Exploring AI Matters
Our mission is to help the policy community understand the breadth and richness of AI and the potential for such technologies, wisely applied, to augment all sorts of human endeavors.
Some AI tools are able to assist humans in performing tasks faster, more accurately, or more efficiently. Some, however, are inaccurate and unreliable. Who or what we hold accountable for these flaws, and what incentives we do or do not create for their correction will influence AI’s hand in how we work.
In this series we will refine, sharpen, and clarify your understanding of AI.
Exploring AI Matters
Episode 9 - Too Many Stars! Too Many Galaxies! How to choose what to look at?
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Send us comments and questions.
The Vera C. Rubin Observatory, located high in the Andes in Chile, is expected to be operational near the end of 2023. The Rubin’s wide-field reflecting telescope will scan the entire southern sky every few nights. In this episode of Exploring AI Matters our guest, Harvard Professor of Astronomy Edo Berger, a leader in the Rubin Observatory’s work will be our guide to what is intriguing in deep space, and how AI will help astronomers navigate this ocean of images. [2023-03-09]
Welcome to Exploring AI Matters. This podcast series, previously known as Mind the Gap, Dialogues on Artificial Intelligence, will continue to appear in the ABA series to the extent that in addition, all of the episodes, old and new, will now appear under our new podcast name, Exploring AI Matters. Thank you.
SPEAKER_04Near the end of Men in Black, Agent K looks up at the night sky and says, They're beautiful, aren't they? Stars. I never look at them anymore, but they actually are beautiful. Other than seeing the stars rise, progress, and set, which is the effect of the Earth's rotation, most of us looking at the night sky don't observe that, which excites the curiosity of astronomers. Like the billions of galaxies other than the Milky Way, the collision and merging of galaxies and catastrophic transient events like supernovas and the collisions of binary neutron stars and black holes known as kilonovas. Back in the 1950s, astronomers relied on Earth-based telescopes like Mount Polymar's 200-inch mirror. But the faintest objects they could observe with such telescopes were, by today's standards, not very good, not very faint. Subsequent decades brought the advent of new instruments to detect, observe, and record other kinds of signals. X-ray, radio, and infrared signals emitted from stars and their black hole remnants. In 1990, when the Hubble Space Telescope became operational, astronomers could see much deeper into space, detecting, capturing, and analyzing fainter and hence more distant signals. Since the signals were further away and take longer to get here, observing more distant objects means seeing events much further back in time. In the next two years, if all goes as planned, three new observatories will dramatically extend the observational reach available to astronomers and astrophysicists with powers of ten deeper into space and back in time. One is the James Webb Space Telescope, launched in December of 2021 and expected to be operational later this year in 2022. Webb will provide infrared resolution and sensitivity over Hubble and extend observations to objects with signals 100 times fainter than any Hubble could detect. Another is the Vera C. Rubin Observatory, located high in the Andes in Chile and expected to be operational near the end of 2023, whose wide field reflecting telescope will scan the entire southern sky every few nights. Third, are the LIGO or LIGO gravitational wave detectors that allow physicists to detect the collisions of black holes in neutron stars. The next science run of LIGO is scheduled to begin in December of 2022. Previously, detection of observational enigmas like supernovas or binary neutron stars happened in few enough numbers for astronomers to keep track of the data coming in about them. With the Rubin, the observational data captured becomes so enormous that only about one-tenth of one percent can be studied in the traditional way. How will astronomers decide which data to study? What phenomena will they focus on? How will AI help them sift through this deluge of observational data? It's an exciting time to be an astronomer. Not only because Web and Rubin will, we hope, become operational, but because astronomers will be shaping artificial intelligence into tools that can help them manage and make sense of what the new amazingly powerful observations will bring. We are fortunate to have as our guest today Professor Ado Berger, who is deeply involved in the Rubin Observatory's work, and who we will ask to be our guide to what interests him in deep space the collision of neutron stars and black holes, the explosions of massive stars, and other transient catastrophic events. Hello, I'm Charles Palmer, a computer scientist.
SPEAKER_01Hello, I'm Alma Adams, a national security lawyer. We are your hosts for this episode of Mind the Gap Dialogues on Artificial Intelligence. In addition, we have two more hosts.
SPEAKER_03I'm Mark Donner, another computer scientist.
SPEAKER_05And I'm Roland Trope, also a national security lawyer.
SPEAKER_01Each episode will normally be led by two of us, with the other two adding impromptu questions and comments as the spirit moves them. So, Dr. Berger, we are delighted to have you here with us today. And to put some grounding for our audience, you are a professor of astronomy at the Harvard University Department of Astronomy and the Center for Astrophysics. So that leads me to the following question: Are astronomy and astrophysics the same thing? And if not, how should a lay person like me be thinking about their differences?
