Hosted by Lester Nare and Krishna Choudhary, this two-story, 2.5-hour special sets the table for Nobel Prize Week with deep dives into two recent Nobel-winning domains - gene editing (CRISPR) and gravitational waves (LIGO) - and how AI is accelerating both. We trace CRISPR from bacterial immunity to Stanford’s new “CRISPR-GPT” lab co-pilot, then pivot to how machine learning upgrades are pushing LIGO past its noise limits to capture new classes of gravitational waves. Summary • CRISPR, from bacterial immune memory to RNA-programmable genome editing • The 2012 Science breakthrough: guide RNAs unlock programmable editing • The patent saga and the 2020 Nobel Prize in Chemistry • Stanford’s CRISPR-GPT: an AI “co-pilot” trained on expert lab threads and papers • Experiment planning, guide design, and safety guardrails for CRISPR-GPT • Biosecurity and ethical guardrails around AI in biology • LIGO’s foundations: Einstein’s equations, binary pulsars, and interferometer engineering • The “noise budget”: seismic, environmental, and quantum limits • AI-driven denoising and template generation: unlocking earlier inspirals and tougher detections • Funding, leadership, and the global policy race to keep LIGO competitive • Big picture: AI as an amplifier of discovery in both medicine and physics Show Notes Stanford Medicine - AI + CRISPR Breakthrough Nature Biomedical Engineering - AI-CRISPR Original Paper Caltech - AI Helps LIGO Science - LIGO Machine Learning Paper
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
Hello, Internet. This is your captain speaking, Lester Nare, joined as always by my co-host and our resident PhD, Krishna Chowdhury. This is From First Principles. We have an exciting episode. It'll be a little bit of a special episode this week as we gear up for Nobel Prize Week, which is going to be our episode next week. We're going to do full coverage online. all week long, all three days, detailed deep dives. So be sure to subscribe to catch our Nobel Prize Week coverage. But in advance of that, we wanted to cover two of our favorite recent Nobel Prizes in a little bit of a grab bag of coverage that is anchored by two new papers about how AI is being used for these two fundamental discoveries that got the Nobel.
The first of which is going to be how AI-powered CRISPR could possibly lead to faster gene therapies. This is out of Stanford Medicine. And they put their paper in Nature. And then the second story is going to be about LIGO. Again, how can we apply new algorithms to hush the noise, which may lead to more black hole discoveries. This is out of our favorite Caltech, The Legends. And it was published in Science. So this is going to be a great prep episode for Nobel Prize Week. You're going to love it. This is From First Principles.
my friend how's it going episode 10 we've made it to double digits double digits baby thanks for everyone who's tuned in we're also on double digits on instagram and tiktok it's it's just get us there on youtube though yes yeah we are on youtube so definitely check us out on youtube if you're not watching this on youtube uh we really appreciate all all the support i think people really like science content yeah uh We do have Nobel Prize Week coming up next week. That's going to be huge. You're going to get so much content from us.
We know you guys like the history and the current, so we're going to be combining both. But this week, in preparation... We're going to talk about CRISPR and LIGO. Yeah. Two of our favorite recent Nobel Prizes, let's say, in the last 10 years. Yep. And our first story comes out of Stanford Medicine. And this is related to their new CRISPR-GPT, a large language model that is accelerating gene editing processes and increasing accessibility to CRISPR.
So we've talked about CRISPR, CRISPR-Cas9 on the podcast quite a bit. Yeah. And... Yeah. And this one is like, it's an AI co-pilot. Yes. For CRISPR. For gene editing. For CRISPR researchers. It's like, you know how like, you know, all the coders out there probably know about co-pilot. Yes. And you can just like verbally tell it what to... code and then it'll just code it. This is our favorite new term, vibe coding. Vibe coding, yeah. Now you can vibe code with CRISPR research.
Okay, it sounds funny, but it's actually going to be really, really good for the scientific community and for the medical community in terms of finding new therapies, all sorts of stuff. It's very cool. Vibe gene editing. Yeah. And so we talked in previous episodes, you made this kind of reference to this idea of like a bacterial immune system as a way to think about CRISPR. That's how it started, yeah.
And so, help me, can you like re-explain like what you meant by that as we kind of talk about this combination of AI and CRISPR? Yeah, I mean, you know, in preparation for Nobel Prize Week, I think it's fitting that we talk about the Nobel Prize winners and how they got that discovery and how they got there. And it's an incredible story of like... collaboration in science and also in curiosity-driven research, not application-driven research.
Right? Because CRISPR now, obviously, it has so many applications, therapeutic, agricultural, all of these other things. But, you know, it really started off as a curiosity-driven scientific research program. So let's briefly start with what the new paper is about, and then we can sort of tell the story about how we got here. Okay. So the new paper is basically taking this pretty mature technology, which is CRISPR-Cas9, which is this ability for us to snipe and cut DNA at wherever we want.
And it's taking that technology and it's – it's training in AI to do that technology, and then it's letting all of these researchers who might be novices in the technology but want to use it, Have a helping hand. And that's where this co-pilot thing comes in. It's like I can, as a researcher, speak in my natural language. And it will then convert that natural language into sort of the technical implementation. Yeah, yeah, exactly.
And that's what that current paper that came out in Nature Biomedical Engineering is doing. Yes, yes. And so how did we get here? Yeah. The question is, how did we get here? Because this is crazy. We're playing God at this point. We're playing God at this point, and Christopher lets us play God, right? And I think it's nice to see how it gets there and what the human stories are. So we're actually going to start back in 1987. Ooh, good year. Why is that a good year?
I don't know. Oh, okay. The 80s, baby. The 80s. Everything was good in the 80s. That's what they say. Everything was good in the 80s. That's what they say. Yeah, we were winning in the Cold War. Yeah, that's true. That's true. Ronald Reagan was president. Look, look. So everyone was like, it was fine. Take one for the Gipper. Even though he had Alzheimer's at the time, whatever. In 1987, he had full-blown Alzheimer's at the time. Well, he shouldn't have used Tylenol because that's probably the reason why he had that. Yeah, yeah, yeah. It's a whole other story. Turns out, turns out.
Anyways, 1987. Okay. Okay. Gene editing is not a thing, really. Okay. Right. There are these things called restriction enzymes, restriction endonucleases. And what they do is they target a specific... motif of DNA, so a specific sequence, and whenever you have that specific sequence, it'll go in and cut it. Okay? The problem with that is it requires that specific sequence. Okay?
So it was really in its nascent state at the time. And in 1987, some scientists in Japan, led by Yoshizumi Isshoni, okay? I'm going to say that again because that was wrong. It was Yoshizumi Ishinou. I got the vowels mixed up there. But in 1987, he discovered this segment of DNA in E. coli.
E. coli is sort of the bread and butter of biomedical research. Yes. At least when it comes to, you know... this sort of researching bacteria as a model organism to try to understand just basic molecular biology. E. coli is like the goat for that kind of stuff. And what he noticed was, he was sequencing parts of the E. coli genome. At that point, you didn't really have the technology to sequence the whole thing. Now, obviously we have the whole E. coli genome. But back then he noticed that there was this part of the genome
That had a very weird structure. Okay. What it had was it had these repeating sequences. Okay. And in between these repeating sequences were non-repeating unique sequences. Okay. So it was kind of like there was like A, which is the repeating. And there would be A, B, and then A, C, and then A, D, and then A, E, and so on and so forth. And each of these parts were about 30 base pairs long. It was very weird.
Highly structured. The A was always the same. And at this point, this is very new research where we're starting to sequence large parts of genomes of organisms. And whenever you find this kind of nice mathematical repeating structure, it always like raises flags. Right. Right. So this guy, he publishes a paper in 1987 where he's like, you know, I find these consistently repeating segments and then with non-repeating segments.
And everyone started focusing on the repeating segment because that's the part that gets highest signal. Whenever you're doing any kind of thingy on the genome, you're going to get some signal on how much of a match you're getting. And the repeating part is getting this higher signal. So the repeating part is kind of crazy because... It's also a palindrome. You know what a palindrome is? Yes. It's like a race car. Yes. Race car is a palindrome because you read it front or from the back. It's the same thing.
Yep. Well, the same thing was happening with these repeats. Okay. Interesting. These repeats had... If you, you know, DNA is double-stranded. Yes. So there's a top segment, for example, and then there's a bottom segment. Yes. And the top segment you'll read from left to right. Yes. That's what they call from the three prime to the five prime. And that'll be, you know, some sequence of ACTGs. But if you read the bottom segment from the other way, it'll be the same exact thing. So it's already like, not only is it repeating and has this highly conserved structure.
Yes. Yes. But it's also within the repeats, there's this structure. It's this palindromatic structure. Yeah, yeah, yeah. And it's already, so already it's like, you know, E. coli. Yeah, yeah, yeah. E. coli has this. It's like, that's crazy. Yeah, right. And people thought it was interesting, but didn't really think anything of it. Okay. Then came Francisco Mojica from Spain. Mm-hmm. This is back when bioinformatics was just a nascent field and people had, you know, started to sequence not just E. coli, but they were starting to sequence viruses.
Specifically, these things called bacteriophages. Have you ever heard of these things? Yes. Yes, I remember from my... They do. They look like alien motherships. Yeah, with these little, like, long, fatty long legs. Exactly. Like, they look like the Martians that come down and try to... But they're at the molecular level. It's just like the aliens in War of the Worlds. Exactly. When they come down, they look very similar to that. Exactly. Bacteriophages look exactly like that. They've got this capsule, which is this really nice polygon, and that has the DNA material, and then it's got the neck or the shaft, and then it's got these prongs, these legs that attach to the bacteria.
Yes. And they're viruses that attack bacteria, hence bacteriophage. People started... finding the DNA sequences of the stuff that's inside bacteriophages. And what they noticed, Francisco Mojica specifically, what he noticed was the stuff that people weren't thinking about in CRISPR, which are, you've got the repeats, the A, A, A, A, but you also have these B, C, D, which has low signal. Those interspaced unique parts were exactly the same DNA that,
as the stuff that we were finding in bacteriophages. Okay? So now it's like, and he comes up with this really cool hypothesis, which is that CRISPR, which, okay, I'm not going to say CRISPR because right now we don't have the acronym yet, historically. Sure. But what he comes up with is he says this segment of DNA, of this A, B, A, C, A, D, This is some type of bacterial immune system. Okay. Because how does our immune system work? Okay.
We get struck with a virus. Let's say the flu or the COVID or whatever. Right? The COVID. The COVID. Yeah. What ends up happening is you've got these T cells, which create antigens. No, they create antibodies that try to bind to the antigens of whatever pathogen that comes in. And the way that our adaptive immune system works is the T cells create these... sort of antibodies that try to match with the antigens.
This is the lock and key. Exactly. It's the lock and key where the antigen has a specific key or specific lock. We try to make the key that fits and the ones that get the right fit, those are the ones that actually proliferate and create more of themselves. You have the sort of mini evolution that goes on within your body to try and remember the the stuff that came in so the next time it comes in you can be ready and you don't have to have a fever it can immediately recognize that pathogen and kill it yes right it can tag it and then other stuff comes in and kills it yes okay so that's this adaptive immune system that we have as humans yes so mojica had this crazy idea of what if we had an adaptive immune system for bacteria where this is a way because obviously you've got these dna segments right that are
natively found in viruses but now you're you you've got a memory mechanism yes where you're keeping track of the stuff that you maybe saw in the past in your own dna and maybe maybe that serves as a kind of template where you can compare it to stuff that comes in and be like that's bad yes i'm gonna get rid of that yes okay so so he had this idea that it's a bacterial immune system This A-B-A-C-A-B. This repeating pattern that we are seeing that has repeating components interspersed with non-repeating components.
Yes, exactly. And he was like, you know, the fact that they match means that this is some kind of adaptive immune system. Funny story, though. He found this thing in 2001, and he tried to get it published. All the big ones, nature, science. We always talk about nature and science papers. All rejected him. Oh, my goodness. Because it was either like you're wrong or like bioinformatics was so new. It's like you're probably – I don't care. We don't have enough of you. Yeah, so he finally published in 2005 in a journal that's way lower tier than Nature and Science.
But it's a fundamental discovery because – That's what led to all of this CRISPR research. With this little insight. Yeah, it's a little insight into, okay, the fact that this matches and the fact that the bacteria keeps track of this, like why would it keep track of this unless it was for remembering the stuff in the past, right? And making sure that it doesn't get duped again, right? And now we know actually that it is a bacterial immune system. What ends up happening is, You know, you have a pathogen, let's say, that comes in and it infects a population of E. coli.
99% of them are going to die. In fact, only one in a million are going to survive. But that one in a million that's going to survive is going to be like, that was awful. I don't want that again. So it's going to take – it's going to take – like after it survives, it's going to take whatever part of the viral genome that is left and it's going to integrate it into this A, B, A, C, A, D. Now it's going to make it A, E or A, F. It's A. Adding to this chain of history of next time you see, we've added this new component part so we know what to do.
We have that lock and key mechanism in order to, it's like the self-defense piece. Yeah, it's like now I can be like, oh, this matches with you. That means you are not me. Bad guy. Which means you're a bad guy, right? Where's your citizenship? You don't belong here. Exactly. And you have to remember, like, in biology, especially with E. coli, right, they don't have eyes. They don't have, like, reading mechanisms, right? So the way that, you know, at a molecular level, you can actually sense whether something is different and whether something requires...
intervention is to actually have a template of DNA that matches the thing that you're trying to get rid of. Yes. Yes. Right? The idea is the lack of a sensory system. Yeah. This is their sensory system. This is their – it's memorializing in the DNA layer because they don't have sight, smell, hearing, touch. This is what they got. Yes. Right? It's like if I recognize, then it can create a cascade of other – processes that then takes care of that 100% right which is which is kind of which is kind of cool to even think about so now francisco mojica establishes that this thing is a bacterial immune system and this is around where we start getting the acronym crispar okay okay um or crisper sorry guys um someone's gonna yeah we've got enough comments about it someone's gonna say something but crisper stands for clustered
Because it's clustered. It's all in the same genetic locus. Okay. Regularly interspaced. Yeah, because of that spacing. Because of the regular interspacing. Yes. Short palindromic repeats. Which is literally what you just described. Yes. Which is the idea that the A, which is the part that's repeating, is palindromic. Yes. So it's regularly interspaced and it's interspaced with these viral genomes. And it's short. Yeah, it's short. It's only about 30 base pairs. Yep. Right? Yep. Yep. So...
