industries.
In the automotive sector, AI is primarily used to power autonomous cars, with these systems
expected to become standard in new vehicles in the medium to long term. Even though most
automotive manufacturers are developing AI-powered cars, Google and Tesla are currently the front-
runners in the field.
Healthcare is another major area of AI disruption, with applications in the fields of diagnostics, drug
discovery, personalized treatment plans, and robotic patient care. One of the major breakthroughs
brought about by AI is the shift in approach from treating diseases to preventing them.
Artificial
intelligence
Automotive industry Entertainment
Healthcare Finance
Education
47
AI disrupts many elements of the
automobile industry’s ecosystem
Automotive industry: overview (1/2)
The application of AI in the automotive industry can be traced back to 1962, when
the first industrial robot started working at a General Motors (GM) plant in New
Jersey. At the time, Japan was the only other country to recognize the true
potential of this new technology and the benefits it would provide.
One of the major breakthroughs was made in 1981, when Takeo Kanade developed the world’s first
direct drive arm, with motors installed directly into the joints of the robot arm. This marked the
beginning of an era where robots began to replace humans in the car manufacturing process on a
large scale. Since then, advances in computing power and the proliferation of big data have resulted
in deeper penetration of AI in the automotive industry, from driving to design to manufacturing.
Areas such as automotive assemblies, displays on dashboards, and even overall vehicle
conceptualization are all experiencing robust innovation aided by faster computing systems like
GPUs.
Today, many cars are equipped with sensors in the front, back, and sides, along with rear- and front-
facing cameras, to capture real-time data. This data is then used to identify anomalous patterns,
which the AI system uses to relay precautionary measures to the driver in real time. The AI-powered
adaptive cruise control systems are another feature gaining significant traction, according to Anand
Rao, partner and innovation lead of PwC Data & Analytics.
If you set your speed to 70 miles an hour and the car in front of you is going
50, as you approach it within a safe distance, you’ll start slowing down
automatically.
Anand Rao, partner and innovation lead, PwC Data & Analytics, 2016
AI is not only changing the entire driving experience but also every other element of the automobile
ecosystem, including on-demand mobility, connected cars, insurance, electric vehicles, advertising,
etc. For example, the pay-per-use consumption model pioneered by companies such as Zipcar, Uber,
and Lyft is evolving towards an autonomous model. Ford has earmarked a US$1 billion investment
over the next five years in AI company Argo AI to develop autonomous vehicles and use them in
areas such as ride sharing and ride hailing. In addition, insurance companies are gradually using real-
time data on driving behavior to deliver highly personalized policies.
48
Shipments of AI-based systems for
automotive industry are growing
Automotive industry: overview (2/2)
The worldwide number of unit shipments of AI-based systems for the automotive market is expected
to register robust growth up to 2025. The global automotive market is expected to ship 121.5 million
AI-based systems in 2025, growing at a CAGR1 of 24% from 22 million units in 2017.
Unit shipments of AI-based systems for automotive industry in millions
121.5
109.5
99.0
85.5
+24%1
72.0
57.0
42.5
31.0
22.0
2017 2018 2019 2020 2021 2022 2023 2024 2025
49
Source: Statista estimates, IHS
Deep learning makes fully
autonomous cars a reality
Automotive industry: autonomous driving (1/5)
According to a study by Garret/Galland Research, the number of self-driving cars on the road is
expected to reach 10 million by 2020. Another report by IHS estimates the annual sales of self-
driving cars to reach approximately 33 million units by 2040, with nearly 76 million vehicles with
some level of autonomy sold by 2035. Even though the required hardware, such as sensors,
cameras, radars, and ultrasound systems, has contributed a lot to making autonomous cars a reality,
it is really the developments in AI that play the most pivotal role.
Impact of deep learning on autonomous driving
Level 1 Level 2 Level 3 Level 4 Level 5

Driver Partial Conditional High Full
assistance automation automation automation automation
The vehicle can The driver is The vehicle is in These vehicles The vehicle can
assist the driver allowed to full control of are designed to perform all
in selected disengage from driving respon- perform all aspects of
functions, e.g., some tasks, with sibilities under safety-critical driving under all
steering or the vehicle certain specified functions and circumstances
acceleration/ taking over, but conditions, but respond to that are usually
deceleration must always be a driver is many dynamic managed by a
ready to take expected to take driving human driver,
control and is control when situations, without a driver
responsible for required requires minimal being present
most safety human
functions intervention
High Human intervention No intervention
Source: SAE International, KPMG, Techrepublic, Medium, IHS
Deep learning architectures enable cars to learn from their experiences and adapt to real-time
situations without human intervention. This is particularly important because it is impossible for
software engineers to write programs to cover every variable or driving situation a car may face. In
fact, deep learning is solely responsible for the evolution of cars from driver assist technologies to
fully autonomous vehicles.
50
Source: Garret/Galland Research, IHS
Overview of autonomous cars and
their autonomy level
Table Title Range in Launch

Company Model Autonomy level miles year
Alphabet Waymo Level 4 n/a 2018
aicon Level 5 500 Undecided
A9 e-tron Level 4 311 2020

Audi
Q6 e-tron Level 4 310 2018
Elaine Level 4 311 2019
Baidu-BAIC Group n/a Level 4 n/a 2021

Level 3 at launch, upgrade
BMW iNext 435 2021
to level 4 a year later
Faraday Future FF-91 Level 4 300 2018
Fisker EMotion Level 4 400 2019
Future Mobility Corp Byton Level 3 250 2019
General Motors Chevrolet Bolt n/a 238 2020

Jaguar Land Rover /
Jaguar I-Pace Level 5 298 2020
Waymo
Kia Niro EV n/a 238 n/a
Mercedes-Benz V-Class Level 3 n/a 2021
Navya Autonom Cab Level 4 n/a 2018
Porsche Mission E Level 4 310 2020
SYMBIOZ Level 4 n/a Concept
Renault Easy Drive Level 2 n/a 2019
Easy Drive Level 4 n/a 2022
Toyota Concept-i Level 5 186 2020
I.D. VIZZION Level 5 n/a n/a

Volkswagen
I.D. Buzz Level 5 370 2022
51
Source: autocar.co.uk, autocarpro.in, Bloomberg, carwow.co.uk, businessinsider.in, CNN,
digitaltrends.com, greencarcongress.com, greentechmedia.com, theverge.com, topgear.com
Google and Tesla are leaders in
driverless car development
For deep learning to be successful, it is imperative that the algorithms are fed with large amounts of
data. One of the most important trends driving automation in the industry is the growth in the
number of connected cars. As autonomous cars share driving experiences and readings with each
other, algorithms use that data to adapt to various situations without having to experience them
firsthand.
Google, the global leader in self-driving technology, recently launched the testing phase of its self-
driving car called Waymo, which is powered by over 2 million miles of real-world driving data. Elon
Musk, co-founder and CEO of Tesla, has also announced the rollout of a completely autonomous car
by the end of 2017, equipped with Nvidia’s Drive PX 2, a supercomputer that uses deep learning to
teach the car to handle itself.
Even though Google and Tesla are the most visible brands chasing driverless technology, a host of
auto manufacturers and other technology companies are also partnering with each other to develop
cars with varying levels of autonomy. A few key examples can be found in the following table.
Deep learning is the best enabling technology for self-driving cars. You
hear a lot about all these things on a car: the sensors, the cameras, the
radar, and LIDAR1. What you need are the brains to make an autonomous
car work safely and understand its environment.
Sameep Tandon, CEO and co-founder of Drive.ai
52
1: Light Detection and Ranging is a remote sensing method that uses light in the form of a
pulsed laser to measure ranges
Car companies’ activities in
autonomous driving (1/2)
Other
Company Description AI platform partnerships
▪ Launched a level 3 ▪ Nvidia – DRIVE PX 2 n/a
autonomous vehicle
in 2017 and is
expected to a level 4
vehicle by 2020
▪ BMW’s iNEXT project ▪ Intel Nervana – incl. ▪ Mobileye1 – EyeQ5

to be launched in machine & deep processor, 360°
2021; will consist of learning training and camera system,
fully electric and simulation and a sensor
connected 7 series infrastructure fusion solution
cars with level 4
autonomous
▪ Intel – Intel GO
automated driving
technology
solutions
▪ GM spent US$600 ▪ IBM’s Watson – ▪ IBM – to develop

million on OnStar Go OnStar’s touch
autonomous vehicle screen interfaces
development in 2017
▪ In 2016, it spent
US$581 million to
acquire Cruise
Automation, a start-
up that specializes in
developing software
to operate self-
driving cars
▪ Had 30 self-driving
all-electric Chevrolet
Bolt vehicles on the
road in October
2016
53
1 Mobileye was acquired by Intel in March 2017
Source: Company sources, press releases
Car companies’ activities in
autonomous driving (2/2)
Other
Company Description AI platform partnerships
▪ Spending US$1 ▪ Yui – an AI assistant ▪ NTT – to develop
billion on developing made by Toyota technology for
self-driving cars connected cars
utilizing fifth-
▪ Launched the next-
generation signals,
generation
big-data analysis,
autonomous
and AI applications
customized Lexus in
March 2017 with two ▪ Stanford and MIT –
platforms: Chauffeur to develop AI
– Level 4 and 5 applications for use
autonomy; Guardian in its autonomous
– driver-assist cars
▪ Launched its first ▪ Nvidia – Drive ▪ Mobvoi – voice

autonomous concept recognition and
car Sedric in March
▪ PX 2
language
2017 processing
technologies
▪ Its 2025 strategy
focuses on ▪ Mobileye1 – to inte-
developing electric grate its Road
vehicles and Experience Mgmt
autonomous driving mapping service
technology into VW cars by
2018
▪ Launched a fleet of ▪ Nvidia – DRIVE PX 2 ▪ Microsoft – to

100 autonomous install Skype for
Volvo XC90 SUVs by Business on new
the end of 2017 Volvo 90 Series
cars
54
1 Mobileye was acquired by Intel in March 2017
Source: Company sources, press releases
Cloud computing augments AI in
the automotive industry
Automotive industry: cloud computing
Rapid advancements in technology have resulted in a closer relationship between AI and cloud
computing. Since AI works primarily by analyzing huge volumes of data, companies (including AI
developers and end users) are quickly realizing the benefits of moving away from the confines of
internal IT structures and into the vast computing power and data that the cloud has to offer.
According to Eric Schmidt, chairman of Google’s parent company Alphabet, companies tapping into
cloud computing will also get access to crowdsourced data, which, when combined with advanced AI
technologies, will lay the platform for truly disruptive businesses across many industries.
Therefore, companies in the automotive industry are using data from the cloud to deliver various
services, including marketing, navigation, and payments.
Examples of such collaborations are below:
▪ GM and IBM have partnered to enable IBM’s AI supercomputer Watson to power GM’s upgraded
onboard car system called OnStar Go. This platform now uses data from the cloud to offer
recommendations to drivers on places to eat, shop, or fill gas. Moreover, when drivers are nearing
a preferred location, say a fast food joint, they have the option to pre-purchase their order before
their arrival.
▪ Researchers at the NYU Tandon School of Engineering are developing an AI system which enables
self-driving cars to combine the information retrieved from cloud-based maps with their onboard
sensors to achieve enhanced navigation, maneuverability, and ability to adapt to changing road
conditions.
▪ Nvidia and Baidu have partnered to develop an AI-powered cloud computing platform for
autonomous vehicles. They plan on integrating Baidu's cloud platform and mapping technology
with Nvidia's self-driving computing platform to create solutions for high-definition maps, Level 3
autonomous vehicle control, and automated parking.
55
Source: cloudwedge.com, company sources
AI impels transition from reactive to
proactive healthcare
Healthcare: overview
Artificial
intelligence
Early diagnosis & Precision medicine &

preventive healthcare personal genetics
Surgical Drug discovery

assistance
Recovery &
rehabilitation
In the past, innovations in the healthcare industry have been restricted to the
digitization of manual tasks in order to save on time and effort. However,
developments in the field of AI and machine learning have not only accelerated
the pace of innovation in the industry but are also changing entire operating
models.
One of the major changes brought about by AI is the shift from mere treatment of diseases
(reactionary mindset) to a treatment that focuses on diagnosing illnesses at an early stage or before
they occur (preventive mindset). According to Praveen Soti, general manager for healthcare and life
sciences at Wipro, the future of healthcare lies in moving away from the present conditions, in which
hospitals are incentivized to fill beds, prescribe tests, and perform surgeries, to actually keeping
patients away, with AI playing a pivotal role in this transition.
As computing power becomes more advanced and algorithms become smart enough to spot
patterns in digital data and images, the process for diagnosing and treating illnesses is becoming
more of a data-driven practice.
▪ KenSci: a Washington-based start-up which is using AI to track the onset and progress of a disease,
along with predicting chronic and critical illnesses within a group of patients.
▪ Pathway Genomics: a genetic testing lab which is developing an AI-powered application to provide
customized preventive health advice based on a user’s genetic makeup.
▪ IBM and Under Armour: They have partnered to develop an AI-powered ‘cognitive coaching
system’ which processes data from over 200 million people to develop personalized advice on
sleep, fitness, and nutrition.
56
Source: ai100.stanford.edu, Wipro
AI enables personalized medical
advice without seeing a doctor
Healthcare: new market creation
AI is also creating new markets in the healthcare industry. For example, Norwegian company
Your.MD has developed an AI-powered mobile app that matches a patient’s symptoms to publicly
available data collected from various sources and offers them personalized advice regarding their
ailments. This mainly works by filtering out people who do not need primary care and letting the
doctors focus on those who need the most attention.
This has resulted in the creation of a new healthcare market called ‘pre-primary care’, in which the
patient ends up doing more for themselves and acts as a driver for large-scale behavioral change.
In 20 to 30 years, we really will be living in the Jetson era. By then, big

