tribune. As applications of artificial intelligence (AI) proliferate on a large scale after decades of stagnation, its ability to produce intelligence equivalent to ours is called into question, and the very term of intelligence is denied to it. The major players in the discipline confirm that we are still far from the goal. Beyond the current euphoria, would AI therefore be at an impasse?
All RNs have the same shortcomings: hyperspecialization, lack of autonomy, inability to manipulate meaning and create a representation of the world. The fact that these limits have been manifested since the work of the pioneers in the 1960s gives the feeling that the discipline is going in circles. Like an enchanted horizon fleeing from researchers, the Grail of universal intelligence escapes them every time they think they are reaching it.
Despite advances in cognitive science, intelligence remains a mystery and, strangely, there is no basic research dedicated to the development of a theory of intelligence. The fact that AI benefits from hundreds of billions of investments in the absence of any theory of intelligence is questionable. There is not even a precise definition of intelligence and for good reason: a definition is a summary …
However, the amount of knowledge accumulated by the cognitive sciences is considerable. Why would a synthesis not release principles useful to AI? The absence of a theoretical foundation can also explain why by focusing on specialized skills, AI experiences successes that it is unable to generalize. In the absence of a theoretical framework, the researchers progress without precise direction, they approach each new technique in the hope that it will lead them a little further. It works sometimes, but decades are wasted investigating dead ends.
Researchers are like ants exploring the small strip of land delimited by technology, without the possibility of concentrating on the points of rupture proposed by a theory. They hope to one day reach the top of the hill which will reveal to them the path to true intelligence, but the risk is great that this will happen only after having explored all of the conceptual dead ends which surround them. There can be no slower and more expensive method and once at the top of the hill, it is not certain that they realize it for lack of conceptual benchmarks.