That's sort of how CRISPR is established. And CRISPR is really this bacterial part. Conceptually, this is the birth, this is the origin of the idea. Yeah, and when people say CRISPR, what they really mean is CRISPR-Cas9 or Cas13 or Cas any of these other... Yes. technologies. And the CAS9, the CAS13 part, that's where our two heroes come in. Got it. Emmanuel Charpentier and Jennifer Doudna.
Doudna. So this is interesting. Does that make sense? It does make sense. There's a whole... We'll get to it in a moment, but yes, I'm tracking. Yeah, you're tracking so far. So... Now let's talk about our two Nobel Prize winners. Yes. So we got Emmanuel Charpentier. Yes. So she did her PhD at the Louis Pasteur Institute in Paris, founded in the late 1800s by Louis Pasteur himself.
You know, one of the foremost... Yeah. So it must just be like, you know, it just sprouts. And now we're like, like, it's like, it's like, that's so stupid.
But like back then, I mean, you know, the observation is I let out rotting food and now there's randomly there's flies. Right. Right. Right. The amount of experiments that he had to do to actually convince people that it's not that is actually insane. And I think we should do a separate episode on that. This is people forget that it takes a very long time for new ideas that come from sort of scientific research to actually permeate and settle in the zeitgeist. And in most cases, it is...
very aggressively pushed back. Yeah, yeah, yeah. Exactly. And so Louis Pasteur Institute, she gets her PhD. She starts a lab at University of Vienna. And that's where she gets really into studying the Streptococcus pyogenes bacteria, which is a pathogenic bacteria in... It causes tonsillitis, and it's known as a flesh-eating disease because it causes strep throat and it actually starts damaging the soft tissue.
And she was really interested in trying to find some way to circumnavigate this pathogen. And while she's working, she actually discovers that there are small RNA molecules that are part of this streptococcus bacterium, that are doing something very close to what the CRISPR sequence is doing. Okay. Okay? They match very closely. And the function, because they match very closely, right, the function should be about the same.
Because in biology, the structure means that your three-dimensional structure is about the same, which means you're probably doing the same thing. So she's trying to do this research, right? Doudna. Yes. She actually grew up in Hawaii, did her college in Pomona College, which is... Oh, yeah, it's Sun Street. In Southern California. And, you know, thank goodness for her French professor. Okay. Because...
She was majoring in chemistry, and it was getting hard, as science gets hard. It does sometimes. Science gets hard. And she went to a French professor, and she was talking about, you know, maybe switching her major from chemistry to French. And her French professor was like, no, no, no, no. Which is pretty awesome. Shout out to Jean-Brun Doudna's French professor. Thank you. Yeah, shout out to the humanities French professor at Pomona College to be like, no, stick with chemistry. Chemistry is dope. Because that's how we get the Nobel Prize.
Yes. And so she stuck with chemistry. She did her PhD with... the Harvard Medical School professor, Jack Sostak, who also won a Nobel Prize in 2009 for his research in telomeres, which is the stuff that... It's at the end of chromosomes. Yes. It's at the end of your DNA, basically. There's a bunch of repeating AT-AT-AT-AT-AT-AT that just goes on forever.
And that's basically a way to... circumnavigate the fact that when you age, your DNA gets shorter and shorter. It's a little buffer. Yeah, it's a little buffer. It's a bunch of nonsense DNA. Not really nonsense because obviously it's got functions, but it's a bunch of stuff that doesn't really code for anything to secure the stuff in the middle, right? It's like when you're copying stuff, the stuff at the very ends is going to get wrecked. Right. So you want to have a giant buffer. Yes. So he was really instrumental in that kind of research.
She did her... research, PhD research under this guy. Yes. And she was really interested in RNA interference. Okay. Which is this idea of like small types of RNA actually interfering with DNA and influencing gene expression, shutting stuff down, things like that. So she's already in this sort of... molecular biology mindset. The idea is that there's these cases where this RNA is almost acting like a mechanic of some kind.
It's coming in, house cleaning, moving stuff around. Yeah, and this was brand new research back in the day. They'd all been... grown up in the days of the central dogma. Right. Right. DNA makes RNA makes protein. Yes. And the RNA is just a messenger. Yes. That is an intermediate between the DNA that's inside the nucleus and the part that makes the protein, which is the ribosome, which is outside the nucleus. So this RNA becomes a messenger. But she was sort of growing up in her academic career at the time when RNA was finding all of these new uses that people had...
Not really tried to figure out before. It's not simply that RNA is a translator and a transporter. It has active function within the cell. So she's researching RNA interference at the time. She's at Yale at the time. Boo. Yeah, Boo Yale. She ditched Yale. Good. She ditched Yale for Cal Berkeley because Berkeley solved the two-body problem. Yeah. Do you know about the famous two-body problem in academics? Yes. It's basically you're married, your spouse is an academic, and it's one of the toughest problems to solve in academia, the two-body problem.
But Berkeley solved it by hiring both her husband and her. And good for them. Yeah. You know? They are winning. So they switched to Cal Berkeley in the, I think, early 2000s. Very good choice. Good choice. I think they also liked the fact that, you know, Berkeley compared to the Eastern institutions, which we went to one of them. Yes. The Eastern institutions are very formal, if I may say, right? They're very... You know, formal.
Upright and straight. Yes, upright. They've got something in there. But, you know, in California, we're a little bit laid back. Yeah. And they liked that. And so they came to California. And that's where... Jennifer Dodna first encountered CRISPR. Okay. Okay. She had a brief sojourn into actually private tech. So she worked for Genentech, which is this company. It's a big company. She worked for Genentech.
She wanted to see what private industry research would be like. She lasted two months. She wrote this article about her, and it was just like, no. No. this thing for me. But because she had left, right? She had sort of left all of her teaching and administrative obligations behind. So when she came back, she could just focus fully on her research. Freedom! Right? Freedom. Freedom. And this is when one of her colleagues, Jillian Banfield, contacted her.
Jillian Banfield was a professor at Berkeley. Mm-hmm. She was working with trying to sequence sort of the metagenomes of samples. So she'd go to like mines and try to collect bacteria there, or like thermophils and try to collect bacteria there, soil samples, and try to do a metagenomic analysis where you sequence a bunch of genomes and you try to find, you know, stuff in the genome. Okay. And what she had noticed was, and this is something that was in the literature as well,
You kept coming up with this CRISPR sequence. Okay? She found it in bacteria in the mines. She found it in bacteria here, there. Not only did she find it in bacteria, she also found it in archaea, which is kind of like bacteria. They're, you know, sort of indistinguishable from bacteria if you look at it. in a microscope. They kind of look like bacteria, single-celled organisms. But if you look at the phylogenetic tree of life, you know, where you have some primordial first-life ancestor and then that fans out, one of the first two things that fanned out was actually bacteria and archaea.
And then from bacteria outcrop the eukarya, which is us. Eukaryotes are the multicellular organisms that have mitochondria living in them and chloroplasts living in them and a nucleus living in them. The powerhouse of the cell. Yes. And all of these. So the higher complex organisms came out of the bacteria. But very early on. In the evolution of life came out bacteria and archaea. And the fact that the CRISPR sequence exists in both bacteria and archaea is very, very interesting.
Because this means this happened very, very early evolutionarily. Very, very, very early. It says two things. One is viruses were around like... Even back then, viruses were like, ooh, I got to get in on that. Yeah, yeah, yeah. And the other thing is this thing is preserved, which means it really might have this immune capability. And it might be nice to figure out how it works. Yes. Yes. Now, Jillian Banfield at Berkeley, she thought it was something to do with RNA interference. Okay. It's some kind of interference mechanism where it creates an RNA and then that RNA interferes with these viral DNA to sort of stop it from happening.
Yes. So what she literally did was she Googled RNAi Berkeley. Okay. And out pops Jennifer Doudna's DNA. Yeah. Right? Yeah. She's like, okay, this is someone who works in R&AI. It's at Berkeley. So, gives them a call. Yes. And they set up a meet cute at the cafe, which is the free speech movement cafe in Berkeley. Yes. Yes. Very famous. It's a cafe in one of the libraries in Berkeley.
It commemorates the free speech movement back in the 1960s and 70s. A movement we are trying to reinvigorate here in the United States. Yeah, it's probably like... On its high tails right now, right? We really need that. But they met at this free speech movement cafe. Yes. They have a little bit of banter. And then Jillian Banfield gets right into CRISPR. It's like, this is what CRISPR is. It's kind of crazy. Jennifer Doudna is like immediately hooked. Yes. Like, this is crazy. Yeah. Right? Yes. Palindromic.
Yeah. Regularly interspaced. Because people had seen these sort of regular structures before or these palindromic structures before, but not in this sort of nice... like organized form. It was too perfect. It was too perfect for it to be just some random thing. Yes. Okay? Yes. So that's when Jennifer Doudna starts thinking about CRISPR. Okay? And she starts investigating the CRISPR system. She starts mapping it. And what, at the time, people end up finding is...
There's the CRISPR part of the genome, which is this A, B, A, C, A, D. And then there's another part of the genome that's always there right next to it. Okay? We talked about how promoters are right next to the part of the genome. Yes. So this is kind of like that. It's not a promoter, but it was an associated region because it would always be there right next to the CRISPR system. It was a crony. It always hung out. It would always be hanging out right before. They started calling it CRISPR-associated. And it became Cas.
So now we're getting CRISPR-Cas. So the CRISPR is that regularly repeating interspace sequence. And then the crony that's always with the little mob boss. That's the Cas. That's the CRISPR associate. Yes, and it becomes the Cas. Yes, that becomes where Cas and CRISPR-Cas9 sort of... Okay, I'm trying. So now we've got to get to the 9. Yes. Because we've already fixed the puzzle with CRISPR. Yes.
We fixed Cas. Yes. Now we've got to get to the 9. Right. This bacterial immune system and this associate of that sort of immune system structure. Okay, yes. So she starts researching these Cas systems. It turns out there's two types. There's a class one system, and that requires many different proteins to function. They'd figure that out. And then there's a class two system that was sort of simpler. Okay. Now let's get back to Emmanuel Charpentier.
Yes. Okay. So meanwhile, while Doudna is doing this in her lab... Sharpeian TA starts looking at the streptococcus genome. Yes, the soft tissue eating. And she starts... Yeah, the soft tissue streptococcus tensilitis causing bacteria. And she starts discovering a previously unknown RNA molecule that was very closely associated with CRISPR. Okay? And she called it the tracker RNA for trans-activating CRISPR RNA. It's something that was required for...
the CRISPR RNA, which is the RNA that is made from transcribing the CRISPR DNA, right? CRISPR, remember this ABA, ACAD thing. Yes. This is in the DNA. Yes. But that gets transcribed by RNA polymerase to become an RNA molecule. Yes. Right. And then that RNA molecule, presumably for all the other gene stuff, that RNA molecule goes outside and makes a protein. But in this case, what it was doing was it's mediating this immune system. Okay. And what she realized was there's this other, RNA molecule called tracker RNA that is required to make that bigger CRISPR RNA work.
Got it. And do its thing. Okay. Okay. So she starts looking at this tracker RNA and she's like, okay, this thing is essential for making whatever CRISPR is doing work. She publishes that in 2011. Okay. Got it. Okay. And then we have the chance meeting in Puerto Rico. Ah, look. Great, great. In San Juan, Puerto Rico. In the U.S. In the U.S. They meet at a conference, and they get introduced to each other with one of Doudna's colleagues.
They have a short meet-cute, and then they decide to... explore the city together, the old San Juan city together the next day. And on this walk in the old San Juan city, they talk about what they've been working on. And Sharpen TA says, okay, why don't we collaborate? Because you're a biochemist. I'm sort of the structural person who does bacterial research. Perhaps we can help each other. They agree.
And so Jennifer Doudna now, they start like... collaborating via Skype and their postdocs start collaborating. RIP Skype recently shut down. Yeah, yeah. Now it's called Zoom, but I guess I still use Skype. That just shows how old I am. But they start collaborating on the interweb. Yes. And – They start focusing on a class 2 system. Remember, Doudna was focused on class 1. The more complex. Charpentier is like, no, no, I got a class 2. Yeah, yeah. It's quite nice.
Yes. It's this thing in the streptococcus, and it requires only a single protein, not the multiple that you need. And the single protein, they're calling it Cas9. Mm-hmm. I see. So now we're getting to Cas9. It's the ninth associate. It's the ninth associate. It's sort of named on the... the order in which it was discovered. Which makes sense. So it's CRISPR-Cas9. She's like the single protein that's required. And they suspected that Cas9 was a scissor that cut viral DNA.
So what CRISPR would do is CRISPR would be this record of all the viral DNA that the bacteria had seen in its past. It would be like a historical map. Yes. Or in its ancestors' past that had been passed down genetically. And... The Cas9 system would take this, match it to a viral protein, and then cleave it. Yes. Okay? So they did an experiment where they combined the Cas9 protein, they combined the CRISPR RNA, which is the thing that gets transcribed from the ABA-CAD.
