Southampton asks volunteers to vet AI-sifted telescope data for hidden black holes

The project works as a chain of filters. The AI goes first, cutting millions of possible locations in the Milky Way down to the places where black holes may be hidden ^[4]. People go second. Southampton's team turned the data that survived into simple graphs and animations for volunteers to inspect ^[10]. Telescopes go last. Dr. Adam McMaster of Southampton's physics and astronomy department said volunteers would be "narrowing down precisely where we need to point our telescopes next, and you might be the one to discover a black hole." ^[12]

The order of those steps tells you what a volunteer's flag means. McMaster said black holes "often leave clues because of their strong gravity, which can bend light as they orbit around stars, creating an effect called gravitational self-lensing." ^[7] "Once we detect this, we'll know an invisible black hole exists," he said ^[8]. In Southampton's account, that detection comes after the telescopes have looked. The volunteer's click comes before it, and it puts a target forward for follow-up.

The data were collected for a different job ^[9]. The team says its AI searched in days what used to take years ^[5]. That figure is how long the screen took to run. The release does not say how the model was trained, how many candidates it passed to volunteers, or how often it drops a real signal or keeps a false one.

I think putting people after the machine is a sound design, provided the shortlist is small enough for volunteers to read closely and broad enough to keep the odd cases. Grace Clarke, a 20-year-old physics with astronomy undergraduate at Southampton, has signed up ^[13]. "The data's easy to understand, it's just pattern recognition, and you don't need a degree in astronomy to read it," she said ^[14]. Volunteers learn from examples the team shows them ^[11]. The AI was also doing pattern recognition, so the human pass is a second classifier, taught by example and run on the machine's output ^[4][14].

Then there is scale. Even if the "thousands" Southampton estimates meant only 1,000, that would be about 14 times the roughly 70 black holes whose locations are known ^[1]. "With potentially thousands of new discoveries, we need all the help we can get," McMaster said ^[15]. His phrasing turns an estimate of what is hidden into a forecast of what the project might find. By his own description, the self-lensing clue comes from black holes orbiting stars ^[7]. Any population TESS can reveal this way is therefore a subset of that estimate.