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 5 - Can AI Uncover the Origins of the Biggest Black Holes?
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.
A supermassive black hole is estimated to be a million or a billion times bigger than the Sun, but no one truly yet understands the origins of these massive phenomena. In this episode of Exploring AI Matters, Dr. Becky Smethurst, an astrophysicist and research fellow at Christ Church College, Oxford University, describes how AI is helping to answer fundamental questions about the formation and evolution of black holes and their connection to the development of galaxies. [2022-09-08]
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_05For thousands of years, we Earthlings have been fascinated with the night sky and its myriad stars. Beginning with Galileo's first telescope and accelerating ever since, astronomical instruments have dramatically expanded our ability to examine the things we see in the night sky. For example, an old issue of Scientific American, published near the end of the First World War, contains an article entitled The Heavens in August 1918. It reports on a great new nova in the constellation Aquila. A nova is a star that suddenly and for a relatively short time becomes much brighter than in the past. This report notes that the only other nova of modern times had been discovered 17 years earlier in 1901. In a YouTube video dated July 22, 2021, our guest for this episode, Dr. Becky Smithhurst, explains that the classic type of supernova is a massive star 10 times the size of our sun that runs out of fuel at the end of its life. Essentially, gravity starts to crush it inwards. Finally, all of the outer layers of the star rebound off the inner core and get thrown out in a huge explosion. Today the rate of discovery of novas has increased dramatically. One of the reasons our instruments survey so much more of the sky, both in angular area and depth, which means we capture more such events. Greater depth means farther away in space and thus further back in time. For example, the VeraC Rubin Observatory Telescope is scheduled to be launched in October 2022 and to become operational about a year later. It is estimated that the Rubin telescope will photograph the entire available sky every few nights. Experts predict that it will discover three to four million supernovas over its projected 10-year service line. This deluge of data has exceeded the capacity of traditional astronomical analytic processes, in which trained experts, usually graduate students, car over and analyze the raw data. Astronomers and astrophysicists have lately been exploring two new pathways crowdsourcing using systems like Galaxy Zoo and artificial intelligence. The flood of recorded observational data enables us to ask questions about the stellar universe or cosmos that 100 years ago would have been incomprehensible to anyone hearing. Like, which comes first? The formation of a galaxy or of a black hole within it? In this episode of Mind the Gap, Dialogues on AI, we talked with Dr. Becky Smethurst, a research fellow at Christchurch College, Oxford University, whose research concerns questions like do galaxies and black holes evolve together? And what influence does a supermassive black hole have on matter near its event horizon and on the galaxy it is at the center of. We are particularly interested to ask her how astronomers are contemplating the use of AI to sort through and analyze the data. Dr. Smethurst is at the leading edge of the use of artificial intelligence in astronomical research. We encourage you to take an evening and read her short book, Space, Ten Things You Should Know, or watch some of her YouTube videos published weekly that explore questions about stars, galaxies, and black holes. Hello, I'm Roland Trope, a national security lawyer.
SPEAKER_04And I'm Charles Palmer, a computer scientist. We are your hosts this week for this episode of Mind the Gap Dialogues on Artificial Intelligence.
SPEAKER_03In addition, we have two more hosts: Mark Donner, a computer scientist, and Alma Adams, a national security lawyer who can't be present today.
SPEAKER_04Each episode will be led by two of us, with the other chiming in with impromptu questions as they are moved. So, Dr. Smithhurst, I've been looking forward to this for weeks, ever since I discovered your YouTube channel. Please tell us uh a little bit about your area of research and why is it of interest?
SPEAKER_01Um, so my area of research is all about supermassive black holes. So these are the black holes that are millions to billions of times the mass of the sun. And we think that they live in the center of every single galaxy in the universe. And my question is, how do black holes grow? And most people sort of look at me like I'm stupid when I say that because they're like, well, aren't they just sort of the behovers of the universe, you know, sort of like the vacuum cleaners of the universe? But I just think it's really difficult to grow a black hole. And the way that we've thought that happen in the past is that two galaxies merge together and their black holes merge together in the middle, and the two grow together. Um, but my research looks at galaxies that have never merged with another galaxy, and we find that their supermassive black holes are just as big as all of the rest of them. So it's sort of thrown the cat amongst the pigeons a little bit on how black holes grow. Um, that's my research, how I got into it. Um, I like to tell people that I was a why child. Um, I'm one of those kids that constantly is like, why this? Why that? Why this? And I think I never lost that as uh as I grew up. And I think all scientists will probably say the same that you you have to have that natural curiosity to keep asking questions. And for me, the thing that fascinated me the most was space. It was just that the endlessness of it and the fact that you could see it with your own two eyes at the same time, these things that were so far away. And for me, it was always the mystery of it that intrigued me the most. So the fact that I've ended up studying black holes is it's probably because the mystery of it always intrigued me the most. Black holes will always be these mysteries to us because they are these areas of space where there is so much matter that the force of gravity is so strong that light cannot even escape from them. So there's no way we can get any information about what's in inside a black hole, if you will. And so we have to be clever about how we study them. We have to study the stuff around them to figure out what's going on.
SPEAKER_04What a puzzle. Like you, as a child, I had always uh been interested in astronomy and uh uh considered majoring in it. And I don't think I was quite good enough. My telescope was great, though. But I always imagined if I did this as a living, I would be an astronomer up on a hill, freezing to death uh while looking through a periscope late at night. In fact, my mother even said, you'll never meet girls. You will always be on the mountain. Is that how things go now? Are you up on a mountain?
SPEAKER_01You you can become like a professional astronomer that that runs an observatory. So, like, for example, um, there are people at La Palma in the Canary Islands or in Hawaii or the Atacama Desert in Chile that spend months at a time there operating the telescopes. But my role is more that I pitch to use those telescopes for say three nights, five nights in a year. And I might do that once or twice a year, get the data I need, and then squirrel back away to my office to then do the science on it that I want. So there are nights that you spend sort of with the people who are experts in using these, you know, billion-dollar facilities that we build in order to observe the sky. And they are some of the best nights as well. You know, you literally switch to working nights and you get to set the telescope off, um, sort of taking an image that's like 20 minutes exposure or something. And you think, oh, I'll just pop outside while it does that and I'll go look at the night sky. And that's incredible. But the majority of my time, uh it's an office job, right? I'm sat at my desk, I'm at my computer, I'm crunching through data, but it's data that I want to crunch through, right? I want to be at the desk and I want to be there because the the answers you get out at the end is is what you're in it for, right?
