This article discusses the differences between artificial intelligence, machine learning, and deep learning. [1] Artificial intelligence is the broad field of creating machines that emulate human intelligence through capabilities like reasoning and problem solving. [2] Machine learning uses statistical algorithms to find patterns in large data sets and make predictions without being explicitly programmed. [3] Deep learning is a subset of machine learning that uses artificial neural networks to automatically extract features from data for tasks like speech recognition. While related, AI, ML, and DL each have distinct definitions and applications.
This article discusses the differences between artificial intelligence, machine learning, and deep learning. [1] Artificial intelligence is the broad field of creating machines that emulate human intelligence through capabilities like reasoning and problem solving. [2] Machine learning uses statistical algorithms to find patterns in large data sets and make predictions without being explicitly programmed. [3] Deep learning is a subset of machine learning that uses artificial neural networks to automatically extract features from data for tasks like speech recognition. While related, AI, ML, and DL each have distinct definitions and applications.
This article discusses the differences between artificial intelligence, machine learning, and deep learning. [1] Artificial intelligence is the broad field of creating machines that emulate human intelligence through capabilities like reasoning and problem solving. [2] Machine learning uses statistical algorithms to find patterns in large data sets and make predictions without being explicitly programmed. [3] Deep learning is a subset of machine learning that uses artificial neural networks to automatically extract features from data for tasks like speech recognition. While related, AI, ML, and DL each have distinct definitions and applications.
This article discusses the differences between artificial intelligence, machine learning, and deep learning. [1] Artificial intelligence is the broad field of creating machines that emulate human intelligence through capabilities like reasoning and problem solving. [2] Machine learning uses statistical algorithms to find patterns in large data sets and make predictions without being explicitly programmed. [3] Deep learning is a subset of machine learning that uses artificial neural networks to automatically extract features from data for tasks like speech recognition. While related, AI, ML, and DL each have distinct definitions and applications.
Artificial Intelligence, Machine Learning, and Deep
Learning: Are They All the Same? Sai Mannam
Artificial Intelligence (AI) is a field of computer science and
computer systems that emphasizes frameworks to perform tasks that conventionally are perceived as requiring human cognition and intelligence. It is an industry that has been progressing and integrating into our daily lives through the technologies we use. With recent scientific developments, Machine Learning (ML) and Deep Learning (DL) have been prominent names associated with AI. While there is an over- lap among all of them, they are not the same.
WHAT IS ARTIFICIAL INTELLIGENCE?
As the name implies, the goal of Artificial Intelligence is to create machines that emulate human intelligence “artificial- ly.” To be able to understand what this means, we must first example, there are autonomous robots that function inde- learn what constitutes intelligence within a human. Capabili- pendently from human intervention like the iRobot vacuum ties that we often take for granted, including our reasoning cleaners and there are augmented robots that enhance hu- abilities, linguistic competence, problem solving, visual and man capabilities like robotic prosthetic limbs. auditory perceptions, and how we learn, are broad catego- rizations that constitute our intelligence. These in fact are WHAT IS MACHINE LEARNING? quite difficult to simulate in artificial systems owing to their Like the subsets of AI discussed above, Machine Learning complex computational nature and limitations in transferabil- (ML) is just another subset of AI. Thus, all of ML can be con- ity. sidered AI, but not all AI can be considered ML. Put simply, There are many areas of research within AI that have Machine Learning uses statistical algorithms to understand had practical applications for the general public. Natural large quantities of data and finds patterns that we can use to Language Processing (NLP) is a subset of AI that builds gain insight on the data as a whole. Any digital information computational tools for analysis and modeling of human text, can constitute as data and therefore can be run through a such as machine translation, summarization, and question- machine learning algorithm to obtain these insights. Many answering. NLP is prominent within our daily utilities, includ- of the products we use today have some form of machine ing email filters, search results, and predictive text. Machine learning pipeline implemented into their workflow, including Vision is another subset of AI that aims to simulate the vi- Netflix, YouTube, Spotify, Facebooks, and even voice assis- sual perception component of human intelligence. Through tants like Siri and Alexa. The unique aspect of ML is our abil- a combination of hardware and software tools, machine vi- ity to teach the machine how to learn without explicit, pro- sion aims to analyze images and provide predictive insights grammed instructions. The data is what the machine learns for human guidance. A popular application