What is different between 'AI' and 'machine learning'? I will explain this in an easy-to-understand manner

by geralt

Although "AI" and "machine learning" are often used in parallel, they are completely different. Software engineer Ben Dickson explains how AI and machine learning differ, and often why they are used at the same time.

Why the difference between AI and machine learning matters | TechTalks
https://bdtechtalks.com/2018/10/08/artificial-intelligence-vs-machine-learning/

I have seen the word "machine learning and improved artificial intelligence (AI) to collect and analyze hundreds of data to improve the user experience of mobile applications" on products and websites that use technology There should be many people. However, Dickson notes that there is great confusion in the field of machine learning and AI. In many cases, people do not know the difference between machine learning and AI, or they intentionally ignore because of marketing.

On the difference between machine learning and AI, Mr. Dickson explains machine learning as "what you can do and what you can not do is clearly defined, what you can define", and AI is "change definition".

First, machine learning is a subset of AI, one way to run AI. Machine learning finds common patterns by comparing and testing enormous amounts of data, and acts based on that rule.

For example, if you give a large amount of X-rays and corresponding symptom data to the machine learning program, the program will be able to analyze X-rays. Because machine learning can compare images and find common patterns. And when you know the pattern, when you give new data, the program will tell you if the data contains symptoms studied in the past.

In machine learning, what we trained in algorithms using human-labeled data is called supervised learning , and what gives unlabeled data and finds patterns in the algorithm itself is called unsupervised learning . Reinforcement learning is also a popular way, giving rules and constraints to machine learning algorithms, and letting the algorithm itself learn "how to reach the best goal". For reinforcement learning it is common to use a reward such as "game score", and the algorithm tries to maximize this reward within constraints. AlphaGo developed by Google's DeepMind is a typical one.

Machine learning, especially its advanced subset of deep learning and neural networks, is very attractive, but it is not magical. Machine learning classifies information based on learned data and predicts the future. There are people who compare neural networks and deep learning with the human brain, but Mr. Dickson said both are very different.

On the other hand, AI is a very broad one. Andrew Moore of Carnegie Mellon University says "AI says science and engineering that creates computer behavior about what was considered until recently," AI needs "human intelligence." Mr. Moore's remark is one of the best things to define AI, but its content is ambiguous. The meaning of the word "until recently" changes with the passage of time. According to Moore's definition, computing machines several decades ago can be thought of as AI at the time, but in modern times there should be few people who think that the computer is AI. Besides this, Zachary C. Lipton who studies computer science at Carnegie Mellon University says about AI "A goal that is changing with time and improving based on the ability" human beings are impossible to machine " I am talking.

"AI" which we imagine is a technology equal to human intelligence level that appears in fiction such as movies and novels, but we still do not know how to make such things. The advanced AI that exists as of 2018 is about the same as the ability of a child of a human being, and it is about tasks specialized in some fields. However, with these technologies, YouTube and Netflix will be able to accurately "recommend" according to the user's preference, and it goes without saying that technology is important to our lives. Through intelligent amplification , we make our lives more productive.

by geralt

In other words, unlike definite machine learning, AI is a "changing goal" and its definition will change with the evolution of technology. The argument that "What is AI and what is not AI" is not easy. Decades later, the latest technology which is considered "AI" as of 2018 should be regarded as an "analog" like a calculator.

There are ups and downs how companies dealt with the word "AI". In the early days of research, many researchers stated that "the same level of AI as humans is coming soon", but in fact it was not developed. Since people began to be disillusioned with the word AI, then the word "AI" came to be avoided. Even when IBM developed Deep Blue , IBM says "Deep Blue is a supercomputer and not AI". In fact, despite technically Deep Blue was AI, it is.

However, after that, the word deep learning and neural network came to be used in many fields in 2012. Deep Learning began performing tasks that were impossible with rule based programming, a dramatic evolution in the fields of voice and face recognition, image classification, natural language processing.

And actually, human brain and neural network are greatly different, but explanation will be made that these two structures are similar, once again the word "AI" which is obsolete will come out again. Scientists and engineers including Earon Mask advocate the AI threat theory that "AI will have a serious impact on humanity in the future", and there is a possibility that robots may deprive the employment of human beings did.

by geralt

With these elements, AI's hype is lit, and many companies are trying to profit by using the word "AI" in an ambiguous way of use. The word AI used in many advertisements does not refer to a specific technology, it just adds a mysterious atmosphere to the product. However, as in the last time, if AI does not meet the expectations of people, there is also a possibility that "AI to winter" will come again that will avoid the word AI again.