data, the internet of everything, precision medicine and AI will have
converged. A body scan will verify what we can already predict based on
genetic mapping. Therapy, diet, and treatment will all be personalized to the
individual and treatments like chemotherapy will seem as barbaric as
leeching. Some of the greatest advances will be in neuroscience. We'll have
the ability to map the billions of neuron firings for personalization and
unique mental health therapies.
Al Babbington. CEO of PrescribeWellness, 2017
57
Source: Company sources
AI applications facilitate early and
accurate diagnosis
Healthcare: early diagnosis (1/2)
According to a 2017 study conducted by the Mayo Clinic, a non-profit medical

practice and research group, original diagnoses in the U.S. are revised by a
second medical professional 88% of the time. A previous study conducted by
National Academies of Medicine in 2015 revealed that diagnostic errors were
responsible for up to 10% of all patient deaths and for up to 17% of all hospital
complications.
Advancements in AI have resulted in early detection of diseases through the use of deep learning to
analyze huge amounts of data and recognize patterns, a process especially useful in diagnostics.
Some examples of AI systems designed for early diagnosis are listed here:
IBM Watson
The supercomputer analyzes large volumes of internal and external
data to detect early signs of cancer or vascular diseases.
Google DeepMind
Google has teamed up with the NHS in the UK to build a machine
learning system with the ability to recognize vision-threatening
conditions simply by conducting a digital scan of the eye.
Pathway Genomics
Is developing a simple blood test to detect or predict certain cancers
by tracking minute levels of cancerous cells that have been cast off
from tumors into the bloodstream.
Lumiata
Uses machine learning to identify patients who need early
hospitalization or medication plans.
If you miss something and a patient develops cancer five years later, there's
no systematic routine that tells you how to correct yourself. But you could
build in a system to teach the computer to achieve exactly that.
Geoffrey Everest Hinton, The New Yorker, 2017
58
AI applications are driving early and
accurate diagnosis
Healthcare: early diagnosis (2/2)
Microsoft
It has gone a step further by developing an AI system that interprets
online search engine behavior, e.g., people researching symptoms of
diseases long before they actually occur, and takes necessary pre-
emptive measures.
University of Malaga and the University of Granada

Both have jointly developed an AI technique to detect Alzheimer’s by
revealing brain patterns associated with the disease. The technique
models different regions of the brain and uses deep learning to
compare it with the brain of a healthy person.
Langone Medical Center (New York University)

It has developed a machine learning algorithm to detect Post
Traumatic Stress Disorder (PTSD), simply by listening to an individual’s
speech pattern. The AI has a success rate of 77%.
59
Robots aid in surgical assistance,
rehabilitation, and research
Healthcare: robotics (1/2)
Just fifteen years ago, the field of healthcare robotics was mostly science fiction. A company called
Robodoc, an IBM spinoff, was the first to develop a robotic system for orthopedic surgeries. Even
though the technology worked, the company could not achieve commercial success and eventually
shut down. However, recent advancements in robotics and AI have resulted in a strong increase in
the research and practical use of healthcare robotics.
These systems are increasingly being used in various medical fields and across the entire spectrum
of user populations:
▪ Surgical assistance: This area of healthcare robotics has received the maximum attention over the
years. One of the first companies to gain substantial market traction was Intuitive Surgical, which
launched the ‘da Vinci’ system, a technology initially meant to support minimally invasive heart
bypass surgery, which also went on to aid in the treatment of prostate cancer. Now in its fourth
generation, the ‘da Vinci’ provides assistance in multiple laparoscopic procedures through not only
a physical platform but also a data platform which is used to study the entire surgical process.
Other companies making surgical robots include Verb Surgical (a start-up spun out of Google and
Johnson & Johnson) and Dublin-based Medtronic.
▪ Recovery and rehabilitation: A patient suffering from neuromuscular diseases or injuries, such as
those that occur after a stroke, can benefit from robot-assisted sensory motor therapy. Such
systems are also capable of assisting people with behavioral disorders like Autism Spectrum
Disorder and Attention Deficit Hyperactivity Disorder (ADHD). Another important role is to help
neuroscientists improve their general understanding of brain function. Robots analyze data on
stimulus-response recordings to get insights into the relationship between external mechanical
forces and neural plasticity. An example of such robots is the MIT-Manus rehabilitation robot,
which has helped to bring about improved recovery of both acute and chronic stroke patients.
▪ Research: One of the growing areas of robotics technology is its application to conducting
research for overall human health. To achieve this, scientists are trying to create a robotic system
that is similar to human biology in order to simulate its workings in an artificial setting.
Furthermore, robots, such as those that work with people with neurodevelopmental disorders, can
collect data on patient behavior that can be analyzed to develop personalized treatment.
A report by the Robotics Business Review segments the healthcare robotics market into:
▪ Direct patient care robots
▪ Indirect patient care robots
▪ Home healthcare robots
60
The three market segments of
healthcare robots
Healthcare: robotics (2/2)
Direct patient care robots

▪ Surgical robots (to carry out clinical procedures)
▪ Exoskeletons (mechanical suits that bolster patient strength)
▪ Prosthetics (replacements for lost limbs)

One example is the nursing care robot ROBEAR
Indirect patient care robots

▪ Pharmacy robots (mainly used to reduce labor costs)
▪ Delivery robots (autonomous delivery of medical products)
▪ Disinfection robots (to interact with patients with infectious ailments)
Panasonic Autonomous Delivery Robots such as HOSPI are an example
Home healthcare robots

▪ Robots that provide assistance to the elderly or chronically ill people
An example is the home healthcare robot Pillo
61
Source: Indiegogo, Panasonic, Phys.org
AI systems reduce time and costs to
develop novel drugs
Healthcare: drug discovery
According to a study conducted by the Tufts Center for the Study of Drug
Development (CSDD), the average cost of developing a drug that gains market
approval is US$2.6 billion, and the process takes more than 10 years. Even then,
less than 10% of potential medicines make it to market, according to Jackie
Hunter, CEO of BenevolentBio, the life sciences arm of London’s BenevolentAI.
Apart from being time-consuming and expensive, this process also limits the
number of diseases scientists can focus on.
Machine learning algorithms can play a very important role in reducing the time and cost by using
previously generated data to establish patterns and decipher which experiments need to be done.
Other algorithms can also be used to predict the side effects of certain chemical compounds on
humans, thereby speeding up the approval process.
Key developments in this area include:
▪ Atomwise: This start-up uses machine learning and 3D neural networks to accelerate drug
discovery for diseases like Ebola, multiple sclerosis, and leukemia. In 2015, the company’s solution
discovered two new drugs to combat the Ebola virus, with the analysis finishing in just one day.
▪ twoXAR: It leverages big biomedical data such as gene expression measurements, protein
interaction networks, and clinical records to discover drugs that treat diseases like glaucoma and
liver cancer.
▪ Berg Health: This Boston-based start-up has an inverse approach to drug discovery. It analyzes
data to determine why some people survive diseases and then applies the results to improve
current therapies or create new ones.
▪ Stanford University: For situations where a small amount of data hampers the effectiveness of
conventional deep learning algorithms, scientists at Stanford University have developed ‘one-shot
learning’, a new kind of deep learning that requires only a small number of data points (usually
hundreds). Even though the researchers were skeptical about the efficacy of this method, test
results published in ACS Central Science in April 2017 were very positive.
▪ Insilico Medicine: This company uses a new deep learning technique known as a generative
adversarial network (GAN) that uses historical biological and chemical data to ‘imagine’ new
cancer-like molecules with specific properties.
62
Source: news.stanford.edu, techcrunch.com
AI is used to map the human DNA
to deliver precision medicine
Healthcare: precision medicine and personal genetics
Precision medicine has been available for some time now in the form of low-tech
therapies like allergy treatments and blood transfusions. Even genetic targeting of
tumors has become more commonplace now, spawning medications such as the
famous Herceptin1, which targets the HER2 protein associated with an aggressive
form of breast cancer.
However, AI technologies such as deep learning and natural language processing are now being
combined with big data from consumer wearables, sensors, and connected devices to detect
patterns for mutations and linkages in diseases. This is then used to deliver precision medicine down
to the familial and individual level.
One of the most significant uses of AI in the healthcare industry is to understand the human DNA.
Greater insights into an individual’s genetic makeup are providing the opportunity to deliver
personalized healthcare for each person based on their biological disposition.
Even though much of the research in this field is still in its early stages, a few promising projects are:
▪ Flow Health: is building a knowledge graph of medicine and genomics using over 30 petabytes (30
million gigabytes) of longitudinal clinical data of 22 million veterans collected over 20 years. The
goal is to determine how each gene variant in the genome affects phenotype2.
▪ Human Longevity: co-founded by Craig Venter, the man credited with mapping the human
genome. The company is aiming to sequence at least one million genomes and use the data
generated from them, along with DNA-related information, to offer better precision medicine for
ailments such as cancer and heart diseases.
▪ Deep Genomics: a Toronto-based start-up which applies GPU-based deep learning to determine
how strongly genetic variants affect RNA splicing3, a contributing factor for many diseases. The
company’s database is aiming to be able to explain how hundreds of millions of genetic variations
can impact a human’s genetic code.
▪ University of Toronto: A team is developing a GPU-powered genetic interpretation engine that

could identify cancer-causing mutations for individual patients.
▪ Editas Medicine: specializes in a process called gene editing using a technology called
CRISPR/Cas9. The company aims to use this tool to treat genetic diseases such as eye disease,
cancer, sickle-cell anemia, and Duchenne muscular dystrophy. One of its initial plans is to inject a
virus containing Cas9 into the eyes of people suffering from a rare form of progressive blindness
caused by a specific gene mutation. The enzyme would then cut the faulty sequence, triggering a
natural DNA response in which the cell repairs the deficit itself.
1: Has been used to treat more than two million patients worldwide and has generated
global sales of >US$64 billion for Swiss pharma company Roche 2: The total set of
observable characteristics displayed by an individual under a particular set of environmental 63
factors 3: The removal of certain DNA sequences
Source: techcrunch.com, company sources
AI is driving the personalized
learning industry
Education: overview
Artificial
intelligence
Intelligent tutoring Educational games
Science Resources/courses
simulation
Personalized
learning
The universal approach to education has mostly remained the same, with a single
teacher delivering the same message to a large group of students, with little
attention being paid to their individual progress. It is the desire to create
customized learning programs for each student that is currently the biggest driver
of AI in education and learning.
Machine learning capabilities embedded in online learning programs track the entire learning
process of an individual, diagnose misconceptions, and provide timely guidance, feedback, and
explanations.
For example, if a student misunderstands a particular mathematical concept, the AI system could
send an alert to the teacher before it becomes a bigger issue later in their education.
Recent years have witnessed an influx of heavy investment in the personalized learning space, the
most notable being the US$240 million made by the Bill and Melinda Gates foundation. This initiative
consists mainly of investments in private companies that analyze big data to develop software that
creates individual learning plans for students based on their performance with a special focus on
their weak areas.
Other interactive educational technologies that have gained traction over the last few years include
homework support systems, science simulations, and virtual labs, educational games, online
resources, and open online courses.
64
Source: ai100.stanford.edu
ITS are one of the most popular AI
applications used in education
Education: intelligent tutoring (1/2)
Intelligent Tutoring Systems (ITS) are the most widely used AI applications in the education industry.
They are educational applications of AI and machine learning technologies that emerged as a
scholarly discipline over 40 years ago. Central to every ITS is its ability to gather in-depth data on an
individual level and use it to assess the level of progress and offer feedback to promote productive
learning behaviors such as self-regulation, self-monitoring, and self-explanation.
ITS are now widely used in K-121 schools and colleges, especially in the U.S. The One Hundred Year
Study on Artificial Intelligence by Stanford University in 2016 states that the use of intelligent tutors
and other AI technologies to assist teachers in the classroom and at home in a typical North
American city is likely to expand significantly over the next 15 years.
Examples of intelligent tutoring systems that are already being used are:
▪ Cognitive Tutor: Developed by Carnegie Learning, this software is primarily used in U.S. high
schools to learn mathematics from grade 9 to 12. It has two major analytics components:
‘Skillometer’ and the teacher reports. The ‘Skillometer’ uses data generated from student
interaction with the software to visually indicate the current and future levels of mastery on a
particular subject. Teachers are also provided with detailed information on each student’s
progress.
▪ SHERLOCK: It is used to teach Air Force technicians to diagnose electrical system problems in
aircraft.
▪ University of Southern California: The Information Sciences Institute at the University of Southern
California has created avatar-based training modules to train military personnel being sent to
international posts in appropriate behavior when dealing with people from different cultural
backgrounds.
Other ITS have been used in other disciplines, such as geography, circuits, medical diagnosis,
computer literacy and programming, genetics, and chemistry.
AI will not replace tutors, it will support them and it will guide them to be
better teachers.
Tom Hooper, founder of Third Space Learning, 2016
1: Kindergarten (K) and the 1st through to the 12th grade (1-12) in school 65
Source: ai100.stanford.edu, bdtechtalks.com, ewise.com, forbes.com, Narrative Science,
National Business Research Institute, wired.com
Virtual humans can be used to
coach people in different ways
Education: intelligent tutoring (2/2)
Even though Apple’s Siri, Microsoft’s Cortana, Amazon’s Alexa and Google’s Assistant have already
given us a peek into what virtual humans can do, scientists are still a long way from developing
characters that possess the required degree of social intelligence to perform mainstream tasks.
Two examples of how virtual humans are used in education are SimSensei and Kaspar:
Kaspar – a robot developed by the University of

