It is time to study AI in "new field"

byAlex Knight

It has been pointed out that artificial intelligence (AI), which has been studied mainly by computer scientists and robot engineers, should be studied by opening up a completely new fieldMIT Media LabManuel Cebrian and Mr. Iyad Rahwan. Why are the existing research methods limited, how can the future be changed by a new research field, two researchersNautilusIt explains in.

It's Time to Study AI on Its Own Terms
http://nautil.us/issue/58/self/machine-behavior-needs-to-be-an-academic-discipline

As of 2018, mainly computer scientists and robot researchers are researching and developing AI's behavior. They concentrate on only the functions that they intended, and they are working on development so that AI can act efficiently. The automatic driving car moves unattended from point A to point B in any weather, game competition AI wins human beings, and the mail classification program accurately determines whether it is spam or spam. Those who study the behavior of AI about such specific tasks are very useful in the field of AI and robot engineering, but Manuel Cebrian of MIT Media Lab said that it alone is not enough to adapt to society "Iyad Rahwan says.

This is "If only a physiologist studies human behavior" "Given that only cranial neuroscience studies criminal behavior and is responsible for designing educational curriculum," the problem will be in danger.

He won the Nobel economics award in 1978Herbert SimonIs "Science of systems"among"natural ScienceIs knowledge about natural phenomena and natural phenomena. We need to ask whether "artificial" science, that is, artificial or artificial phenomena can be studied. " According to Simon's way of thinking, even if it is the same "intelligence", since abstract concepts are fundamentally different between humans and machines, from the fundamental level, we will science AI's "behavior" scientifically It is necessary to study. And at this time, human behavior is physiology, sociology, psychology, political science ·Game theory·MacroeconomicsAs should be explained by various scales, such as AI's behavior should also be studied from various aspects.

byVlad Tchompalov

However, there are a couple of points to note when studying the behavior of AI as a completely new field. First of all, you should not think that "AI has acting subjectivity". When the dog blows someone, its responsibility is not the dog but the owner, but still studying the behavior of the dog is helpful in "predicting behavior." Similarly to AI, we should assume that AI's responsibility for action is on the side of the person dealing.

Also, since the behavior of the machine is fundamentally different from humans and animals, care must be taken not to anthropomorphize it or make it an animal model. Even when it is useful to use research methods of human and animal behavior, it must be kept in mind that machines exhibit qualitatively different intelligence patterns and behavior patterns from living things.

Apart from these precautions, Mr. Cebrian asserts that "new learning related to machine behavior" is very important and useful. Because it is impossible for a researcher to grasp the behavior of a machine simply by looking at the source code. The latest AI architecture including the deep neural network is complicated and more than "to judge whether it is a good person or a bad guy by looking at the human brain scan image", "AI is optimal just by looking at the code It is difficult to judge whether it is ethical or not. " Also, what is said to be "good" in society continues to change, social scientists can track the change, but the AI ​​community may not be able to do the same.

And AI takes essentially unpredictable behavior and even a programmer who made it can not accurately predict. Computer scientist'sAlan TuringYaAlonzo ChurchEven showed the view that whether the algorithm has a certain property is unknown until actually trying to move the algorithm. In principle there is a limit to discriminating whether or not the code has a safe and desirable trait by "just looking at the code".

The greatest point of studying the behavior of machines from the framework of computer science and robotics engineering is that new aspects such as economics, sociology and political phenomena can be added to the field of AI. People studying computer science and robotics can be said to be the most talented people on the planet, but on the one hand rumors spread in ethnic discrimination and moral dilemmas, stock market stability, social networks We do not receive training in specialized areas such as phenomena. Of course Mr. Cebrian studies about various phenomena phenomena, but in fact when working with experts in each field, we can see how inappropriate and wrong the things they were doing. From these experiences, Cebrian said that they slowed down the way to make a machine learning model based on big data.

byMartin Shreder

It is possible that an algorithm may harm individuals and communities if "together with" the fact that accurate prediction is impossible in principle "and" the possibility of human overlooking "are combined. Also, it is problematic that the person who influenced the basis of algorithm decision making is uncertain and the AI ​​system is black boxed. However, the point that AI has a potential possibility of adversely affecting people and society tends to be rebounded in the AI ​​community.

As of 2018, the content being discussed about AI is based on evidence of poor evidence gathered by experts. This pointed out that mathematician Cathy O'Neal "lacks the obvious field of academic research that understands and criticizes the role of algorithms that make a lot of decisions in technology, especially in our lives." AI's Social scientists and computer scientists need to cooperate seamlessly in order to conduct an integrated and measurable scientific approach on behavior.

In addition, Mr. Cebrian believes that substantial economic value will be born if studying the behavior of machines can prove that a particular algorithm meets ethical, cultural and economic standards. Just as consumers seek ethical and environmental certification for what they consume, if AI is given objective "authentication", it is possible to put an algorithm on the market.

In the near future many of the software and hardware should be driven by AI. Although it is unknown whether it is an anthropomorphic form like an SF movie, human beings need to understand AI, which plays a new role, and control the influence on humanity. Mr. Cebrian insisted that we need to open up a new scientific field and study more deeply about the behavior of machines in order to build a healthy and interrelated relationship between humans and AI.

byJehyun Sung