SPEAKER_04Well, is that one of the main differences between an astronomer and an astrophysicist?
SPEAKER_01Yeah, exactly. An astronomer is someone that observes the sky and says, Oh, look, there's a supernova that's gone off there. That wasn't there the day before. An astrophysicist is someone who go, okay, we've observed the supernova. It has these properties. What is the physics behind what's going on? I am an observational astrophysicist. So I actually use telescopes. I am an astronomer and an astrophysicist, but I have colleagues that probably couldn't even point out to you where Orion is because what they do all day is they they simulate the universe uh on computers, or they're very theoretical and they're working with equations on a blackboard or a whiteboard, I guess, these days. Um, and so you know, there are astrophysicists solo and there are astronomers solo, and then there's some of us that sort of sit in the middle between between both of those.
unknownGreat.
SPEAKER_05And you mentioned that you pitch to get a few nights a year. How how can that few, that small sample give you so much to be able to then squirrel away, as you say, in your office studying it?
SPEAKER_01Yeah, so that would be in telescopes that are designed for that. They're designed for what we call follow-up from huge surveys. So, for example, one of the largest astronomy collaborations is the Sloan Digital Sky Survey, um which is a US-based telescope in New Mexico. And it's been running for 20 years, essentially just observing the night sky night after night after night after night, taking images, little postage stamps that it adds up, gets more and more light from each part of the sky each night. Then we can catalog that and say, okay, if you're interested in something specific, you can find them in that catalogue. But that might not be the observation you need. You don't just need an image, you might need what's called a spectra where you take the light, split it through a prism, and look at all of the individual colors and wavelengths, the fingerprint of light from that galaxy or that star to tell you more. And so then that's when you might pitch from another scope, say, I have picked out these 10 things or these 50 things in this survey that I then want to follow up more details. And that's what you'd pitch for a telescope for. And sometimes, you know, the the newer telescopes or the newer instruments are very oversubscribed. Like the number of nights they have to give and the number of like nights that people apply for is sometimes, you know, 10 times oversubscribed for some of the more popular telescopes. So it's competitive as well.
SPEAKER_05What would you say is the most remarkable discovery by astrophysicists in your lifetime?
SPEAKER_01In my lifetime, there's a to be fair. I there's a couple I could pick. I think for me, the discovery of exoplanets, the fact that there were planets around other stars in the Milky Way that are not the sun, was monumental. I think everyone had always sort of assumed they had to be there, but actually finding it in the mid-90s. And then by the late 2000s, when the Kepler Space Telescope was launched, finding thousands of them after that, that was monumental. I don't remember it though, because I was a young child when the first one was found in something like 1995. I think I was probably watching Toy Story or something instead. The one I remember that I think is the almost bigger than that in a way is the discovery of or the detection of gravitational waves. So this idea that space is actually a thing, that you know, mass can curve it. And if you have two big black holes that merge together, they're gonna send ripples through space. And we can detect those ripples as sort of space compressing and stretching. A, the fact that we can detect that is amazing. And B, it was the first time we'd ever observed something in the universe, got information about something in the universe that wasn't in the form of light. It was a whole new way of observing the universe that we had never had access to before. And I think almost that's more monumental in the fact that, you know, because oh, it's the idea of other planets exist, or we're not that special. It's almost like just a reminder to us. But this was a whole new way of observing the universe that we never had before. And I think that will revolutionize the field more than anything else.
SPEAKER_05And it changes your research also?
SPEAKER_01It will, yeah. It has some implications. Um, not so much my area of black holes, because I went with supermassive black holes. The waves that they detected were from um what we call stellar mass black holes. So, you know, maybe like 10 to 30 times the mass of the sun, rather than millions or billions of times the mass of the sun. But it has implications for like, okay, well, how how small of a black hole can you actually have form from like a supernova or something like that? You know, instead of getting what's called a neutron star instead, it tells you what that limit is, um, which is really interesting in sort of all the ideas of how do black holes form in the first place.
SPEAKER_05Now you said something just a few moments ago about black holes uh merging or colliding. And I have to say, when I was a young child and I loved looking at the sky and tried my hand at amateur astronomy, I never would have imagined that anything I was looking at either had in the past or would in the future collide. Could you explain something about that galaxies possibly collide and maybe even our galaxy with Andromeda Galaxy might someday collide?
SPEAKER_01Yeah, so our nearest, our nearest big galaxy neighbor is Andromeda. If you live in the northern hemisphere, you will be able to see it. It's sort of a fuzzy patch in the sky, which is amazing that you can see that. Um, but it's actually about the size of sort of like six full moons across in reality, if you could see the whole thing. And it is coming towards us in the next two to five billion years or something, it will collide with the Milky Way. Now, I use the word collide very loosely there because space is very big. Wasn't that what Douglas Adams like? Space is big. So you won't believe just how mind-boggingly big it is, right? The space between stars compared to the space that the stars in the galaxy take up is tiny. It's almost like you know how you learn how the majority of an atom is empty space, like 99% of an atom is empty space, right? The same thing is almost true for a galaxy. The majority of it is empty space. So when the two collide, the better word is probably merge, because no two stars will collide when the galaxies collide. Like I think I worked out the number, like the probability ones for a YouTube chat, for a YouTube video, and it was astronomically small. It was tiny. Um, and so when the two do come together, it'll be more like a redistribution of everything that's in the galaxy, um, just sort of getting sloshed around a little bit. But black holes can collide, can actually fully merge and become one black hole. And that's usually because so our sun is actually quite rare in the fact that it's by itself. So 85% of stars are not by themselves. Um, so it's quite rare in that fact. Most of them are in what we call a binary or a tertiary system of two stars or three stars all orbiting each other. So you can imagine if you've got two massive stars that form together and are orbiting each other, and they both go supernova and become black holes one day, those black holes are also still orbiting each other. And so that uh eventually those two things having such strong gravitational fields, they will merge together. They'll spiral in and merge together one day over millions and millions of years, because these processes take a lot of time. Um, but that is what we think happens, and that sort of mitigates the problem of space being very big and the idea of collisions being very rare, because you've already started with the system that that makes it so that that um that merger will be very likely. And it's why we've, I think we've discovered over 30 or so now of these instances of either a black hole merging with another black hole or a neutron star merging with another black hole, a neutron star being like the pre the pre-stages of a black hole. If you add enough mass to a neutron star, it will collapse and become a black hole.
