Should we model AI learning on children?

Adopt developmental psychology in the development of this technology

There are more and more virtual assistants that can answer questions. AIs that chat can be found for free online. Yet, you only have to start interacting with them to see the limitations. They don't have the concept of meaning, which sometimes makes exchanges difficult. For Jean-Christophe Baillie, AI researcher, the issue of semiotics could potentially be learned by algorithms but by an approach similar to that of learning language in children.

In fact, if most scientists have tried to give an adult mind to artificial intelligences, Alan Turing submitted already in 1950 the idea of designing instead AIs with the mind of a child. This one could develop and with the right curriculum become more mature. For example, the question of language is interesting. Developmental psychology has shown that the child learns terms by experimenting with the physical and social world around him. By exchanging, he acquires the meaning of colors, objects, and even intangible concepts.

This could be repeated with AIs. Many researchers are trying to do this with robots but they are big money investments with risks of mechanical disorders. So an alternative solution would be virtual reality where the algorithm learns by manipulating elements and gradually develops meaning.

Time: 34min43