IP Issues in the Era of Artificial Intelligence

Prof. Joseph Straus, Munich

3rd EU China IP Academic Forum

Shanghai, November 22, 2018

© J. Straus 2018 1
 Points to Consider

• The Fourth Industrial Revolution

• Artificial Intelligence (AI) – Evolutionary Technology

• Investments in AI Technologies

• AI "Products"

• AI – Challenges for standard patent law concepts

• Lessons to Learn

© J. Straus 2018 2
 Navigating the Next Industrial Revolution

Source: World Economic Forum

© J. Straus 2018 3
 The Fourth Industrial Revolution

• The advent of cyber-physical systems – involving entirely new

capabilities for people and machines
[N. Davis, WEF, ]

• The marriage of physical and advanced digital technologies,

such as
– analytics, artificial intelligence, cloud computing and the internet of
things
[Forbes Insights with Deloitte]

© J. Straus 2018 4
 Artificial Intelligence (AI)

• "The science and engineering of making intelligent machines"

[John McCarthy, 1955]

• A machine's adaptation of cognitive functions that are

associated with the human mind – such as understanding of
language, problem solving, and learning
[https://www.techpats.com/artificial-intelligence-potential-implications-patents/]

• In the sense of patent law: Mathematical algorithms allowing

computers to simulate intelligent human behavior
European Patent Office

© J. Straus 2018 5
 Artificial Narrow Intelligence (ANI)

• Specialized in a specific area – e.g. IBM's Deep Blue® Super Computer (Chess), or China's
Tianhe-II – 34 quadrillion calculations per second!
• Can solve complex problems extremely fast – but have no perception of things other than the
information provided to them by the creators
• Examples: Intelligent thermometers (Nest®), Apple's "Sirr"®, video games, search engines, etc.
• Cannot imitate thought process outside the scope of their pre-determined operation
• Efforts on the way on creating an AI processing information with software based on human
physical, biological and chemical thought process; electronic neural networks, cognitive
computing algorithm and artificial neo-cortex through software.

[Gurkaynak, Yilmaz, Haksever, 2016]

© J. Straus 2018 6
 Artificial General Intelligence (AGI)

• Represents "Human-level AIs", computers as smart as humans – in every aspect

and capable of performing all intellectual tasks humans can

• Performing tasks involving complex calculations requiring substantial effort,

time dedication for humans – very simple for AIs

• AI by now succeeded in doing essentially everything that requires "thinking"

but has failed to do most of what people and animals do "without thinking"

• May be reached around 2030

[Gurkaynak, Yilmaz, Haksever, 2016]

© J. Straus 2018 7
 Artificial Super Intelligence (ASI)

• Represents AIs "much smarter than the best human brains in

practically every field, including scientific creativity, general
wisdom and social skills"
• It is expected that AGI once established it will evolve itself into an
ASI very quickly – as a result of an exponential growth
(phenomenon of "intelligence explosion" or "singularity)
• ASI major forms: speed super intelligence, collective super
intelligence and quality super intelligence – anyone of the three
capable of creating the other two
[Gurkaynak, Yilmaz, Haksever, 2016]

© J. Straus 2018 8
 Artificial Intelligence Explosion

Hutson, Science 18 May 2008

© J. Straus 2018 9
 Some Technical Foundation of AI

• Tools that can process vast amounts of data, detect and interpret
patterns – previously impossible to calculate, identify or even imagine

• They enable machine prediction, diagnoses, modeling and risk analysis

• AI – an essential element enabling effective use of large data volumes –

not manageable manually – and algorithms no longer efficiently
reprogrammed by hand

© J. Straus 2018 10
 Machine Learning – An Important AI Tool

• A method of data analysis that automates analytical model building

• Uses algorithms that iteratively learn from data

• Allows computer to find hidden insights without being explicitly

programmed where to look

• Google's "AlphaGo" – masters Go – "Deep learning is killing every problem

in AI"
[J. Schaefer, 2016]

© J. Straus 2018 11
 Investments in AI Technologies

• China – US $ 2.1 billion on an AI industrial park

• EU - € 1.5 billion to AI research through 2020

• France - € 1.5 billion to AI research through 2022

• US Government – hesitant to engage

© J. Straus 2018 12
 Q4 2017 TOP 10 Deals of VC Investments Split
Between China and US in AI

© J. Straus 2018 13
 Comparison United States - China

United States China

Years experience of the More than half have more than Forty percent have less
nation’s data scientists 10 years. than 5 years.