Hertfordshire that helps autistic children deal with
personal interactions. It uses touch sensors and the
ability to detect gestures and eye gaze to teach them
how to make eye contact and when it is appropriate
to touch others
SimSensei – One of the pioneering efforts to create

socially intelligent virtual humans is the SimSensei
program being developed by the University of
Southern California. It is an autonomous system that
is able to interpret a person’s mood just by
observing their behavior, like where one is looking,
the tilt of the head or if someone is frowning. It
essentially records and analyzes behavior in order to
establish a personal rapport with humans. SimSensei
has been used by people to further their public
speaking skills and their handling of job interviews.
The U.S. military has also used it for leadership
training.
Within 15 to 20 years, artificially intelligent tutors will be as common as car

insurance, and these tutors will engage with a child’s emotions through
speech interaction, support problem-solving, and develop much more
targeted learning paths for each student.
Tom Hooper, founder of Third Space Learning, 2016
66
Source: ai100.stanford.edu, bdtechtalks.com, forbes.com, ewise.com, Narrative Science,
National Business Research Institute, wired.com
Its wealth of date makes the finance
industry perfect for the use of AI
Finance: overview
Artificial
intelligence
Wealth management Personal finance

management
Insurance Banking
Fraud detection
Owing to its heavy reliance on large volumes of numbers and data, the financial
services industry is ripe for the disruption offered by AI. Large volumes of
historical data on banking, insurance, mortgages, and financial trading have been
integrated with deep learning algorithms to automate routine tasks, mitigate risk,
prevent frauds, and generate new insights.
According to a survey conducted by Narrative Science and the National Business Research Institute,
32% of financial services executives surveyed confirmed using AI technologies such as predictive
analytics, recommendation engines, voice recognition and response.
Besides general AI applications in the insurance and banking industry, two areas that are gaining a lot
of traction are especially suited for AI:
▪ Wealth management: Private wealth management used to be only an option for an elite group
with lots of funds because an actively managed portfolio usually involves high management fees.
By using AI, there is now the possibility of offering wealth management with very low fees as the
algorithm, often called a robo-advisor, decides or suggests changes in the portfolio.
▪ Fraud detection: Detecting fraudulent cases was always a high priority in the financial industry. By
using algorithms that analyze a huge amount of data, potential fraud cases can be found a lot
easier and faster than before.
67
Source: Narrative Science Report, medium.com
AI wealth management solutions
offer higher personalization
Finance: wealth management
Even though the rapid growth of robo-advisors is a good example of the

increased penetration of AI in the wealth management industry, it represents only
a fraction of the technology’s potential.
Robo-advisors have simple and rule-based algorithms that select exchange-traded funds (ETFs)
based on information such as age, risk appetite, income, etc. However, the next generation of AI-
powered wealth management solutions will also have the ability to self-learn and therefore offer
higher degrees of personalized advice for each consumer.
Australian bank ANZ was among the first to explore the potential of AI when it started using IBM’s
Watson supercomputer to aid financial advisors. Since then, many other firms, such as Goldman
Sachs, BlackRock, UBS, Deutsche Bank, and Bridgewater Associates, either have built their own AI
engines or are investing in third-party developers.
Below is a list of some key developments:
Bridgewater Associates: The world’s largest hedge fund is

building an AI engine to automate the entire functioning of
the company and eliminate human emotional volatility.
Called the “Book of the Future”, the system is being run by
David Ferrucci, one of the leading developers on IBM’s
Watson computer.
Aladdin (BlackRock): It is based on open-source technology

and uses NLP to analyze large volumes of data from
documents such as news stories and broker reports. For
example, it analyzes data on trade activity in order to detect
complex patterns and predict the transactions most likely to
fail. It can also gather satellite images to see how full a
retailer's parking lot is and then correlate that data to the
company's revenue and stock price.
Kensho (Goldman Sachs): It uses machine learning to help

find correlations responsible for movements in stock and
currency prices. The engine answers questions such as
“How do defense stocks react to terrorism incidents in
Europe?” or “How do populist votes affect local currencies?”.
Sqreem (Wells Fargo, BlackRock, UBS, and Deutsche Bank):

It uses deep learning to analyze data on people’s digital
activity to predict which products and services they are
most likely to want. It also protects companies against
financial crimes through algorithms that can detect
anomalies relating to illicit behavior.
68
Source: americanbanker.com
Cybercrime costs the economy
several hundred billions
Finance: fraud detection (1/2)
The proliferation of connected devices and mass digitization of companies has

increased the risk of fraud, hacking, data compromise, and other cyber-
vulnerabilities. According to McAfee, cybercrime costs the global economy over
US$400 billion annually, with credit card fraud accounting for a large portion of
this cost.
PwC’s 19th Annual Global CEO Survey found that 69% of financial services CEOs reported that they
were either somewhat or extremely concerned about cyberthreats, compared to 61% of CEOs across
all sectors.
In order to combat this and detect patterns of anomalies, many financial institutions are turning to
machine learning techniques such as logistic regression, decision tree, random forest, neural
networks, and clustering. These AI techniques help financial institutions to study the buying behavior
of each customer and then compare it to other indicators to build a complete picture of a
transaction.
Detecting fraud is especially important in areas such as online shopping, online payment, and credit
card usage, and there are many examples of companies in those areas.
We estimate that in the U.S. alone, the value of false declines is more than
13 times the total amount lost to actual card fraud. Applying machine
learning to decision-scoring is a new way of creating a positive consumer
experience, while also minimizing fraud.
Al Pascual, senior vice-president, research director, and head of fraud and

security at Javelin Strategy & Research, 2016
One good example is PayPal, which has used machine learning to bring its fraud rate down to just
0.32% as compared to the industry average of 1.32%. It applies machine learning to study users’
purchase history and detect patterns, which can then be used to implement new rules that prevent
scams being repeated.
More examples can be found on the next page.
69
Source: aibusiness.org
Several finance institutions use AI to
combat and prevent fraud
Finance: fraud detection (2/2)
CO-OP Financial Services has partnered with Feedzai, a

provider of AI-powered payment applications, to develop a
machine-learning-based risk management tool. The
company works across all payment types, including cards,
vouchers, prepaid card tokens, or bitcoin.
Lloyd’s Banking Group recently partnered with Pindrop to

combat fraud. Pindrop’s software creates a sort of audio
profile by identifying 147 different features of a human
voice from a single call. This is then used to detect any
unusual activity and potential fraud.
Mastercard has launched Decision Intelligence, an AI-

powered decision and fraud detection service. It leverages
account information such as customer value segmentation,
risk profiling, location, merchant, device data, time of day,
and type of purchase made to examine how a specific
account is used over time to detect normal and abnormal
shopping spending behaviors. According to Johan Gerber,
EVP of Security and Decision Products for Enterprise
Security Solutions, the company has “seen a 40% increase in
the accuracy of detecting fraud with Decision Intelligence,
compared with the old algorithms, and a 50% reduction in
false positives”.
ThetaRay provides an AI-powered platform to combat

lending frauds, ATM hacks, money laundering, and cyber
attacks. The company has recently partnered with ING to
detect SME lending fraud.
Westpac is working with start-up Red Marker to use NLP

techniques to detect content at risk of breaching legal
regulations as it is being created.
70
AI tools ‘learn’ patterns to deliver
personalized financial advice
Finance: banking, personal financial management (1/3)
Robotic Process Automation (RPA): One of the most widespread applications of AI

in the banking industry is RPA, which eliminates much of the analysis work
currently carried out by both junior and senior employees. Even though these
processes are highly standardized and formulaic, they still require a large number
of people performing low value-added tasks such as reconciliation and
consolidation.
RPA helps with lower transaction processing time, enhanced productivity, redeployment of staff to
higher-skilled roles, and elimination of manual errors.
However, such basic automation, which is the work of business process management and rule
engine software, is limited by its inability to adapt to change. An emerging trend, however, is the
combination of RPA and cognitive technologies such as machine learning and NLP to automate
perceptual tasks usually requiring human intervention. This integration is helping financial institutions
extend automation to all areas of their business, such as lending, card operations, risk & compliance,
and core banking operations, as they transition towards a fully digital operating model.
A few examples are:
Feedzai: It helps with transaction monitoring and fraud

prevention by recognizing behavior patterns that could
indicate fraudulent payment activity.
FinGenius: an NLP system used not only to interpret and

answer employee or customer questions but also to
suggest answers.
Fonetic: It offers speech recognition technology to automate

customer interactions on the telephone.
71
Source: Ewise.com, Narrative Science, National Business Research Institute, wired.com
AI tools use predictive analytics to
offer real-time financial advice
Personal Financial Management: Another area in financial services where AI is

making a mark is personal financial management (PFM). Over the last two years,
there have been a number of PFM tools which mostly use cloud-based AI,
predictive analytics, and chatbot messaging to track consumer purchases and
offer real-time advice on spending and saving habits.
For example, if someone makes a purchase, an AI application can let that person know how this
affects their monthly budget or how their current monthly expenditure stacks up against their
historical average. This is especially applicable to millennials who have low trust in banks, have a
mobile-first expectation regarding their products, and are generally more willing to try new products
and services.
Pefin: Launched by Ramya Joseph, a former investment

banker, the online platform can sync its data with a
consumer’s financial institution to get a more holistic view of
how much they should spend and invest.
Simple: uses machine learning combined with behavioral

economics to understand a user’s individual earning and
spending patterns and offer personalized advice.
Wallet.Ai: an app that monitors a consumer’s spending habit

and alerts them on how to restrain their needs and improve
their saving behaviors.
72
Source: Ewise.com, Narrative Science, National Business Research Institute, wired.com
Cards and payments are the leading
banking sector regarding AI usage
The banking industry is using AI for various purposes, depending on the feasibility and complexity of
the processes. According to a study by Consultancy.uk, 84.2% of cards and payments divisions used
artificial intelligence in 2017.
Banking sectors using artificial intelligence worldwide in 2017
Cards and
84.2%
payments
IT 63.5%
Finance and
40.4%
accounting
Marketing 31.4%
Customer service 30.8%
Sales 25.0%
Corporate level 22.4%
R&D 19.2%
HR 16.7%
Planning and
16.0%
development
Distribution and
13.5%
logistics
Procurement 7.1%
Manufacturing
7.1%
or operations
Legal 2.6%
73
Source: Statista estimates, IHS
A major use of AI in entertainment
is the creation of original content
Entertainment: overview (1/2)
Artificial
intelligence
Movies Games
Advertising Music
Personalized content
The entertainment industry profits from AI applications in various segments, such

as movies, advertising, personalized content, music, and games. The applications
range from creating original content and personalizing the offered content to
developing serious opponents in the gaming context. There are several possible
applications of AI in the movie segment. One of them is the use of AI to support
with tasks which used to be primarily manual, e.g., reviewing a movie to select the
clips for a trailer. Another application is the creation of original content and scripts
based on the analysis of previous movie performances.
Two examples are:
▪ IBM Research and 20th Century Fox – recently collaborated to create what they called the ‘first
ever cognitive movie trailer’, for the movie Morgan. IBM’s Watson platform analyzed hundreds of
horror movie trailers to determine what type of content elicited the fear response in people. It
then used the insights gained to select the 10 clips used in the trailer, after which a human editor
created the final version. Moreover, the entire process took only 24 hours, as opposed to the
usual 10–30 days of manual editing.
▪ Benjamin – Jack Zhang is taking the use of AI in movies even further. His company Greenlight
Essential has developed an AI software called Benjamin that has co-written the script for a horror
movie titled ‘Impossible Things’. According to Zhang, this is the first instance of AI creating original
content in collaboration with humans. The software essentially used NLP to analyze thousands of
movie plot summaries in correlation to box office performance. Sunspring is an example of a movie
entirely written by AI without any human intervention. The software is a recurrent neural network
that is usually used for text recognition. The movie was placed in the top 10 out of hundreds of
entries at the Sci-Fi London contest in 2016.
74
Source: theguardian.com
AI enables gaming characters to
develop their own personalities
Entertainment: overview (2/2)
Netflix is a good example of AI driving personalized content in the entertainment

