In a world where artificial intelligence is becoming increasingly central, a new experiment recently astonished researchers. The AI ChatGPT was confronted with an age-old challenge, a mathematical riddle from Plato’s dialogue Meno, and its result surprised more than one expert. Alternating between a display of clumsiness and a burst of creativity, ChatGPT revealed an unexpected facet, where error could well be the breeding ground for deeper insight. A 2,400-Year-Old Riddle In a daring experiment, researchers confronted ChatGPT with a 2,400-year-old mathematical problem. This situation highlighted the unexpected capabilities of AI, demonstrating that a simple refusal to answer can sometimes be more informative than a correct answer. This challenge is not just about numbers, but raises fascinating questions about machine reasoning and learning. Ancient HistoryTo appreciate the significance of this test, we must go back to Antiquity, where Plato, in his dialogue Meno, presents a challenge whose subtlety remains both striking and educational. Socrates asks a student to double the area of a square. The obvious answer, to double the length of the sides, leads the student down the wrong path. The true solution lies in understanding the diagonal, an aspect that, to this day, prompts us to reflect on our approach to knowledge. ChatGPT Put to the Test ChatGPT’s modern take on this challenge was orchestrated by researchers from Cambridge and the Hebrew University of Jerusalem. Imitating the Platonic scenario, the AI was tasked with solving geometric problems. When faced with the square, it brilliantly reached the correct conclusion, but when it came to doubling the area of a rectangle, the AI made a gross error by deeming the task impossible. This blunder intrigued experts, revealing a little-explored facet of artificial intelligence. A Spoken Error This clumsiness should not simply be considered a technical flaw. On the contrary, it demonstrates a cognitive effort on the part of the AI, an attempt at reasoning beyond the simple restitution of information. The researchers evoke the notion of the zone of proximal development , a concept where learning occurs through error. ChatGPT, in its development, has indeed exhibited a form of creative initiative, similar to a student struggling with logic and trying to find a way out.Inspiration and Creativity When observing this reaction, one can’t help but draw parallels with the adventures of iconic characters, such as Tony Stark and his first Mark I armor. Although far from perfect, this prototype demonstrates a higher potential in the making. Like a Dungeons & Dragons player who creates precarious but newly original strategies, the AI here reveals an exploratory skill that deserves to be encouraged. A step towards transformed learning This experience raises relevant questions about the future of artificial intelligence in various fields, notably education. Far from considering AI as a simple generator of answers, its use could be oriented towards supporting human thinking. By acting like a patient teacher, who lets their students make mistakes in order to learn better, we could radically transform teaching methods.Avenues for research For researchers, ChatGPT’s improvisation could open the way to new ideas, even from apparent failures. Although it does not think like a human being, this machine gives the impression of thinking in progress, a characteristic that could enrich the academic field. By imagining a debate table bringing together Socrates, Jarvis and Rick & Morty, we see a mixture of absurdities, anecdotes and unexpected flashes of inspiration. The interaction between ancient knowledge and technological innovation is thus strengthened.
To read Arthur Mensch : portrait d’un pionnier façonnant l’intelligence artificielle en Europe
