AI and the future of lens finding

55:22on Vera Rubin as well. And the impact that AI is going to have in the detection and characterization and data processing pipelines, again, with still a human in the loop, but if you're taking 800 terabytes of a data set and you can narrow it down to candidate targets that then can go to human review, that seems like it's going to have huge impact in terms of trying to more aggressively and quickly kind of get from the 30 lens candidates to the thousands number. I mean, what do you How do you feel about the role that AI will have in your research particularly? >> So, it's definitely going to have a huge

56:04impact on astronomy. I think it's too early to say exactly what it's going to look like. But I think it's safe to say that it's, you know, it's definitely going to be around and it's going to fundamentally change the way science is done. Uh you know, AI is already excelling at finding lenses because they really don't look like that much other stuff in the universe, right? A ring with four really bright point sources. >> Mhm. >> Uh you know, it could be maybe four stars or something, but you know, AI's are really good at finding lenses and and telling the difference between four stars and four images of a quasar, for example, and they can do it much faster

56:45than like a poor grad student. >> going to say a poor grad student. >> has to look through like terabytes of data to find these little things, right? Uh people also use uh machine learning to model lenses. You know, some people have have tried generating lots of examples with clumps of dark matter in these simulated uh lenses, showing them to neural networks, and then essentially showing that neural network a real lens and saying, "What is dark matter? You know, tell us the properties of these clumps." >> Yeah. >> So, people are are trying all of this kind of stuff. I don't know yet uh to what degree it's it's going to be successful. Uh but it's I mean, extremely

57:26interesting. I mean, in general, astronomy has been a great uh test bed for AI because there's so much data right >> Uh and the parameter space is so enormous. Uh, yeah, I think that the challenge for AI, in my opinion, is doing it in a way that humans are going to believe. >> Yeah. >> So, if the AI tells you something, it did some really complicated analysis that maybe you don't really understand how it it drew the conclusion that it did. Uh, and so making it a believable tool for scientists, I think, is the challenge. But, people are working on that. So, they're they're trying to understand how the AI is drawing, you know, if it if it makes some interesting

58:06statement about the properties of clumps in a lens, for example, how did it get to that conclusion just from looking at the lens? >> It makes total sense. I mean, it's obviously going to be continuing to impact a variety of areas of science and I I I just this is so fascinating and I have so many questions, but I'm going to try to have us land the plane here with a couple of just uh, clarifications. I'm going to come back to this. And so, when you wake up in the morning and you grab a cup of coffee and you head to the lab, you know, a large part of your day is like, you know, working and designing with the theoreticians around what like understanding you have to understand the dark uh, matter theory quite well.