industry. The company’s AI framework called Meson provides recommendations
based on the viewer’s previous viewing activity and behavior.
Apart from simple recommendations, Meson also uses data such as subscription history, previous
interactions with content, type of device, and time of day to create personalized homepages for each
user.
The company recently announced that it was using AI to enable users to enjoy uninterrupted viewing
even on a slow internet. An AI-powered system called Dynamic Optimizer reviews each frame of a
video and compresses it only as much as required, without compromising on image quality. This
differs from previous technology that compressed the entire stream but could cause fuzzy, pixelated
or unclear images.
Gaming: Until recently, AI in gaming was restricted to characters following a

predetermined script. However, developers are now trying to create AI-powered
characters that actually think, learn, and have the potential to develop their own
personalities. Two examples are:
▪ Google AlphaGo used machine learning to beat Lee Sedol, one of the world’s best players of Go,
an ancient Chinese board game, a feat that was never previously achieved by a computer
program. The main reason why this win was important was that, in addition to advanced
calculation, the algorithm also exhibited original and creative thinking.
▪ Libratus Poker AI, developed at Carnegie Mellon University, defeated four of the world’s best
heads-up poker players by a huge margin. What made this win particularly interesting for the AI
world was that Libratus did not utilize neural networks but reinforcement learning, where it
learned from trial and error at a massive scale.
75
Source: theguardian.com
Start-ups: funding and M&A
With the rise of AI, more and more start-ups venture into the market. Most
work in the field of machine learning algorithms, followed by natural language
processing. The annual global funding of AI start-ups experienced an average
growth rate of almost 70%, increasing from US$0.6 billion in 2012 to US$4.9
billion in 2016. The corresponding number of deals grew by 47% in the same
time frame, from 151 in 2012 to 703 in 2016.
In terms of M&A, the number of deals jumped up by 38% in 2017. Looking at
the most recent M&A deals, big tech companies like Google, Apple, Amazon,
Microsoft, IBM, or Facebook have often been the acquirer. But Chinese tech
giants like Baidu or rising stars in the start-up world, like Twitter, Uber, or
Spotify, also acquire AI companies.
76
Highest number of AI start-ups in
machine learning applications
Overview (1/2)
With the rise of AI, more and more start-ups venture into the market. Most work in the field of
machine learning applications, followed by natural language processing. As of March 2018, start-ups
focusing on applications for machine learning also received the most funding with almost US$19
billion.
The annual global funding of AI start-ups experienced a high growth of almost 72% CAGR1, from
US$1.7 billion in 2013 to US$15.2 billion in 2017. The corresponding number of deals grew by 44% in
the same time frame, from 310 in 2013 to 1349 in 2017.
Number of AI start-ups as of October 2018
Machine learning
866
applications
Natural language
350
processing
Machine learning
321
platforms
Computer vision
263
applications
Virtual personal
232
assistants
Computer vision
221
platforms
Smart robots 192
Speech
183
recognition
Recommendation
113
engines
Gesture control 64
Context aware
33
computing
Video content
28
recognition
Speech-to-speech
21
translation
77
Source: Venture Scanner, as of Oct 2018
Funding of AI start-ups grew at a
CAGR1 of about 72% up to 2017
Overview (2/2)
AI start-ups: annual global funding in billion US$
15.2
+72.1%1
6.3
4.6
3.5
1.7
2013 2014 2015 2016 2017

Source: CB Insights
AI start-ups: global deal share U.S. vs. non-U.S. deals
U.S. deal share Non-U.S. deal share
23% 25%
32%
38%
50%
77% 75%
68%
62%
50%
2013 2014 2015 2016 2017

Source: CB Insights
78
AI start-ups ranked by funding
(1/13)
Table Title Funding

# Company in mUS$ Investors Headquarter
TCV, General Atlantic, SoftBank,
1 ByteDance 3105 KKR & Co., Primavera Capital Beijing, China
Group
Alibaba Group, Fidelity
International, HOPU Investment
Management Company,
2 SenseTime 2600 Qualcomm Ventures, Temasek Beijing, China
Holdings, Tiger Global
Management, SBCVC, Silver Lake
Partners, Suning.com
CDH Investments, CITIC
Securities, CreditEase, Green
UBTECH
3 940 Pine Capital, Haier Group, Telstra, Shenzhen, China
Robotics
Tencent Holdings, Qiming
Venture Partners
AID Partners Capital, ARCHina
Capital Partners, Blackbird
Menlo Park,
4 Zoox 790 Ventures (Australia), Composite
California, U.S.
Capital Management, Draper
Fisher Jurvetson (DFJ)
Caffeinated Capital, Founders
Fund, GIC, Khosla Ventures,
5 Affirm 720 California, U.S.
Lightspeed Venture Partners,
Morgan Stanley, Ribbit Capital
Alibaba Group, Ant Financial,
Boyu Capital, SK Group, China
State-Owned Assets Venture
6 Face++ 607 Beijing, China
Investment Fund, Foxconn
Technology Group, Qiming
Venture Partners
Ant Financial, CCB International,
Foxconn Technology Group,
7 Megvii 607 Russia-China Investment Fund, Beijing, China
Sinovation Ventures, Sunshine
Insurance Group, SK Group
79
Note: Data as of December 2018
Source: CB Insights, Crunchbase
(2/13)
Table Title Funding

Andreas, BoxGroup, Credit
Suisse, Deep Fork Capital, Fabrice
8 Dataminr 577 New York, U.S.
Grinda, GSV Capital, Richmond
Global Ventures
5 Prime Ventures, AME Cloud
Ventures, Data Collective, Draper
Fisher Jurvetson, ICONIQ Capital,
9 Zymergen 574.1 California, U.S.
Innovation Endeavors, Obvious
Ventures, Prelude Ventures, SciFi
VC, SoftBank Vision Fund
Accel, CapitalG, Cloud Apps
Capital Partners, General Atlantic,
10 CrowdStrike 481 IVP (Institutional Venture California, U.S.
Partners), Instant Scale Ventures,
Telstra Ventures, Warburg Pincus
3W Partners, Amgen Ventures,
ARCH Venture Partners, Ireland
11 WuXi NextCODE 455 Strategic Investment Fund, Massachusetts, U.S.
Polaris Partners, Sequoia Capital,
Temasek Holdings, YF Capital
Accel, CapitalG, Credo Ventures,
Earlybird Venture Capital, Kleiner
12 UiPath 448 New York, U.S.
Perkins Caufield & Byers, Sequoia
Capital
Baillie Gifford, New Enterprise

13 Tempus Labs 320 Associates, Revolution Growth, T. Illinois, U.S.
Rowe Price
Allen & Company, Baillie Gifford,

Casdin Capital, First Round
14 Flatiron Health 313 New York, U.S.
Capital, GV, LabCorp, Roche,
Stripes Group, SV Angel
Amadeus Capital Partners,
Atomico, BMW i Ventures, C4
15 Graphcore 310 Ventures, Dell Technologies Bristol, UK
Capital, Microsoft, Samsung
Catalyst Fund, Sequoia Capital
80
(3/13)
Table Title Funding

Blackstone, Capital One Growth
Ventures, Dell Ventures, DFJ
Growth, Draper Nexus Ventures,
16 Cylance 297 California, U.S.
Fairhaven Capital Partners,
Insight Venture Partners,
TenEleven Ventures
Google, Sequoia Capital, SIG
17 Mobvoi 252.7 China, Volkswagen Group, Beijing, China
ZhenFund
Acadia Woods Partners, EPIC
Ventures, HWVP, Irish Strategic
Investment Fund, Kleiner Perkins
Caufield & Byers, Microsoft, Polar
18 InsideSales.com 251.2 Utah, U.S.
Capital Management, Polaris
Partners, QuestMark Partners,
Salesforce Ventures, Shares Post
Investment Management, USVP
Hoxton, Invoke Capital, Insight
Venture Partners, KKR & Co.,
19 Darktrace 230.5 SoftBank, Summit Partners, Cambridge, UK
Vitruvian Partners, Taliz Capital,
Ten Eleven Ventures
Breyer Capital, Sutter Hill
20 C3 IoT 228.5 Ventures, Pat House, Pat House, California, U.S.
The Rise Fund, TPG Growth
Accomplice, Atlas Venture, IA
Ventures, Intel Capital, Meritech
Capital Partners, New Enterprise
21 DataRobot 224.6 Massachusetts, U.S.
Associates, New York Life
Insurance Co, Recruit Strategic
Partners, Right Side
Baidu, Eastward Capital Partners,
Flybridge Capital Partners,
Fortress Investment Group,
Founders Fund, Kensington
22 ZestFinance 217 California, U.S.
Capital Partners Limited,
Matrix Partners, Northgate
Capital, Oakhouse Partners
81
(4/13)
Table Title Funding

Daimler, France Telecom (now
Orange S.A.), Global Catalyst
Partners, Hyundai Motor
23 SoundHound 215 California, U.S.
Company, Kleiner Perkins, Midea
Group, MKaNN, Nvidia GPU
Ventures, RSI Fund
Lansdowne Partners, Lundbeck,
Upsher Smith Laboratories,
24 BenevolentAI 207 London, UK
Woodford Investment
Management
Alibaba, CAS Investment
Management, China State-
Owned Assets Venture
Cambricon
25 200 Investment Fund, CMB Beijing, China
Technologies
International Capital, Group,
Oriza Seed Capital, SDIC Venture
Capital, Turing
China Bridge Capital, Tencent
26 iCarbonX 200 Holdings, Zhongyuan Union Cell Shenzhen, China
& Gene Eng
Daniel Klueger, Frederick Ryan,
Global Asset Management, Ivan Washington D.C.,
27 Afiniti 197.2
Seidenberg, TRG (The Resource U.S.
Group), Zeke Capital Advisors
28 Cybereason 188.6 SoftBank, Spark Capital Massachusetts, U.S.
Allianz Ventures, BMW i Ventures,

Draper Nexus Ventures, General
Motors Ventures, Greylock
29 Nauto 182.6 California, U.S.
Partners, Index Ventures,
Playground Global, SoftBank,
Toyota AI Ventures
Andreessen Horowitz, C4
Ventures, Index Ventures, JP
30 Anki 182 Morgan Chase & Co., Silicon California, U.S.
Valley Ventures, Two Sigma
Ventures
Access Industries, BlackPine PE
Sentient
31 174.4 Partners, Horizon Ventures, NJF California, U.S.
Technologies
Capital, Tata Communications
82
(5/13)
Table Title Funding

Advantech Capital, CDIB Capital,
Credit Suisse, IDG Capital
Rokid
32 158.3 Partners, Linear Venture, Hangzhou, China
Corporation
Temasek Holdings, Vision Plus
Capital, Walden International
Baseline Ventures, EPIC
Ventures, Focus Ventures, JetBlue
33 Shape Security 132 Technology Ventures, Kleiner California, U.S.
Perkins, Norwest Venture
Partners, Singtel Innov8
Foxconn Technology Group,
34 CloudMinds 130 Keytone Ventures, SoftBank, California, U.S.
Walden International
FANUC, Hakuhodo, Hitachi,
Mitsui & Co, Mizuho Corporate
Preferred
35 129.9 Bank, Nippon Telegraph and Tokyo, Japan
Networks
Telephone Corporation, Toyota
Motor
Cherubic Ventures, China Media
Capital (CMC), Hearst Ventures,
36 Liulishuo 128.9 Shanghai, China
GGV Capital, IDG Capital, RTA
Capital, Trustbridge Partners
Brain Qualcomm Ventures, SoftBank
37 125 California, U.S.
corporation Vision Fund
Accel, Cathay Innovation,
Columbia Pacific Advisors,
Deutsche Börse, Ericsson,
38 Trifacta 124.3 California, U.S.
Google, Greylock Partners,
Ignition Partners, New York Life
Ventures, Ridge Ventures
Felicis Ventures, Fenox Venture
Vicarious Capital, Iconical, Khosla Ventures,
39 122 California, U.S.
Systems Morpheus Ventures, Samsung
NEXT, Wipro Ventures
Alpha Intelligence Capital,
Declaration Partners, Hawk
Equity, iNovia Capital, Georgian
40 WorkFusion 121.3 New York, U.S.
Partners, Guardian, iNovia
Capital, New York Presbyterian
Ventures, NGP Capital
83
(6/13)
Table Title Funding