SPEAKER_05Could you give us an idea of how much information astrophysicists will be gathering once the Rubin and Webb observatories are in space, operational and collecting data, assuming all goes well?
SPEAKER_01Yeah. Um, so the Rubin Observatory is actually a ground-based telescope. Um, it's actually going to be uh yeah, in in Chile, which is quite exciting. Um, and so um it still will collect a lot, a lot of data. So the survey um that I mentioned before, the Sloan Digital Sky Survey is the one we have now. That surveyed the northern hemisphere sky to you know a certain depth, not very deep, you know, just getting sort of the nearby stuff. And that gave us a million galaxies, which was already far too many galaxies for a person to look through. And when the survey was completed, sort of mid um 2000s-ish, that was far too many for the AI at the time to also classify and could not do a good job on it, basically. Um the Vera Rubin observatory, um, which is hosting the large synoptic uh sky survey um is thought to be going to be able to detect about a billion galaxies. So, I mean, that's not just an order of magnitude, that's a that's a couple of orders of magnitude larger, right? Even if you can classify all those galaxies, the shape of them, um, and you manage to get 99% of them classified to a high enough standard that you're happy with it, that's still a million galaxies that you've got left over. If you've got a sample of a billion. Just because the telescope is is observing um further out, you're gonna collect more and more lights. You'll see fainter things at larger distance in greater detail as well than we've ever been able to see before. So it's a lot of data. It's a huge amount of data. It's the point where people are worried about where on earth are we gonna store this data and how are we going to let people access it at any time from anywhere in the world that's scalable? It's a very interesting problem. One that raises, yeah, not just data storage, but also what are you gonna do with a billion galaxies? Are they actually all going to be eyeballed at some point by someone? Because a lot of the stuff that's in the Sloan Digital Sky Survey has still remained unyeballed because it's been sat on computer archives, because a million is a lot. So there's the question of you know, this idea of big data reducing stuff to numbers, and you can work with statistics and population, but you you can't help but wonder how many of those will actually be seen by human eyes.
SPEAKER_05So, with all of that enormous amount of data, which is harder, I think, for most of us to even quantify or or imagine in our heads, when did astrophysicists start to use AI to grapple with all of this data? And how have they done so?
SPEAKER_01Yeah, so very early on, I think astronomers were sort of an adapter of AI for what it could do. I think because you know, we've we've been simulating the universe for so long, it was one of those sort of steps of like, okay, we are the imaging science. So using AI to classify images is is kind of the thing that we should pioneer in a way. Um the problem that you have is that imaging specifically is very difficult for machine learning and artificial intelligence algorithms to do. You know, I could show you a picture of a zebra right now, and a four-year-old would know that it was a zebra and not a lion, but a machine learning algorithm doing that with images is very difficult. It's why when you have those I'm not a robot checks on the internet, that the recaptures, they're always images because we're still trying to train machines to recognize certain things in images. Um, and so to train a machine, um, you need uh huge, what we call gold standard data, sort of expert data that's classified that it can be trained on to say, this is a spiral galaxy, this is what you should be looking for. This is an elliptical galaxy, it's a big blob, but this is what you should be looking for. The thing is, there's so many uh in-betweens with images that you can have that don't fit into those nice, neat little categories, which is where a machine struggles. So there was uh a citizen science project called Galaxy Zoo that was set up back in 2007 to actually classify the shapes of those a million images of galaxies with the public's help. And, you know, that thought it would be about 10 years before that would be done before getting through all a million images, and it took something like six months. And that was fantastic because that solved that problem of well, we can't have the machine do it. Okay, the public has managed to do it, and 20 odd people have looked at each image and classified all a million galaxies, and great. That also then means that you have this training set. And so um, there was what's called a kaggle competition. Those who are very um big coders will will understand and know what a Kaggle competition is, but it's essentially a competition between coders to say who can write the best code to do this. And there was essentially one of these to say who can write the best machine learning algorithm to classify these galaxies to within agreement with. With the public who've previously classified them of say, let's say over 90%, 95%, 97%, how high can you push it? And everyone who entered that competition was finding that the training set just wasn't big enough. They were they were saying a million galaxies is not enough to teach the machine to what to look for. And the person that won it actually realized that the classification of an image didn't change if the image was mirrored and reflected. And you could do that, you could do that sort of four times in sort of like a rotation of the image and still have the same classification. And so he managed to boost his training set to four million. And that was just about enough to get you somewhere in the range of like 95-96% agreement with the public's classifications of it. And so I think that probably helps understand what where the struggle is coming from. It's just that there's so many different nuances and images, especially with astronomy images, where you could have a satellite fly across the image, or you could have a cosmic ray, which is very high energy particle from space, hit your detector, and you could have this big blowout of on your detector or something like that. You could have two galaxies merging, which then affects your image as well. Because then the computer's like, which which one do I um which one do I classify? And then obviously the biggest thing with machine learning is what about the things you don't tell it to look for? You know, all are those things gonna remain undiscovered in the data because you only told it to look for certain things. And the things that look weird, the computer won't know that they look weird and so won't flag it like a human would. Um, so I mean that's kind of the the difficulties that we have in astronomy.
SPEAKER_03So, how confident are we in the quality of the classifications that were done by the by the humans in this crowdsourced thing? I mean, that's a I mean, you didn't eyeball all million of them.