SPEAKER_00Well, thank you for having me. It's a pleasure to be with you today and to talk about astronomy and astrophysics. Um, I think the roots of astronomy um uh go back many centuries, um, and and for centuries and decades up until the early 20th century, um, astronomy was really the study of cataloging the sky. So uh counting stars, watching uh where they're located relative to each other, uh, watching planets move and calculating their orbits, um, and so on. Um, I think the study of astronomy became astrophysics uh when astronomers started asking questions about the fundamental processes that govern how stars operate, um, when they discovered that there are other galaxies outside of our own, that the universe is expanding, um, that different um chemical elements produce different uh signatures in the spectra of stars and galaxies. Uh, and so I think for many years, uh people have shifted the terminology from astronomy to astrophysics to kind of designate this growing understanding. Uh things have gotten a little bit complicated in the last few decades when uh we are now studying uh molecules in space. And so there's a field of astrochemistry. Uh, people are searching for biological signatures in the spectra of planets outside of our solar system, leading to the study of astrobiology. And so I think as the field has progressed in that sense, uh, in some sense, astronomy has now become again kind of a capture-all term. Uh, and so for me personally, I'm happy to use astronomy or astrophysics interchangeably.
SPEAKER_01Thank you for that. So, would it be a fair summary to say, and hopefully I don't blunder this, but would it be fair to say that astronomy perhaps focuses and observes the position and movement of specific things in the universe? You know, stars, planets, and maybe astrophysics applies physics and chemistry to better kind of understand different astronomical observations, like how the universe is created, sort of at a high level. Would would that be a a way for people to be thinking about the two um distinctly, but noting as you just said, there is some intersection?
SPEAKER_00Yeah, I think I think that's that's a fair statement. Um, but but really, you know, and you can see it reflected in uh, as you mentioned, I'm in the Department of Astronomy and at the Center for Astrophysics. Uh, and so these terms, even within the same institution, are used interchangeably.
SPEAKER_01Exactly. But that helps clear up any confusion that I had. Now, part of your research has focused on what we what we're referring to as catastrophic transient events. Um, so what what qualifies something as a catastrophic transient event? And how has your research sort of uh evolved or shifted in this space? No pun intended.
SPEAKER_00Yeah, that's an excellent question. I think when when people think about the universe, they think of it as this enormous place that has been around for uh more than 13 billion years. That, you know, we use the term astronomical to designate something that is very large or very long-lived. And so we have this view of the universe as unchanging. We go out at night and we can see the same stars that uh the ancient Greeks uh saw. Um, you know, the the stars appear to move, but that's because the earth is rotating. Uh different stars appear, but that's uh during different parts of the year, but that's because we're orbiting this the sun. Uh but it turns out that there are objects in the universe that actually evolve and change on human timescales. And generally that has to do with um either objects within our solar system that are moving, like asteroids uh or planets, um, as well as things that explode. Um and there are catastrophic events uh such as supernova explosions, uh, which take place on human timescales. Um, you know, typical supernova will last for uh a few weeks, maybe a few months. Uh, there are transient events that last even shorter periods of time, some uh as short as a few milliseconds. Um, and so it turns out that the universe actually is teeming with these um rapidly changing uh catastrophic events. Some of them repeat and are periodic, some of them are one-time uh completely catastrophic, and the object that produced them uh uh completely destroys itself. Uh but that's always been um of interest to me to understand these explosions, both because I think they're uh provide us with a way of studying the universe in a different mode than what we traditionally do. Uh you know, we we can't necessarily study one of these objects by saying, oh, you know, it's okay, it's gonna be there next year or next decade or next century. Um we have to act very quickly. We have to use telescopes uh underground and in space very rapidly to respond to these kinds of events. Uh, but also these events, because they're so energetic, um, they play a really critical role in the evolution of the universe. So, for example, we know that supernova explosions synthesize different chemical elements that end up being incorporated into next generations of stars and into planets and into life. The iron in our blood was produced by a supernova explosion. Um, and these events also uh generate an enormous amount of energy, and that energy as and material outflowing from the site of the explosion um can trigger uh next generation of stars to be born. It impacts the evolution of galaxies on a large scale. Um, the other reason I find these events interesting is it turns out that when a star explodes, it releases so much energy in such a short period of time that it can rival and even outshine its entire galaxy where it occurred. And so for a brief period of time, maybe a few minutes or a few hours or a few days, uh, we can see these individual explosions to very large distances across the universe. And so they provide us with a signpost and essentially a lighthouse uh at very large distances, allowing us to probe parts of the universe that are very difficult to see in other ways.
SPEAKER_01Okay, so when we're talking about supernovas, um just kind of walking us back through that very helpful explanation and sort of picture that you were presenting for us, we should be thinking of them as sort of the an explosion of a star, right? At a sort of a baseline level. Is that what we should be thinking of when we talk about supernovas?