Yes. And then they combined the target DNA, and they're like, okay, it should work. Didn't work. Okay. Okay? They're like, what the hell? Yes. Okay. And then after a lot of soul searching and everything, Charpentier is like, well, you know, the tracker RNA that I had earlier, I thought it was just doing this job of like preparing the CRISPR RNA for this task. But what if it's like needed for something else? So they put in the tracker RNA. This other thing you were talking about. This other thing that I was talking about.
That was sort of just like in the sidelines. Right? In the sidelines. They put in the tracker RNA and all of a sudden, We're getting scissors. Yes. We're getting scissors that are like actually cutting stuff. Yes. They're like, oh, okay. So we got something that cuts. And this is meaningful because still in and around this time, this idea that RNA has this active... like aspect to it is, is in its nature.
Like it's still kind of being worked out. Yeah. Like it's not quite obvious. It's not quite obvious that like, that like RNA would be such a huge part. Right. Of the activity of an enzyme. Right. Like Cas9 is a protein. Right. But like it needs this tracker RNA to actually do its job. Yes. Right. And now it gets interesting. Mm-hmm. Because, The way that this Cas9 system is identifying the thing that it needs to cut is through the CRISPR RNA.
Right. But the next question is, what if I just make the CRISPR RNA whatever I want? So, okay. And now that is the key insight. We've discovered that CRISPR is this repeating pattern that's kind of this historical map of past interactions that bacteria were having that helped it to remember how to fight back. We have these associates, these cronies, the bag men that go out and kind of do stuff, one of which is Cas9.
Yeah. Right? Right? And Cas9 had one job. And that one job was to cut people. Actually cut people. Yeah. Cut them. But how did it know what to cut? Right. Right. And... Could you change its instructions of what to cut? Yeah, because it's almost like, as you say, okay, let's go with the cronies thing, right? Yes, yes. Like the cronies got a photo. Yeah, right, right. Of like who he needs to cut. Yeah, this is who you got to go after. And then he looks around, right? What if I just replace the photo with whatever I want?
Yes, put a little deep fake. Yeah, right? Exactly. And that's what they figured, right? That is so fascinating. The first insight was, okay, maybe we can give it. Our own version. We can repeal and replace the initial CRISPR sequence with something else so that Cas9 goes after what we tell it to go after. Yeah. Not what it was maybe historically X, Y, and Z. Yeah. The other thing that they did that was, again, huge and part of the discovery. Because already, this would be a crazy discovery. Yeah.
Okay, already the fact that a Cas9 system, what it's doing is it's matching this DNA with whatever DNA that comes in and then it's cutting it out that place. Already that's insane. Yes. Okay. The next thing they did was they noticed that we'll track her RNA. which is the sort of helper that comes in. Yes. Let's say in the crony example, it's the driver. It's the driver, yeah. Which gets, like, just as a loose analogy. Exactly, yeah. It's the driver, right, that comes in. And then you've got the photo. Yes.
Which is the thing that lets it recognize, right? These are two separate components. Right, right, which are important. Yeah, which are important. So what they figured was, what if we made them a single component... Oh, okay. Called guide RNA. Okay. This was a big one. Okay. Because what they did, because now it's just way simpler. Anyone in a lab can just make a single piece of guide RNA. It's a single piece of RNA. Yes. That I can just make. Yes. I know how to make RNA. Yes. And I can just make that and be like, this part of the RNA is the thing that I want to cleave.
Yes. That's what they did. They combined, they added a nucleotide sequence that sort of stitched these two together. The CRISPR RNA, which is the part that's recognizing what I want to cleave. Yes. And the tracker RNA, which is what's sort of preparing the Cas9 protein to do the cleaving. They combine these into a single thing. And it's not as trivial as just like, oh, I just like staple them together. Because if I staple them together, that's going to change the morphology of them. Yes. And then the Cas9 protein, your crony is not going to be happy.
Right. He's going to be like, I don't know what to do with this. I can't recognize this photo. Yes. Right? Yes. They did it in such a way that the tracker RNA part of the structure was intact, and the CRISPR RNA part of the structure was intact, but they were connected together in a way that didn't get in the way of the Cas9. Yes. Yes. That's what they did. They created a guide RNA. So, and again, to continue this really bad analogy, whenever you're trying to commit crimes, you want to minimize the amount of people who are your co-conspirators.
Exactly. And so instead of having a driver and then the crony who's doing the cutting... Yeah. You just have the driver and the crony be the same person as one more or less point of failure. And it's hard to find that person that can do both things. And it's hard to make that person and train him. But you got it. That's exactly right. And so that is what became the CRISPR-Cas9 system. Okay. Okay. It's really the guide RNA. That was the sort of key aspect that made it move from an observation of something that's happening naturally to a programmable platform or concept or system that could then effectuate specific instructions dictated by...
Yes, very good point. And to that point, in 2012, they published in Science. Yes. And in their abstract, they could already tell this was going to be big. Right. Because they wrote in their abstract that this is an RNA programmable genome editing technology. It's so funny because, I mean, that's immediately what I think of. It's just like, well... You can now... Because before we used to have these things called zinc finger nucleases. Okay. Right? That was the best we could do where you have these like zinc fingered domains of proteins. Yes. That would recognize three nucleotides at a time and then actually like try and cleave stuff.
And that was the best we could do with programmable stuff. It was really hard to make, dude. It's really hard. Like, for each zinc finger is, like, attached to three. You got to make, if you want to do a much larger recognition, you got to do three times three times three. And they've got to be the right structure. And, like, making these proteins is really hard. Making RNA is super easy now. Right? Even back then, it was, like, way easier. Which goes to show, like, right after they published this, right? Then there was an MIT group from Feng Zhang who published how to do this with mammalian cells and human cells.
And then since then, I mean, the technology has just completely blown up. Yes, 100%. Now we're making crops that can withstand drought. Yes. We've got... Shout out David Freeberg and the All In Boys. Exactly. We've got CRISPR technologies that do... Genome editing in humans. But it's happening. I mean, it's curing sickle cell disease, which is really nice. Big for black people. Yeah. It's also being used by the Chinese.
In one instance. I'll give them that. It was one instance. That we know of. That we know of. Yeah. But they literally said they took out a genome that gives... HIV a leg up in trying to infect someone. But no, at the end of the day, you like, you edited the genome of like a human being. And now that human being is out there. You know, like reproducing and stuff. That's kind of crazy. It gets very – Like I did not sign up for that. It gets very messy. But like in any case, like CRISPR – I hope you now see the sort of – 100%.
The thread of discovery, right? Where it starts from just curiosity-driven. Yes. What is A, B, A, C, A, D doing in E. coli? Why is it in Archaea? Yes. To now we've got a technology that lets us play God. What's so important about this story, there's so many important aspects of this story, but I think the thing that sticks out for me is a lot of times folks don't understand how a scientist in a lab with an electron microscope or looking at really tiny stuff –
They don't get the through line for how that ends up creating something that is going to be one of the most profound scientific discovery terms of impact on medical outcomes for people that we've ever made. I mean, this is like having a programming language with infinite idea space. space yeah yeah yeah for the fundamentals of like what makes us yeah and like it's it is the blueprint we can now edit blueprints of buildings right the buildings are us and the buildings are us it's insane it's it's it's it's insane there's an interesting before we get to
The paper. The paper. Because now we've created the base of understanding how did we get to this new paper that's coming out of Stanford Medical. That's right. About creating CRISPR-GPT, which just feels like yoga babble of new thing CRISPR, new thing chat-GPT. Just smash it together. It doesn't actually mean anything. Yeah. But now that we have that background, there's an interesting side story to these characters that you just talked about and touched on at the end. The president of the United States and his good friend Jeffrey Epstein are not the only ones who have suffered from years of long legal battles over who came first.
Two of America's most elite research universities have been locked in a legal dispute over who was first to file. Yes. That's file. F-I-L-E. File. And we're talking about patents. Yes. The other P word. We're talking about patents. Yeah. Because whenever you make something big like this. The immediate reaction from people who are capitalistic is like, that's multi-trillion dollar market. Yeah. In terms of personalized medicine, gene editing. I mean, we're talking, it's like at least one or two NVIDIAs.
And so just as a brief kind of sidecar to the background you just gave, we talked about this. Let's start at the beginning, which is the discovery. So Doudna and Charpentier, they make this discovery in 2012. Publish it in science. Publish it in science. And then months later in early 2013, Fang Zhang at the Broad Institute at MIT is the one who made the leap to show that CRISPR worked in eukaryotes. Yes. This is the mammalian cells, animal cells, which matters because that's how you can create a product.
Right. Yes. Now it becomes something that the health people want. Right. It's not just this observation about our natural environment. You can actually productize it. Yeah. So the idea here is that, you know, Berkeley proved that the tool worked and then the Broads showed that it can work in cells that matter for medicine. Yeah. Yeah. that's when the patent race begins. So it gets a little bit messy. So Berkeley filed their patent application first in May of 2012, covering CRISPR broadly, no pun intended. And then the Broad filed later in December of 2012, so just a couple months apart.
But MIT fast-tracked their application with the patent office, which meant they actually got the first CRISPR patent in 2014 covering use in human and animal cells while Berkeley's application was still making its way through. Oh, my gosh. Which is like, okay, okay, okay. So now the first legal challenge, Berkeley challenged that, obviously. So now in 2016, fast-forward two years after MIT got that initial patent for human and animal cells. The patent office held this special interference trial, not to be confused with RNA interference, to decide if the Broad's patent overlapped with Berkeley's patent.
And then in 2017, the ruling was a little bit of a surprise where the patent board said both can stand. Both patents can stand. And it was because – They argued that CRISPR moving from a test tube to human cells was not obvious. Like, this is sort of the term I use with patents. Like, is it obvious? Yeah, yeah, yeah. I see. So they were arguing that it's not obvious, and it counted as a separate invention. Okay. So this obviously gave MIT the upper hand for commercial applications.
Yeah, yeah, because, like, that's where the big money is. That's where the money is. I mean, sure, you can use it for agricultural stuff, but, like... You know, which is important. Yeah, which is important, yeah. But it didn't end there, not satisfied, right? So round two. Round two. So Berkeley came out with new filings aimed at this eukaryotic cells issue. Right. And they got a second interference case. Now, in 2022... Berkeley lost the claim outright because they were arguing that the Broad hadn't in fact shown that they were able to make it work in human cells, but they were.
So it was kind of like, eh, it didn't really... So the loss then had this big financial and reputational hit, because it's like, you're begging now. Right, right. Like, you lost, take your L, take your toys and go home. But, obviously, this is worth hundreds of millions, billions of dollars, licensing fees. So like, you know. Okay. Here's what's interesting, and this goes back to our Nobel Prize discussion. Yeah. In 2020, right, the Nobel Prize ultimately went to... Jennifer Doudman, Emmanuel Chagbite.
Chagbite at Berkeley, despite not getting the patent. So the credit for the discovery went to the original founders. That's right. And they didn't, you know, the Nobel Prize Committee could have easily done it for three people. Because the three is the limit, according to Alfred Nobel's will, of how many people can win a Nobel Prize. But they chose to give it to Charpentier and Doudna, right? Which is like, look, the OGs, you were first. I mean, as I was saying, right? It's like patents are, I think, local in time, but Nobel Prizes are forever.
Yes. You know? Yes. Unlike diamonds. No, I'm kidding. So there was a split between who got the glory, like the prestige, right, to be immortalized forever. Yeah. And who's going to make their shareholders hundreds of millions of dollars. Exactly. Yeah. So this kind of has shaped the biotech startup playing field in terms of developing CRISPR therapies because you likely had to license it with the Broad Institute. And all of Berkeley's allies were kind of boxed in at least commercially. Yeah. Yeah. Now, just this year, in 2025, this is still ongoing.
It is. Because a federal appeals court actually reopened the case. Oh, my God. Saying that the Patent Office may have applied the wrong legal standard back in the 2022 case. So this is not entirely over. Wow, it's not even over. In terms of, you know, who owns CRISPR itself. in human cells. It's still being litigated. And again, I think hundreds of millions and billions is underestimating the total addressable market impact when these gene therapies get to scale, not only in the US and other Western, but globally.
So it was just this... Dude, that's crazy. I mean... To think – I don't even know – so all of these people who have been using CRISPR, there's so many companies that have been using CRISPR on the MIT license, right? Let's say for sake of argument that Berkeley wins this appeal in 2025, what's going to – This is – we're going to start our new From First Legal Principles sideshow. Yeah, yeah. No, there's a really interesting point there. Right? It's like – Is it – what are the retroactive remedies for Berkeley saying, look, we've lost out on –
existing revenues, it could potentially be a very big landmark case if ultimately Berkeley ends up winning this reopening of the case, which is... I'd love to see MIT just pay Berkeley out of pocket because I just hate MIT. Yeah. Look, we are the best university on the East Coast. They're always chasing us. We get it. We get it. It's hard to be the best all the time. But I mean, to Doudna's credit, though, I saw a...
interview of her with dan rather yes back in 2017 back when you know this was at the height of it and um dan rather in his journalistic voice is like i wouldn't be doing my job if it weren't if i wasn't asking about the legal dispute and and she said very very calmly and quite nicely that you know that legal dispute is between the two institutions right mit and berkeley right and Her opinion was that science shouldn't slow down because of a legal dispute.
So she was totally for all of these companies using the existing MIT patent to go and forge ahead. You know? Yeah, yeah. Which is how it should be. Which is, I think that was really cool to see that. And in good spirit of it. I think what we all mean when those of us who are passionate about science and research, experimental design and growth of this understanding of the fundamental knowledge of the world around us and applying it in a way that impacts people's lives every day. Yeah. Not to poo-poo on the theorists.