of machine vi- from. sion is for autonomous vehicles: self-driving cars use mul- Data is important for learning; thus, the quality and quan- tiple cameras, lidar, radar, and ultrasonic sensors to process tity of data both play an integral role in the predictive insights the visual field around the car in order to make decisions that can be extracted. We can train the algorithm by using on driving. Another major application of machine vision is a subset of our data, which we will call the “training set”. in healthcare. From biomedical image analysis to surgical Once the algorithm goes through a set number of iterations guidance, machine vision enables healthcare providers to through the training set, we can use the rest of the data not detect illness earlier and create preventative measures for shown to the machine as a “testing set.” This is analogous to optimizing health outcomes. A subset of AI that sees over- a student in a classroom learning a new subject. The teacher lap with machine vision is robotics. Although we perceive teaches them, the student absorbs the information and tries the field of robotics as encompassing humanoid robots like making connections as the material is presented to them, Sophia, there are many other subfields within robotics. For and then they are tested on their comprehension of the sub- Journal of Young Investigators Science News
ject through an exam. Likewise, the algorithm going through
the training set is like that student learning the material in class and the testing set is like the exam. there are multiple different methods within deep learning) to The main types of ML that we will go through are Super- extract features from the data shown to it. The above section vised Learning, Unsupervised Learning and Reinforcement discussed that in machine learning, we have to tell the ma- Learning. In Supervised Learning, the data is labeled to tell chine the specific features in which various parts of the data the machine the exact patterns it needs to look out for. Sup- can be separated. The neural network, on the hand, runs pose you have a classification problem. There is a dataset through the data and automatically extracts the features with images of apples labeled as “apples” and oranges la- that best represent the data. The beauty in deep learning is beled as “oranges” and the task of the algorithm is to clas- that as we advance as a civilization and have more access sify the image associated with apples and oranges correctly. to data, the performance of deep learning models also in- This is a Supervised Learning example because the training creases. Traditional Machine Learning usually plateaus after data already consists of input vectors and their correspond- a certain quantity of data and its performance flattens. This ing targets. In Unsupervised Learning, there are no labels has been a major reason for why Deep Learning has be- on the data and the algorithm must draw insights without come popular in recent times, where technology has driven this added crutch. Suppose now the data contains images of improvements in the data we acquire and store. apples, oranges, and bananas without any labels on them. An application of Deep Learning is Speech Recognition, This would likely be a clustering problem where the algo- something that is out of the scope for ML. Additionally, DL rithm parses the data and finds patterns among the images has played an integral role in Self-Driving Cars. In addition with different fruits and subsequently groups them together. to using ML principles, as these cars obtain more data and Reinforcement Learning is based on the maximization of re- tracking measures about driving behavior on the road, they ward as the algorithm learns what to do and how to map situ- improve and become more efficient as autonomous agents. ations to actions. In this form of learning, the machine learns For that reason, the reason for software updates for Tesla through trial and error. The agent is subsequently rewarded vehicles is so that the model gets trained with the extra data or penalized based on the answer. Thus, as the model gains collected from Teslas driving on the road to improve overall more reward points, it trains itself in the correct path to fol- performance. low. AlphaGo, a program created to compete against play- IMPLICATIONS AND NEXT STEPS ers in the game Go, used reinforcement learning to beat the best human competitors. As Dong et al. 2020 writes in their Review of Human Intel- ligence and AI, “AI, which simulates human operational intel- WHAT IS DEEP LEARNING? ligence, is far superior to human beings in computing speed, Deep Learning (DL) is a subset of Machine Learning. It is a capacity, and accuracy” which will allow AI to “help humans method that relies heavily on data representations in con- in more fields and more profoundly.” Although AI, ML, and trast to task-specific algorithms. As a result, the computers DL are still in their budding stages of conception, there are learn from experience and understand the world in terms of many advancements owing to progress within these fields a hierarchy of concepts. Inspired by the human brain, deep that have helped humans lead better lives. Enabling more learning mainly utilizes artificial neural networks (though people to understand the conceptual frameworks around Journal of Young Investigators Science News
these principles will allow diverse perspectives to approach
the field and thus become better utilized by more people. My hopes in this lecture series are to at least introduce or make the concepts within AI less daunting and more approach- able.