Cerebras
41 112 Benchmark California, U.S.
Systems
Bessemer Venture Partners,
Columbia Capital, Edgemore
42 Endgame 111.4 Capital, Kleiner Perkins, Paladin Virginia, U.S.
Capital Group, Tech Operators,
Top Tier Capital Partners
Advantech Capital, Northern
Light Venture Capital, Oriza
43 Petuum 108 Pennsylvania, U.S.
Ventures, SoftBank, Tencent
Holdings
Business Development Bank of
Canada, Canadian Government,
Data Collective, Hanwha
44 Element AI 107 Montréal, Canada
Investment Corp, Intel Capital,
M12, National Bank of Canada,
Nvidia, Real Ventures
Centerview Capital, Citi Ventures,
Draper Nexus Ventures,
45 Ayasdi 106.3 FLOODGATE, GE Ventures, IVP, California, U.S.
Khosla Ventures, Kleiner Perkins,
Caufield & Byers
Bill & Melinda Gates Foundation,
CRV, Data Collective DCVC, EPIC
Recursion
46 105.4 Ventures, Felicis Ventures, Lux Utah, U.S.
Pharmaceuticals
Capital, Menlo Ventures,
Mubadala Investment Company
Barclays, BNP Paribas, Credit
Digital Suisse, Goldman Sachs, HCA, In-
47 104 Tennessee, U.S.
Reasoning Q-Tel, Lemhi Ventures, NASDAQ,
Square Capital
Bain Capital Ventures, Battery
Ventures, Chris Dixon,
48 BloomReach 97 California, U.S.
New Enterprise Associates
Alexandria Venture Investments,
Collaborative Fund, GV, iSELECT
Benson Hill
49 94.7 FUND, Lewis & Clark Ventures, North Carolina, U.S.
Biosystems
Mercury Fund, Mercury Fund,
Prelude Ventures LLC
84
(7/13)
Table Title Funding

Goldman Sachs, HCL
Technologies, Northgate Capital,
50 MOOGsoft 92.9 California, U.S.
Redpoint, Singtel Innov8, ST
Telemedia, Wing Venture Capital
Ben Brigham, CIBC,
Green D Ventures, Kennet
51 Conversica 87 Partners, Providence Strategic California, U.S.
Growth, Savano Capital, The Yard
Ventures, Toba Capital
Clal Insurance Enterprises
OrCam
52 86.4 Holdings, Intel Capital, Meitav Jerusalem, Israel
Technologies
Investment House, Moshe Gaon
Alumni Ventures Group, Blue Ivy
Ventures, Collaborative Fund
Collaborative Fund, Green D
53 Upstart 85.7 California, U.S.
Ventures, First Round Capital,
Khosla Ventures, Rakuten,
Third Point Ventures
Adam Balon, Demis Hassabis,
Hoxton Ventures, Jon Wright,
54 Babylon Health 85 Kinnevik AB, Mustafa Suleyman, London, UK
NNC Holdings, Richard Reed,
Vostok New Ventures
Deutsche Telekom Capital
Partners, Global Founders
55 Dynamic Yield 83.3 Capital, Innovation Endeavors, New York, U.S.
Marker, Naver Corporation,
Union Tech Ventures, Vertex
Ventures, Viola Growth
EDBI, FirstFloor Capital, Media-
Tek, Naver Corp, LINE Corp,
56 Appier 81.5 Pavilion Capital, Qualgro VC, Taipei, Taiwan
Sequoia Capital, Translink Capital,
UOB Venture, WI Harper Group
Balyasny Asset Management,
Clearvision Ventures, CME
57 Orbital Insight 78.7 Ventures, Geodesic Capital, GV, California, U.S.
Intellectus Partners LLC, Lux
Capital, Sequoia Capital
85
(8/13)
Table Title Funding

Andreessen Horowitz, Anne
Wojcicki, Asset Management
Ventures (AMV), Data Collective
58 Freenome 77.6 California, U.S.
DCVC, Founders Fund, GV,
Innovation Endeavors, Polaris
Partners, Section 32, Verily
GGV Capital, HOF Capital,
InnoSpring Seed Fund, New
Enterprise Associates, Manicv
59 Drive.ai 77 California, U.S.
Mobility, Northern Light Venture
Capital, Nvidia GPU Ventures,
Oriza Ventures
Accel, Aglaé Ventures, Alven
Capital, Clark Valberg, Point Nine
60 Algolia 74.3 California, U.S.
Capital, SaaStr Fund, Storm
Ventures
Capital One Growth Ventures,
Nexus Venture Partners, Nvidia,
61 H2O.ai 73.6 Paxion Capital Partners, California, U.S.
Transamerica, Transamerica
Ventures, Wells Fargo & Co
Accenture Technology Ventures,
Chevron Technology Ventures,
CICC, Eight Capital, GE Ventures,
62 MAANA 71.2 Global Bridge Capital, Intel California, U.S.
Capital, Saudi Aramco Energy
Ventures, Shell Technology
Ventures, Sino Capital Group
GE Ventures, GV, Granite Hill
Capital Partners, Hewlett Packard
Ventures, MassMutual Ventures,
63 Tamr 69.2 New Enterprise Associates, Pear Massachusetts, U.S.
Tree Partners, SBI Investment,
SineWave Ventures, Thomson
Reuters, Work-Bench
Ericsson Ventures, Kleiner
Perkins Caufield & Byers, Mitsui
64 Drawbridge 68.7 California, U.S.
& Co, Northgate Capital, Sequoia
Capital
86
(9/13)
Table Title Funding

Alameda Ventures, CME
Ventures, The Entrepreneurs'
65 Spark Cognition 63.2 Texas, U.S.
Fund, US Department of Energy,
Verizon Ventures
Accenture, Accel, AirTree
Ventures, Cisco, Deutsche
Telekom Capital Partners, EDBI,
66 Paxata 61 California, U.S.
EDB Investments, In-Q-Tel,
Synapse Partners, Toba Capital,
Walden Riverwood Ventures
Accel, Morgan Stanley Alternative
Investment Partners, Rincon
67 Invoca 60.8 Venture Partners, Salesforce California, U.S.
Ventures, StepStone Group,
Upfront Ventures
Christian Faes, CreditEase
Fintech Investment Fund, David
68 Onfido 60.3 Rowan, Jon Reynolds, Khaled England, UK
Helioui, Plug and Play, Salesforce
Ventures, Taavet Hinrikus
Airbus Ventures, Intel Capital,
69 AEYE 59.1 Kleiner Perkins, R7 Partners, California, U.S.
Taiwania Capital, Tyche Partners
Canvas Ventures, Industry
Ventures, M12, Promus Ventures,
70 CrowdFlower 58 California, U.S.
Salesforce Ventures, Trinity
Ventures
Canvas, Felicis, Harmony
Figure Eight
Partners, Industry Ventures, K9
71 (formerly 58 California, U.S.
Ventures, Microsoft Venture,
CrowdFlower)
Promus Ventures, Trinity
Ventures
Bank of America Merrill Lynch,
Bloomberg, Goldman Sachs.
Versive (formerly
Formation 8, Madrona Venture
72 Context 57 Washington, U.S.
Group, New York Life Insurance
Relevant)
Co, Rolling Bay Ventures, Work
Bench
87
(10/13)
Table Title Funding

Emergence Capital Partners,
73 Chorus.ai 55.3 California, U.S.
Georgian Partners, Redpoint
AME Cloud Ventures, Andreas
Bechtolsheim, Data Collective
74 Mythic 55.2 DCVC, DFJ, Lockheed Martin Texas, U.S.
Ventures, Lux Capital, SoftBank
Ventures Korea
Accomplice, Atlas Venture, Knight
75 Captricity 51.9 Foundation, Social Capital, White California, U.S.
Mountains Insurance Group
AME Cloud Ventures, B Capital
Group, Baidu Ventures, Data
Collective, Draper Associates,
Draper Fisher Jurvetson (DFJ),
76 Atomwise 51.3 California, U.S.
Dolby Family Ventures, Farzad
(Zod) Nazem, Grand Challenges
Canada, OS Fund, Tencent
Holdings, Y Combinator
aMoon Fund, Aurum Ventures,
Deep Fork Capital, Dolby Family
Zebra Medical
77 50 Ventures, Intermountain Shefayim, Israel
Vision
Healthcare, Khosla Ventures,
OurCrowd-GCai
Battery Ventures, In-Q-Tel, Jump
Narrative
78 47.87 Capital, Northwestern University, Illinois, U.S.
Science
Sapphire Ventures, USAA
B Capital Group, Day One Capital
Fund Management, Draper
Associates, Inventure Oy, Nvidia,
79 AImotive 47.5 Budapest, Hungary
Prime Ventures, PortfoLion,
Robert Bosch Venture Capital,
Samsung Catalyst Fund, Tamares
Darling Ventures, Dell, General
Electric (GE), Honeywell Venture
Foghorn Capital, Intel Capital, March
80 47.5 California, U.S.
Systems Capital Partners, Saudi Aramco
Energy Ventures, The Hive
Group,Yokogawa Electric Corp.
88
(11/13)
Table Title Funding

Aleph, Alibaba Innovation
Ventures, Mosaic Ventures,
81 Nationwide Ventures, Ibex
Nexar 44.5 Tel Aviv, Israel
Ventures, True Ventures,
Expansion Venture Capital, Tusk
Ventures
11.2 Capital, Bold Capital
82 Partners, Data Collective DCVC,
Kindred Systems 44 California, U.S.
Eclipse, First Round Capital,
Tencent Holdings
DNA Capital, Emergent Medical
Partners, GE Ventures, MedTech
Innovator, Morado Ventures,
83 Arterys 43.7 New York Presbyterian Ventures, California, U.S.
Northwell Ventures, ORI Capital,
Temasek Holdings, Varian
Medical Systems
Battery Ventures, Clear Ventures,
84 Reflektion 42.8 Hasso Plattner Ventures, Intel California, U.S.
Capital, Nike
Bessemer Venture Partners
IAngels, La Maison Compagnie
85 Applitools 41.8 d'Investissement, Magma Tel Aviv, Israel
Venture Partners, OpenView,
Sierra Ventures
Corazon Capital, GV, LDV Capital,
86 Clarifai 40 Lux Capital, Menlo Ventures, New York, U.S.
Union Square Ventures
IBM, Intel Capital, M12, Microsoft
87 Ventures, Norwest Venture
CognitiveScale 40 Texas, U.S.
Partners, The Westly Group,
USAA
360 Capital Partners, Bosch, CEA
Prophesee Investissement, iBionext Growth
88 (formerly 40 Fund, Intel Capital, Nissan- Paris, France
Chronocam) Renault, Robert Bosch Venture
Capital
89
(12/13)
Table Title Funding

Accel, Elaia Partners, General
89 Shift Technology 39.8 Catalyst, Iris Capital, Microsoft Paris, France
Accelerator Paris
Greenbay Ventures,
FOUNDER.org, MIG, New
Enterprise Associates,
90 KONUX GmbH 38.5 Munich, Germany
Unternehmertum Venture
Capital Partners, Upbeat
Ventures
Correlation Ventures, Crosslink
Capital, Cultivian Sandbox
Ventures, Data Collective,
91 Descartes Labs 38.3 Expansion Venture Capital, New Mexico, U.S.
March Capital Partners, Milliways
Ventures, TenOneTen Ventures,
ValueStream Labs
First Fellow Partners, Quantum
Strategic Partners, Tom Glocer,
92 AlphaSense, Inc. 37.1 California, U.S.
Triangle Peak Partners, Tribeca
Venture Partners
Benchmark, Binux Capital, Plug
93 Aquifi 36.8 and Play, Shenzhen O-film California, U.S.
Technology
Ascent Venture Partners,
Earlybird Venture Capital,
94 RapidMiner 36 Longworth Venture Partners, Massachusetts, U.S.
Nokia Growth Partners (NGP),
OpenOcean
Alibaba Capital Partners,
Gigi Levy Weiss Korea Investment
Partners, MizMaa Ventures,
95 Twiggle 35 Tel Aviv, Israel
Naspers,Sir Ronald Cohen, State
of Mind Ventures, Yahoo! Japan,
YJ Capital
415, Acequia Capital (Acecap),
Greg Gladwell, Ignition Partners,
96 Tractable 34.9 Insight Venture Partners, Plug England, UK
and Play, Scott Roza, Zetta
Venture Partners
90
(13/13)
Table Title Funding

Andreas Bechtolsheim, Canaan
Partners, Data Collective DCVC,
97 PerimeterX 34.5 California, U.S.
Vertex Ventures, Vertex Ventures
US
Fenox Venture Capital, Horizons
Ventures, Kantar Consulting, Waltham, Mas-
98 Affectiva 34.3
Kleiner Perkins Caufield & Byers, sachusetts, U.S.
Myrian Capital, WPP
Accel, Collaborative Fund,
Houghton Mifflin Harcourt,K9
Ventures, Mattel, Sesame
99 Osmo 32.5 California, U.S.
Workshop, Shea Ventures, StartX
(Stanford-StartX Fund), Upfront
Ventures
100 Mya Systems 32.4 Emergence, Foundation Capital California, U.S.
91
Number of M&A deals grew by 38%
from 2016 to 2017
Mergers & acquisitions
After a robust growth from 2015 to 2016, the number of M&A deals jumped up by 38% in 2017.
Looking at the most recent M&A deals, big tech companies like Google, Apple, Amazon, Microsoft,
IBM, or Facebook have often been the acquirer. But Chinese tech giants like Baidu or rising stars in
the start-up world, like Twitter, Uber, or Spotify, also buy AI companies.
The biggest deal (with announced deal size) has been Microsoft’s acquisition of GitBub – a company
that provides a large code repository that is used by developers and companies for hosting projects,
documentation, and code for US$7,500 million in October 2018.
Number of M&A deals
38%
119
87%
86
12%
78%
46
41
23
2013 2014 2015 2016 2017
92
Source: CB Insights
Recent M&A deals (1/9)
Deal value
Company Description in mUS$ Acquirer Date
Provides a large code
repository that is used by
GitHub developers and companies for 7500 Microsoft Oct 2018
hosting projects,
documentation, and code
Provides an AI-based engine
which automatically locates
DataFox and pulls the most current n/a Oracle Oct 2018
information available on public
and private businesses
Provides easy-to-use deep