We love you too. Yeah. You guys are important. Important. We get it. Practical applications are nice. Yeah. Yeah. And yeah, I thought it was a really cool answer. That's good. No, that's good. Institutions are always going to battle over the dead carcasses. Yeah. So that's kind of how it goes. I mean, to me personally, I think Berkeley won because they got the Nobel. Yeah. you know what what really matters all the now and now she gets a you know at berkeley the tradition is every nobel prize winner gets a really nice parking spot right next to their building look like everybody else has to park in the parking structure and walk yeah but like there's a it says reserved for nobel prize winner oh it's so great just like that she's just like roll up and park right next to it there's a whole studio episode about a parking spot like just this exact same story that's right
Yeah, yeah, yeah. But I love that in academic institutions, they're like, no, she wants to know. Yeah, no, she's not going to the garage. What? She's not a peon. Yeah. That's for you PhDs and lab rats. Okay, so we've now, we have a really strong foundational understanding from first principles about what it is, what are we talking about when we say CRISPR-Cas9? Yeah. And kind of the history of how this osmosis and collaborative nature brings us to a point of actual application.
So now Stanford Medicine put out this new paper about CRISPR-GPT. Yes. Which is now taking all of this decades, all the way back to 1987 where we started the story of understanding. Yeah. To now applying something that's relatively new in the zeitgeist, large language models, and applying it to this Cas9 use case. So what does that look like? Yeah, so now it's been about 13 years since the discovery of CRISPR-Cas9, and it's been used in a variety of different applications.
But at the end of the day, it's actually quite intricate. The whole technology requires intricate decisions. By now we have Cas9, we also have Cas13, we have all of these different associated proteins that you can use in these different use cases. And it's hard for a novice to know which one to use. There's a high failure rate. There's safety concerns. You can get bottlenecked in all sorts of little tiny things that happen in the scientific process. And to actually...
Make that better for researchers. Stanford, along with Princeton and Google DeepMind, they actually developed this co-pilot, which is CRISPR-GPT. That's what they're calling it. It's a co-pilot for gene editing experiments where researchers can literally talk to the CRISPR-GPT and be like, you know, I want to use CRISPR on this... human lung cancer cell line, and I want to target these locations.
How should I do my experiment? And the GPT will output all of the steps that you should do, how you should design your experiments, what to look out for, how you should analyze the data. All of that, right? It becomes like this sort of lab mate. Yes. That, you know, you wish you had. I wish I had. I was going to say. Back in my PhD days, right? And it can really expedite and democratize the use of CRISPR. Yes. In all of these systems by researchers who might not be totally versed in how to do it in the first place, right?
And I think that's really cool. Yes. It was trained – so the training is pretty interesting. It was trained on 4,000 real-world discussion threads. So there was this thing called a CRISPR-focused Google group, which is around 3,000 question and answer sessions with really – Talented CRISPR scientists. And what this GPT did, it was actually a CRISPR Lama 3. So it's based on the Lama 3 from Meta, I believe. And it was a specialized model that was trained on these discussion threads, along with a lot of scientific papers that were out about CRISPR.
Lama 3 is old. They could have used... That might have been... Anyway. Yeah, I mean, it was probably like something that they started way back. 100%. It's probably something that required, you know, less compute resources and all this other kind of stuff. 100%. It's kind of interesting that DeepMind was involved in this, but they still use Lama 3. But the AI was trained on these nuances of scientific discourse. That you don't really get with reading papers. Like you don't get troubleshooting. Right.
You don't get problem solving. You don't get like, this is what I tried and it didn't work. That's something that you get in these discussion threads. Yes. Right? So similar to how the normal chat GPT that everyone uses is trained on all of the internet, including all these Reddit threads. Yeah. All we're saying is the training data for this specialized model used a version of a Reddit thread with only very high-level experts that are talking about this outside of the explicitly data-driven or experimental context. It's sort of this gray area, that human element.
Yes. And then you get an AI that sort of talks and thinks like a human scientist. It approximates that context. Exactly. Yeah. And they compared it against just, you know, GPT-3.5, GPT-4.0. Because obviously, I mean, I can ask GPT-4.0 about CRISPR and it'll be pretty good. Yes. Is this any better? It's a lot better. Which makes sense. Like when it comes to experiment planning, this was near perfect. Okay. And I've actually tried experiment planning with GPT-40, and it's very, very bad.
It's not good. It's a lot of noise in there. Yeah. It's a lot of, like, it's stuff that an undergraduate would, like, come up with. That's like, oh, that's cute. But, like, no. There's, like, five reasons why that experiment won't work. But this one was very good at it. It was also really good at guide RNA design. You know, the guide RNA that we were talking about. Yes. Yes. It's not as simple as just, oh, this is the sequence that I want, right? You want to design a sequence that will only go to your specific location in the genome and not get misidentified with other parts of the genome, right?
Because at the end of the day, it's a lock and key. Right. But if the lock and key closely matches elsewhere, then you're going to start cutting other parts of the genome. Right. And then your whole experiment fails. So it's really good at guide RNA design. And it's also really good at scientific question and answer. Like if you want to learn more about how this stuff works or what is the logic behind a certain experimental step, you can ask it and it actually gives it back to you. So the Stanford lab that developed this is actually already using it with its PhD students. And now it's out like in the open.
And I think it's going to be really cool for expanding the broader impact of CRISPR, Cas9, and these CRISPR technologies. There's something like 7,000 to 10,000 single gene disorders out there. And in order to work on them, if you can make it easier for the researcher, that's a huge win. There's going to be, and again, obviously, we will be remiss and get escoriated in the comments if we don't talk about the obvious third rail related to all of this stuff, which is you're combining these two extremely powerful but also extremely dangerous frontier technologies.
And they have this accelerant effect when they're now combined and also open sourced. And so the obvious kind of the safety, you know, this is why people talk so much about AI safety. It's not always the most obvious, like, oh, it's just going to be in a robot and then we're going to have drones with computer vision killing everybody. It is more so if you have, you know... in some place, a group of expert scientists who are in this arena, but maybe are under a repressive regime that wants to do crazy stuff.
And now they have accessibility to better tools of discovery. Yeah. There's a whole like human piece to it. There is this technology piece in its entirety. Like how do you decide how to Stanford decide to release it or not release it? That's right. I mean, they did. So to be fair, they did have like guardrails in place. One of the guardrails was like, you know, if you're trying to like do CRISPR on like COVID or HIV, it'll just be like, no. What are you doing?
Why are you doing that? If you give it an identifiable DNA sequence, so like something that's more than like 30 base pairs long, which is something that you can maybe identify with individuals. If you want patient confidentiality, again, it'll lock down. I mean... Obviously, there's ways to get around this, right? And it is concerning. I just want to make sure we at least touch on that. And I agree with you. Like, I think it's a very important point. It's not the focus of this discussion. Yeah. But, you know, as we talk about all these things, everything has sort of this double-edged sword.
And it goes back to the whole kind of quote where every technology is agnostic. It's about how ultimately we as humans choose to implement it. So it's a human... CRISPR doesn't kill people. People kill people. Sure, bro. But it is a worthwhile discussion point. And if there's interest in this topic of CRISPR and us kind of continuing on the sort of safety conversation, just let us know in the comments. And we'll definitely come back to touch on this. This is, I think, the fourth or fifth time. And we talk about safety every single time. Yeah, every time.
Maybe not as in depth. But these are things that, again, folks at the Frontier do have to start thinking about. Because, you know, we already seen what happened with nukes. And so it's the same story. Yeah. History doesn't repeat, but it rhymes. Yes, exactly. But this is really, really fascinating. Yeah. And I'm, you know, 2020. So five years ago, the Nobel Prize went to them. One of my favorite Nobel Prizes over the past 10 years, definitely, is Christopher. I actually remember, you know, in 2012 at Princeton as an undergrad,
I was just having a conversation at a dinner table with a friend of mine who was in molecular biology and she was talking about this paper that had just come out in science about programmable DNA. And that's where I... At the dinner table is when I first heard the word CRISPR. I was like, like CRISPR, like crispier, like food. And then she was like, no. And then she described it to me and I was like, oh. Oh, shit. Yeah.
Yeah. This is insane. We're going to start referring to people like KFC Chicken. Are you original or crisper? Yeah. Because people, it's going to start being a thing with everyone. And I still remember that day when it was like, it was sophomore year, 2012. Yes. And I was like, whoa. Yeah. No, big deal. Whoa. This is a big deal. Big deal. Yeah. Very, very big deal. This is our first Nobel Prize reference story of the day. Again, Nobel Prize week. It's next week. Be sure to tune in. We're going to be doing crazy coverage. The best coverage from First Principles.
Nowhere is going to be better. The best is going to be the greatest. It is the best coverage. They say it's the best. Okay, so we just finished talking about CRISPR, which was our first favorite Nobel Prize. We're going to talk about our second one, a great achievement for American science. This was all America. This is us. Yeah. America. And which is the sort of LIGO. And again, we're tying it from the AI story. that just came out this year, to the history of this project and how AI is making this initial tool and kind of area of research even better.
So the headline on this LIGO story, new algorithm hushes unwanted noise in LIGO may lead to more black hole discoveries. So this is from researchers at Caltech. And their idea was we want to make it easier for LIGO to find black holes with eccentric or oblong orbits, as well as catch mergers earlier in the coalescing process. Yes. They published their paper in Science, which is this AI kind of context. But we've talked about LIGO on the podcast before. And I remember you talking about how this is sort of this, like, long tail of this initial contribution of what Einstein predicted.
100 years 100 years ago yeah and so we're gonna do the same thing again which is kind of look at this story of the scientific curiosity journey from that initial prediction now to how all these new ai things are actually making ligo better better yeah it's kind of crazy 100 years ago Actually, more than 100 years ago. So, 1915. Okay. Einstein publishes his General Relativity.
Yes. Which is a new way of thinking about gravitation, not as a force in the sense of Newton, where there's a force between two bodies, but actually as a disturbance in the actual fabric of space and time, where... Now you have this unified quantity called space and time, which he had sort of already unified in 1905 with special relativity. But special relativity dealt with inertial reference frames, which means a reference frame here, let's say us sitting stationary, and then a frame of someone in a car that's moving in a straight line.
with constant velocity. The idea is any experiment that he does in there is the same as any experiment that we do stationary. This is kind of, he's taking Galilean relativity from Galileo's time and taking it very, very seriously. Right? And he's combining that with the fact that the speed of light is constant to get special relativity. All these consequences about how time is really relative, space is really relative, and there's this metric that is a combination of the two that is the thing that is invariant.
So now we get special relativity. General relativity is where he generalizes it to say, what about accelerating reference frames? What if that train was not moving in a constant velocity, but was moving around? Was turning, was speeding up or slowing down? How would that affect... these calculations that I've been making. And he comes up with general relativity, which is this idea that gravity and acceleration, gravity and this change in space and time are actually the same thing.
We don't have to get into the whole details, but at the end of the day, it created this sort of nice... mathematical framework for how we identify and understand space and time in a very nitty-gritty, precise way. Which impacts us, you know, every day. Every day. And I mean, you know, GPS, for example, is impacted by general relativity. The fact that GPS satellites are so far up means that they have a smaller gravitational field due to the Earth compared to us down here, which means that their clocks...
actually run a little bit faster than ours do. Right. You would think because of special relativity that their clocks would run slower. Yes. Right? Because they're moving so fast. Yes. But actually the general relativistic effect is something like four times... The effect from special relativity. So special relativity, the fact that they're moving so fast means that the clock is slowing down. But the fact that they're in such a smaller gravitational environment, right, where the gravity is so less, means that they're actually moving faster.
Which is why there's a correlation. These things matter. Yeah, and actually measuring that and accounting for that is essential to how GPS works. How we get precision. Yeah, because at the end, GPS can... localized me to within a few meters of where I actually am, right? That few meters, if you think about it, the way they actually do it is from the light traveling between us and the satellite and the time delay between us and several different satellites, right? And then triangulating from those several different satellites to where we are.
Well, in order to actually measure that time difference, you need to have extremely precise clocks on that clock. GPS satellite. Yes. That precise clock is going to be so precise that it is going to feel these relativistic effects. And so you need to have an active mechanism that accounts for the fact that that clock is going faster than this clock by exactly the amount that Einstein's field equations say that they will go faster by. And then you do the math and then you like, oh, you're here, you know, and you need to go over there and so on and so forth.
That's actually a really good example. So it's very much involved in... Right, right. It's a good example of how GR relates to a very everyday use case that everyone can understand. Everyone, yeah. Everyone's used GPS. Yes. Okay, so now we're getting to the fact that Einstein publishes these field equations. People start trying to find solutions to the field equations, okay? Which means to say... The field equations give us a way to understand what matter does to space and what space does to matter.
The great saying by John Wheeler, who we'll get to later on, is that... Space tells matter how to move and matter tells space how to curve. Okay. That's effectively what Einstein's field equations are. They're a way for us to mathematically describe the stretching and turning of space time. Okay. So space time becomes this sort of. like field in some sense.
Okay? It's like everywhere and it's got a way to stretch and stuff. And one can imagine people start asking, can I make waves? Okay? In this field. In this sort of field of space-time. You can imagine, right, like if I've got ropes or if I've got a rubber band, right, I can create waves in my rubber band. If I have a bunch of rubber bands lined up together and I stretch one and then that sort of propagates on the rest, you can create these waves in rubber bands. And if space-time is this sort of elastic thing, then I should be able to create waves in space-time.
The thought, my initial thought on that is... You must need an immense amount of energy. Energy, yes. An insane amount of energy. In order to disrupt the fabric of space-time. Yeah, yeah. I mean, and you're exactly right. You do need an insane amount of energy to disrupt the fabric of space and time. But before we even try to think about what could cause waves in space and time, you've got to first show that the Einstein field equations...
Admit a solution that looks like the wave equation. Okay. Okay? The wave equation is something that's been around for hundreds of years. And it's very well defined, you know, sines and cosines, traveling. The wave equation is a very simple second order differential equation that's basically saying that the amount of curvature that I have in space tells me how much acceleration I have in time. And that's... Basically how the wave equation works, right? There's a second derivative in space. There's a second derivative in time.