SPEAKER_01No, exactly. So that was a huge thing as well. Um, so you can almost it's there's like two stages to this. You have gold standard expert data. So someone like myself would classify a few of them, and you would see for each person, you would show them that image, do they agree with the expert classifiers? And each user essentially has like a trustworthy score, if you will. It's not a score of how well they're doing, it's just that you can really easily filter out trolls this way. So you have a couple of different types of trolls, a troll that comes along and randomly answers. And you can pick those out because they'll never agree with the experts, and they'll also never agree with the crowd hardly either. So you know to downweight their classification basically and not sort of think it's that important. There's also trolls that will come along and say the exact opposite of what it is every time. And hilariously, they're really useful because all you have to do is flip their response and you care the right answer and you can add the crowd. This is also why you get more than 20 people to look at each image as well, though, because then you have statistics on each image. So sometimes an image is too hard to classify, you know, it will be something very fuzzy, something very far away. You can't really discern any detail. And you can usually pick that out in the classifications because the vote essentially has been split across the crowd. You know, we call it crowdsourcing, these classifications of these images through citizen science. And it this is really what it is. So there's lots of different ways that we can A, filter out classifications, or B, know when something just isn't classifiable because of the nature of the image itself. And those unclassifiable ones tend to be the ones that computers struggle with as well. Um so you know, there's lots of things that you can do to make sure that you know you you you can trust the classifications you have.
SPEAKER_05When you speak of classifications, though, is this a fixed set of categories or is that set constantly growing as you look at more and more varieties of constellations?
SPEAKER_01Uh so we're not looking at so constellations is is just a No, I didn't mean constellations, I mean galaxies. Okay, sorry. Um yeah. So as we with Galaxy Zoo, what we did was we um provided people set answers. So we'd ask specific questions, is this a spiral galaxy or is it smooth and roundish? And people would have to decide, and then you you'd go through sort of like a decision tree, you know, like a Cosmo quiz almost like this or this, this or this, right? Um and you and they people would make their way through that. With machine learning that we use, obviously, then a machine is trained to then classify with the exact same um classification system. That would obviously be trained machine learning. There's something else you can do though, which is um unsupervised, it's called, um, rather than what I just described it which was supervised learning, where you tell it to look for certain things. Unsupervised learning, you essentially would just feed everything you've got to an algorithm, and the algorithm would decide how to group things. And so it would hopefully group things as here's all the spiral galaxies, here's all the blobish galaxies on sort of like two different clusters. Um, so there's different ways you can go about this, but with what I do, we're really interested in specific shapes because we know that that's how we'd classify them. It's how Edwin Hubble classified them back in the 1920s, and it's sort of stuck um in the same way that the biologists have their, you know, classification animal tree, everything like that. You know, we have a similar thing for galaxies. And so it's we want to know those classifications because we're looking for specific things. There are other people though who are like, I want to find the weird stuff that doesn't fit into any um cluster that a computer can find, and it will spit out. Here's all the weird things. And so that's where you would maybe then use something like unsupervised learning.
SPEAKER_05And is there some sort of set number of these classifications, just a ballpark figure?
SPEAKER_01Yeah. Um, so the the first iteration of Galaxy Zoo was a million images classified at least 20 times by each uh classified by 20 times and like 20 different people. But then there's a certain amount of clicks, obviously, within the decision tree that you have to make your way down, the maximum of which I think is about 12. So it's at least 20 million classifications, but number of clicks is gonna be what 240, no, 2400 million. So I guess that's 2.4 billion in in US speak. Um yeah, it I mean it's a crazy number of of clicks that that was sort of crowdsourced. Um, but in terms of like classification, if if a computer or an algorithm was doing this, then yeah, it would be if you had a million galaxies, you'd get a million classifications out the bat. But I mean, the decisions to make that, you know, if you're thinking about algorithm jumping around stuff, perhaps following a chain, then then you're talking way more than that.
SPEAKER_05Could you compare for us? Sorry, Mark, did you have something you wanted to ask me?
SPEAKER_04Go ahead, that was me. Go ahead.
SPEAKER_05Oh, sorry, Charles, go ahead.
SPEAKER_04You um you're sounding more and more like an AI researcher because uh these are the kind of topics I discuss with my students. Sure. Um, so how exactly, I mean, is this how you're using AI in your work? Or uh is it a little different?
SPEAKER_01Or at the minute, stuff has changed actually. So um, this was very early work with Galaxy Zoo and and everything we were doing. Now things have changed a little bit in the fact that we are utilizing the machines and humans together. So now we have an AI on the back end of the Galaxy Zoo website where we ask people to classify these images. Because, you know, even though those first million were done, us astronomers are not going to stop taking pictures of the sky. So it's constantly more that needs to be classified. Um, and what we have is a machine working in the background. And this is um, for those who uh speak machine, it's a convolutional neural net. Um, and it's active learning. So the machine is actively learning as more classifications come in all the time. It already has learned from a different classification set that might be from a different telescope or something like that. And it's already probably gotten rid of the ones that it is very confident that it can classify. So the things that are very obviously just big blob of a galaxy, you know, nothing special, just quite round, very smooth, very flat. And the things that are very obviously giant, big, beautiful spiral galaxies. You know, it will be very confident of its classification. And this is the thing that's so special about this algorithm is that it does give you a confidence level on its classification as well, because of this active learning. And so those, it will probably never show to a human because it's like, I don't really need the help with this, I've got it. The machine will make a decision, though, about what galaxies to show the humans and to put on the site. And the fun thing is that, you know, as we launch this, people worry, well, I'm not gonna see all of them. What if I miss something cool? The cool ones are usually the ones that the machine can't classify. It doesn't need help with the boring, fuzzy blobs. It needs help with the stuff that looks weird and wonderful that it's like, I have no idea what this is. I need a human to look at this. And so this is how moving forward we think we're gonna tackle something like the Veer Rubin Observatory that we think is gonna give us a billion galaxies after the survey is complete. Okay, maybe the machine can do 99% of them, of the easy ones, leaving still a million very more difficult, weird, and wonderful things for the humans to look at. And so this is um what we're doing right now with Galaxy Zoom. And it's really helpful to me because the stuff I study is the galaxies that haven't had mergers. And you spot those because they're not these big blob of galaxies that have been formed when stuff or gets rearranged in a merger. They're the beautiful, pristine sparrow galaxies that have been left alone their entire life. And so picking those out, they're very rare, is obviously something you need either humans or a machine to do. And at the minute, out of those a million that we had originally, we have a hundred of these things. Just goes to show how rare they are. So the fact that we are getting a sample that's not northern hemisphere, like the original survey we had was, but southern hemisphere with the Vera Ruben Observatory, different galaxies, a billion of them, not a million of them, it's gonna increase our sample size a huge amount. But picking them out without, you know, this machine that we've trained on all the previous surveys to do this was going to be so difficult. And so we we can see this future where it's sort of like, you know, humans and machines working together. Sort of you got this friendly robot behind the scenes being like, I got this, don't worry.
