In trying to find solar systems, we do a lot of different things. We look for planets to pull their stars to and fro in ways we can see in stellar spectra. Basically, astronomers use fancy police speed guns to look for stars moving under the influence of planets. We also look for the light of stars to dip as planets pass between us and them.

And in the rarest kind of discovery possible, we can sometimes catch star systems passing between us and distant stars, and the star and planets in those solar systems can momentarily brighten the distant star through an effect called gravitational lensing. This effect comes straight from relativity and describes how mass can act as a lens and bend light intended for another part of the universe toward us, allowing us to see an object as brighter than it would otherwise appear.

The math behind relativity is not as hard as you might think; I’d much rather solve for relativistic effects than calculate how dust clouds affect how we see things.

The thing is that the maths do get complicated when you are dealing with the effects of both a star and planet on background light, and researchers had been making a pair of different assumptions to deal with what they thought were different kinds of events.

But the computers told them that was silly. Researchers trained an AI system to find and categorize gravitationally lensing planetary systems, and the AI unified the two systems in a way that demonstrated our simplifications caused us to miss how what we saw as two different situations are actually one situation that can appear in a variety of forms.

Sometimes, in our new reality, it appears that computers are going to have a thing or two to teach their programmers.

More Information

UC Berkeley press release

“A ubiquitous unifying degeneracy in two-body microlensing systems,” Keming Zhang, B. Scott Gaudi, and Joshua S. Bloom, 2022 May 23, Nature Astronomy