SPEAKER_00Yes. When we talk about a supernova, we're talking about the catastrophic explosion of a star. Um there are essentially two flavors of supernova explosions. There are explosions of very massive stars, much more massive than our own sun, uh, about 10 times more massive. Um, and those stars have a very short lifetime, and uh they they burn through their um nuclear fuel. Uh, within a few million years, they eventually run out of the ability to burn any more fuel, uh, and therefore they lose support against their own gravity, uh, leading to the core of the star to rapidly collapse on itself, uh, send out a shock wave that explodes the rest of the star, uh, and will generally leave behind either a neutron star, uh incredibly dense, very compact objects, or even a black hole.
SPEAKER_01Not all stars explode, right? Some perhaps fade or they're not, they don't, that supernova again is not something that is sort of a signature of all stars.
SPEAKER_00That's correct. So a star has to be at least eight or ten times more massive than our own sun to go through that process and eventually produce uh one of these so-called core collapse uh supernova. Um stars like our own sun will leave behind a white dwarf as they kind of slowly fizzle out and fade away. Uh, some of these white dwarfs that are that are left by uh by these less massive stars like our own sun uh could, through some processes of um um gaining mass from a companion star, for example, uh reach a critical mass threshold, about one and a half times the mass of our sun. And at that point, they essentially ignite like a giant nuclear bomb uh and they completely explode. And those kind of explosions uh are incredibly bright, uh, they're incredibly uniform in their properties. And those types of supernovae called thermonuclear supernovae, uh, in analogy to a thermonuclear uh bomb, um are the ones that are being used actively to map out the expansion history of the universe and led to uh the discovery of the accelerated expansion of the universe and dark energy, for example.
SPEAKER_05Could I ask before you go much further, could you distinguish for us the difference between a neutron star and a black hole and why some stars become one and big instead of becoming the other?
SPEAKER_00Yes, that is an excellent question. So um both neutron stars and black holes are the remnants of these massive stars um exploding. And whether a star will produce a neutron star or a black hole has to do with the mass uh of the core. So there is some critical point above which um the core becomes so massive that it can no longer even be supported as a neutron star and will essentially collapse into a singularity and produce a black hole. So that is somewhere between two and a half and about three times the mass of our sun. Uh if a star has a core that massive, uh it will produce a black hole. Uh a neutron star uh is, you know, in the in the case of a neutron star, the collapse of the core is halted by something called neutron degeneracy pressure. Essentially, the star cannot be squeezed uh beyond a certain point due to quantum mechanical effects that essentially provide support against any further collapse. And so that collapse is halted when the core reaches a size of um only a few miles, about seven or eight miles. So about the size of a of a typical city. Uh, but that core holds as much as two or two and a half times the mass of our sun. So these are incredibly dense objects, but not as dense as you know this infinitely dense singularities in in black holes.
SPEAKER_04Wow. Well, so let's let's pull back a little bit to uh something a little closer in. The Rubin Observatory is is coming up. Um can you tell us what you'll be doing with that? And and is it still on schedule, all that stuff?
SPEAKER_00Yeah, the Rubin Observatory is is going to be uh a phenomenal um project. It is. It is essentially one of the or perhaps the largest ground-based astronomy project uh of US astronomy of this decade. Um and the idea of the Rubin Observatory is it's a it's a large telescope, it's about an eight-meter diameter uh mirror telescope with a very wide field of view. Uh, so it can rapidly map the entire sky in order to essentially carry out a survey where every few nights the entire sky visible from the southern hemisphere will be mapped out. And so you can think of this as instead of large surveys that happened in the past, which predominantly built up a single image of the entire sky over many months or years or even a decade, here we will be able to essentially produce a movie, a running movie of what the sky is doing. So map the entire sky every few days. Um, and the goal is twofold. At the end of the survey, to essentially take all of those images that were collected over about a decade and collect and add them together to make one of the deepest maps of the sky that, or the deepest map of the sky that's ever been made. But at the same time, by comparing images of the same part of the sky every few nights, uh, we are going to look for objects that are either moving or changing in brightness. Um, and that is the part that I am particularly interested in in terms of the Rubin uh uh survey, which is going to be called LSSD.
SPEAKER_04So this is, as we were talking earlier, this is gonna generate a lot of data, um, more than's ever been deliberately collected before, I suppose. Um who's gonna work on this and you know how many hundreds of astronomers are gonna do it? How are they gonna get the data? Uh we don't have a big hose coming uh from the new observatory.
SPEAKER_00Yeah, so I think in terms of uh quantity of data, this is going to generate an unprecedented amount of information. So just to kind of put things in perspective, um, the images that will be generated by the telescope every single night will be about 20 terabytes of data uh coming out every single night. In addition to this, um every image that's collected by the telescope, as soon as it is um read out from the telescope, will be compared to a previous master image of the same part of the sky to search for any object that has shifted in position or changed in brightness. And then those objects will be sent out as a stream of alerts to the community, uh, providing astronomers with information about every single object that changes every single minute. So there will be millions of these objects every night that are either asteroids that are moving, uh supernovae that are have just happened or that have happened a few days before and are continuing to change in brightness, or other types of transient events that are different from supernova explosions, and hopefully ones that we've in fact never seen before. Now, in terms of who is going to receive all of this information, there are there are indeed, I think, hundreds of astronomers in the US and around the world who are eagerly waiting to get their hands on all of this data. Um the data will be open to the US astronomy community. Um and I think the hope is that as as these data products uh are delivered uh in real time and in yearly releases, that uh people will download the parts of the data that they need and find creative ways to work with such an enormous stream of information.