SPEAKER_04Well, that's sort of like uh the the folks who say AI stands for augmented intelligence, where it's a combination of uh human and machine. Uh I have to ask, how many people were involved in the zoo uh thing?
SPEAKER_01Yeah, the Galaxy Zoo. So Galaxy Zoo actually was the first of these citizen science projects. It actually spawned a huge number of them called the Zooniverse, um, which has everything from astrophysics projects to zoology projects about the Serengeti and penguins to now historical projects where they've scanned in like World War I diaries and are asking people to what's the word?
SPEAKER_04Uh transcribe.
SPEAKER_01Transcribe, thank you. So everything to like historical documents that have been scanned in, you know, World War I diaries where they're asking people to transcribe, to digitize what they see, right? Uh in terms of dates and locations and everything like that. Um, so the Zooniverse is, I probably can give you a number of researchers that are involved. Um, there's probably thousands around the globe. In terms of the number of volunteers, the Zooniverse is over a million. I think it might be 1.5 million now. It's it's a huge number of people worldwide. Galaxy Zoo, I know, is about 300 or that 300,000 people that volunteered worldwide to do the classifications. In terms of number of researchers, we're probably around about 30 that have various different projects. And they're based um in the UK, the US, uh, Australia as well. You know, real sort of like international uh project, also in Europe, too, uh in the Netherlands. In the US, we're based uh at the Adler Planetarium in Chicago, is sort of like the headquarters, uh, if you will, for those who've been. Um, but it's also people in uh Arizona and Minnesota as well. Um, so it's a really, really great collaboration, actually. Um, and I it's a really friendly collaboration as well. Um, and a lot of astrophysicists, but also we we've brought in, you know, some computer scientists as well, with this notion of, you know, we want to make sure we do this uh AI machine learning stuff right.
SPEAKER_04Yeah, you're eating my mind again. I was about to ask uh, how closely are you collaborating with computer scientists to push forward?
SPEAKER_01Yeah, very so um we just had a PhD student finish up at Oxford whose project was essentially machine learning with this universe. Um he essentially wrote a lot of this back-end code for the the friendly robot at the back of the website that decides which galaxies to show people. Um that he was jointly supervised by um someone, a professor in the astrophysics department, but also a professor in the professor, but also a professor in the computer science department as well at Oxford. So that was a really nice joint collaboration um where he really got to do both of getting to grips with all the astrophysics side of things and the astronomy side of things, of like what's important in terms of these images, the image processing, um, you know, are all the artifacts in the image, what's real, what's not, that kind of stuff. But then also the computer side, computer science side of things as well. Um and he'd applied his his knowledge that he gained in this PhD, not just to this, but also to um if you heard of fast radio bursts, it's sort of the the biggest mystery at the minute in astrophysics, just trying to explain these. And there's a telescope in Canada called Chime that's essentially been, you know, switched on for a couple of years now, just listening out for these fast radio bursts at all time, picking them out though, of just that noise of radio data is very difficult. And again, you know, machine learning can be used there to pick those out. So there's lots of different applications in astronomy, and we were, you know, very lucky that um we could a, you know, fund a PhD student to do something like that, but also get someone to do it for this problem of of you know, we're gonna have too much data in a couple of years.
SPEAKER_04Well, one of the one of the cool things that seems to happen when you start applying AI in new areas is that you discover, and you may have mentioned this earlier, um, you discover weaknesses in the AI with or or techniques that uh might work for identifying zebras, but not so much for uh you know black holes. So are are there weaknesses that uh machine learning is, I mean, it's not just benefiting astrophysics, it's benefiting machine learning too.
SPEAKER_01Yeah, one of the big things that we noticed was that um some of the techniques that can be applied in other areas can't be in astrophysics because, for example, image machine learning can learn a lot on color. And the thing is, color is very important in astrophysics because it tells you a lot about the astrophysics of what's going on in the galaxy. If you have a very blue galaxy, that means it's full of a lot of blue stars. Blue stars are the hottest stars, the youngest stars and the biggest stars. So it tends to mean you form some stars recently if you've got some young stars in there. Red colors, or what we call red, it's an astronomer's red, it's it's very yellowish, really. Um, is is more of a sort of a dying embers of the fire. It's the smaller, cooler, older stars that live longer. Um, you know, the stars are very much like live fast, die young. If you're bigger, you burn through, you have to burn through your fuel quicker to resist the stronger force of gravity crushing in. So even though you've got more fuel, you have to use it quicker. And your lifetime is something like a hundred thousand years, as opposed to a much smaller star like the sun, which will live 10 billion years. Um, so when you see yellow, you think, okay, well, that's not formed a lot of stars recently because all the blue ones, the young ones, have died off. And the thing is, if you see something that's maybe a little bit far away, so it's a little bit fuzzy and it's yellowish in color, people would probably assume that it is also not a spiral galaxy. Because the other thing that we see is that when two spiral galaxies merge and they become this big blob of a galaxy, this is what we call elliptical galaxies, or everything's been redistributed. A lot of the gas that makes stars, the hydrogen gas, gets used up in sort of a big bender of star formation, essentially. And you get left with what we call a red and dead galaxy that is also elliptical in shape. So color can bias your shape classification. If you see a reddish color, you might think, oh, well, then that must be one of these elliptical galaxies and not a spiral galaxy. And for a long time, people were biased by that color. And it was only with the first iteration of Galaxy Zoo that we were like, actually, 30% of red things are spiral galaxies. And that was because the we didn't tell the users, the volunteers, anything about color. So they did not have that bias going in. And so therefore, people could pick this out. Again, it was part of the sort of big population of things as well. You had that benefit of being able to pick out those big numbers. But still, that was a lot of you know, a lot of galaxies, more than people thought. Machines also learn based on color. It's part of an image, it's part of the information that is in an image. And so again, if you're especially if you're feeding it classifications that are perhaps also biased by color, then the machine learns the same bias. So that is one thing that we've we've struggled with with astronomy is that if you have that kind of color information that that can bias um things. Also, as well, you can be hugely skewed if something, you know, if if you're only focused on, say, the galaxy in the middle of an image, that's what you care about, but it's in a very dense field of galaxies, like it's perhaps in a very dense area of the universe where you have like a cluster of galaxies together. How do you tell a machine to just focus on the one thing in the middle? It will get constantly distracted. Um the other thing about those dense fields as well is that they act as these big lenses. You know how if you if you take uh a wine glass and you have that stem with the with the ring at the bottom, and you can put it in front of a candle, and you can see the the the flame from a candle make this beautiful ring because of the the wine glass stem acting as a lens. Galaxies can act as the same thing, but instead of you know glass, it's it's gravity that's doing the bending of the light. And so they bend the light from more distant galaxies. More distant galaxies are from the earlier universe, whether it was denser, there was more gas to make more stars. So the galaxies are bluer. And so you end up with these blue arcs in your image. And a machine learning algorithm, if you just feed it that thing, it's like, oh, something blue, it must be a sparrow galaxy. And you're like, no, no, no, those are those are the these are these arcs of these background galaxies, you know, these are these lensed objects that that aren't galaxies in the cluster that you're bothered about. And so there's so many of these different biases that you have to try and mitigate in your algorithm somehow, or somehow take into account and get rid of these biases. And that's quite difficult uh in AI.