And they're related by the velocity of propagation. So Einstein sets out to see if this works in his... in his equations of special relativity, in his field equations. So at this point, it's in the 1930s. Einstein's already at Princeton, New Jersey. He's already at the Institute for Advanced Study. At this point, I don't think the institute had actually been made, and so he had an office in Frist Campus Center, which used to be the old physics building and the math building on campus.
Spent a lot of 2 a.m. nights. Yeah, yeah, yeah, yeah, with the late meal. Yeah, late meal was classic. So he's sitting there and he tries to find a solution to the wave equation in his field equations. He does the math, and he actually comes up with the fact that what he notices are these things called singularities, which he can't quite get rid of when he's trying to do this analysis.
And so he writes a paper saying that the field equations do not admit wave solutions. He writes this up, and he sends it over to the physical review. The physical review at this point was... The editor was John Tate, and he is kind of a stickler for the peer review process. At this point, Einstein had already sent in a bunch of papers to the peer review. There was the very famous EPR paper, Einstein, Podolsky, Rosen, which is the one that was talking about local variables solving quantum mechanics.
He was saying that, you know, this whole quantum stuff is weird. Maybe there's some like... hidden variable that explains quantum mechanics. That was the famous EPR paper. There was also the Einstein Rosen bridge paper that he had published in the physical review. So he's had a pretty good relationship with the physical review. So he sends this one in. And he says that, you know, my equations do not allow for wave-like solutions. John Tate sends it over to his buddy H.P.
Robertson, who's also in the Princeton physics department. So, you know, Einstein sends it to John Tate. John Tate sends it back to Princeton. And Robertson finds an error in the paper, okay? He finds that the singularities that Einstein is so worried about and... the reason why he thinks that wave equations won't exist, are actually just coordinate singularities. They're not actually real physical singularities. Coordinate singularities are something like, you know, trying to find the longitude of the North Pole, something like that.
It's an artifact of, like, our... version of how we're keeping track of latitude and longitude, right? But the North Pole could be anywhere, right? It's just we like the North Pole at the top because that's how it's spinning. But you can easily imagine a coordinate system where the North Pole and the South Pole are on the opposite sides of the equator, and we just have really convoluted equations of mechanics for how physics on the Earth works, right? But at the end of the day, it's a choice that we've made as humans. on where to put the zero of our coordinate system, right?
It's an artifact of the coordinate system. So Robertson actually looks at Einstein's paper, and he's like, actually, these singularities are coordinate singularities. They're not physical singularities. So he writes back to Tate. Tate then... pushes back on Einstein and he's like, hey, I got these peer reviews. I was wondering if... Very nicely. The letters are still there. Not like reviewer number one. No, not like reviewer number one from the Mars paper. Yes. Where it's like, you need to... This needs to get rejected. It's like, this is Einstein we're talking about. So obviously you have to be like, you know, they have these comments.
Can you take a look? And Einstein is just livid. Yeah. He's like, I can't believe you showed my paper to someone without my consent. Given this, I'm just going to publish somewhere else. He retracts his paper. The goat. He never publishes in physical review again. And John Tate goes down as the guy who editorialized Einstein. But he was a man true to his word.
He was like, I don't care if you're Einstein or God, you're going to go through the same peer review. And hey, if Robertson, who's a great mathematical physicist in his own right, says something is wrong, maybe you need to pay attention, right? Seems reasonable. Seems reasonable. Einstein retracts his paper. He gets word from his assistant elsewhere about Robertson, because Robertson goes to his assistant. It's like, you know, you guys are having trouble with these singularities. Like, you know, there's one way that you can just like do this and then blah, blah, blah.
And then the math trick works out. And so Einstein gets word from his assistant that, you know, there's a way to counteract it. He goes back to the drawing board. And Einstein, true to his physics side, right, he sets aside his ego and he's like, okay, that was bad. And he corrects his paper and he publishes it in a Philadelphia journal. I think of the Benjamin Franklin Society or something, some random journal. You know, not the physical review. Right, right. Right. And it's a totally new paper.
Okay. And it's a new paper that admits that gravitational waves can, in fact... be allowed in the space-time of Einstein's field equations. It's so funny, right? Because it's like this guy, like he, you should have just read the peer review. right he could have just read the peer review and like but but he had to go through all this he chose violence he chose violence and like at the end of the day he even admitted to himself that he was wrong and like corrected his paper but he never went back to the peer review that's a level of petty i aspire to be yeah yeah it's like i know i'm wrong but how dare you that's einstein dude you don't yeah anyways he never published in the peer review the the physical review ever again but that paper sort of established this idea that okay
We can have gravitational waves. You can have waves in this system of space-time where space-time is this fabric. Yeah, that governs the dynamics of all things. Okay. So, all right, fine. Now we've got a paper that says gravitational waves do exist. Right. Okay? Just like electromagnetic waves. Yes. Okay. And just like electromagnetic waves, in order to create gravitational waves, you need accelerating bodies. So you can't have a thing that's just moving at constant velocity. Because it's not going to disturb the space.
Because it's not going to disturb space in the way that it propagates out. Yes. Okay. It's going to create a disturbance, but that disturbance is just going to like sort of be local to it. Yes. Okay. And you're not going to get this like radiating effect. It's like if you're in a boat and you're stationary, you don't create a wake. Yes. But if you're moving... But in a boat, even a moving boat creates... This is a big... Okay. This is a fine detail. A boat that's moving at constant velocity is still going to create waves. Yes, that's fair.
That's fair. A charged particle that moves at a constant velocity will not create a light wave. Light is only created from accelerating particles. So something that's moving in a circle means that it's changing velocity. That is going to create a radiating effect. Something that's speeding up or slowing down is going to create a radiating effect. But something that is moving at a constant velocity is not. There's actually a very simple argument for why this is the case. It's one of my fundamental... It's one of my favorite arguments from Einstein's special relativity.
Okay? So, suppose... Not, right? We're going to do this by a proof of contradiction. Suppose not. Suppose that a charged particle that moves at a constant speed does create light. What if I were to boost myself into a reference frame that I'm moving with the charged particle at a constant speed? In my reference frame, that charged particle is stationary. Yes. So I'm not going to see any light. But now there's a contradiction. Yeah, right. Immediately, got it. Immediately, there's a contradiction.
Because a stationary observer observed light, but me moving with this particle does not observe light. It would be something if the stationary observer observed a particle with some light, and I observed it at a different energy, right? Yes. Maybe it was like boosted in ultraviolet or down in infrared or something like that. But the fact that I observed no light... Yes. ...is unphysical. Yes. Because both me and the stationary observer should observe the same physics. Yes. Right? Yes. So...
It's a consequence of relativity. The constant velocity motion does not radiate. The same thing happens with gravity, right? Suppose there's a gravitational object that's moving at a constant velocity. If I boost myself into that reference frame, that object is now stationary, and I shouldn't observe any gravitational waves. Well, if I don't observe any gravitational waves in my... in my frame of reference with the body, then a stationary observer should also not observe gravitational waves. On the other hand, if it's accelerating, if it's moving, if it's accelerating and it's speeding up or it's slowing down, then no matter what inertial frame I choose, it's also going to be either speeding up or slowing down.
So I am going to observe some form of gravitational radiation, or in the case of charged particles, some form of electrical radiation, light. Yes. You know, that's kind of interesting. No, that's a great... It's such a great, like, simple argument you can make using just the symmetries of physics. Yes. So anyways, we need something that's accelerating, okay? So for the longest time, gravitational radiation was just this thing that, you know, it's hypothesized. But even Einstein said, you're never going to find it. Yeah.
Never. Because you can do a back of the envelope calculation about how much stuff would stretch and squeeze by. Right. And it's on the factor of like 10 to the minus 20. Meaning like the actual fabric that would be measurable by us would be so... Any length would be corrected by 10 to the minus 20. Right? Like, if you take a meter stick, an atom is 10 to the minus 10 of that meter stick. Right? So it's an atom of an atom. It's just like...
Even Einstein's like, yeah, okay. Right. Like, this is a cool, like, you know, theoretical exercise, but... So the point being that his view was like, we won't be able to get to a level of experimental verification because of the... The scope is just so outside. Like it was an engineering problem for detection. But the engineering problem required physics that was like stupid. Yeah, yeah, yeah. Right? Yeah, yeah. You know what I mean? It's like a cool thing. Yeah. But it's a cool thing. Yes.
Okay. 1970s actually is when we first got a confirmation that gravitational waves exist. Okay. It was by Taylor and Holtz. Actually, one of their plaques is on the Princeton University physics thing because he won the Nobel Prize for it. I think he was a postdoc at the time at Princeton or maybe a PhD student. So they used the Arecibo. radio telescope.
Have you seen this thing in Puerto Rico? It's a giant radio telescope that's like in a bowl created by the geography of the mountain. Wasn't it recently kind of had like a hurricane came through or something, had a little damage to it? Yes, it did. And then they had to... And they had to fix it and everything. But it's a bowl created by the earth that we then put concrete over. And then it basically just scans the sky with the earth and tries to find radio sources. So they were trying to look for pulsars at the time with that radio source.
And what they found was a pulsar. And they were very happy because back in the day, pulsars were very new, so they found a new pulsar. But this was a very weird pulsar. It was a pulsar that had a cycle of 60 milliseconds. So 17 times a second it would go down. And a pulsar is basically a neutron star that kind of acts like a lighthouse. It's got these strong magnetic fields, so it's got these jets of radio light that are coming out.
And it's a lighthouse in the sense that as it rotates... Right? Yes. The part of the beam, the radio beam, is going to just like go through the earth. Yes. Every 60 milliseconds. Yes. Right? Which is quite fast. It's quite fast. And it's a really precise clock. Okay? It's one of the... These pulsars are some of the most precise clocks... ever made in the universe right because it's really hard to slow down a neutron star that's the size of several suns that just because of the conservation of angular momentum it's just going to keep going okay so most pulsars they're very reliable they'll just be like clockwork this pulsar was different okay okay because this pulsar every once in a while it would the the pulses would arrive
three seconds too late or three seconds too early. Remember, it's going at 60 milliseconds, but you can plot the trajectory of how their delay is. Yes. And that delay was happening three seconds too late or three seconds too early. And what they figured was this pulsar is actually rotating with another neutron star. That other second neutron star is also a pulsar, but we're not seeing it because the lighthouse isn't crossing our line of sight.
So it's maybe doing one of these and we're missing that lighthouse effect. But this one is coming right at us and it's revolving in this binary pulsar system. Got it. This became a really nice test bed for general relativity. Because now all of a sudden you had rotating objects that were massive that were accelerating right yes because they're they're revolving around each other which means they're constantly correcting their motion because of the gravity between them which means if they're close to each other right three seconds in light distance like the moon is about one and a half seconds away so this is imagine like it's the earth and like
Another two moons away. Yes. Two stars that are like the size of the sun and more. Yes. That are revolving around each other, right? So the point is they're quite close. They're quite close. So their gravitational impact, their gravity independently is going to have a high impact on each other because of proximity. Exactly. Yeah. Which means that according to Einstein's GR... these should be radiating gravitational waves. If they're radiating gravitational waves, then they should be losing energy because it takes energy to deform space and time.
Yes. And if they're losing energy, they should be falling into each other and they should be rotating faster. So you can do all of the math. I was going to say you can do the math. You can do the math, and you can figure out, according to GR, how fast should that decay be, right, of their total energy. Yes. And it's a beautiful plot that they have in their paper, where it shows a line that shows the theoretical prediction, and then it shows dots of the experimental. And the dots are lining up exactly exactly. To Einstein's GR.
Which means the math in the GR paper is now, which was theoretical at the time, there's now an experimental observation that was verifiable to what the underlying sort of statement of work is around GR. Yeah, exactly. And it's like, where could this energy be going? It has to be going in the gravitational waves, which is what's... predicted by this equation. And the equation also predicts exactly the decay, and that's exactly what we see. And now we've been watching this pulsar for the past 40 years now. Which makes sense.
And it just continues down. It's like the perfect experiment in the sky. Yeah. For any number of different types of observational research, this being just won. Yeah. And they won the Nobel Prize in 1993 in physics for this discovery. It was a very big one, right? It established the reality, the experimental reality of gravitational waves. Okay. Now, I just want to make a note that what's interesting is it required no specialized measurement equipment at the time. It was just... What we already had, radio telescopes?
Yeah, it was just a radio telescope. Radio telescope and math. Arecibo, yeah. Arecibo and math, right? And so this established at least the existence of gravitational waves, of the fact that space and time undulate in this wave-like manner, and it carries energy, the same exact amount of energy that Einstein's equations predict. Okay? Now, it's something to go from these things exist to... to let's actively try and sense them. Right. Right. Right. Because that's what Einstein was going at.
He was like, this is going to be impossible to sense these. This was an indirect confirmation. Yeah. It's an indirect observation. But it was not a direct observation. It's not like, oh, that was, I just saw one. Yeah. Right. It's just like, oh, this thing is just, putting it out into the sky, and we're seeing it because it's losing energy. Right. So that's the indirect observation. By the conservation of energy, that energy has to go somewhere. Where does it go? Into these gravitational waves. Yep. Okay. So now we're getting to... So this is the 1970s.
They find this thing. Also in the 1970s, people start thinking about how do we actually make direct detection of gravitational waves. So that's where we start getting into people like Feynman, actually. Feynman starts... He gave this talk about sort of a thought experiment of how to actually detect gravitational waves. And he showed that, you know, if you had like... a sort of rod attached to springs and stuff. And if a gravitational wave of insanely high capacity went through, then the rod would actually undulate, because the springs would be attached to that space and time.