SPEAKER_04So with other telescopes, you had to schedule time on the telescope. You had to get in line. Uh it sounds like, since this is a survey uh device, is that no longer the case that everybody gets the same thing and and nobody gets to point it in a particular direction?
SPEAKER_00Uh that is that is correct. So uh the the vast majority of the time on the Rubin Observatory will be dedicated to this uh survey that will essentially run itself um day after day, night after night, uh week after week, year after year. Um and all astronomers will have equal access uh to this data. And I think there is going to be so much of it uh that you know, I think the hope is that people will be incredibly creative about what kind of questions they're going to ask and uh find interesting new ways to answer those questions with such an enormous volume of information.
SPEAKER_04Well, you you you lead me to the next thing I was wondering is surely astronomers like yourself or as astrophysicists like yourself are looking at new ways to process this data, being able to process it in our lifetime. Uh are there some new techniques that are coming up that'll help this?
SPEAKER_00Uh yes, I think this is going to require in many of the scientific questions that that the Rubin Observatory data could answer will require new techniques and new ways of um analyzing the data. Um, you know, for example, in the study of supernova explosions, um, you know, until about a decade ago, there were only a few hundred supernova explosions discovered every every year. Um, and so you could go through them uh one at a time, you could, in fact, and I think some people do, remember the name of each individual one during a year. Um And you could study them painstakingly, you know, with a group of students or uh or individually. Um, we are now in a situation of in the last few years where we are uh finding several tens of thousands of supernova explosions every year. And so that's already challenging our traditional approach of studying these events. But with Ruben Observatory, we're talking about finding several millions of these supernova explosions every year, uh, as well as things that we have presumably never seen before. And so that will require completely different ways. We we're no longer going to be able to uh sit in front of a computer and uh look at one object at a time. We are going to uh require approaches based in machine learning and artificial intelligence to sift through the data and essentially uh let us know what are the most interesting objects that we should follow up with other telescopes that we should study in a more dedicated fashion, as well as to construct large samples of events and study them in a statistical way and in a way that we have not been able to do before.
SPEAKER_05Can I ask one sort of um oblique question? Because you you've been saying discovery of uh a certain number of supernova in a year. Earlier you talked about how quickly you have to respond when a supernova happens. But I'm I'm a little confused about the sense of time, because by the time we observe the event, it's happened, you know, hundreds or thousands of years ago. Is it important to an astrophysicist to keep track of somehow organizing things in terms of when they happened, or is it only matter when we observe them? Does that make sense?
SPEAKER_00Yes, I think that's that's an excellent question. And it's it's it's true. We are finding these supernova explosions in other galaxies uh that are incredibly uh distant from our own Milky Way. Uh and so some of these events that we're seeing happened um hundreds of millions and even billions of years ago. And one of the questions that we can answer as we find these uh supernova explosions to larger and larger distances is how did the universe essentially we have a snapshot in time and we're moving back in time as we look further and further away. So we can ask, how has the universe evolved during its lifetime? For example, as I mentioned earlier, uh these thermonuclear supernovae that we can find to quite large distances allow us to map the expansion history of the universe. So we can tell now that the universe expanded a bit more slowly in the past compared to the present. The rate of expansion is accelerating with time. And that tells us something about the composition of the universe as a whole. Uh, similarly, we can look at um the chemical signatures in these supernovae and we can ask the question: how did the stars that produce these supernova explosions evolve as the universe itself was evolving? As more and more heavy metals were being incorporated into new generations of stars, did that change the properties of those stars? Did it change the properties of the explosions? And so one of the exciting, I think, experiments that will come out from the Rubin Observatory data is both the ability to look further back in time and also to have such a large number of objects that we can have large statistical samples where we can ask the questions are supernovae in our local patch of the universe today similar to supernova explosions that happened five or six or seven or eight or ten billion years ago early in the universe? And that is right now quite an open question.
SPEAKER_01Dr. Berger, we were talking a little bit about the new techniques that astronomers may be developing or deploying or looking to deploy or use in preparation of the information or sort of the flood of new data coming from the observatory. But AI is not a new technique to you in your work, right? Could you tell us a little bit about how and why you've been using AI to date?
