Facebook's AI Can Solve University-Level Math Problems In Just A Few Seconds

AI is getting better at solving math problems. And recently, a Facebook's machine learning system has managed to solve calculus problems of university levels in just a few seconds.

Guillaume Lample and François Charton of Facebook AI Research feed the artificial intelligence data of tens of millions of problems relating to calculus, which were generated randomly by a computer. These problems were mostly mathematical expressions with the involvement of integration, a calculus technique to calculate an area under a curve.

The AI made use of NLP (natural language processing), which is commonly used to analyze languages. What it did was to consider the mathematical problems as sentences with nouns and operations are variables and denoted x. The next step is to translate the problems into solutions.

When the two scientists conducted tests on the AI with 500 calculus problems, what they achieved is the accuracy level is 98%, the highest rate a math-solving machine has ever managed to get to.

The AI also has to work on differential equations, which require not only integration but also other techniques. The artificial intelligence only correctly solved 85% of these equations, as well as 40% of harder ones.

However, Facebook’s AI is still capable of solving math problems that trouble other equivalent programs.

This achievement has limited practical application in real life but it is the first step to create a system that can tackle problems that are too hard for human mathematicians to solve, according to Charton.

Artificial intelligence systems can be of great help for mathematicians when it comes to time-saving; for example, theorem-proving, according to Nikos Aletras at University of Sheffield