Lobe n/a Microsoft Sep 2018
learning development tools
Focused on deep learning

Vertex.AI compilation tools and n/a Intel Aug 2018
associated technology
Provides cloud-based, AI-

powered marketing
Datorama intelligence and analytics n/a Salesforce July 2018
platform for enterprises,
agencies, and publishers
Offers an AI platform that

Bonsai empowers enterprises to build n/a Microsoft June 2018
and deploy intelligent systems
Provides conversational AI
technology to be used in
Semantic Machines n/a Microsoft May 2018
customer service and home
automation
Cloud workload mobility
software that helps
Velostrata n/a Google May 2018
organizations move workloads
to and from public cloud
93
Source: CB Insights, company websites
Deal value
Provides a data analytics
Sqrrl tool to combat 40 Amazon Jan 2018
cybersecurity threats
An online platform of tools

Pop Up Archive for organizing and searching n/a Apple Dec 2017
digital spoken words
A conversational commerce
platform to connect
Banter n/a Google Nov 2017
businesses with customers
via messaging apps
A platform that enables its

users to design and develop
Init.ai n/a Apple Oct 2017
natural-language
conversational mobile apps
Regaind Uses AI to analyze pictures n/a Apple Sep 2017
Cosmify Provider of data analytics n/a Meltwater Aug 2017
AI-based news and data

Algo n/a Melwater Aug 2017
tracker
A neural network-based AI
platform and SDK to quickly
AIMatter n/a Google Aug 2017
detect and process images
on mobile devices
A company that provides
solutions based on artificial
Halli Labs n/a Google July 2017
intelligence and machine
learning
An intelligent assistant for

Ozlo n/a Facebook July 2017
iOS and the Web
94
Deal value
A machine learning start-up
that works on natural
KITT.AI n/a Baidu July 2017
language understanding
technologies
Provides eye-tracking
SensoMotoric
technology useful for VR n/a Apple June 2017
Instruments
and AR applications
Uses AI to turn dark or
Lattice Data unstructured data into 200 Apple May 2017
structured data
A machine learning start-up

Niland specializing in music search n/a Spotify May 2017
and recommendations
Develops vision perception

software and hardware with
x Perception n/a Baidu Apr 2017
applications in robotics and
virtual reality
An online service that hosts
Kaggle data science and machine n/a Google Mar 2017
learning competitions
Uses AI to automate the
extraction of data from
Wrapidity n/a Meltwater Feb 2017
unstructured web-based
content
A cybersecurity and
machine learning firm
RealFace 2 Apple Feb 2017
specializing in facial
recognition technology
A company that specializes
Raven Tech in AI voice assistant platform n/a Baidu Feb 2017
development
Maluuba Natural language processing n/a Microsoft Jan 2017
95
Deal value
Provides cybersecurity
Harvest.ai 20 Amazon Jan 2017
solutions
Duck
Develops enterprise data Creek
Yodil n/a Dec 2016
management software Techno-
logies
Geometric Develops machine learning
n/a Uber Dec 2016
Intelligence techniques
An AI platform that solves
the data integration needs
General
Bit Stew Systems of utilities, aviation, oil and 153 Nov 2016
Electric
gas, and manufacturing
industries
Develops machine learning
applications for the General
Wise.io n/a Nov 2016
customer experience Electric
market
A computer vision company
that develops accurate
Zurich Eye position information for n/a Facebook Nov 2016
robots to navigate in indoor
and outdoor environments
Tuplejump Provider of big data n/a Apple Sep 2016
Provider of natural language

Angel.ai n/a Amazon Sep 2016
processing
Provides a conversational
UX platform enabling
API.AI n/a Google Sep 2016
natural language
interactions
Manufactures computer
vision processors used in
Movidius n/a Intel Sep 2016
drones and virtual reality
devices
96
Deal value
A machine learning
Tuplejump company that simplifies the 20 Apple Sep 2016
handling of big data
Provides natural language
Genee processing to schedule n/a Microsoft Aug 2016
meetings
Provides self-driving
Otto 680 Uber Aug 2016
technology
Provides computer vision
SAIPS n/a Ford Aug 2016
solutions
Nervana Provides deep learning

350+ Intel Aug 2016
Systems algorithms
Designs and develops
BeyondCore automated analytics 110 Salesforce Aug 2016
software
Provides a machine learning
platform to help other
Turi developers embed its 200 Apple Aug 2016
capabilities into their own
applications
Provides technology to
predict consumer buying
Sales Predict n/a eBay July 2016
behavior and sales
conversion
Develops machine-learning-
Mood stocks based image recognition n/a Google July 2016
technology for smartphones
Magic Pony Provides computer vision

150 Twitter June 2016
Techno logy solutions
A machine vision company

that builds algorithms that
Itseez n/a Intel May 2016
help vehicles avoid
obstacles and collisions
97
Deal value
Develops cloud-based data
technology solutions that
Expert maker help automate and optimize n/a eBay May 2016
data-driven decisions and
processes
Provides machine-learning-
Crosswise based cross-device mapping 50 Oracle Apr 2016
data
MetaMind processing and image n/a Salesforce Apr 2016
recognition solutions
Provides an open-source-
Prediction IO based machine learning n/a Salesforce Feb 2016
server
Provides AI-powered
SwiftKey 250 Microsoft Feb 2016
smartphone keyboards
Develops dialogue and

audio analysis products and NICE
Nexidia 135 Jan 2016
technologies for optimizing Systems
audio and video media
Uses AI to detect emotion

Emotient n/a Apple Jan 2016
from facial expressions
Provides recommendation
Hunch and predictive analytics n/a n/a Jan 2016
technology
Provides a cognitive
Saffron
computing platform for n/a Intel Oct 2015
Technology
industries
A deep learning company

Perceptio that uses AI to classify photos n/a Apple Oct 2015
on smartphones
98
Deal value
Provides AI software that
helps computers and people
VocalIQ n/a Apple Oct 2015
speak to each other in a
more natural dialogue
Offers a cloud-based image
analysis solution that
Orbeus n/a Amazon Sep 2015
identifies faces, scenes, and
objects
Provides geospatial software
deCarta n/a Uber Aug 2015
platforms
Develops machine learning
Whetlab technologies based on n/a Twitter June 2015
artificial intelligence software
Taboola A content discovery platform n/a Baidu May 2015
Uses machine learning to

Timeful help people manage their n/a Google May 2015
schedules better
Designs and develops
Tempo AI enterprise mobility n/a Salesforce May 2015
management solutions
Operates a personalized
predictive marketing platform
that helps companies
TellApart n/a Twitter May 2015
leverage customer data to
drive sales through
personalized marketing
Provides a secure cloud
Explorys computing platform for the n/a IBM Apr 2015
healthcare industry
processing services through a
AlchemyAPI SaaS API for various n/a IBM Mar 2015
businesses and computing
applications
99
Deal value
Provides actionable social
Sociocast intelligence for web and n/a AOL Mar 2015
mobile platforms
Wit.AI n/a Facebook Jan 2015
solutions for apps or devices
Provides prescriptive
Granata
analytics technology to
Decision n/a Google Jan 2015
organizations and
Systems
individuals
Develops text analysis
Equivio software for the legal 200 Microsoft Jan 2015
market
Enables developers to build
Wit.AI a Siri-like speech interface n/a Facebook Jan 2015
for apps and devices
Provides object recognition
and text recognition
Vision Factory n/a Google Oct 2014
systems based on deep
learning
Engaged in deep learning
Dark Blue Labs technologies to understand n/a Google Oct 2014
natural language
Creates visual city guides by
analyzing photos from social
Jetpac media to infer key n/a Google Aug 2014
information about the
places they were taken
An AI-powered assistant
that can schedule
Emu n/a Google Aug 2014
appointments and sets
reminders
Develops revenue analytics
Freshplum solutions for eCommerce n/a TellApart July 2014
companies
100
Deal value
Develops a visual artificial

intelligence software that
Madbits understands, organizes, and n/a Twitter July 2014
extracts relevant
information from raw media
Provides predictive analytics

products to Fortune 500 HERE
Medio Systems n/a June 2014
and emerging growth Holding
companies
Provides a cognitive
Cogenea computing and n/a IBM May 2014
conversational AI platform
Provider of voice
Novauris
recognition software for n/a Apple Apr 2014
Technologies
mobile devices
An AI platform that
performs a variety of AI
DeepMind tasks; it is not pre- 600 Google Jan 2014
programmed and can learn
from experience
Develops software that

connects, assembles, and
Causata transforms customer data n/a NICE Aug 2013
into portable
real-time intelligence
Develops voice and image

DNN research n/a Google Mar 2013
recognition solutions
101
Competitive landscape
Companies from various industries are currently developing AI and related
applications. Google, IBM and Microsoft are leading AI innovations in the IT
industry, whereas Amazon and eBay are investing in AI to improve their
eCommerce platform, and ridesharing company Uber is using AI for
autonomous driving, food deliveries, and mapping research.
Collaborative development is on the rise, and leading companies such as
Amazon, Apple, Facebook, Google/DeepMind, IBM, and Microsoft are currently
working in partnership towards developing AI applications. The acquisition of
small-scale AI companies in relevant fields by tech giants like Apple, IBM, and
Microsoft is decreasing the learning curve. Other leading companies include
Baidu, Facebook, and Salesforce.
102
Most of the leading AI companies
are from the U.S.
Company comparison
Location of selected leading AI companies
Comparison of selected leading AI companies
Revenue
1
Company Headquarter in bnUS$ Key AI areas
Text-to-speech, computer vision, deep
Amazon Washington, U.S. 177.9 learning, NLP
Apple California, U.S. 265.62 Machine learning

Baidu Beijing, China 13.0 Machine learning, robotics
eBay California, U.S. 9.6 Predictive analytics, cloud-based AI, big data
Language technology, machine learning,
Facebook California, U.S. 40.6 computer vision
Machine learning, deep learning, automotive
Google California, U.S. 110.9 industry
IBM New York, U.S. 79.1 Machine learning, cognitive architectures

Microsoft Washington, U.S. 110.43 Machine vision, machine learning, healthcare
Salesforce California, U.S. 10.54 Machine learning, analytics
Voice and image recognition, machine
Uber California, U.S. 7.5
learning, automotive industry
103
1: As of Dec 2017 2: As of Sep 2018 3: As of June 2018 4: As of Jan 2018
Source: Annual reports, corporate news letters
Amazon works in various areas to
advance Alexa's capabilities
Amazon (1/2)
Amazon.com, Inc., incorporated in 1996, is a Seattle-based company that primarily sells a range of
products and services through its online retail platform. In addition, the company also manufactures
and sells electronic devices, including the portable reading devices Kindle eReaders and Fire tablets,
Fire TVs, and Echo. It provides online direct publishing platforms for independent authors and
publishers to publish books in the Kindle Store, an online repository of eBooks.
In November 2016, Amazon launched its AI platform, which consists of new services such as
Rekognition, Amazon Polly, and Lex for enterprises. The company has invested US$100 million in the
Alexa Fund, which has so far invested in 22 start-ups working at various stages of product
development. In addition, Amazon has created the Alexa Prize in the field of conversational AI,
challenging university students to develop bots that can hold human conversation.
▪ Year incorporated: May 1996

▪ Revenue: US$177.9 billion (Dec 2017)
▪ Profit before tax: US$3.8 billion (Dec 2017)
▪ Market capitalization: US$563.5 billion (Dec 2017)

▪ Employees: 566,000 (Dec 2017)
R&D areas
Image analysis Amazon Computer vision
Natural language
Machine learning Deep learning
understanding
104
Source: Annual report, company information, press releases
Amazon’s AI research is used in
several services and features
Amazon (2/2)
Amazon‘s current AI service initiatives are focused around natural language understanding (NLU),
automatic speech recognition (ASR), visual search and image recognition, text-to-speech (TTS), and
machine learning (ML).
The company aims at using the outcomes of AI research in the following services:
Amazon Rekognition: Launched in November 2016, mainly for enterprises,

it provides image recognition, categorization, and facial analysis. It can, for
example, recognize human faces, detect emotion, and tell whether a
particular subject is wearing glasses or not.
Amazon Polly: a deep-learning-powered text-to-speech (TTS) service,

mainly for enterprises, that converts a text input into an MP3 stream that
sounds like an actual conversation.
Amazon Go: In December 2016, Amazon launched Amazon Go, a retail

store which uses AI techniques like computer vision, sensor fusion, and
deep learning to expedite the shopping process. Using the Amazon Go
app, the shopper simply walks into the store, scans the products, and
walks out, paying via the Amazon Go app.
Amazon Alexa: a natural language processing system that can answer

questions, provide weather reports, play music, and control the lighting in
the house. The voice assistant powers various Amazon products, including
Echo (hands-free speaker), Echo Dot (hands-free, voice-controlled device
with a built-in speaker), Amazon Tap (a portable Bluetooth and Wi-Fi
enabled speaker) and Amazon Fire TV.
Cybersecurity: In January 2017, Amazon acquired harvest.ai, a provider of

