One intelligence can mask the other

Researchers at DeepMind and those at Meta (formerly Facebook) have claimed new exploits in recent weeks for their AI models. DeepNash, DeepMind’s artificial intelligence, is up to the level of human experts in the game Stratego. Cicero, in his Meta, is among the ten best players in diplomacy. These two games require the combination of two skills that have hitherto been difficult for algorithms to reconcile: strategic planning and decision-making under uncertainty.

Is artificial intelligence close to human intelligence? Taking achieved performance within a given game as a sign of intelligence, even if it is challenging even for humans, is shorthand. Rather than borrow it, mathematician Guillaume Lajoie, principal member of the Mila (Québec Institute for Artificial Intelligence), prefers to focus on the dynamics of the artificial neural networks that make up AI systems, convinced that “there are fundamental mathematical principles that underlie intelligent behavior, observable on a much larger scale.” Symbolic or Boolean “Laws of Intelligence”, in the form of mathematical models, which would apply both to the functioning of the brain and to the functioning of artificial networks. »

Many researchers, at the frontiers of neuroscience and computer science, have set out to find the principles that likely underlie the common computational base of human and artificial intelligence.

They are not the only ones who follow the traces left by thought. Historians Victor Gisemberg and Emmanuel Zing, of the Léon Robin Center at the Sorbonne University in Paris, sought to make a forgotten manuscript speak in the library of Saint Catherine’s Monastery in Egypt. Emerging from the many layers of this palimpsest, mixing centuries of re-inscriptions, is nothing less than the oldest astronomical catalog: that of Hipparchus the Greek.

Exploring intelligences is also a journey through time.