Artificial intelligence has been given new capabilities never seen before, computer scientists have claimed.
Researchers were able to use AI to help prove and suggest new complex mathematical theorems.
Scientists hailed the breakthrough as a major step towards giving major new capabilities to such systems.
“Problems in mathematics are widely regarded as some of the most intellectually challenging problems out there,” said Geordie Williamson, a mathematician who was a co-author on a new paper describing the breakthrough.
“While mathematicians have used machine learning to assist in the analysis of complex data sets, this is the first time we have used computers to help us formulate conjectures or suggest possible lines of attack for unproven ideas in mathematics.”
The results have been described in a new paper published in Nature. It saw DeepMind – the Google-owned AI company – work with top mathematicians to create a framework that would allow researchers and artificial intelligence to collaborate together.
Typically, much mathematical work has relied on looking examples and finding patterns or relationships. That requires a combination of creativity and calculations – and while computers have helped with the latter for decades, the former has always been the preserve of mathematicians.
In the new research, however, scientists were able to show that they could use artificial intelligence to some of that creative work, previously confined to humans. It is a task central to research that was thought to be reliant on human intuition, and so could apply to a whole host of different areas of maths.
To show that the system worked, the researchers then applied it to two distinct areas of mathematics, identifying previously unknown relationships in knot theory and combinatorial representation theory. That shows that it is not only hypothetically possible but has already done mathematical research that was previously unknown.
It is the first time that significant mathematical discoveries have been completed using machine learning, DeepMind says. And the breakthrough also represents a model that can be used to do even more.
“Neither result is necessarily out of reach for research- ers in these areas, but both provide genuine insights that had not previously been found by specialists. The advance is therefore more than the outline of an abstract framework,” writes Christian Stump, an expert who was not involved with the research, in an accompanying article in Nature.