SPEAKER_00Yeah, I think that um a primary motivation for this has been this realization that even today, uh, we are getting detailed information only for about 10% of all supernova explosions. So that means that we are uh potentially taking 90% of the information and putting it aside and not treating it or learning from it and not maximizing its potential. Um that will be exacerbated when Ruben Observatory starts operating because that number will drop from 10% to 0.1%. So we don't want to spend all this effort finding millions of supernovae only to, at the end of the day, throw away 99.9% of them and only study 0.1%. So that is that has been a key motivation for incorporating uh machine learning techniques into the study of supernovae. Uh and we have looked at this from different angles. So at the simplest level, um, I mentioned earlier that there are two predominant types of supernovae. There are the supernova that come from massive stars and the ones that come from white dwarfs. Uh, because massive stars are uh so heterogeneous in their properties, their uh final configuration at the end of their life depends a lot on the mass of the star and what kind of environment it lived in, if it had another companion star or not. Uh and so these stars are incredibly diverse. And when we look at the explosions of these stars, we find that they fall into a lot of different categories of supernovae that we have kind of built up over the years, this taxonomy of these um supernovae. And the way we place objects into these different groups currently is using a technique called spectroscopy. Uh, so we take a spectrum of the supernova, not just study its brightness. And that is very expensive to do. Um, as opposed to a big telescope like Rubin Observatory, being able to detect a supernova in about a minute, uh, getting a spectrum of a supernova takes about an hour. And so that's clearly a bottleneck for our approaches of how to classify supernovae into different categories. And what we need to do is to essentially learn how to use the Rubin Observatory data directly, the data stream that we get essentially for free, quote unquote, to place these supernova into different categories. And that's something we've been doing over the past few years using test data that are very similar to what Rubin will produce uh in a couple of years from now. And trying to find out how well we can we can carry out this task.
SPEAKER_01And in terms of the sort of training of the machine learning, you talked about sort of some of the different categories that you might be using. I think you had mentioned brightness. Curious, sort of, how you've been able to train in terms of machine learning to help better classify supernovae to this extent right now.
SPEAKER_00Yeah, and so what we've done is um we used um a survey uh that that took place a few years ago that um was much smaller in scope than what Ruben Observatory will do, but otherwise had very similar characteristics using uh the same frequency of observations, uh almost the same sensitivity of observations, and so on. Uh and we spent a lot of telescope time um classifying spectroscopically, using spectra, uh several hundred of the supernovae that came out of the survey. And we've used that as our training set. So for those, we know what category they fall into, uh, and we can train different types of algorithms. And part of the exciting aspect of this work is that when our group did this, we were the first to actually use real data to assess this problem. People have made simulations of uh supernova data, and they've tried to, they've essentially injected fake supernovae into surveys and then um and then essentially tried to classify them knowing what the answer is because they injected them in. Um and so that was quite a complex process. And um, you know, real data is always more complex than simulated data. So I think our approach was to actually use real observations, real supernova explosions, and explore how well we can classify them into these different categories. And what we found is that it is not a trivial process. Um, some types of supernovae uh stand out in some unique ways that allow us to classify them with very high confidence and very high success rate. Uh, some are a little bit more tricky. They um easily masquerade as another type of supernova, uh, and they're a little bit more difficult to pull out of the data and classify correctly. Uh and so one of the things we found is that it is not going to be just a single algorithm uh that will work equally well across all types of supernovae when we're trying to do this in real, in real life, in real time. Uh, but we're going to need a different set of approaches.
SPEAKER_01And you you had talked about um sort of the the observatory and and framed it in the construct of a running movie, I think is the term that you used in saying how it would perhaps provide up to 20 terabytes of data a night. So thinking about all that volume of data, which which you'll be able to kind of review and assess. Um, what are the kinds of things perhaps that you're you're hoping you'll be able to observe or learn from this data, either confirming prior observations or sort of new emerging observations? Just, you know, what are the possibilities out there with respect to the types of data that you might be able to access now?
SPEAKER_00Yeah, I think I think the projects we've been thinking about kind of fall into three different categories. The first one is that um, you know, we have certain types of supernovae that account for the bulk of the supernovae that we find in the universe. So when we go and we do a survey, uh about 95% of what we find are things that we've seen now for many years and many decades. But we've seen hundreds of them or a few thousands. Um and that limits uh our ability to ask questions like the one I mentioned earlier, which is are these uh supernovae and therefore the stars that produce them change as we look back into the history of the universe? Uh, when we have millions of these supernovae instead of a few thousands, we can start slicing up the data in really interesting ways. We can look at supernovae of the same class that happened in different types of galaxies or at different uh epochs in the history of the universe and start asking the questions about how they change or whether they change. Um that's one category of investigations. Uh, the other one is that um over the last decade or so, we've discovered new, fairly exotic types of explosions that are incredibly rare. So there are some categories of supernovae that we've seen for the first time, and we only know of a few dozen cases. Uh and so far, those have been mainly kind of a curiosity. You know, we collect them one at a time, like rare stamps, and we study each one individually, and each one is a little bit different from the other. Um, but we don't have a complete understanding of what stars produce them, what processes they represent. Um, with Ruben, we'll be able to go from a few dozen of these events to tens of thousands of these events, and therefore put them on an equal footing in terms of our understanding of these events. And then perhaps the most exciting thing is discovering new things that we've never seen before. You know, in the entire history of supernova observations, we have not reached a million supernova yet. Uh, we're actually pretty far from finding a million. And so there are going to be these one in a million events that will come into view with the Rubin Observatory data. And the question is, how do we find them? How do we find these rare needles in the haystack? Uh, we we can't do it by eye. We can't go individually through millions of events and and look for outliers. Um, and so part of what we're doing is is training algorithms that are designed to find anomalies in the data, things that we have never seen before, without us specifying what it is that we're looking for. Uh, but just kind of telling the algorithm, you know, surprise me, find find me something new before. Surprise me.
