Randy Scott
What if AI could discover new black holes and unlock the secrets of the universe?
Hello, and you're listening to Where What If Becomes What's Next from Carnegie Mellon University, where we're exploring what AI means to industries, government, society, and you.
On this episode, we're diving deep into the cosmos and back in time to the birth of the universe to explore how artificial intelligence is accelerating astrophysics and astronomy at warp speed.
How will AI help discover new black holes in galaxies? What can AI teach us about exploding stars? Can AI prove the Big Bang Theory of the origin of the universe once and for all? So buckle up as we embark on a journey to the stars and beyond. Joining us today are astrophysicists and astronomers who are using AI to explore our cosmos in groundbreaking ways.
Ever since the early 1600s, when Galileo pointed his handmade telescope at the heavens, we've been exploring our skies and wondering, what's out there and where did we come from?
Today, we're starting to answer those questions with advanced telescopes, such as the Hubble and the James Webb Space Telescopes probing the cosmos and new wide-field observatories that are taking pictures of the entire sky every night. Where once we were only able to see a handful of planets and thousands of stars in our night sky, now we are counting thousands of exoplanets and billions of stars, and distant galaxies and astronomical phenomena such as black holes from the far reaches of the universe.
And from these better telescopes comes better and bigger data – massive amounts of data that can only be analyzed and understood by AI, revealing new knowledge about the cosmos and the laws of physics itself. Even though AI is revolutionizing astrophysics and astronomy today, it's actually been in use in astronomy for a while. What's different now?
Rupert Croft
So it's correct that AI has been used in astrophysics research for decades.
Randy Scott
We're speaking with Rupert Croft professor of physics and a member of the McWilliams Center for Cosmology and Astrophysics at Carnegie Mellon University.
Rupert Croft
And so maybe 20 or 30 years ago people said, okay, how about instead of just pointing a telescope at one object at a time, we just get a telescope and we point it at every position on the sky over a period of years and we build up a so -called survey of the sky. And if you do that, you go from having to say one or two galaxies that you're trying to understand to millions and now we're at the stage of billions or tens of billions. And so it turns out that AI tools were perfectly suited to this revolution in astronomy of big data and surveying the sky. And so that's basically where this thing started. And so the AI was used to do relatively simple things back then and now it's becoming more complicated.
Randy Scott
Okay, then let's start with something really complicated – black holes. A black hole is born when a massive star dies. The star explodes some of its matter out into the universe as a supernova, and the matter that remains collapses in on itself and forms a black hole. Traditionally, detecting these invisible giants involved observing the movement of stars near them or seeing how light bends around them. And then, in 2019, the Event Horizon telescope made history when it captured the first fuzzy image of a black hole at the center of galaxy M87.
But a few years later, something happened to that blurry image.
Rupert Croft
That image has been processed by AI to remove noise, if you like, and actually maximize the signal in it. And so you can take an AI and train it on noisy images of models of black holes. And then it will figure out how to get rid of the noise and give you what a real black hole looks like.
Randy Scott
John Wu is an assistant astronomer at the Space Telescope Science Institute, which operates NASA's Webb, Hubble, and other space telescopes. John, how exactly does AI imaging work?
John Wu
So black holes are actually very, very compact because they're so dense due to gravity. And what's happening is, let's say, computational imaging or AI can be used to reconstruct the image on extremely, extremely small scales.
That is, they can provide exquisite high -resolution imaging, and this is by using a pretty complicated technique called interferometry, where you measure the interference patterns of multiple telescopes together. And it turns out that you can have a lot of different possible images that give rise to the types of patterns that are seen by all these multiple telescopes. And so using AI to quickly filter through all the possible...
images that could give rise to what we do end up observing and then imposing some statistical priors or some physical guesses as to which ones are more realistic than others. That's generally the approach that's taken to derive this kind of stuff.
The AI enhanced image of the M87 black hole was breathtaking. Additionally, scientists are using AI algorithms to sift through massive amounts of data from telescopes and space observatories to identify potential black holes faster than ever before, especially when they collide with each other. Rupert?
Rupert Croft
So we've seen this in the last 10 years, that there are black holes out there in space, in our galaxy and other galaxies that are merging together. When they merge, they make these ripples in space -time called gravitational waves that we can detect on Earth. And those ripples, they can be analyzed with AI. The AI can look at the ripples and then tell us what kind of black holes created them.
Randy Scott
So Rupert, why is it so important to find and study black holes?
Rupert Croft
So they used to be some sort of marginal object that people like, those sort of curiosities, but gradually over the last few decades, they've become more and more important. And people realize that in fact, they actually govern a lot of what happens in the universe. You could imagine that maybe 20 years ago, we discovered that every single galaxy like our own has a black hole at the center of it. Turns out that those black holes, because they make huge amounts of energy, they're able to expel star -forming gas from galaxies. They can actually change how galaxies form, how stars form. They can basically stop stars and planets forming in those galaxies. So it turns out that in part of a life cycle of a galaxy, black holes actually can dominate the history of how the galaxies actually form and the stars in it form.