SPEAKER_04Yes. And in fact, I saw an article um uh the beads, the little uh little diamonds that uh Einstein predicted uh we're just seeing for the first time. Fascinating. So, okay, so you're you're you've found some some weaknesses in uh machine learning techniques and and they're learning from you as well. If you could pick an improvement or two, uh, what sort of improvement would you like to see?
SPEAKER_01Hmm. I mean, it's funny because we're reaching the point where we can have 99% agreement with like, you know, a public or expert classifications. And you think that that's, you know, what what kind of improvement are you are you getting at that stage? It's that idea of, you know, it's it's it doesn't have to be perfect. You know, don't strive for perfection, strive for done. Um, but at that stage, as we're talking about these uh surveys that are of a billion galaxies, and 99% need you'll still have a million left. It's almost like how can you reach that improvement level where you where you reach a number that isn't quite as scary as still having a million left to classify? We know it's doable, um, but if it's scalable in that sense is is difficult to know. Um I definitely like to see improvements on picking out much fainter features um in images. So one of the things people look for is yes, okay, you've got the glaringly bright galaxy, but perhaps you might have something like um the Magellanic clouds around this galaxy. So for those in the southern hemisphere won't know what I'm talking about, these are two little dwarf galaxies that orbit the Milky Way, and they're slowly sort of being torn apart by the Milky Way's gravity. But they're so much fainter. Like someone, you know, observing who's in Andromeda right now. Say they're in a planet in the Andromeda galaxy, looking back at the Milky Way, would probably find it very difficult to see though those trails, right? And so you have to observe for a very long time to be able to see them. Getting an AI to look for them is difficult when they're that faint. You you obviously have to feed them, feed and train the AI on you know, very obvious examples, very faint examples as well, and getting to pick that out. I'd I'd love to be able to see an AI that could pick out something like that, or even pick out the weird things. Like, here's something that doesn't look like what you told me to classify. You know, I think it's those the the unknown knowns that are buried in the data that it would be great if an AI could pick out, but I I just don't know how it would do it.
SPEAKER_05I want to go back again, I guess, to what's familiar for me, which is what I remember as a child. I would look up at the sky and I would think, I'm looking directly at anything I'm seeing. There's nothing between me and what I'm looking at. But from your description, there we are sometimes looking at one galaxy through other galaxies. How do you train AI to deal with that?
SPEAKER_01Yeah, that's another thing, is it's almost like distraction, isn't it? So there's a really famous object. Object called Hogue's object. And it's a very strange galaxy. It's what's called a ring galaxy. It has sort of this core to it, and then it's surrounded by this big bright blue ring. And it's perfect. The ring is unbroken. We see it completely face-on as well. And it's an incredibly rare type of object. And inside that ring, in what you would see, think of like as the empty space between that sort of core and the ring, there is another ringed galaxy in the very distance. It's phenomenal. And you wonder whether, okay, well, if you trained a machine on that image, would a machine only classify a ring galaxy if it had another one, you know, in the empty space of the ring? Or would it always miss those little ones in that empty space and stuff like that? So that sort of like distraction of an AI, or you know, maybe perhaps you're looking for something specific in a galaxy, but it was been missed because it's merging with another one. I think distraction is a huge, a huge thing in AI. The fact that you could be looking for one thing, but because it's near another in the sky, it gets ignored.
SPEAKER_05And I have a question also that just comes from my fascination with cameras. Different cameras and different lenses render light differently. In creating these data sets, do you have to deal with biases that may enter because different telescopes gather the light, different cameras were capturing the image? How do you do to what extent does that cause a problem? And how do you deal with it?
SPEAKER_01I mean, that's basically like the the history of astronomy and the history of CCDs. It could be here forever, right? One of the main things is um is aberrations through the lenses. So you don't get a perfect image. You might get what's called an airy disk, where you get sort of rings surrounding your main image. That's a huge problem, but that's been solved for a long time. It's something that Fourier transforms, the magic of Fourier transforms, uh, anybody who knows what that is will know they are magic, um, can remove. Um, and so there's lots of things like that. There's also obviously atmospheric turbulence to deal with when you've got a ground-based telescope. You can think of even, you know, when the Hubble Space Telescope was first launched, the mirror was deformed. And so it wasn't giving us uh, you know, a clear enough picture because the light was bouncing off something that was deformed. There were ways to correct that, but it still wasn't great. But one thing that amazes me now is something called adaptive optics, which are on a lot of telescopes. Have you ever seen an image of a telescope that's firing a laser into the sky? And you're like, why is that telescope firing a laser into the sky? It's what's it's basically creating a fake star. And we know that that laser should be a perfect point of light that the telescope will observe, right? But instead, it will be distorted in some way, and that's because of the atmosphere that it's traveling through. Your turbulence on a plane, right, is is caused by different pockets of air that are of different temperatures and pressures, and that the light gets, you know, disturbed in the same way as it passes through that air towards us. And so if you can record the disturbance that the laser is experiencing going through that atmosphere, and you can correct your mirrors to account for that distortion. You can literally deform your mirrors to be like, oh, well, that light's gonna bounce, and then that'll give us that clear picture of the laser again. You know that the light coming to a telescope from something else, from a star, from a galaxy, is gonna have the same distortion on it and then be corrected in the exact same way. And usually that had to be done after the observations, but they can now do it in real time. They have real-time deformable mirrors on the back of telescopes to account for that kind of stuff. And again, that's that's usually like an algorithm that's running in the background making those decisions. And that still doesn't fail to blow my mind every time I think about it.