SPEAKER_03Um, so so at any given time, there's uh sort of a collection of unresolved questions. When when I was uh a young scientist being trained, black holes were a theoretical thing that had not been observed, but that were predicted. Now we've observed them. Are there a collection of of unresolved or unanswered questions that that Rubin and uh Webb will give you a chance to probe and answer?
SPEAKER_00Yeah, I think I think there are um there's a range of questions, but I think one of the most interesting ones is as we have um as we've been building up these um knowledge of different classes of supernova explosions, we we are starting to understand that that catastrophic moment at the end of a star's life essentially gives us a view of what happened to that star in its life. Um and we want to essentially over time build up a mapping between saying when I see a supernova of this type with these characteristics, I know that it came from a star that started with that mass and was in a system with a companion, and it has lost a certain amount of mass during its life, and then it went through this process, and then it left behind a neutron star. We want to kind of have a story uh for these stars uh so that we don't just look at these explosions as the final moment in the life of the star, but we use them to kind of track back the complex history of how stars form and evolve uh during the history of the universe. So I think we have a really unique opportunity uh just because these stars, when they explode, become so much brighter. We have kind of a moment to capture uh all this information about their life. And uh these are, you know, there are certain phases at the end of lives of these massive stars that that are just impossible to simulate on a computer. We we don't have a detailed enough understanding. So we can kind of predict certain aspects of what a star will do as a function of different properties, but when it comes to the final century or decade or year in the life of the star, we simply have no idea. And this is our turn chance to learn uh about that.
SPEAKER_04So if if if you think about the what AI does, and we've talked about this in other podcasts, it's a prediction. You train it so that it looks at things the way you do or the way astronomers do, and then it predicts what you might see next. Is is that aspect of AI or you know prediction machines uh come into play here?
SPEAKER_00Yeah, I think there is an element of that in the sense that um, you know, so far I've talked about uh taking this data and and putting these supernova in different categories. Um of course, you know, for some questions that we we want to answer, we can we don't have to do this in real time. We can collect the data, we can have the full data set for a given supernova, and then we can put it into the algorithm and it will tell us, oh, it's a type 2 or a type 1A or type 1B and so on. Um but another aspect of this is that we want to do this in real time because some of these events are going to be incredibly interesting, incredibly rare, maybe something we've never seen before. And then to gain a complete understanding of these objects, we can't just rely on the Rubin Observatory data. We might want to follow up that supernova with James Webb or with another telescope on the ground, or we might want to collect X-ray data about this event or radio data. Uh, and so in that sense, uh, we are trying to make a prediction of what we think will be interesting events uh where additional supporting observations are actually going to delve deeper into the physics. And so maybe that's again a place where the it's astronomy versus astrophysics. You know, it's not just about classifying objects and putting them in different categories, it's understanding the underlying physical processes that that govern them and how they exploded.
SPEAKER_05Can I ask you to go back to something you mentioned much earlier, or or we mentioned, and you talked about very briefly, which are um the detectors of gravitational waves from uh mergers of black holes and neutron stars? First, the concept of a black hole and a neutron star merging, given all the seemingly empty space out there, is hard for me to imagine. So if you could explain that, that would be helpful. But also, what's a gravitational wave and uh what's its relationship to these events and our observation of it? What does it tell us about it? And lastly, how will AI help you pursue your studies? Because this is something you recently co-authored an important article on, and we'd like to know more about what you're currently looking into.