So this thing would move, and then part of the energy of the gravitational wave would actually go into the spring, and then if you measured the spring, you could actually measure the gravitational wave. So he showed that mathematics of something like this, A system that could do the direct observation and what the structure of that system would be and why it can work. Exactly. And so given that sort of theoretical work of showing that a system like that would work, this guy Joseph Weber from the University of Maryland in the 1960s and 70s, he created a gravitational wave detector, which was sort of something that he could fit inside a laboratory.
And he said that he found some. Now we know that he probably didn't because people couldn't replicate it as his experiment. And science is all about replication, right? And probably he didn't find any. And it sort of marred the field because, you know, you've got this guy who's saying, I found him, but he didn't. And he was an honest scientist. But, you know, even honest scientists can get sort of hung up and not see the sources of their errors, right? So it was kind of lacking, let's say, the field.
Okay, so this is where the two heroes of our story come in. Kip Thorne and Rainer Weiss. Okay? And to really understand where they come in in the 1970s and 80s, we got to go back to their roots. Okay? Kip Thorne gets his PhD at Princeton under John Wheeler. Okay? John Wheeler is a...
seminal figure in the field of physics. He's kind of like, I would say, the American Arnold Sommerfeld. Remember Sommerfeld that we talked about? Sommerfeld being this guy who tutored amazing physicists in their own right, but never won the Nobel Prize himself. John Wheeler is one of those guys. He tutored Kip Thorne, won the Nobel Prize. He tutored Richard Feynman, also got his PhD under John Wheeler. He tutored Hugh Everett, who is the guy behind the many worlds interpretation of quantum mechanics, this idea of multiverses and all the, you know, Marvel stuff.
He also tutored Jacob Bekenstein, which is a guy who did seminal contributions to black hole thermodynamics, the idea of, like, entropy in black holes and information in black holes. He sort of revamped... He made general relativity cool again after the war. Yeah, okay. Yeah, because... General relativity was sort of this field that was not cool, okay? It wasn't something that you could do direct detection on. Einstein had sort of closed the book by finding the Einstein field equations.
And then people thought it was sort of just, oh, like, you know... It's just book work at that point of like trying to find new directions with GR. Is it similar to how when people say string theory is cool on paper, but we can't do anything experimentally to really see the things that are being argued? Yeah. Like a parallel. It's kind of parallel. GR, I wouldn't compare to string theory, obviously, because GR actually has insane experimental, even back then. Okay, fair. Right, even like Mercury's orbit was explained by GR.
Fair. One had done the detection of bending of light around a solar eclipse. That was Arthur Eddington. So GR had its confirmation. So you're saying string theory is trash. I'm kidding. I'm kidding. Well, okay. Actually, speaking of string theory, string theory actually started with John Wheeler. John Wheeler... came up with this idea of the S matrix, which was, I don't understand much about it, but it's some way of like defining a new way of dealing with quantum mechanics that, that had to do with combining gravity and quantum mechanics.
S matrix was then taken by Werner Heisenberg and developed into a theory. And then that sort of. form in the string theory later on. So John Wheeler is even, in fact, responsible for the genesis, the very, very embryonic genesis of string theory. He did a lot of really crazy stuff, dude. He did S-matrix theory. He coined the term wormhole. He did the one electron universe. I don't know if you've heard about that, where it's like every electron is just one electron in many different... Like, people love trying to go into, like...
the science fiction of this kind of stuff. He popularized the term black hole. He also actually was one of the first people to foray into quantum gravity, of trying to combine quantum mechanics with gravity. So this incredible, incredible physicist sort of became the, you know... this godhead status at Princeton physics, attracted the talent, created talent with Kip Thorne and with Feynman, but he never himself won the Nobel Prize.
He was the prince who never became king. Yeah, but I mean, to those who know physics, like John Wheeler is a very big deal, right? Yeah, he was one of the few people who... He defended Freeman Dyson when Freeman Dyson was going up against Oppenheimer. And yeah, it was like... He's one of the few guys that can go against the bigwigs, even go up against Einstein and be like, Nah, that doesn't make sense. And Einstein would actually listen instead of being like, I'm never going to publish here ever again. You know?
Yeah. And Kip Thorne was Wheeler's PhD student. And Wheeler sort of started this idea of revamping general relativity. Okay. Kip Thorne grew into that. They wrote textbooks together. And Wheeler was the sort of theoretical part of the wing. Mm-hmm. And then there was this experimental part that was Robert Dickey. Robert Dickey is another one of these guys, great prince and greats. He's responsible for the invention of the lock-in amplifier, which is an electrical way of finding small electrical signals that are of some characteristic frequency in a bastion of noise.
He also... was the discoverer of the cosmic microwave background. Okay, the CMB, the radiation from the Big Bang. He was one of these first sort of... Applying general relativity to cosmology problems. And him and his student Wilkinson and Peebles published the paper right after the experimental discovery. They were actually planning to make a dish to observe the cosmic microwave background. And then meanwhile, unbeknownst to them...
Penzias and Wilson were neighboring in Bell Labs, which is right down the street from Princeton. They were stuck with a microwave antenna. They're trying to do research with microwave antenna, and they keep getting this noise. Yep. Yep. And they're like, what the hell is this noise? And they're trying to find where the noise is coming from. One of them actually crawled in and started scraping the pigeon poop. That was in the antenna. Because they're like, I don't know, maybe it's the pigeon poop. And they're like scraping. They got rid of the pigeon poop. The pigeon, the noise was still there. And they're like, dude, what is this noise that we're getting?
And they called up Robert Dickey. And they're like, we're getting this noise. You guys are physicists at Princeton. And Dickie's like, he hangs up the phone and he tells his collaborators, boys, I think we got scoop. Because they're trying to make a thing. And these guys have done it. Penzias and Wilson immediately recognize what it is. They immediately publish. Dickie, Peebles, and Wilkinson published their paper on the CMB, the theoretical background, right after theirs.
So you can see when they published, actually, it's like back to back in the same issue of the journal they're publishing, right? And Penzies and Wilson go on to win the Nobel Prize for the discovery. It really should have been a three-person with Dickey. But actually, years later, I think in the last two or three years, Jim Peebles, who was the guy who published the theoretical paper, he actually won the Nobel Prize. And it was kind of a validation. It was kind of a posthumous...
Nobel Prize for Robert Dickey, who had already died. Yeah, yeah, yeah. But it was kind of cool that, like, you know, that collaboration finally won the Nobel. It got recognition. Yeah, it got recognition. Peoples won it for all of the other stuff that he did with cosmology later on. He actually showed that, you know, the CMB should look like this with this bumpy, wavy nature when you look at the power spectrum. And then when people did the power spectrum, they found the bumpy, wavy nature. Wilkinson, who was the other guy who did the... paper with them.
The WMAP satellite, which is the Wilkinson anisotropy probe that was sent into space to actually capture the cosmic microwave background in excruciating detail, that was named after Wilkinson. So, you know, they got their credit in their own right, but it's unfortunate that Dickey didn't get his... recognition with the Nobel Prize. But one of Dickey's postdocs was Rainer Weiss. And Rainer Weiss still, to this day, says that, you know, one of his biggest influences in terms of physical thinking was Robert Dickey at Princeton.
There's a beautiful photo of Kip Thorne and Robert Dickey standing right next to a plaque that commemorates... Sorry, there's a beautiful photo of Kip Thorne and Rainer Weiss, the two Nobel Prizes that were awarded for LIGO. They're standing next to the... plaque that commemorates Robert Dickey at Frisk Campus Center, right? There was a big celebration of his work, and both of them were in attendance, along with Wilkinson, who was in attendance.
So, you know, it's this big tradition that sort of started at Princeton of, like, let's take GR seriously. Yeah, yeah. Okay? They move on to their different things. Kip Thorne goes to Caltech. Rainer Weiss goes to MIT. And they start thinking more earnestly about gravitational research in the experimental domain. Okay. And they're like, can we actually do this? Can we actually... The thing that Einstein said we couldn't do. Like, say that Einstein couldn't do. Can we actually do this? Okay. Reiner Weiss starts thinking about it.
He's more on the experimentalist side. So he's from Dickey's group, right? So he's trying to actually think about the experiment. And he comes up with this idea of the interferometer, which is really what LIGO's design is today. The idea is you've got an L-shaped instrument. where I have light moving in this direction and light moving perpendicular. It's the same setup that Michelson-Morley used back in the late 1800s to show that light was moving in the same speed in both directions, regardless of an ether and all this other stuff.
That same thing, what if we just made it big? Okay? Okay. What his main contribution, what Rainer Weiss's main contribution was with that paper in the 1970s and 80s was to figure out what are the types of noise that I'm going to run into when I make an instrument of this scale and I want to measure something this precisely. Okay? And he came up with a list of all of the noise sources that they're going to have to deal with. There's going to be seismic noise.
Right. Earthquakes. Yes. Even an earthquake that's on the other side of the world is going to create low frequency vibrations on the earth. At the level that we're trying to measure. And that's going to create stuff. The waves from the ocean. That's going to create noise. Cars and trucks, people walking by, that's going to create noise. Elephants in the Sahara. Yeah. Like all the stuff that you could think of. He created a list of all the stuff that we would have to come up with. And at the end, he had something called quantum noise.
Okay. Which is literally the Heisenberg uncertainty principle. Yes. Okay? Yes. It's like there's noise that we're not going to get rid of ever. Yes. Because that's just how it is. Yes. And we'll get back to that later because the AI story actually ties into that. So he made a list of all of these noise sources and he came up with sort of the ways in which we can make this happen. Kip Thorne, on the other hand, he starts doing theoretical calculations of what are the sources in astronomy that could create...
these gravitational waves, right? That we would be able to see. How often would these go off? Right. Given our assumptions about the universe. Right. And theoretically, is it possible, like that Heisenberg uncertainty limit, what is that limit? Right? Are we hopeless? Or is the Heisenberg uncertainty limit just below the sensitivity that we need to actually find these things? Right. Okay? So you've got these two, Rainer Weiss is spearheading the...
Approach from the experimental side, and Kip Thorne is spearheading the approach from the theoretical side. Yes. Okay? They think that, you know, maybe this could actually work. Okay? They're like... It's just so crazy. They're like, maybe this could actually work. The audacity. Yeah, dude. And in the 70s, they decide to pitch it. to the nsf okay national science foundation that's the national science foundation they pitch it to the nsf and the nsf is like this is this we have this amount of money yeah we have this amount of money and what you're saying is something that einstein said wouldn't work
So they're like, all right, we're going to give you a little bit of funding. Why don't you try to get down to these sources of error? We're going to give you enough funding where you're not going to find gravitational waves. But you can at least start making a prototype to figure out if you can mitigate these sources of error. Right, right, right. Right, right, right. Before we even get down to that 10 to the minus 20, let's get down to 10 to the minus 10. All right. Where it's like, can you even theoretically like and experimentally make such a... instrument.
Yes. Okay? So they start going into it... Caltech gets a telescope that's about like 90 meters, which is pretty big even then. MIT also gets an instrument and they start experimenting with this kind of stuff. And it seems promising. It seems like they're actually working. So then Kip Thorne recruits Ron Draver, who's from Scotland, who had also worked on the early parts of this gravitational wave research.
And he gets them together and he's like, all right, we're going to pitch this thing. This is going to be big. It's going to be a big deal. All right, boys, we ready? So, they start doing it. Upper management has turmoil. Personal reasons, Ron Draver doesn't get along with this guy Voight, who's also part of the experimental corroboration. They just really don't get along. You hate to see it. You hate to see it, but the two of them are like, I'm out. I'm out. So now Kip Thorne and Rainer Weiss are just like, okay.
Okay. It's just us again. So it's just us again. NSF halts funding. Oh, God. Because they're like, you guys are like headless chickens. You have no idea what's going on. Um... We're going to halt funding and we're going to, if you guys come up with a strategy that actually makes sense and that isn't like petty, then maybe we can talk. So that's when Kip Thorne invites Barry Barish, who was at Caltech. Okay. Barry Barish is, in 1994, he's brought on as the new director of the project.
And Barry Barish is the reason why producers get the best picture award in the Oscars. Okay? The Oscars give the best picture award not to the directors or the writers. It goes to the producer. It's the guy who's, like, putting it all together. Barry Barish is that guy. This is so random. It's so random. But he's really an experimental physicist who... who does particle physics research.
He was involved with the superconducting supercollider, which had failed in Texas. So he had already learned his lessons there. He knew how to deal with people. He knew... how to figure out what he wanted. He knew how to talk to people to get what he wanted. And he knew how to get people on board. He was the politician of the group. He was the guy. Yeah, he was the guy who saw the big picture. Yes. And he's the guy who, when he was brought on... He transformed LIGO from a 40-man operation to this thousands of people.
That is what is required to make that detection. He became the wartime CEO. Dude, and actually you say that, but somebody actually said to him, like when they were advertising, like who Barry Barish is, at the time, you know, the war in Iraq wasn't very popular. Yes. And they said, you know, if Barry Barish was in charge of the war in Iraq, it would have been over by now. That's how they advertise him. He really was the wartime CEO. Yeah. He was the guy who was like, okay, we're going to build this thing. Yes. We're going to require two detectors.
Right. We're going to have one in Hanford. Yes. In Washington State. Yes. This is actually the original site where we enriched plutonium for the Manhattan Project. Now it's being used for this. There's going to be another one in Livingston, Louisiana. Yes. Okay. We're going to build them two. And he really started this collaboration and it started taking a life of its own. And even Kip Thorne and Rainer Weiss were like, thank God for Barry Bearish. You know? Because he just full steam ahead.
Like, we're going to build this thing. I'm not going to have another failure like the SSC on my hands. All credit to him. He created that collaboration that built this. Every project always needs the operations guy. It's funny. My dad works in stage one drug discovery. And he was over in Scotland. And there was a researcher at one of the Scotland universities that had a really interesting fundamental discovery. He went to go to VCs to say, hey, give me millions of dollars to make this a thing.
