Abstract
The performance of artificial intelligence systems, particularly large language models (LLMs) - the best-known, but not the only one, being ChatGPT - is astounding. While there is widespread agreement that they are not yet truly intelligent, many users and specialists are convinced that the progress they are making heralds the imminent arrival of fully intelligent systems, endowed with what is known as "artificial general intelligence". But there are skeptics, who believe that the progress we can expect from available techniques is limited: the impression that the goal is close at hand would be a mirage. For my part, I would argue that artificial intelligence systems by their very nature lack an essential dimension of human intelligence: they only seem to approach it through a false connection.