SPEAKER_05It sounds like a partial answer to the question I'd like to ask next, which is what are the most exciting new instruments coming to your field in the next several years? And how will they improve the data situation? For example, you mentioned in a May 2021 YouTube talk that the James Webb telescope is set to quote revolutionize nearly every field of astrophysics. Could you tell us why and what other instruments might be on the horizon?
SPEAKER_01Yeah, so James Webb is is going to be a huge one. Um, it uh is sort of been plugged as like the successor to Hubble, but it's nothing like the Hubble Space Telescope. First of all, it's much bigger than Hubble. Hubble is only um a mirror of two meters across, so that's its collecting area. I love the t-shirt, Mark. Nice. Mark's got a James Webb t-shirt on. Um Hubble is only two meters across, whereas James Webb has a six-meter telescope, six and a half meters. What that means is that its collecting area for light is bigger. So it can collect more light, which means it can see fainter things. So that's great. You see fainter things with James Webb than you do with Hubble. Hubble looks in optical light, mostly optical, a little bit of UV, and just brushes the infrared part of the spectrum as well. So the bits that sort of um, you know, bookend the optical visible part of the spectrum that we actually see with our eyes. James Webb is just infrared. And most people are like, why then? It's seeing stuff that we actually wouldn't be able to see with our own eyes, but actually it can tell us way more information with infrared. First of all, dust blocks infrared. And when I say dust, I mean like gas that has heavier elements in it, like iron and carbon that might have clumped together to give you something bigger than just an atom, right? It's a little particle of maybe carbon clumps or something like that. The the less glamorous term for it supernova poop, because this stuff's made in stars and then thrown out by supernova. Um, that's what I call it anyway. So dust blocks visible light. Uh, it doesn't block infrared light, though. The infrared light just goes right around the dust, the supernova poop. So that's great. It's gonna let us see inside very dusty regions. Dusty regions are where stars form. So that's one thing that James Webb is gonna be able to do. Also, because the universe is expanding, light gets red shifted. So it's galaxies in the very early universe give off their light in visible wavelengths. By the time it reaches us, it's infrared. So there's literally a limit to how far you can see with Hubble, because at that point you won't receive any more light from that galaxy because you physically can't see it anymore because it's all in the infrared. So James Webb will be able to see way back further than Hubble ever has been able to, and we hope back to sort of the formation of the first stars in our universe. It also in the infrared can pick up um water absorption features. So all elements and atoms and molecules absorb a specific wavelength of light, it's sort of like they're the light that they're akin most to, right? It makes it gives them the energy they need at a very specific wavelength to to vibrate in a certain way. Um, and so if, for example, you're observing the light from a star that has passed through the atmosphere of an exoplanet orbiting around it, and just that little wavelength of light that we know water absorbs just disappears in the light that you detect with the James Webb Space Telescope, you know water is there on that exoplanet. So everything from planets to seeing the most distant things in the universe is what James Webb is going to do. And this is why people are so excited about it. Uh, it now has a launch date of the 18th of December. So that'll be a very nice Christmas present if uh all things go smoothly for astrophysicists. And James Webb, I think, is the one that gets talked about the most. But we have so many other facilities coming online, the Vera Rubin Observatory that we've talked about, but also the extremely large telescopes, both the the European um and the one that's being uh built in in Maunicaia, that's had um some legal uh troubles to it, rightly so. And um also on top of that, you've got a lot of uh radio telescopes coming online very soon. The square kilometer array is coming online very soon uh across South Africa and uh Australia. They're gonna revolutionize radio astronomy like has never been seen before. Like the most incredible amount of detail on things that give out radio emission, which are some of the most energetic things in the universe as well. Things like black holes and neutron stars and stuff like that give off radio waves. So that's gonna massively transform things. And then the one thing that I'm very excited about, we have these gravitational wave detectors on the ground at the minute. They're basically big L-shapes of mirrors that detect when the um distance between those mirrors gets squashed or shrunk by gravitational waves. But they're only sensitive to certain frequencies of gravitational waves, the frequencies that are given off by stellar mass black holes merging. But supermassive black holes merge too, but give off different frequencies. But to detect those, you need a bigger detector, bigger than you could probably build on Earth. So we are planning to build one of these huge big L-shaped detectors in space. It's gonna be called the Lisa Observatory. I can't remember what Lisa stands for off the top of my head at the minute, but it's a collaboration. I think it's the ESA's uh uh observatory that they're planning to build, the European Space Ugency. And I'm so excited for it. The idea that we could detect gravitational waves from supermassive black holes, my my babies, what I study first, but also that we'd have this huge contraption in space that we'd we'd manage to build. And that's quite far in the future. I don't think that's gonna be until the late 2030s that that's that's really realized. But the fact that that is sort of being decided on now and and sort of being pushed forward is is really exciting.
SPEAKER_05Because Web has had so many delays, I have to ask just a couple of other questions about it. Um, how is it gonna filter out the infrared noise creation from the Earth and from, of course, our sun?
SPEAKER_01Yeah, yeah. So, first of all, from Earth, it's actually going to be many millions of kilometers away from Earth. Um, so that helps. It's at something called L2, which means it's essentially going to trail Earth in its orbit around the sun. So it'll be protected from Earth's infrared source. Of course, the sun is not something you really escape in the solar system. Um, so it also is equipped with a essentially giant sun shield, is what you can call it. It's about the size of three tennis courts, it's huge, it has to be folded up inside the rocket, and when it gets to the location in space that it's going to orbit the sun at, it will unfurl very, very slowly. And this is one of the reasons that people are quite nervous because things could go wrong with that unfurling, and it needs that sun shield to operate because all of the instruments that detect this infrared light are cooled down to you know a couple of Kelvin essentially to be able to function. Um, and so shielding and that shielding actually working is very, very important for this telescope. And it's, I think it's many layers thick as well, this sun shield, and it should be able to filter out all the infrared from the sun so that that does not interfere with its observations.
SPEAKER_05And is space sufficiently large that we don't have to worry about things hitting it and puncturing it?