SPEAKER_00Yeah, excellent question. So um uh it is true that um stars, you know, single stars are very far away from each other in even in a galaxy, and it's very hard for them to find each other. Uh, we think that uh quite a lot of these um objects that lead to the merger or the collision of two neutron stars or a neutron star in a black hole or two black holes actually formed initially as a system of two stars. Uh in fact, we know if you look up at the night sky, almost every star is in a system with a companion. So more than half of all stars uh have a companion star. Our sun actually doesn't, and it's it's actually in a minority in that sense. Uh and that's especially true for very massive stars. Very massive stars like to be born in pairs, and sometimes even triples and quadruples. Uh and so if you have a pair of these very massive stars, each one uh eventually will die and leave behind either a neutron star or a black hole. Uh and those two uh neutron stars or black holes at the end of that process uh will essentially orbit each other. Um, and as they orbit each other, they will lose um the orbit will shrink due to the radiation of gravitational waves. So gravitational waves are these uh ripples in space-time that that travel due to um um due to this motion, uh, for example, of two very compact, very dense, very massive objects uh in a binary system. And so that process might take billions of years, but eventually those two stars, those two neutron stars, or those two black holes, or a black hole and a neutron star, uh will collide into each other after they lost uh all of their orbital energy uh to gravitational waves. The gravit the way we detect um gravitational waves is an incredible new technique. And the first gravitational waves were discovered only in 2015. So that's a very recent development. Uh and what we can detect with these um detectors on Earth on the ground is just these final few orbits in the life of a system like that, just before these objects uh collide, and then we eventually see the collision itself. Uh so these gravitational wave signals only last for a few seconds, uh maybe a minute, uh, and that's that's what we see, this kind of final burst of gravitational waves. So these um these objects, these systems composed of of these two uh neutron stars or two black holes were you know we know of some, we know of these pairs of neutron stars in our own Milky Way galaxy. We we see them through different techniques, but it turns out that they're uh you know quite readily detectable once you have something like LIGO. Um and by now we have uh detected about a hundred of these um collisions uh coming from very large distances outside of our own Milky Way. Uh, I think the place where um AI will play a role with this is as these discoveries are being made now, people are starting to think about the next generation of gravitational wave detectors that will be so sensitive that they will essentially see every single black hole or neutron star collision in the entire history of the universe. So now we're talking about instead of finding a few hundred uh in a year or in a few years, we're again going to be in the same situation that we are with Rubin Observatory. There'll be this deluge of millions of these collisions happening every year. And the question there is not only how do we classify them into different possible configurations, but also how do we effectively measure the properties of these black holes and neutron stars, which we can extract from the gravitational wave data. So currently it takes days to analyze each one of these collisions. So if we're going to find millions of them every year, we can't spend days. We need new techniques. And so one thing we've been doing is developing um analysis algorithms that are based in deep learning by essentially again teaching an algorithm, giving it millions of examples of what these collisions might look like, and training it in advance to extract information about these collisions.
SPEAKER_05One quick follow-up to that. Um the collision of these uh neutron stars and uh black holes, otherwise known as kilanovas. You wrote back a few years ago about how that's how the universe creates gold. Could you explain that for us?
SPEAKER_00Yeah, yeah, that's uh that that was that was kind of a fun project. And it's it's kind of an ongoing question. So um, you know, as we look at at objects in the universe, we look at the at the periodic table, we can ask, you know, where do different elements come from? What kind of cosmic process produce them? Um and um going back to the 1950s, uh there was a fairly good understanding of where the lighter elements come from, you know, hydrogen, helium, and then things like carbon and nitrogen and oxygen, uh, all the way up to iron and cobalt and nickel and things like that, um, we know that those. Come from different processes. Some of these elements are synthesized inside of stars during their life and they're expelled out. Some of these elements, the heavier ones like iron and nickel and cobalt, are uh produced during supernova explosions. And then there was this kind of open question about where do the heavier elements in the periodic table come from? Things like gold and and you know platinum and uranium and so on. And there was a fairly good understanding of the process that produces them, but not where that process takes place. And for I think decades the assumption was well, if you know if supernova are energetic enough to produce elements like iron and cobalt and nickel, then surely uh at least some of these supernova explosions probably also produce these heavier elements. Uh, but there was this competing theory um following the discovery of pairs of neutron stars in our own galaxy that collisions of neutron stars might actually um have the um right conditions to uh in a very short period of time produce large amounts of these uh heavy elements. And um back in 2017, we uh produced the the actual, for the first time, observational data uh that showed that process happening uh in the first neutron star collision that was discovered by by LIGO. Uh and we actually were able to measure how much of these uh heavy elements were produced. And it was really staggering. We found that um there was the equivalent in that one collision of about 10 times the mass of the Earth just in gold. Um so it's you know it's a staggering amount of these heavy elements that are produced in these neutron star collisions. And I think that has kind of revised our way of thinking about these heavy elements, that they're not produced in small amounts in large numbers of supernovae explosions. They are produced in very large quantities in these very rare neutron star collisions, instead. Did you use AI to help solve that problem? Uh no, that was that was brute force using uh an enormous amount of of time on um um more than a dozen telescopes around the world, including the Hubble Space Telescope and so on. So the Philosopher's Stone is a pair of neutron stars? Uh I guess so. Yes. We're now on the hunt for more of these events. Uh, you know, as I mentioned, this was back in 2017. Uh these collisions are quite rare. So um the best way to find them is to wait for LIGO to pinpoint one of these collisions and then observe that same location, not in gravitational waves, but using traditional telescopes, using radio waves, using visible light, using X-rays, using gamma rays, and so on. Uh and so this is kind of an ongoing project. But you know, there these events are rare now because we can only see them to very small distances, uh, relatively speaking. With these next generation gravitational wave observatories, we think we're going to find hundreds of thousands of these collisions. And we will be able to map out the history of how the universe was enriched in these very heavy elements.
