Google Dipmind’s artificial intelligence won the gold medal at the World Mathematics Olympiad!

Researchers at Google Dipmind say alphaageometry2 artificial intelligence has been able to work at the gold medal level in solving the International Mathematical Olympiad (IMO). A few months ago, Dipmand’s artificial intelligence performed at the silver medal at the Olympiad.

According to the Scientific American report, in a matter of months, Dipmind researchers have been able to upgrade their artificial intelligence system, Alphaageometry2, at the International Mathematical Olympiad from the silver medal to gold.

At the Olympiad, high school students need to solve very difficult mathematical issues. “I think it is not long before the computers get a full score on IMO,” says Kevin Buzzard, a mathematician at the University of Imperial College in London.

Extraordinary performance of artificial intelligence in Mathematics Olympiad

Previous research has shown that when artificial intelligence can solve geometry problems, it can also be used in more complex applications; Because artificial intelligence needs a high level of reasoning and selection from possible steps to solve the problem to solve geometric problems.

To this end, Dipmind researchers have focused on the development of artificial intelligence specialized in solving more complex geometry problems. Their first version of their artificial intelligence was introduced January 2024, called Alphaageometry. The second version is also called Alphaageometry2. At the 2024 Summer Olympiad, the model had a few errors, and now the researchers have been able to improve the performance of the system.

In the new study, researchers evaluated the system’s ability to solve various IMO problems over the past 25 years. The system was able to solve 84 % of all geometric issues offered in the IMOs over the past 25 years. That was 54 % for the previous model. Also, the Dipmide artificial intelligence system had to solve 50 issues of the new IMO test. According to the researchers, Alphaageometry2 was able to solve 42 issues properly, slightly above the average scores of gold medalists in the race.