And they were like, OK, great. And he wanted to be CEO. And they were like, we need a... very bearish we don't need uh you and my dad came in as that operations person who like understands the fundamentals yes but understands the way in which the business world and all these things that are not about the science yes necessarily yeah do matter a lot exactly about these huge large-scale projects yeah yeah when you're talking about that this is something that requires something beyond like knowing how the einstein field equations were right Which obviously Barry Barish does.
He's a trained physicist. Right. But his real expertise is in this ability to manage. Yes. A thousand people towards one goal. Which is not trivial. It's not trivial. No. And this is the guy who did it, right? So fascinating. And he, you know, so he started construction. Yes. He got the money from NSF. Yes. He convinced NSF this is going to work, guys. He got the money from NSF. It took five years to build the buildings. in Livingston and Hanford in Washington and Louisiana, where, you know, you've got these three kilometer long tunnels, right?
And the tunnels are so long, they have to be kept completely straight that they're accounting for the curvature of the earth, right? Where they have to actually go under a little bit and then get over, right? Yes. Um, In the entire tunnel, you've got to create a vacuum that's better than outer space in the entire three kilometer tunnel. And it's like the stupid stuff was actually hard, like creating a tunnel that's better than anything we've ever had. Yes. Right. All of the stuff was completely brand new. Right. You had to create the best mirror of all time.
To have these laser interferometers where the laser is going down both and then bouncing back. That mirror needs to be near perfect. And it is near perfect. It's near perfect glass. It's the best mirror that we've ever made. This mirror, if you left it ringing, right? Not that you would want to do this. But if you rang it and had it vibrate, that thing would vibrate for 100 million times. Before actually stopping. Because there's nowhere for the energy to go in this vacuum. It just keeps self-propagating. Yeah. So it just stays within itself and becomes this like perfect spring pendulum thing that just vibrates, right?
So you've got to create all of these new technologies, right? Just to create the environment. Just to create the environment. And finally, in like 2001, you had LIGO in its first iteration. This is when the technology is there to find if you're lucky... gravitational waves. And this was just a proof of concept to say that, okay, we've reached this far where we've reached the sensitivity level where if we get really lucky and there's like some massive merger that's like right next to us in the galactic neighborhood, we're going to find it.
They didn't find anything. But that's science, right? The probability was very low. They solved a lot of the engineering problems and this was the crawl, walk, run approach and we're like in between the crawl and walk stage. Yeah, they were in the walk stage at this point. Got it. Okay. And then they upgraded LIGO. To this LIGO 2.0, to this higher sensitivity. And this is right around the 2010s. Okay. They're upgrading. And this is now the run. Now we're getting to a point where it is likely that we will catch gravitational waves.
If it were to pass the Earth, we are now able to detect at a sufficiently small level... That it will detect it if it is there. That's right. Yeah. And so then comes the detection. Okay. This is September 14th, 2015. It's at 4. Yeah. It's at 4.50 a.m. Louisiana time, the LIGO computers at both Louisiana and Hanford detect something.
Okay? They detect something and they've got an automated email system that sends it out to like a core group of people. It's like that scene in movies when they detect aliens for the first time. Exactly. All the sirens go off, everyone gets an email. Wait, wait, wait. Is it? Is it? Yeah. In this case, so this was still, they were still in the middle of their run phase, where they were still testing stuff. They were like ramping up. So the PA system in the actual control room had not been like configured yet. So funny. So it just, the email went out.
Right. Okay. And the people in the control room had no idea. That's so funny. Okay. So LIGO had just had this $200 million upgrade, right? It was called Advanced LIGO at the time. But they were still getting the bells and whistles of this upgrade together. And there was a likelihood, but people weren't thinking it was going to happen so soon, right? And so the alert goes out, and it's only a handful of data analysts. They see this thing. There's a guy in Germany. There's a guy in University of Florida.
There's obviously guys at Caltech and MIT. They look at this. They're like, yeah, right. Yeah, right, right, right. Yeah. Yeah, right. Because it's so beautiful. Oh, no. The signal is like something, you know, like you do simulations and you're like, this is what it should look like. And then out comes an email saying, hey, this is what I found. It's like, okay. It's exactly like the signal. It's like exactly like what you're saying. So it's like, no, right? And all of the big scientists are like, no, this is too good to be true. The guy at Caltech actually was just like, I don't have time for this.
He looked at the email. He's like, and then he just went on about because he's got other stuff to do. It's advanced LIGO. And, you know, his grad students and postdocs are like, did you see the email? And he's just like, dude, it's you young people. This is not how it works, okay? The world is not idealistic. Yeah, yeah. This is not how it works. And it just wouldn't go away. Because usually, you know... They used to have these things called, they would inject noise, blind injections to see if the system works.
Yes. So maybe there's a guy who's injecting the fake data in there to see if the email alert works. Right. So they call everybody who's in charge of that. And he's like, no, dude, I didn't do that. Yeah. They go back into the nitty gritty of the data, of the raw data. They're like, it's in the raw data. Yeah, yeah, yeah. And they're like, okay, so now they just start getting real paranoid. Yeah, yeah, yeah. They're like, who is doing this? Yeah, right, right, right, right. Okay, who is the evil genius that like went into the Rod? And it turns out there's only a handful of people that know the instrument at such a level where they can, they've covered all their tracks.
Right, right, right, right, right. Right? And pretty soon they start thinking like, actually, it can't be a single person. Yeah. Because it happened at both Washington and Washington. Louisiana. So it must be a bunch of people. So now it's like there's a giant conspiracy and everyone's looking at each other like, okay, what's going on? Then it turns out, it turns out, so this happened at 4.50 a.m., right? That night on September 14th, there were two postdocs outside of the Livingston facility that were trying to test the system.
With local noise. And what they were doing was they got in their car and they were just driving down the arms. Oh, yeah, yeah, yeah. And just like accelerating and then braking. And accelerating and then braking. They were trying to see if this would inject noise into the Livingston facility. Yes. But they had stopped at 4 o'clock because the guy had to catch a flight at 6 a.m. He's like, I need to get to the airport. Yes. So... So they call up these postdocs and they're like, what were you doing at 4.50?
It's like, 4.50? Actually, no. I know for a fact that I was driving to the airport. Yes. So it definitely wasn't me. I was nowhere near. And also, like, why would that happen at the Washington facility? But at this point, everyone's, like, real paranoid, right? So they're trying to be their own worst enemy, trying to see who faked the signal. Because this signal is just too good. Yes. Right? Like, God isn't this kind. Yes. And at the end of the day, it's like, okay, is this a prank? All of the dudes that could have done this get in a room.
And they're just like, all right. Did you do it? Did you do it? What about you? What about, did you do it? And everyone's like, no, we didn't do it. And it's like, shit. Silas descends across the room. Yeah, it's like, dude, what? Okay, so then... You say, okay, so nobody did this. And it's kind of like that Sherlock Holmes quote, right? Where it's like, when you eliminate the impossible, what remains, however improbable is the truth, right? Is the reality. They're like, dude, this is crazy. So now they start having, they have a blind data augmentation technique where they have a blind team that tries to decipher...
How likely is this to be chance that the detectors just like randomly, maybe there was an earthquake that was right in the middle that like arrived at the same time or something, you know, or like the detectors themselves like did this like fluctuation randomly. How likely is that? So they start going down this rabbit hole of like, okay, what if, how much of this is a fluke? Because you got to figure out a confidence, right? Remember the confidence is five sigma. So these guys are in charge of that.
And they go through and there's this big unblinding party where they open the box. It's like a gender reveal. So what happens is they do all of their analysis without actually looking at the real data. Got it. And then they open the box and now they take the real data and say, okay, given our analysis, where does the – given our distribution, where is the real data? Okay. They don't even know. with the real data. It is, right, right. Okay? They've been doing this blind. So they have this unblinding party where they open the box, and everyone knows when this is going to happen.
And the party happens, you know, their Zoom meetings are only set up for, like, 50 people, but, like, everybody in the collaboration shows up. So it crashes. Right, right. And so then they're scrambling, they're like, okay, we've got to get the Zoom meeting up, finally get it up. They unblind. And this is well above Five Sigma. Oh, my God. And everyone is like, that's when the champagne comes out. Yeah, yeah, yeah. That's when they're like, wow, we got him. We got him. We got him. We got him. We got him. We got him.
a single gravitational wave that was like so ridiculous, right? Because had those, had those two postdocs been going on their thing for like through four 50, we wouldn't have, we wouldn't have caught this thing. It's such a, it's, it's now become like part of like the LIGO lore of like, and those two postdocs are now like famous because like, like they decided to like call it early at like four o'clock in the morning, you know, it's so cool to think about that. Like,
Yeah, like we were listening to... You know, gravitational radiation from billions of light years away from two black holes that merged. And to think about the incredible, like, sensitivity of this instrument. We really haven't gotten into just how incredible this is. I threw around a number like 10 to the minus 20. Yes. Right? To give you a sense of how small that is, that's like measuring the distance from the Earth to the sun.
And then you add an extra atom. Ah, yeah, yeah. To that distance. To that distance. Yeah, yeah, yeah. That's stupid. Yeah, no, I don't... This is where... This is why Einstein was like, no. Yeah, no. Because it's so outrageously arrogant for us to think... To think that we could measure that. We could measure a disturbance at a degree that is so small... And be able to make it verifiable. Yeah, yeah. 10 to the minus 19 meters is what we actually did, which 10 to the minus 15 is a proton.
So this is 10,000th the distance of the diameter of a proton, the size of a proton. So we have these two 3-kilometer tunnels on different sides of the country. And we're measuring a difference. Of a tenth of a proton. No, a 10,000th. A 10,000th. Of a proton. Yeah. Right? Like the detectors themselves are made out of atoms that are made out of Protons. And the protons are like a ten thousandth the size of the atom. And now you've got... So it's an incredible achievement, right?
It took a thousand people. Experimentalists, theorists, NSF bet big on this. The National Science Foundation really like bet the farm on this thing working. And it worked, right? It took decades. Yeah. And we got them. And we got them. It's a new way to look at the universe, right? Before we were limited to electromagnetic radiation. We had pulled all the stops when it comes to electromagnetic radiation. We'd gone all the way from radio, which is long wavelength, all the way from radio to infrared, microwave, infrared, visible, the optical spectrum, ultraviolet, X-ray, gamma ray.
We've got gamma ray telescopes that are looking at gamma ray bursts. We've done particle detectors, right? We've had neutrino detectors, cosmic ray detectors. Many of which we've talked about previously on this podcast. On this podcast. And now we were listening to the universe, not just looking at it, right? We were listening to the space-time ripples. And what's crazy is like the ripples that come in – especially for this first discovery, was in the human audible range. And so you could actually just hook up an amplifier and hear it.
And hear these black holes coming in. It would be a chirp because the frequency would increase with time. So it would go, like that. Because the black holes are coming together and as they get together, conservation of angular momentum, they're going to speed up. And as they speed up, they're going to release higher frequency gravitational waves. And so the... the signal that we were seeing. And that's why these guys were like, there's no way this is real. Right. Because if you look at the spectrogram, the frequency versus time, you see this characteristic thing.
It's like exactly like in the simulations. You're just going, come on. There's no... There's no way this is real. Right? And so it took them five months from detection to, like, being confident enough where they were like, all right, let's announce it. Because as we've talked about, you don't want to put something out that, you know, is not prepared for – because everyone's going to come at you and not. Yes. Yeah. And especially because in the past, like with Weber and his gravitational wave detector, right, those were fake. But like this, at this point, it's like they had done all of the work and the International Corral Collaboration had done all the work.
They really weren't very confident until they saw the second one. They saw a second one and they're like, oh, okay, this is real. We're like literally listening to the universe. Then the Italian crab collaboration, Virgo, came online. Virgo is a detector very similar to LIGO. It came online and the three of them, so the two LIGOs and Virgo, detected a neutron star merger. And that was a big deal because now that you've got... Three, you can actually do triangulation. Triangulation, right. Okay? With only two detectors, the only signal you really have is the time difference between the arrival here and the arrival here.
Now, gravitational waves propagate at the speed of light. So, for example, if they both arrived at the same time, then you know that it's on this plane. Yes. Right? And it came from this angle. Yes. But if it arrived here before and then here later, then it came from this angle. Yes. Right? And you get this sort of arc in the sky of where it could have come from. Right, right. But that's a giant arc. Arc, right. Okay? If you've got three, then you've got... Three arcs. You've got an arc here, an arc here, and an arc here.
And wherever those three arcs intersect has got to be the spot that it came from. So then once we had the signal from all three, we could point our telescopes to it and actually see what is generating it. The gravitation. And that was crazy because what would end up happening is the gravitational wave comes in first because there's nothing that's impeding gravitational waves. It's just going. Neutrinos actually came in right after because neutrinos also, they're very, they don't interact with anything. So they're just coming through. Yes.
Light was actually later because light was interacting with all the stuff. The stuff. And then there was a tiny delay. Yes. And so we had just enough time to point our telescopes to this thing and see the neutron stars colliding. And that's where we got our first sense of where these heavy elements, the precious metals, like platinum, gold. These guys, they actually come from these neutron star mergers. More than like supernovae and things like that. It's from these neutron star mergers where we get the gold that's in my wedding ring and things like that, right?
You know what's so... And so it opened up on... It's just so cool. A great story to look back and listen to is our story covering Ice Cube, the neutrino detector in Antarctica, because it kind of explains this neutrino travel time and light travel time and why there's a difference between those two, which the implications of which... dovetail with this gravitational wave detection first neutrino right after light then shortly after why is that happening go back and listen to our ice cube story because we kind of touch on the mechanics exactly of light traveling and neutrinos traveling yeah i just it's funny because all these things again these are just they're all connected
Yeah, they're all connected. And now we've got a fourth one, Kagura, in Japan. We did a story on how all four found the largest black hole merger. That was another story on the podcast. So now we've got four, so we can triangulate even better. I think India and Australia are in plans to make their own gravitational wave detectors. So now we're going to have this entire collaboration. And now... The trick is, so we've discovered gravitational waves.
