Transcript slides
What Do You Mean by “AI”? Pei Wang Temple University Philadelphia, USA What this is about My paper is about the definition of AI … Oh! Not again! We all have been bored to death by this kind of philosophical nonsense … Why not to talk about NARS? Why to raise this issue … because it is an inevitable question for every AI/AGI researcher. A working definition of AI establishes the goal of the research, as well as basic assumptions and evaluation criteria. Nobody can really do AI research without a working definition of “AI”. Why we should care Many problems and debates in traditional AI research come from confusions and misunderstandings in research goals. What distinguishes AGI from the current mainstream “AI” is primarily in research goals. Many differences among exiting AGI approaches origin in their concrete goals. What we do agree The best example of “intelligence” is human intelligence, which comes from the human mind. AI should be similar to the human mind, in some sense. AI cannot be identical to the human mind in all senses — AI does not aim at “artificial person”. What we don’t agree In which sense AI is similar (or even identical) to the human mind? Structure: building brain models Behavior: passing Turing test Capability: solving practical problems Function: having cognitive faculties Principle: reaching rationality/optimality What the difference is Each type of working definition sets a valid research goal, and the research produces results of theoretical and practical values. However, these goals and values are different, and do not equivalent of each other (though they are related). They are not different trails to the same submit, or different parts of the same whole. Which is the “correct” picture Why not all together The aspects of human mind become separated when reproduced in computer. We cannot achieve all of them with the same efficiency and easiness. We should not rule out other forms of intelligence. What I’m not suggesting To look for a perfect working definition of AI at the current time — our understanding of intelligence will surely get deeper. To debate until we all agree on a working definition of AI — that will not happen soon. To treat every working definition as equally good — their long-term values are different. What I’m suggesting Select your research goal carefully. Make your working definition clear when using “AI”. Acknowledge alternative working definitions when evaluating other projects. AGI treats “intelligence” as a whole. We’ll keep the diversity of the field, while encourage confrontation of ideas.