Randy Scott
Okay, so a black hole can perhaps unlock some secrets to the cosmos. But how can AI help? Can AI in effect go backward in time to the birth of our universe, about 14 billion years ago, and teach us something about the Big Bang?
Rupert Croft
So people have tried doing that kind of thing, where they would take the present universe and say, let's ask AI, can you run the laws of physics backwards and come up with something, right?
And see what were the beginning, what were the initial conditions that created the current universe that we can see around us. The Big Bang obviously happened when it was zero minutes, seconds, whatever old. And so going back that far, you know, it's difficult, right? So I think the AI maybe could help us in different ways. It could help us maybe with coming up with different techniques to find out about what's going on at those very old times, things that humans have not thought of yet. It can help with designing experiments. It can also help with making the measurements we make more precise on whether there's something wrong in our current picture or not. I don't know, right? Obviously everyone hopes that there's something new and exciting, but obviously all the knowledge we have, or at least all the understanding we have shouldn't be discarded.
Randy ScottWe're talking about AI, astronomy, and the cosmos. When we come back, we're going to talk about how AI will help a new kind of observatory explore deeper into the universe than ever before.
Sponsor Break
This podcast is brought to you by Carnegie Mellon University. Located in Pittsburgh, Pennsylvania, CMU is a private international research university, consistently ranked among the top universities in the world. Its trailblazing research is focused on creating, defining, and accelerating the future of artificial intelligence. For more information, please visit ai .cmu .edu.
Randy Scott
Welcome back. You're listening to Where What If Becomes What's Next from Carnegie Mellon University. And today we're talking about how AI is revolutionizing astronomy and astrophysics in the study and exploration of our cosmos.
In 2025, on a dark mountain top in Chile, a much anticipated new observatory will start surveying the entire visible sky every three days for the next 10 years. That's a lot of data, an astronomical amount of data.
Rachel Mandelbaum
Yeah, AI is going to be essential for analysis of that data. The data set will be incredibly rich.
Randy Scott
We're talking with Rachel Mandelbaum, a professor of physics working on cosmology as part of the McWilliams Centers for Cosmology and Astrophysics, at Carnegie Mellon University.
Rachel Mandelbaum
So the observatory that's being constructed with a new telescope and camera and so on is called the Vera C. Rubin Observatory. It was previously called the LSST project and it was renamed in memory of Vera Rubin a couple of years ago. And it's going to carry out a 10 -year survey of the sky called the Legacy Survey of Space and Time, LSST.
And so it should be starting commissioning, starting the initial observations of the sky later in the summer of 2024 and starting the survey at some point later in 2025.
Randy Scott
Rachel, how is this new observatory different than the James Webb Telescope?
Rachel Mandelbaum
So the James Webb Telescope is optimized towards looking at the very distant universe, trying to discover some of the earliest supermassive black holes and some of the earliest galaxies that ever formed.
And so to do that, it's helpful to be in space because you can measure infrared light much more effectively there. Whereas the Vera Rubin Observatory and the new telescope and camera are optimized quite differently. So they're optimized essentially to be able to look at a really large area of the sky, like something like 10 square degrees, so orders of magnitude larger area than the James Webb Space Telescope, and to be able to very quickly shift, like look for 30 seconds in one place and then move to a different place in the sky and look 30 seconds there and so on. And so it covers an enormous area very quickly and then it comes back to it a few days later. So it's essentially really good at mapping out like a color movie of the southern sky because it's going to go back to each point, you know, many, many times, approximately a thousand times for each point on the sky.
Randy Scott
And that's why it's going to gather massive amounts of data about our universe, right? How much data?
Rachel Mandelbaum
We're talking, you know, tens of petabytes of data. Another way to think about the amount of data is to ask how many measurements will it make of time varying objects? And so the answer is every night it should identify approximately 10 million objects that seem to have changed. So it will put out what are called alerts, essentially like, something changed over there, something changed over there, but 10 million of them per night every night for 10 years.
So it's really going to be quite amazing for a lot of areas of astronomy, but especially for studying the time varying aspects of the universe.
Randy Scott
And how are you involved with the observatory?
Rachel Mandelbaum
CMU has partnered with the University of Washington to establish a project called Link Frameworks. And the goal of Link Frameworks is essentially to develop software infrastructure that can be used by the entire LSST science community to more effectively analyze the very large data sets that will come from LSST. So I'm the co -lead for that project here, and we have counterparts at University of Washington. And the team has software engineers and research scientists kind of working together on how can we implement some of the really, really promising algorithms for analyzing the data in a way that they work on such a large data set. And they work robustly, and they're stable, and that we trust the answers coming out of them with the idea being that the broader community can then build on those tools.
Randy Scott
What kind of discoveries will come out of this?
Rachel Mandelbaum
So the amazing thing about LSST is that it's going to allow advances in a few different areas of astronomy. So, for example, in the area of cosmology, so understanding what physics covering how the universe has evolved, it should help us understand why the expansion of the universe seems to be accelerating.