AI patent applications,
15,317 (First in world) 8410 (Second)
2010–2014

Number of workers in AI
850,000 (First) 50,000 (Seventh)
positions

Percent of private AI
66% (First) 17% (Second)
investment (2016)

[Data: Astamuse; Linkedin; Mckinsey Global Institute]

© J. Straus 2018 14
 Comparison United States - China

• "For traditional scientific fields, Chinese [scientists] have a long way

to go to compete with the U.S. or Europe. But for computer science,
it's a relatively new thing. Young people can compete. Chinese can
compete."

• "China played no role in launching the AI revolution, but is making

breathtaking progress catching up."

[Eric Lander, President of the Broad Institute in Cambridge, Mass.]

© J. Straus 2018 15
 AI-Powered Drug Discovery Captures Pharma Interest

Source: Nature Biotechnology, Vol. 35, No. 7, July 2017

© J. Straus 2018 16
 Machine Learning Classifies Cancer

Tumour classification using a machine-learning

approach. Capper et al.1 used a machine-
learning approach to classify brain tumours on
the basis of genome-wide patterns of a type of
DNA alteration called methylation. The
computer was trained using methylation data
for tumour samples that had been diagnosed
by pathologists using standard microscopy-
based analysis or analysis of selected genes.
After training, the computer was given 1,104
test cases. The authors compared the
diagnoses made by the computer and by the
pathologists. Although the machine was
unable to diagnose all specimens, of the
specimens that it classified, the machine-
based diagnosis was more accurate or could
assign tumours to more-specific subcategories
than the classifications made by the
pathologists.

Source: Nature, 22 March 2018

© J. Straus 2018 17
 AI Designs Organic Synthesis
- A System in Which an AI Program Learns the Rules for Itself -

Retrosynthetic Analysis

Source: Nature, 29 March 2018

© J. Straus 2018 18
 Multibillion Investments in AI & Resulting Products and Processes
- Require Efficient Patent Protection –

Patent Eligible

• EPO, JPO, SIPO & USPTO – by and large – apply same standards as for
"traditional" computer implemented inventions
• Patents available for inventions where computer programs [i.e. algorithm-
related inventions] make a technical contribution, e.g. in the fields of:
• medical devices; the automotive sector; industrial control;
communication/media technology; automated natural language
translation; voice recognition and video compression;
computers/processes itself

© J. Straus 2018 19
 US Patent No. US 8,126,832 B2 of February 28, 2012 for
"Artificial Intelligence System" of Cognitive Code Corp.

"The present invention extracts concepts from text and/or speech and utilizes
numeric representations of concepts and their relationships. The extraction
of concepts allows expression in various patterns to be understood by the
system. The processes allow the system to bypass human language
constraints in order to think in concepts. The present system may also
dynamically construct output so that it can generate intelligent responses in a
number of grammatically correct ways. Because the system utilizes numeric
representations of concepts and their relationships, the system is language
independent and may even function with a plurality of languages."

© J. Straus 2018 20
 US Patent No. US 8,126,832 B2 of February 28, 2012 for
"Artificial Intelligence System" of Cognitive Code Corp.

© J. Straus 2018 21
 US Patent No. US 8,126,832 B2 of February 28, 2012 for
"Artificial Intelligence System" of Cognitive Code Corp.

© J. Straus 2018 22
 US Patent No. US 8,126,832 B2 of February 28, 2012 for
"Artificial Intelligence System" of Cognitive Code Corp.