AI-powered cybersecurity solutions. The company uses machine learning
to track changes in user behavior, business systems, and applications and
to prevent data theft.
105
Source: Company information, press releases, CB Insights
One focus area of Apple’s AI
initiatives is enhancing battery life
Apple
Apple, headquartered in Cupertino, California, is a technology company that designs, manufactures,

and markets personal computers, related software, services, peripherals, networking solutions,
portable digital music players, mobile communication, and media devices, as well as third-party
digital content and applications.
Over the years, Apple has kept its AI projects and research publications shrouded in secrecy.
However, in December 2016, the company published its first AI paper, describing how image
recognition training can be enhanced by using computer-generated images over real-world images.
Since then, Apple has joined the Partnership on AI, which has Amazon, Facebook, Google/DeepMind,
IBM, and Microsoft as its members. Apart from advances in Siri, Apple currently uses AI and machine
learning to improve the battery life across its product line, based on consumer usage patterns and
data on how to manage power consumption at a component level.
▪ Year incorporated: Apr 1976

▪ Revenue: US$265.6 billion (Sep 2018)
▪ Profit before tax: US$72.9 billion (Sep 2018)

▪ Employees: 132,000 (Sep 2018)
R&D areas
Neural networks Apple Voice analysis
Machine learning Personalization
106
Baidu’s Big Data Lab works in
several AI areas
Baidu (1/2)
Baidu was incorporated in 2000 and is a Beijing-based technology company that primarily provides a
Chinese-language internet search platform. In addition, the company also offers, among other things,
social networking, photo sharing, entertainment, security, music products, software, mapping
services, and mobile applications. The company sells its products through its direct sales team and
third-party distributors.
Baidu formed its Big Data Lab in July 2014, which focuses on developing large-scale AI algorithms and
applications in areas such as predictive analytics, large data structure algorithms, and intelligent
systems research. The lab’s AI projects include core search technologies, robots, and knowledge-
based AI.
▪ Year incorporated: Jan 2000

▪ Employees: 39,343 (Dec 2017)
R&D areas
Search advertising Baidu Speech recognition
Robots Knowledge-based AI Automotive industry
107
One area of Baidu’s AI research is
robots
Baidu (2/2)
The AI projects of Baidu’s Big Data Lab include:

▪ Core search technologies: Current research includes fundamental hashing techniques, hashing-
based large-scale machine learning, hashing-based indexing, and fast similarity search, image
search, web search, and ranking.
▪ Robots: Baidu is developing intelligent robots that can not only have a one-on-one conversation
with humans but can also ask questions to get a better understanding of their latent intentions.
DuNurse is an example of a robot developed for healthcare applications. It can diagnose potential
medical conditions and make recommendations on the line of treatment.
▪ Knowledge-based AI: using voice and language recognition to analyze large volumes of
unstructured data for use in domains such as professional query-answering, high quality web
search, and automated legal advice.
▪ Other initiatives:
− Baidu is developing a speech recognition engine with an expected accuracy of 99%. Its latest
product, called Deep Speech 2, uses deep learning to recognize words spoken in English and
Mandarin and is currently used in several Baidu apps such as Duer (Siri equivalent) and Melody
(a chatbot that assists doctors with recommendations and treatment options).
− Partnered with zaijia.com, a manufacturer of home appliances, to launch an AI-based home

assistant robot. Apart from communicating with humans, the robot offers security monitoring
and entertainment services.
− In April 2017, Baidu open sourced its self-driving platform, called Project Apollo, to enable other
automobile companies to develop their own autonomous driving systems.
− In February 2017, Baidu acquired Raven Tech, an AI start-up developing a voice recognition
assistant.
− Baidu developed PaddlePaddle, its open-source deep learning platform, in 2016.

− In September 2016, Baidu partnered with Nvidia to jointly develop a platform for semi-
autonomous vehicles.
− In September 2016, Baidu launched a US$200 million venture capital fund to invest in AI
projects.
108
Source: Company information, press releases, CB Insights
eBay’s AI efforts aim to improve
communication in eCommerce
eBay (1/2)
eBay Inc., headquartered in San Jose, California, provides an online platform that connects buyers
with sellers worldwide. The platform enables registered sellers to create a virtual online store from
where buyers can search and select products to buy. The company offers a retail platform via the
web and also through mobile applications. In addition, the company also sells tickets to sports
events, concerts, and live shows through its online ticketing platform stubhub.com.
eBay aims at improving buyer/seller communications through the use of AI. The company made
many AI-related acquisitions in 2016 in the areas of predictive analytics, cloud-based AI, and big data.
▪ Year incorporated: Sep 1995
R&D areas
Deep science eBay Voice analysis
Machine learning Predictive analytics
109
eBay‘s AI initiatives focus on buyers
and sellers
eBay (2/2)
eBay’s current area of focus in AI is to achieve real-time customer behavior analysis to deliver
personalized experiences to its customers.
Its key AI initiatives are:
eBay ShopBot: an AI-powered smart personal shopping assistant,

currently in beta stage, which is integrated with the Facebook messenger.
It interacts with buyers to help them select the best deal for a particular
product.
Inventory solutions: an AI-powered retail platform that helps sellers to

identify inventory gaps for a particular product and sends alerts to stock
up.
110
Source: Company information, press releases
Facebook created its AI research
group in 2013
Facebook (1/2)
Facebook, founded in 2004, is a technology company that provides social media, mobile application,
and communications platforms. It owns several other popular social media platforms, such as the
photo-sharing platform Instagram and the instant communication application WhatsApp, which
allows people to connect and interact with each other. The company also offers a virtual reality
platform, Oculus VR, that allows users to play games, consume content, and connect with others in
an immersive and interactive environment.
In 2013, the company created the Facebook Artificial Intelligence Research group (FAIR), which is run
by neural networks expert Yann LeCun and has over 60 researchers and scientists. In 2015,
Facebook launched the Applied Machine Learning (AML) group, which is helmed by Joaquin Candela
and integrates its overall AI efforts into the company’s products. The division looks after language
technology, core machine learning, computational photography, computer vision, and image
recognition.
▪ Year incorporated: Feb 2004


R&D areas
Machine learning Facebook Language technology
Deep Text
111
Facebook’s AI initiatives focus on
enhanced text understanding
Facebook (2/2)
Facebook is currently working on the following areas in terms of AI research and development:
FBLearner Flow: is its machine learning platform, which can reuse
algorithms across various products to deliver a unique personalized
experience for every user. According to a company post, the AI system,
which is used by more than 25% of its engineering team, is used, among
other things, to rank and personalize News Feed stories, filter offensive
content, highlight trending topics, and rank search results.
Language technology: One of Facebook’s core area of focus with regards

to AI integration is language technology. The company started to develop
its Language Group after the 2013 acquisition of Mobile Technologies, a
developer of voice recognition and translation tools. Currently, the group
is focused on developing Facebook’s first AI-powered speech recognition
and conversational understanding capabilities.
Deep Text: The company’s latest AI offering, DeepText, is based on deep

neural networks and helps it to analyze each news story, public post, and
even text messages, to determine the context of the content and then
make suggestions accordingly. For example, if a user posts a status
update saying that he/she needs a ride, the AI system will suggest using
services such as Uber or Lyft. DeepText uses deep neural network
architectures such as convolutional and recurrent neural nets.
112
Google’s AI efforts touch various
industries and services
Google (1/3)
Google, a global technology company founded in 1998, offers search, advertising, operating systems
and platforms, enterprise, and hardware products. The company operates as a subsidiary of
Alphabet Inc. and is headquartered in Mountain View, California. Even though Google started its AI
research many years ago, it was the acquisition of British AI company DeepMind in 2014 for over
US$600 million that marked the beginning of the next era of AI development in the company. Since
then, Alphabet has been one of the biggest spenders on AI development in the world. DeepMind is
currently focused on developing AI applications to help manage energy infrastructure, hone health-
care systems, and improve access to clean water.
Google Brain handles all aspects of AI development and application, with the technology now
touching almost all areas of its business, including vehicles, healthcare, translation, photography, and
advertising.
▪ Year incorporated: Sep 1998
R&D areas
Search advertising Google Healthcare
Translation Voice assistance Automotive industry
113
Google is using AI to upgrade its
traditional offerings
Google (2/3)
Google is currently working on the following areas in terms of AI research and developments:
Search advertising: Google Search, which generated over 90% of the
company’s revenues, is reinventing itself through AI and machine learning.
In 2015, Google launched a deep learning system called RankBrain, which
delivers more accurate search results by teaching itself to interpret
queries based on a person’s intent as opposed to a predefined program.
The algorithm is essentially capable of guessing words or phrases with a
similar meaning and of filtering the results accordingly, thereby enabling it
to deliver accurate results, even for novel queries.
Google Translate: In September 2016, Google launched the Google Neural

Machine Translation (GNMT) system, which uses deep learning to translate
entire sentences and not just phrases, as was the case earlier. The main
difference in this approach is that the system now does not only translate
parts of a sentence piece by piece but instead looks at the sentence as a
whole to figure out the broader context and deliver the most accurate
translation. It also keeps learning continuously as it gets access to more
data, thereby resulting in improved and more natural translations.
Voice assistance: Google takes on Siri (Apple), Cortana (Microsoft), and

Alexa (Amazon) with the launch of its own voice assistance service called
Google Assistant. It essentially takes the features of Google's existing ‘OK
Google’ or ‘Hey Google’ voice controls and fuses them with an AI-powered
system which is designed to deliver a more human-like interactive
experience. Google Assistant is currently available on the Google Pixel
smartphones, Google Home (a smart speaker), and the Google Allo app
(smart messaging). The company plans on launching it on Android Wear
devices via Android Wear 2.0, on Android TV, in a few cars, and on
smartphones running on Marshmallow and Nougat (new Android
operating systems).
114
Besides healthcare, Google’s AI
efforts also include automobiles
Google (3/3)
Healthcare: Google DeepMind has formed a research partnership with the

radiotherapy department at University College London Hospitals (UCLH)
NHS Foundation Trust, a provider of cancer treatment. The main goals of
this partnership are two-fold: to use machine learning to reduce the
amount of time needed for radiotherapy treatment for head and neck
cancers and to develop a radiotherapy-segmentation algorithm that can
potentially be used for other body parts as well. Another partnership with
Moorfields Eye Hospital in the UK is applying AI to spot early signs of visual
degeneration just by using images of eye scans. DeepMind has also
developed an app called Streams, which uses deep learning to detect
early signs of kidney injury in a patient’s blood test data.
Automobiles: Google’s autonomous car division Waymo1 recently began

testing its service to members of the general public in Phoenix, Arizona.
The company is essentially focused on building an AI-powered platform
consisting of sensor hardware (vision cameras, radars, and LIDAR), sensor
fusion software, and image recognition systems. According to Waymo CEO
John Krafcik, the company now has access to 2 million self-driven miles,
which translates into about ‘over 300 years of human driving experience’.
This data is what powers Waymo’s self-driving algorithm, and it is directly
responsible for the company witnessing a 400% drop in disengagement
rates, or the number of times human intervention was needed, between
2015 and 2016.
115
1: Waymo is a subsidiary of Alphabet, Google‘s parent company
IBM’s AI efforts started in the 1950s
with an AI checkers player
IBM (1/2)
International Business Machines Corporation (IBM), incorporated in 1911, is an information

technology (IT) company headquartered in New York. The company has five key operating segments:
Software, Global Business Services (GBS), Global Technology Services (GTS), Systems Hardware, and
Global Financing.
IBM’s AI efforts date back to the 1950s, when it created a checkers player that learned from
experience. The company’s AI initiatives gained prominence with the development of the AI-powered
Deep Blue computer platform, which defeated chess grand master Garry Kasparov in 1997.
Currently, IBM’s AI initiatives are focused on search and planning optimization, machine learning,
deep question answering, knowledge representation, and cognitive architectures.
▪ Year incorporated: Jun 1911


▪ Employees: 366,000 (Dec 2017)
R&D areas
Natural language
processing
IBM Image recognition
116
Watson is one of IBM’s main AI
projects
IBM (2/2)
IBM is currently working on the following areas in terms of AI research and developments:
Watson: a supercomputer that applies NLP, information retrieval,
knowledge representation, automated reasoning, and machine learning
technologies to answer questions. The intelligence platform was originally
developed to answer questions on the quiz show Jeopardy. The Watson
platform can now be used in retail, healthcare, hospitality, entertainment,
aerospace, and other sectors.
▪ Watson’s use in CRM: In March 2017, IBM partnered with Salesforce to

enable it to use data from IBM's Watson AI platform, such as the
weather or local shopping patterns.
▪ Watson’s use in cybersecurity: IBM partnered with eight universities in