So the accelerating expansion of the universe was first measured approximately 25 years ago. People got a Nobel Prize for that finding. And we still don't fully understand why that's happening. If we ask really close to home, like what will LSST tell us about the solar system? So it's going to help us identify orders of magnitude, more asteroids and various small bodies at all different places in the solar system so that we can kind of try to understand how did our solar system form and evolve. There are a lot of things we don't understand about that. And it's really fundamental to how did the Earth come to be? And LSST is going to help us answer those questions. And so for all of the scientific questions that I just kind of listed for you, there are places where we're going to have difficulty getting the answers out of the data. And AI is really promising with further development to help us answer those questions.
Randy Scott
As we look at the exciting opportunities created by using AI and astronomy and astrophysics, are there risks, dangers, and ethical issues we should consider? Rupert, let's start with you.
Rupert Croft
Of course, be careful to trust it, right? That's also the big issue is how far can we trust it? So supposing somebody said, I've got this observation and the AI told me that Einstein is wrong, then that's an issue, right? You have to say, OK, hang on a minute. Why did you come to this conclusion? Why should I trust you?
Randy Scott
Are there other risks, dangers and ethical issues we should consider?
Rupert Croft
Yeah, definitely. So there are a whole range of different things.
We actually had a conference a year ago on responsible use of AI in the natural sciences, in science basically. We looked at all these different aspects. And so apart from the biases and the sort of the wrong answers that could be generated, right? They're also the question of reproducibility. I guess everyone always talks about chat GPT, but you know, if you ask chat GPT, you know, the same question three times, it will come up with three different answers. It might be similar, they might be completely different. It's because there's some element of randomness that goes into the generation.
Now, obviously, if you're doing science, that's not good. You can't do an experiment and then get three different answers if you do the same experiment three times, right? And so, you know, we need to be able to deal with that. If you explain to people, I used AI to solve this problem, and then they do exactly the same thing you did, it's possible that they'll get a different answer. And so, you know, again, this is something that people have to use to mitigate when they're developing their AI analysis techniques.
Randy Scott
Rachel, any other ethical considerations?
Rachel Mandelbaum
One major issue in AI ethics is, let's say if you're going to use an AI algorithm to search for something on the web. So some algorithms are based on training data. And if your training data set is biased, that could lead you to make judgments that have some ethical implications, biased against particular groups. In the case of astronomy, if we are using a machine learning algorithm that is based on a training data set that is incorrect, maybe it's missing something, there aren't any ethics implications, there are scientific implications. We might get the wrong answer and don't realize we're getting the wrong answer. But the issue is really with the quality of the scientific result rather than an ethical concern.
Randy Scott
Looking into your crystal ball five or 10 years out, what's next for AI in astronomy and astrophysics? Rupert?
Rupert Croft
You know, you asked me what I think about the future and whether our models of the universe are right. Maybe somebody will discover that they're wrong and it won't be, you know, a professional scientist. It'll be somebody using AI to analyze some data in a way that they couldn't have done years ago.
Randy Scott
Rachel?
Rachel Mandelbaum
I've spent my entire career so far, essentially, trying to understand the history of how cosmic structure has formed and evolved and what that means for how our universe is expanding. And LSST is going to be really amazing at telling us a whole lot more about that than we knew from surveys before now. It's hard to say exactly what problems are we going to solve definitively and what new issues will we discover.
What I can say is that, you know, every survey that I've been part of before has made a number of different scientific advances, but has also opened up new questions that we might not have imagined before.
Randy Scott
John, what's next for you?
John WU
We are very interested in future observatories because NASA did recently green light the development towards the what's called the Habitable Worlds Observatory. So this is really the next flagship.
NASA mission beyond Roman space telescope. So the habitable worlds observatory, one of its goals is to really characterize Earth -like planets around other stars. It's something that's very, very hard to do. And it will also do a lot of other interesting general astrophysics science. Because of its timeline, which is going to be well into the next decade for the development of this mission, I actually do think AI has the ability to help guide and really accelerate the development of the instruments that will go on board the spacecraft and the kinds of science that it will be able to do. So I'm really excited about what those will be.
Randy Scott
And John, any final words?
John Wu
Astronomers often talk about the unknown unknowns. And I think some of the biggest discoveries that will be enabled by AI are going to be these unknown unknowns, things that we haven't even thought of yet.
Randy Scott
Today, we've only scratched the surface of AI's role in astrophysics and space exploration, from discovering new black holes in galaxies to modeling the early universe. AI is a powerful tool that is reshaping our understanding of the cosmos.
Thank you to our guests for offering their insights, and thanks to you for listening. Please check the show notes where you'll find links to additional resources for this episode. And to learn more or contact us with a story idea or comments, please visit ai .cmu .edu slash podcast.
If you are enjoying this show, please rate and review us on your favorite podcast site and share with friends. And be sure to subscribe so you'll never miss an episode.
You've been listening to Where What If Becomes What's Next from Carnegie Mellon University, where we are exploring what AI means to industries, government, society, and you.
Stay tuned and stay curious.