SPEAKER_01Yeah, exactly. Um I mean, it's very small, first of all. It's not like Earth, where Earth might have a stronger gravitational field to pull things in that might be uh nearby, little asteroids, things like that. But also they do have some control over it. So if we think that there's something on a trajectory with it, we could always give it a little nudge and a little push either way with some little thrusters as well. Um so I don't think that's anything we we have to worry about, she says. Fingers crossed.
SPEAKER_05Well, I'd like to circle back to the beginning of our talk because we mentioned this question in our introduction. It's now thought that there is a supermassive black hole at the center of every galaxy in the gravitational driving sea of the whole star system. So, which came first, the galaxy or the black hole?
SPEAKER_01I wish we knew. And this actually might be something that James Webb, the space telescope, can can tell us actually. So, I mean, this is this question comes from sort of thinking about the very early universe, the first stars forming, the first galaxies of stars, these collections of stars. What pulled together that collection of stars? Did the gravity of the stars themselves keep them bound? And then did one star go supernova and become a black hole and sink to the center of that system and became the thing that everything orbited around? Or did say a gas cloud in the very early universe collapse straight down into a black hole and then that shepherded a load of hydrogen gas around it that's the first stars formed from and created a galaxy that's always been orbiting around this black hole. It's the astrophysics chicken or the egg, right? We're not entirely sure which one came first. And the only way we know of that you can make a black hole is if something goes supernova and forms this black hole. Not something a star goes supernova and forms a black hole. So this idea of a gas cloud collapsing straight down is theoretically possible, but we've never observed anything like that happening before. The only way that that would happen is if the gas cloud was prevented from making stars first. And to make stars, well, that unintuitively, you need cold gas rather than hot gas. Everyone's like, but stars are hot. Sure, you'd need hot gas to make stars. It helps them along. But actually, you need very cold gas. You need the molecules in the gas to be traveling quite slowly so that gravity can actually pull them together. If they're traveling very fast, then gravity's got no chance, it's just going to fly off and ping off, and it's not bothered. So very cold gas will bring that together and make it collapse and form a very dense region where you can trigger nuclear fusion and form a star. If you have two clouds of gas next to each other, one of them collapses down and makes stars first, that will be like radiating that other cloud of gas and making it energetic so it stops collapsing under gravity. But then you've still got this rather large clump of gas that still has its own gravity and it'll start collecting more and more and more gas until eventually it reaches some critical mass and could collapse down into a black hole, is the idea. If that gas is being radiated, it should glow. It should glow at very specific wavelengths because it will just be the hydrogen that's glowing at a very specific wavelength. It shouldn't be any stars giving off, you know, light across the whole spectrum of sort of optical light. And so we should be able to pick that out with something like James Webb. So that would be one way that we could prove, okay, if we can find these things, and there was sort of maybe a hint that Hubble kind of found one, but again, it's the infrared, you need to confirm it. If James Webb could find these things, that could answer that question of what came first, the galaxy, the black hole. And if it found them, we could be like the black hole comes first, because here's this gas cloud that's about to collapse and about to become the center of a galaxy, and then stars are gonna form around it.
SPEAKER_05As a person who brings to her subject more enthusiasm and excitement than I think I've ever seen an astronomy, which are you more excited about? The prospects of what these instruments, the telescope instruments, are going to bring to your research or the improvements in AI.
SPEAKER_01The telescopes. I'm so sorry, Charles. I'm sorry, but it's the tell. I mean, the AI is great, but the AI is a tool at the end of the day, right? Yeah, it's like you know, someone's saying, what are you more excited about? Seeing these things or the statistics? Seeing these things, right? I'm not gonna get excited about the statistics, which to me is a simple tool. I think it's very exciting to think about what advancements in AI will get and where that will feed into society. And I think that's always the sort of justification that I give for why do we bother doing astronomy? Why do we fund all of these new instruments and why do we bother asking these big questions about space? You don't know where they might lead. You know, astronomy led to the development of a digital detector, which now everyone carries every time they take a picture, right? In your mobile phone, you have a camera, and that came from astronomy. Uh, radio astronomy needed a way to combine the signals that they were getting from all their radio telescopes so that they wouldn't be really weak. And that ended up giving us Wi-Fi because before that, all the signals from Wi-Fi would bounce around your house and be really weak by the time they got to your computer or laptop. Your computer or laptop uses the same thing in radio telescopes in radio telescopes to combine those signals to get a strong one again. And then all of these image processing techniques, you know, have fed into medical imaging, you know, uh sharpening images, getting rid of noise, you know, in an MRI or a CT scan or something like that. Data storage and data processing, you know, is going to be a huge one going forward as well. And we're the ones taking the biggest data sets, you know, so we're gonna be pushing that forward. So I'm really excited to see if if we do end up pushing one little bit of, you know, machine learning uh techniques forward, what that will then enable us to do, especially with imaging, whether it's maybe, you know, before and after pictures of say where satellite imaging, where a hurricane or an earthquake is struck or something like that, you know, developing machine learning so that that can, you know, actually figure out if there's somewhere on the ground where where um you know first responders need to get to, something like that. Like I'm I'm really excited to see that kind of side of things, but I'm just so excited for the science results that are kind of come off the back of all of the new observatories and facilities that are coming up in the next decade, two decades. And the ones I don't even know to get excited about yet. They're still just uh an idea on a piece of paper.
SPEAKER_05Well, I hope as they do, you'll give us an opportunity to read another book by you, because your first book and when I said can be read in an evening, I probably wasn't fair because I took several because I just enjoyed mulling over each chapter after I read it. But you've been a fascinating guest, uh, and we're we're honored to have you accept the invitation. It's been a delight to talk with you. Thank you so much for taking the time.
SPEAKER_01It's been great. Thanks, guys.
SPEAKER_04Yes, thank you, Dr. Smethurst.
SPEAKER_05The views expressed in these podcasts are solely those of the speakers and not of their employers or organizations. We thank the American Bar Association for their generous sponsorship and support of the production of this podcast. Our theme music was composed and performed by the very talented Ben Rosenbloom. We 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 are writing about a particular episode, please do mention a specific episode number. Please also do 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 Dialogues on AI.
SPEAKER_02Thank you for listening to the AVA Business Law Sections 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.