Major achievement in the field of artificial intelligence: “As good at mathematics as the greatest geniuses”

Trieu Trinh, a recent graduate student at NYU, has been developing AlphaGeometry over the past four years. This artificial intelligence can solve many difficult engineering problems. According to Trinh, who published his work in a trade magazine nature The program was disseminated to the level of the best participants in the Mathematics Olympiad.

Computer scientists did not develop the system alone. From 2021 to 2023, he worked at Google developing AlphaGeometry, specifically at DeepMind, Google's artificial intelligence development arm. DeepMind previously developed AlphaGo and AlphaZero, two programs that successfully solved the games of Go and chess, respectively.

“We are making a big leap”

However, mathematics problems are the most difficult to solve. Trinh said to The New York Times He had difficulty running AlphaGeometry for a long time. Eventually it worked. He says in the American newspaper: “We are not making incremental improvements. We are making a big leap, a big breakthrough in terms of results.”

AlphaGeometry can now solve random geometry questions from Maths Olympiad. The program was able to solve 25 out of 30 questions. In comparison, human Maths Olympiad winners solved 25.9 questions on average. Silver and bronze medals go to people who scored an average of 22.9 and 19.3, respectively.

Human thinking

According to Trinh and the paper's other authors, AlphaGeometry is a “milestone in human-level machine reasoning.”

Other AI tools that have become popular in the past year, such as OpenAI's ChatGPT, are not capable of logical reasoning. One caveat is that AlphaGeometry was developed specifically for Olympiad questions and cannot simply replace mathematicians.

More news about artificial intelligence.