show index hide index
Artificial intelligence (AI), in its incessant quest to understand the world, could well take Plato’s Socratic path, based on exchange and questioning. A Google DeepMind engineer recently proposed a bold approach called “Socratic learning” where the AI agent would improve autonomously through dialogues formed of questions and critiques, distinct from sensory data. Drawing inspiration from fundamental philosophical principles, this method envisages building an artificial superintelligence capable of tackling complex problems such as the unification of the theories of quantum mechanics and general relativity, thus promoting unparalleled theoretical innovation.
The artificial intelligence (AI) of the future could it be defined by a perpetual intellectual exchange, evoking the dialogues of Plato? A Google DeepMind engineer, Tom Schaul, proposes a revolutionary development of artificial superintelligence based on the Socratic method. Through this theoretical hypothesis, the envisaged AI would be capable of acquiring knowledge autonomously through language, without recourse to external sensory data, and of solving complex, unprecedented problems, thanks to this infinite dialogue.
Plato’s imprint on modern AI philosophy
Alfred North Whitehead said that “European philosophy is a series of footnotes to the work of Plato”. The question then arises: tomorrow, will all research in artificial intelligence be a continuation of the same Platonic spirit, or will it move towards new philosophies and approaches? The work of Plato, with its philosophical dialogues, seems to inaugurate a form of reflection which could inspire modern approaches to AI, in particular with this notion of perpetual exchange of questions and answers, characteristic of the Socratic method.
Socratic learning: a model for AI
Tom Schaul’s article, entitled “Boundless Socratic Learning with Language Games”, presents a bold idea: exploiting language as the sole learning vector for artificial intelligence. Unlike researchers such as Yann Le Cun, which support the use of external sensory data for understanding the world, Schaul argues that natural language could be sufficient to create self-improving AI capable of great intellectual innovation. This concept places language at the heart of development and underlines the infinite richness of dialogue for the enrichment of artificial intelligence systems.
A conceptual proposal in search of validation
Schaul’s vision remains until now a theoretical proposition, because it is based on a principle not yet verified. The idea is that the AI agent could improve in isolation, by producing and critiquing its own natural language, without integrating information from the outside world. The Socratic dimension involves perpetual questioning, a constant exchange aimed at refining and generating new knowledge, without recourse to direct physical experimentation. Only purely verbal interaction would allow access to these new conceptions.
Solving theoretical puzzles through language
Among the most fascinating implications of extended Socratic learning, Schaul raises the possibility that AI could potentially solve major, previously inaccessible, theoretical problems. These advances would include the demonstration ofRiemann hypothesis in mathematics or the unification of theories relating to quantum mechanics and general relativity in physics. By subtly balancing Socratic dialogue with scientific rigor, AI could redefine fundamental paradigms in various fields of knowledge.