SPEAKER_04We've been talking a lot about how these new telescopes were going to extend uh the knowledge of astronomy and astrophysics and so on. What about going the other way? And we're using AI to deal with all that. Well, what about looking at it the other way? This new challenge of all this data, all this uh speed and response time that you need to be able to quick go look at it a different way. Do you can you imagine that this drive is going to actually help AI as well and push some of the capabilities of AI even farther than they are today?
SPEAKER_00Yeah, I think I think that's a real possibility. There's already, I think, um, you know, incredible growth in in the astronomy community in using AI and various machine learning approaches in very different contexts. Across essentially all of astronomy, we've built these observatories and these um space missions that gather enormous amounts of data. And so people are using these approaches to ask questions like you know, if I have a survey that finds a million galaxies and I want to classify them into different types, you know, obviously the traditional approach of doing this by eye and saying, oh, this one looks like a spiral and this one looks like an elliptical is not going to work. Um, so that's that's one area. People have been using AI in the detection and discovery of planets around other stars in cosmology. And so there's there's just this flurry of activity of bringing these methods into the astronomy field. And part of it is just trying to understand what are the questions that are most amenable to that approach, uh, and what are the different techniques within the kind of broad realm of AI that are most productive for different astronomical questions. But I think there is this growing, uh so far it's been mostly astronomers using tools that are developed uh independent of their astronomical use. But I think there is this beginning of a conversation and flow of information back and forth because we might need different tools, or there might be some cutting-edge tools that are only being developed now, and we don't want to wait five or ten years until they become mainstream uh to adapt them for use in astronomy. So, one of my colleagues, uh one of my former students, for example, was um recently hired uh into a faculty position where it's a joint position between computer science and astronomy. Uh so there are more and more of these kind of hybrid positions that are coming into place to kind of enable this back and forth of information and developing the techniques and then applying them to data and back and forth.
SPEAKER_05Do you have any concerns about where AI may be taken by developments in astronomy?
SPEAKER_00Um, you know, I think we're on a little bit more stable grounds than um other applications of AI where people have deep and serious and real concerns about the ethical use of AI and and what its impact is on people. Uh for better or worse, we are dealing with objects that are not human and are well outside of our galaxy and mostly happened millions of years ago. Um and so I I think I think in that sense I'm somewhat less concerned. I think one of the things I'm actually excited about, which is the flip side of that, is that um, you know, if you look at the traditional way that we've done astronomy, a lot of it relies on very expensive um equipment, you know, access to very large telescopes, uh, some of them are owned by individual universities or by consortia. But that essentially has privileged uh a small number of large departments that kind of set the tone for the field. And I think that with something like Ruben Observatory providing its data to everybody, and with software becoming more of a significant player in how we think about data, not just big expensive telescopes, I think that is democratizing the way astronomy is being done. And um, it's giving an opportunity for um both you know astronomy faculty, but also for undergraduate students and even high school students across the country to have access to the same information that you know I have at Harvard or somebody else has at Caltech or at Princeton, you know, and so I I think that's really opening up astronomy to a much broader, much broader participation.
SPEAKER_04I can't help but think that uh you know, like the the the New York Post is gonna look at this data and start declaring, there's an asteroid. So mercifully the all the scientists will uh prevent that. Professor, I some of the folks we've interviewed have been very interesting. Some of them are like you, where I think, gee golly, I wish I was a little bit younger and uh I want your job. Uh this is this has been fascinating, and it's it's caused me to think and wonder all kinds of other questions that we don't have time for. But once again, I'd like to thank you for your time and uh all of your advice and explanations and descriptions. Uh, we never could have done this. So again, thank you very much, and hey, good luck with Ruben, and I hope you uh get a bunch of cool stuff from Webb as well.
SPEAKER_00Yeah, no, thank you uh for having me, and thanks for the um exciting questions. And it's it's always fun to talk about astronomy. Um, so I'm happy to do it.
SPEAKER_04We welcome questions and comments from listeners. Send email to comments at mindthegapdialogues.com. We read all comments and questions and will try to respond in the letters section of a future episode. If you're writing about a particular episode, please mention the specific episode number. And please also include pronunciation tips to help us properly say your name when we reply in a subsequent episode. See you next time on Mind the Gap. We thank the Business Law Section of the American Bar Association for their generous sponsorship of the production of this podcast.
SPEAKER_02Thank you for listening to the AVA Business Law Section's podcast series to the extent that the section offers a robust collection of content. To explore more about this topic or to learn about joining the section, visit ambar.org slash bizlaw. That's B I Z L A W.