This is information that you may have seen passing by, without paying much attention: the company DeepMind, a Google subsidiary specializing in artificial intelligence, has developed an algorithm capable of solving geometry problems.
Called AlphaGeometry, this AI represents such considerable progress in this sector that its results were published in the journal
Nature
.
When we look at what she has achieved in practice, however, it seems somewhat insignificant: she simply managed to rise to the level of a champion in the International Mathematics Olympiad, a competition intended… for high school students.
This AI is also limited to Euclidean geometry problems in the plane, that is, the one you learn in college, based on triangles and Thales' theorem.
Read alsoAfter its series of victories over humans, Google retires its artificial intelligence
AlphaGeometry has thus become very effective in the exercise: it has solved 25 of the 30 problems presented since 2000 in this field.
But why should we be surprised?
Or, rather, why be surprised?
After all, AIs now beat the greatest chess players hands down, compete with the best Go players, are capable of responding in a relevant manner and with a language level adapted to complex questions, of generating images of all kinds, to anticipate the weather or predict the 3D conformation of proteins from their chemical formula, which no physical model is capable of doing.
Naively, one might have imagined that finding “small” geometry demonstrations would prove relatively easy.
And, more generally, that…
This article is reserved for subscribers.
You have 86% left to discover.
Flash sale
-70% on digital subscription
I ENJOY IT
Already subscribed?
Log in