January 2017 to provide Watson with the data necessary to expand its
security protocol to be used in the cybersecurity domain. Around 40
companies are now part of Watson’s cybersecurity beta program.
Bluemix (PaaS1) cloud services: Along with Watson, IBM is also working on
machine learning initiatives to expand the Bluemix (PaaS) cloud services.
The company is working towards integrating natural language pattern
recognition, predictive analytics, and image recognition under Bluemix
service offerings.
1: Platform as a Service, in which the service provider rents out platforms such as 117
application servers, databases, and mobile back-end services as a service
Microsoft’s AI efforts include edu-
cation, healthcare, and training
Microsoft (1/2)
Microsoft, headquartered in Redmond, Washington, develops and supports software products,

services, and devices. The company offers a range of server applications, business solution
applications, operating systems, desktop and server management tools, video games, and training
and certification services.
In September 2016, Microsoft announced that it had formed the Microsoft AI and Research Group,
which consists of over 5,000 computer scientists and engineers and includes product engineering,
basic and applied research labs, and New Experiences and Technologies (NExT). Led by Harry Shum,
the division focuses on building AI capabilities across various areas, including agents (Cortana), apps
(Skype, Office 365, and Microsoft Photo), services (vision, speech, and machine analytics), and
infrastructure (AI supercomputer Azur). In December 2016, the company launched an investment
fund, a part of Microsoft Ventures, to make AI-related investments in areas such as education,
healthcare, and training.
▪ Year incorporated: Apr 1975
▪ Revenue: US$110.4 billion (Jun 2018)
▪ Profit before tax: US$36.5 billion (Jun 2018)

▪ Employees: 131,000 (Jun 2018)
R&D areas
Education Facebook Healthcare
Training
118
Various Microsoft AI projects focus
on healthcare applications
Microsoft (2/2)
In healthcare, Microsoft’s AI efforts are concerned with:

▪ Developing machine learning techniques to analyze tumors and design new medication regimes
▪ Developing simulations which detail the progress of cancer in different patients' bodies
▪ Creating programmable biological cells
▪ InnerEye: an AI tool that uses machine vision to give radiation oncologists a 3D view of CT scans of
tumors
▪ Project Hanover: personalizing medicine using AI
▪ Developing a new chatbot with the ability to help people assess their symptoms before going to a
clinic
▪ Healthcare NExT: integrate work from Microsoft's Research and AI units and other industry
participants to reduce data entry tasks, diagnostic procedures, and ease outpatient care
119
Source: bloomberg.com, fortune.com, company information, press releases
In 2016, Salesforce acquired several
companies specialized on AI
Salesforce
Salesforce.com, headquartered in San Francisco, California, is a technology company that primarily

provides cloud-based software solutions on a subscription basis. Its flagship product, on-demand
customer relationship management (CRM) solutions, enables users to connect, engage, sell, service,
and collaborate with their customers. The company, together with its subsidiaries, provides services
to industries spread across the globe, including healthcare, the automotive industry, media, finance,
life sciences, retail, manufacturing, and communications.
Salesforce’s AI platform Einstein integrates its AI capabilities with its CRM tools to make it more useful
for clients to improve their sales pitches. In 2016, the company made significant acquisitions,
including PredictionIO, which specializes in machine learning and predictive analytics, MetaMind,
which specializes in NLP and image recognition, and beyondCore, which develops automated
analytics software.
▪ Year incorporated: 1999
▪ Revenue: US$10.5 billion (Jan 2018)
▪ Profit before tax: US$202.1 million (Jan 2018)
▪ Employees: 29,000 (Jan 2018)
R&D areas
Natural language
processing
Salesforce Image recognition
120
Uber seriously committed to
develop its AI capabilities in 2016
Uber (1/2)
Uber is a technology start-up that connects drivers with people who need a drive through a mobile
application. The company operates in the Americas, Europe, the Middle East, Africa, and Asia-Pacific.
Uber is headquartered in San Francisco, California, whereas its engineering center is based in
Bengaluru, India.
Uber’s acquisition of Geometric Intelligence, a provider of advanced machine learning techniques, in

December 2016 marked the beginning of its serious commitment towards developing its AI
capabilities. The company appointed its then CEO, Gary Marcus, to lead its newly launched AI
research lab in Silicon Valley. The primary aim of the lab is to develop powerful speech- and image-
recognition software which can be trained on less data.
▪ Year incorporated: 2009
▪ Revenue: US$11.1 billion (Dec 2017)1
▪ Profit before tax: n/a
▪ Valuation: US$68.0 billion (Apr 2017) 2
R&D areas
Autonomous driving Uber Fraud detection
Food delivery Mapping research
121
1: CNBC 2: PitchBook
One of Uber’s focus areas regarding
AI research is autonomous driving
Uber (2/2)
Uber’s AI lab is focused on developing advanced algorithms that will enable machine learning with
fewer data requirements and training AI systems using both data and explicit rules.
Below are Uber’s areas of AI focus:
Autonomous driving: Uber’s autonomous car development initiative, led

by the company’s Advanced Technologies Group (ATG), began testing its
service in September 2016 by picking up passengers in Pittsburgh,
Pennsylvania. During the trial period, selected Volvo SUVs were fitted with
sensors (cameras, radars, and LIDARs) and processing units to provide
necessary AI, and a Uber test engineer accompanied the safety driver to
monitor progress with the tests and handle manual driving when needed.
The pilot program expanded to San Francisco, California, in December
2016. However, the company suspended the program in March 2017
following an accident that happened in Arizona. The company is expected
to continue the trials after fixing the errors in the coming months.
Food deliveries: Uber started its food delivery service UberEats as a pilot
program in 2014, which was offered in more than 26 countries in March
2017. The company uses AI to calculate delivery time considering the
traffic, vehicle position, and type of food ordered. In addition, machine
learning is being used to generate restaurant recommendations and to
integrate popular payment options to make transactions quickly and
easily.
Mapping research: The company uses AI for analyzing traffic signs and the
number of traffic signals on a particular road to improve mapping
research, which can be used for predicting the arrival and total travel time
of vehicles.
122
Appendix
123
Glossary
Term Abbreviation Explanation

Artificial intelligence (AI) essentially refers to
computing technologies that are inspired by
Artificial intelligence AI the ways people use their brains and nervous
systems to reason and make decisions, but
they typically operate quite differently.
Medical O2O is the platform connecting those

Medical Online to Offline Medical O2O transforming the medical and wellness world
through cutting-edge technology.
Involves designing new learning algorithms

and improving existing ones to enable
computers to act without being programmed
Machine
Machine learning explicitly. These algorithms allow computers to
learning
analyze large volumes of complex data to
recognize patterns and make predictions and
adjustments.
It is concerned with developing and training
robots to interact with people and the world in
general in predictable ways. However, current
Robotics Robotics efforts also revolve around using deep
learning to train robots to manipulate
situations and act with a certain degree of self-
awareness.
This area is concerned with developing
algorithms that mimic the functioning of the
neocortex area of the human brain, where all
Artificial neural networks
ANN the thinking occurs. This comparison is not
(ANN)
entirely correct because in a human brain,
neurons are not arranged in a linear
sequence, as is the case with ANNs.
124
About the Statista Global Consumer Survey
50+ 46 5,000+ 400,000+

topics & industries countrie int. brands consumers
s
Marketing & Finance & insurance
The Statista Global Consumer Survey offers a global perspective
on consumption and media usage, covering the offline and online social media
world of the consumer. It is designed to help marketers, planners
and product managers understand consumer behavior and
consumer interactions with brands. eCommerce & Health
▪ Cross-tabulation retail
▪ Customized target groups

Food & nutrition Housing & household
▪ Trend and country comparisons equipment
▪ Export in Excel (CSV) or PowerPoint format
Internet & devices Travel
Media & digital media Services & eServices
Mobility Characteristics &

demographics
Find out more on www.statista.com/customercloud/global-consumer-

survey
About the Statista Digital Market Outlook
90 150 7 30,000+
+ +
years (2017- interactive
2023) statistics
The Digital market
Market Outlook presents up-to-date figures on markets of the
regions
digital economy.s The comparable key figures are based on extensive analyses
of relevant indicators from the areas of society, economy, and technology.
What is the size of the eCommerce fashion market in Spain?
How many connected cars are already on the road in China?
The answers to these and many more questions can be found in Statista's
Digital Market Outlook. It provides forecasts, detailed market insights, and key
indicators for the digital economy.
Nine digital verticals: eCommerce, Smart Home, eTravel, Digital

Media, eServices, FinTech, Digital Advertising, Connected Car, eHealth
Direct access & downloads, fully integrated into the Statista database
Market insights, forecasts and key performance indicators
Outlook reports with segment-specific topics (top companies,

trends, deep dives)
Find out more on www.statista.com/outlook/digital-markets 125

About Statista Research & Analysis
Market research – Market Analysis – Data Modeling
Statista Research & Analysis is a provider of comprehensive services in the

fields of market intelligence. Building upon our experience as one of the
world's leading statistics portals, our analyst team can support you in the
collection and evaluation of market, client and competitive information –
tailored to your individual needs. Our team consists of former top tier
management consultants, accomplished market researchers and business
analysts.
Consumer surveys and expert interviews
Market and competitive intelligence
Market sizing and forecasts

Contact us
TEL +49 40 282441 805
E-MAIL [email protected]
Find out more on www.statista-research.com

Authors
Dev Mehta
[email protected]
Dev Mehta has over 10 years of experience working for market

research, legal and consulting companies. Dev worked in various
sectors such as Defense, Digital Marketing, FinTech, Insurance
and Consumer Goods.
Dev completed his Post Graduate Diploma from Massey
University New Zealand, majoring in Business Management and
a Masters in Marketing Management from Middlesex University,
London
Ann-Kristin Hamke
[email protected]
Ann-Kristin Hamke is the VP Strategic Market Insights at Statista

and is in charge of the production of exclusive own Statista
content.
After graduating in Business Mathematics, she worked as a
consultant with the Boston Consulting Group and contributed to
the build-up of the German online fashion retailer About You as
a project manager and by heading the business intelligence
department.
W W W . S T A T I S T A . C O M

Study Id50485 Artificial-Intelligence

Uploaded by

Copyright:

Available Formats

Study Id50485 Artificial-Intelligence

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Study Id50485 Artificial-Intelligence

Uploaded by

Copyright:

Available Formats

What are the main components of the current AI ecosystem?

What are the main components of the current AI ecosystem?

What are some of the major factors driving growth in the AI industry?

What are some of the major factors driving growth in the AI industry?

In-depth: Artificial Intelligence

Questions? U.S.: [email protected] EU: [email protected]

Start-ups: funding and M&A 76

Competitive landscape 102

Competitive landscape (continued)

Current artificial intelligence ecosystem

Reinforcement Recurrent neural

The different types of machine learning are:

Artificial neural networks (ANN)

There are three different types of ANN:

▪ Recurrent neural network

Evolution of artificial intelligence

▪ Cannot use past experiences to influence current decisions

▪ Cannot deviate from assigned tasks

▪ Example: IBM’s Deep Blue

▪ Cannot make significant improvisations based on memories

Partially aware machines

▪ Here, the AI system will be influenced by other entities and

Fully aware machines

▪ Fully self-aware and conscious

1950 1960 1970 1980 1990 2000 2010

Alison Gopnik, cognitive scientist

Automation potential across industries

Manufacturing Finance and Arts, entertainment, Health care and Educational

Automation potential across countries

Rest of the world 50%

Global revenue projection in billion US$

Potential annual GVA1 growth rates in 2035 in %

Baseline state AI steady state

Impact of AI on labor productivity in developed countries in 2035

Impact of AI on industry growth in 2035 in %

Baseline state AI steady state

Information & 3.4

Impact of AI on industry output in 2035 in trillion US$

Baseline state AI steady state

Potential incremental value of AI as compared to other analytics techniques

Oil and gas 79%

Advanced electronics 36%

Illustration of the machine learning process

▪ Supervised learning: specified input

Source: sas.com, McKinsey

Currently, there are the following general types of robots:

There is an exponential pace of improvement in hardware and machine

Uriah Baalke, co-founder of Dispatch, 2016

As human beings, we have contextual understanding and we have empathy,

Examples of companies integrating varying levels of EI into their AI systems include:

Drive.ai: The autonomous car start-up Drive.ai programs vehicles to not

Cogito Corporation: provides EI for healthcare with AI-driven real-time

Nuralogix: has developed technology that claims to be more accurate

Huawei: is looking to develop an emotionally intelligent voice assistance

Beyond Verbal: Founded in 2012, the company aims to develop a

Receptiviti: Founded in 2014, this Toronto-based company has used NLP

BRAIQ: While most technologies related to autonomous cars are

CrowdEmotion: Founded in 2013, this London-based company has

James Kirkpatrick, research scientist at DeepMind, 2017