© J. Straus 2018 23
 AI-Related Patent Publications

© J. Straus 2018 24
 China is Catching Up With the US in AI

© J. Straus 2018 25
 Development of Families with Applications in AI

800
Number of families

700

600

500

400

300

200

100

Filing Year

Source: EPO

© J. Straus 2018 26
 US-AI Patent Champions

• IBM: 2012-2017 – 5.600 AI patents (in 2017 more than 1.400)

• IBM's Watson - a cloud based AI product – provides Application Program

Interfaces (APIs) that can understand all forms of data to reveal business
critical insights "harnessing" the power of cognitive computing –
organized into various products for building cognitive search

• Google: 2012-2017 - ~ 4.500 AI patents

• Google's neural machine translation system developed its own internal

language to represent the concepts it uses to translate other languages

© J. Straus 2018 27
 AI Challenges of Traditional Patent Law Concepts

• Currently applied approaches overlook/ignore [?] that AI technology

itself creates its own technology & applications, i.e. inventions –
solutions of technical problems with technical means
• This requires, inter alia, new assessments of:
– the notion "invention" [identifying a problem, then solved by AI?]
– [non]patentability of mathematical algorithms – the "self-evolving" core of AI
– who is the inventor – can a legal person be an inventor?
– the notion of the “person” skilled in the art – decisive for assessing inventive
step

© J. Straus 2018 28
 Challenges for the IP System Digital Inventor?

Source: Siemens AG 2018

© J. Straus 2018 29
 Deep Dream Image Example

Works of Art?

© J. Straus 2018 30
 Inventorship in the context of AI

• Does the inventor have to be a natural person?

• Although not explicitly mentioned in EPC, the inventor can only be a

natural person according to the current legal interpretation (Articles 60,
62 and 81 EPC)

The inventor is the one who is the author of the claimed invention, i.e.
who has recognised the idea of the invention and has developed it into
a technical instruction in creative activity. Only a human being is
capable of a creative activity in this sense. (Benkard EPÜ, 2. edition,
Article 60, points 9-10)

Source: EPO

© J. Straus 2018 31
 Inventions generated by AI

• In the context of machine inventions, various levels of human

participations are possible

• Although the technical solution can be developed to a large extent by a

machine, the human being is currently not completely eliminated from
the process

• The human being triggering the process of inventing through creative

ideas and/or combining the results delivered by a machine remains the
inventor

Source: EPO

© J. Straus 2018 32
 Could AI be designated as inventor?

• Current situation

– if the applicant designates AI (e.g. a computer) as the inventor, the

requirements of Article 81 and Rule 19 EPC are not met

– if no valid designation of the inventor is filed, the application will be

refused (Article 90(5) EPC)

Source: EPO

© J. Straus 2018 33
 Designation of AI as inventor

• Should inventions be generated solely by AI, it could become necessary

to clarify whether AI could be considered as the inventor

• The role of the inventor and the interpretation of the term inventor would
have to be adapted

• Also the right to a patent in the sense of Article 60(1) EPC would have to
be clarified by the legislator

Source: EPO

© J. Straus 2018 34
 Effect of AI on the skilled person

• If the use of particular creative software becomes common in a certain

field of technology, it may be presumed that the skilled person would
use such software

• Inventions involving AI or machine learning may therefore raise the

skills and knowledge of the skilled person

• This would have an impact on the afore-mentioned requirements of

disclosure and inventive step

Source: EPO

© J. Straus 2018 35
 The UK/Irish Copyright Solution – A Viable Model?

• The copyright in "works generated by a computer in

circumstances such that there is no human author"
• Vested in "the person by whom the arrangement necessary for
the creation of the work are undertaken"
[Copyright, Designs and Patents Act, 1988, c48, § 178 (UK)
Copyright and Related Rights Act, 2000, part 5, § 2 Art. No. 28/2000 (IV)]

• Computer-generated works – not the works of an AI

• Is "non-human" copyright ownership possible? If so – to whom it
should belong?

© J. Straus 2018 36
 Lessons to Learn

• Over centuries the IP-system – by adapting itself to ever evolving challenges

of technological and scientific developments – also those of the 1 st, 2nd and 3rd
industrial revolution – have served the public's well-being well

• The IP-system should develop further – to meet the needs of the 4 th industrial
revolution in general and that of AI specifically – internationally harmonized &
rational & consistent approach necessary!!

• Those challenges may/will have an entirely new quality – thereby the evolving
AI technology will help humans in solving problems beyond known paradigms

© J. Straus 2018 37
 Thank you
for your attention!

谢谢大家听我的发言！

© J. Straus 2018 38