Dimiter Dobrev, Institute of Mathematics and Informatics, Bulgarian Academy of Sciences, Bulgaria (in person)
The modern definition of AI contains an inaccuracy. According to the definition we have nowadays, AI is a computer program which is successful. Indeed, for a computer program to be successful, it must be intelligent, but the opposite is not true. A program can be intelligent but not successful – merely because it pursues different goals and does not aim at the success in question. From a theoretical perspective, the modern definition of AI is good enough because it answers the question “What is AI?” even though it does not encompass all intelligent programs, but only some of them. From a practical standpoint, however, this definition is insufficient. The reason is that we are at the doorstep of creating True AI and among all intelligent programs we must choose the one we will be most comfortable with from now on. Thus, it is not a good idea to choose one of these successful programs. It would be better to choose a program that does not pursue victory at any cost. Such a program could be called a loser because it will not be successful enough. After all, both in humans and in AI relentless ambition is not a positive trait.
Here is the full text.
These are the slides.
You can also listen to Dobrev’s interview on Darik Radio, where he talks about the conference. The interview is in Bulgarian, but you can turn on English subtitles.