We can now listen in on gravitational waves. We're getting signals at about the rate of like tops like five a year. Yes. In the volume of the universe that we can listen to. Yes. Right before it gets like so small that we can't listen to it. Yes. Yes. So they're at some proximally close location. Yeah, yeah, yeah. They're close enough where the gravitational wave hasn't dissipated so much that we're not sensitive. Now, if we make it more sensitive, then we can expand that volume, right?
We can actually expand the radius, and the volume goes up by r cubed. Yes, yes. So just even 10% to 15%. Difference in the radius amounts to twice the volume, right? Because you're going to cube that. Yes. So now the challenge is, how do we make LIGO better? Where we're not only seeing a larger volume, but the stuff that we're seeing, we're starting to notice all of the overtones and all of the stuff that makes a musical note great, right?
If you've heard like, you know, the... The note that comes out of your phone when you're trying to tune your guitar, that's a pure note. It's a C or an A. Not very interesting. Right? But a piano, why is a piano note interesting? It's because there's all of these overtones and nice characteristics from the piano itself. that give it that beauty. Would a way to frame it, like the texture that the physical piano adds to the tone itself, which is a manifestation of its physical construct, like how it physically exists on all of that.
Yes, yeah, exactly. Like a guitar string playing the C is different from a piano string playing the C, is different from a violin resonating at C. And it's all because of that texture of the note. characteristics of that musical note. The overarching note, if you were to do a physics... breakdown of it, the overarching node is still going to be that base C. Yes. But all that stuff on top is what gives it the beauty. So the question is, how do we make LIGO see that beauty?
That piercing. In the gravitational wave. Without any of that extra texture. Instead of seeing just the C. Oh, seeing it in its... I got it. You know what I'm saying? I understand what you're saying. Instead of seeing just the node, which is what it's good for right now, we're seeing I would do it rip, but now there could be stuff on top that would give us characteristics about... the environment that the black holes were in, where the neutron stars were in, right? How long they've been going at it. All of this other stuff. All of the other characteristics. And that's where the recent paper that's out of Caltech and DeepMind comes in.
Got it. Okay? What they're trying to do is push the sensitivity of LIGO to levels... that were unimaginable before. I mean, LIGO itself is unimaginable. Unimaginable. Right? But now they're taking it a step further with the use of AI. Okay. Okay. So why do we need AI, right? LIGO has this problem that's called a wall of noise. Okay? There's a wall of noise right at 10 to 30 Hertz.
Okay? Above 30 Hertz, they're pretty good. Okay? But below 30 hertz at these low frequencies, they're having a real issue. There's too much noise? There's too much noise. Okay. The noise is coming from seismic effects. Like whenever you have a giant earthquake, the earth rings. in these low frequencies and it rings for a while. Okay. On the other, also the moon tugging on the oceans creates these tides and the wave action itself is ringing around the earth creating, so we're pushing that level.
Got it. And we're trying to get rid of that. Now, what they've tried so far is something called feedback control. and you've actually used feedback control in your own life, you have AirPods, right? Yes. The new AirPods, they've got noise cancelling. You've probably used noise cancelling headphones. The way noise cancelling headphones work is they've got a feedback loop that senses low frequency noise, and then that sensor then counteracts that noise and destructively interferes with it to cancel it out.
Mm-hmm. Is this creating an interference pattern so it cancels it out? Yes, exactly. It's creating an interference pattern so that it cancels it out. This works really well on the airplane because the airplane is a low frequency hum, right? Yep. That's the ambient noise. That makes sense because I just was on a flight and I was like, these are really good. Yeah, the reason why is because the airplane is low frequency and you can do that with low frequency because with low frequency you have enough time to respond with your control system. Right. Right? The time scale of that vibration is slow enough where you can actually like counteract.
Yep. But now comes a tradeoff. Right? There's a tradeoff because aggressive control at these low frequencies means you're going to introduce noise from the control system itself. Yes. At higher frequencies. That makes sense. Right? Now, for noise-canceling headphones, it's fine. You might sometimes, if you're really careful, you'll hear a really high-pitched tone in your noise-canceling. Yeah, yeah, yeah. If you're, like, really trying to find it. Right, right. But what you're really worried about is the loud, low frequency.
Right. And so you're getting rid of that, you're happy on an airplane. The level of precision we need at headphones is a very different scale than what we're trying to do when we talk about detection at this level. Yes. At this LIGO's level... Right, right. Now even that is annoying. Yeah. Right? It was not annoying because that's what allowed us... The feedback control on the mirrors themselves, you've got these actuators on the mirrors because the mirrors themselves have noise, right? Yes. They can be tilted in a certain way. They can be up and down. If there's some seismic effect, then the mirrors can wobble.
And then the wobble is going to be greater than the... you know, 10 to the minus 19 meters that you're trying to do. It's incredibly ridiculous, right? How still you're trying to make these mirrors. Yes. Okay. So in order to actually get that, We use these feedback control mechanisms, right? But the feedback control mechanisms inject high-frequency noise. The target would be to get rid of that high-frequency noise and be limited by only quantum noise by the Heisenberg-Uncernib principle.
From that list that we talked about earlier. Yes. We want to be limited by the universe itself, right? And the laws of quantum mechanics. Yes. The quantum mechanics noise comes in two forms. There's shot noise, which is, you know, the laser itself... I mean, at this rate, we're talking about photons hitting the mirror. And the photons are going to hit the mirror at random arrival times. So that's called shot noise. Because the difference in arrival time is noise. The difference in arrival time in the photon is enough to cause the shot noise.
Oh, my God. That's the level that we're working at with these mirrors. These mirrors are like kilograms big. And they're worried about tiny photons hitting it at random times. That's what we're complaining about. Yes. About. The other limit is quantum back action, which is not only is the photon coming at random times, but when it hits and goes back, it's going to transfer momentum. Yeah, to the mirror. And the mirror is going to wobble because the photon has a reaction force. Right. And because there's multiple points, the reaction force is happening everywhere.
Yeah. And it's like, it's like, it's like, so... So that we can't control because the universe is like, you're doing too much. Yeah, yeah, yeah. You know, this is admin privileges only. Yeah, yeah, yeah, exactly. The fact that you're getting this far already, I'm like, wow. But yeah, let's slow down. But that's where we're trying to get to. Makes sense. To having that being the only remaining noise in the data set. It's like, then it's like, okay. And so that's where this scientific AI comes in. Okay. Okay.
What these guys are trying to do at DeepMind and at Caltech is they're trying to quiet LIGO's control systems using a radically new AI approach. Okay. Okay. And this isn't the first time that AI has been sort of used to, you know, help scientific collaboration. Yes. But most of the time AI... has sort of played the role of a data analyst. Right. Right? After the data has been collected. Like we've talked about the Vera Rubin telescope, right? Once the Vera Rubin takes all of these pictures, AI can go through and sift through and find asteroids that move around or new transient effects from supernovas and things like that.
AI can also be used in cosmology for running simulations faster for the universe's evolution and things like that. But this... idea of using AI actively in the data acquisition and in the experiment itself is kind of new. It's not without precedent. CERN uses it at the Large Hadron Collider. They actually use AI algorithms and machine learning systems to actually trigger whether to keep... a certain particle scatter data or whether to discard it.
So there's a machine learning algorithm that basically makes this split second decision on whether a particle collision and the shower that came afterwards is interesting or whether it's something that you don't want to save. The amount of particle collisions that are happening are astronomical. I mean, even though CERN's data storage capabilities are insane, it's not going to be able to save everything, right? So you need some way to sift through that, and machine learning is doing that. It's also accelerating simulations on these particle systems.
But this approach from LIGO is another thing in that vein of using AI actively in the experiment itself. Yep. So what they figured was, okay, we're trying to keep the mirror as still as possible. Yes. Okay? Why don't we use a reinforcement learning agent to train and devise its own strategy on how to keep the mirror stationary? And it's great that we just did an episode on reinforcement learning. Yes. Because it goes very deep into the structure of why. This was in our DeepSeek coverage.
Yeah. Just did these two papers on how they went from human feedback to... Yeah, to just pure reinforcement learning. And so the point about pure reinforcement learning here is that you as the human who is designing the AI system are not giving it feedback about whether the answer that it's going down is correct or incorrect, which enables – I think the point here is like novel ideas about how to solve the problem. That's not limited by our perspective or our perceptions. Exactly. Yeah, exactly. In this case, the reinforcement learning agent is going to just try all these different strategies, see what worked, and then go down that path and make that strategy better.
And that's exactly what they did. They created a reinforcement learning agent. They gave it an environment, which is this highly accurate computer simulation of LIGO. I mean, they can't give it LIGO. LIGO is busy trying to listen. So they made a computer simulation of LIGO, and that's the environment that this agent is now playing around with. Yes. The current mirror was the state, like whatever the mirror was. And then the action it could take was, you know, moving it with all the actuators. Yes. And it could play around and figure out how to keep this mirror as still as possible.
Yes. And you can inject noise into it and see how it would do. And you can do it... Hundreds of thousands, millions of times and different parameters. Yes, different parameters. And you can see and you can give it a reward every time it does a good job. And you can penalize it every time it does a bad job. In this case, the process of actually rewarding it was actually kind of cool. Because you can't just be like, keep it still. Yeah. That's a little too much. Freedom. So what they did was they created three distinct bands in the frequency space.
They called this approach the DLS approach, which is deep loop... Shaping. Loop shaping, meaning this kind of control paradigm where you're worrying about the frequency response of your system and deep from deep learning. Yes. Okay, so deep loop shaping. What they did was they got that frequency band of all of the noise and they split it up into three different categories. So there's a low pass filter for stuff that's below three hertz. Three hertz, meaning three cycles per second. And that's where all of the seismic disturbances, your sort of...
the tides and all of this stuff comes up at low frequency. For people who work in music, this is like all the bass stuff. All the bass, yes. This is the bass. Right, okay. Yeah. And with that bass stuff, it basically said, you're going to get aggressive stabilization. We want to teach you aggressive stabilization. You get a high reward. If you eliminate the bass noise. Okay, great. Then there's the bandpass filter of 8 to 30 hertz. That's the part that we're trying to really optimize. Okay.
That's where the control noise comes in from these guys. And so what they gave was a penalty, a negative reward. If it got any noise in that 8 to 30. In that 8 to 30 hertz. Okay. The low part was just reward. Yes. This has a negative reward. Okay. It's like penalizing heavily. And the high pass filter also penalizing. Okay. But different weights. Yes. And then the crucial thing what they did was they multiplied the rewards at the end.
Mm-hmm. to give a final reward instead of adding. A lot of reinforcement learning techniques have added with additive reward. Nowadays, people are slowly foraying into this idea of multiplicative. The problem with multiplication of reward is it can easily blow up. Right. And like be very dangerous. Right. But there's come mitigation strategies against that. Got it. And so now with multiplicative reward, you force the agent to perform well on all three tasks.
Right. Not just one. Right. Because you could have one in the summation model. You can have one that's super, super high. Yeah. That kind of drowns out. Yeah. And it's just like zero noise in the base. Right. But then you're just like... Completely wrecked in the other ones. That's actually interesting. Through the math, it sort of forces equal distribution of optimization. Exactly. Yeah. Yeah, that's exactly right. Interesting. And so they tested it on this digital twin that they had made of LIGO, and it worked really well. And they're like, okay, this is pretty cool. They injected noise, and it was still able to do it.
And then for one hour, they were given... time time oh yeah you know ligo's real busy you gotta check the schedule i think we can get you one hour yeah they got so they got one hour on ligo um and they installed this this device on ligo and it was a great success the ai controller stably managed these mirrors for an hour and the amount of noise on the actual thing the physical physical livingston observatory they they they got the controls for the actuators yes they attached it to this agent they're like that was the driver of the bus and now now you you take the wheel let's see what you do and for over an hour in real world conditions yes and the noise was like shot like it it like saying it's
Like an insane amount. Right? And it's like, oh. Yeah. This is the future. This is the future. Yeah. There's so many... I could talk about this for a minute. They were getting down to quantum level, dude. They got down to that quantum limit. They were saying the noise level was well below the design goal, being 10 times quieter than the quantum back-to-back action limit. Holy moly. And so now, now that they did it on Livingston, now they're going to make it even better, put it on both LIGOs. They're going to give it to Virgo, put it on theirs.
Give it to Chagra, put it on theirs. And then now we're going to have four... gravitational observatories that are limited by quantum noise. It's so incredible. I think the efficacy and impact. I mean, there's so many levels to this. One is our ability to create a digital twin to such levels of fidelity that it translates from a simulated environment to the real world environment. So well. So well.
Like that in and of itself. Yeah, that itself is great. Thinking about the reward model and using reinforcement learning and finding a way to get that reinforcement learning really tight such that the optimization that it ended up with, again, gets to that degree. And it's basically we're just going to do a software patch. I know that's not literally the translation. Kind of. Existing hardware. Hardware does not change. Yeah. At all. At all. And every one of these four detectors globally is just going to get this 10x noise reduction optimization benefit.
Just a relatively snap of a finger given the time scale it took to the Lego. To be fair, the noise reduction is more than 10x. It's okay. Wow. Right? It's like from before to now, the noise reduction is like 10 to the 3. Oh, my gosh. It's just that now we're below the quantum noise by 10x. Got it. So now the limiting is the quantum noise. And that we just, I don't know what to do with that, right? This is so insane. It's insane. And this has actionable consequences on what LIGO can see. For one, it's more sensitive, so it's got a larger volume of the universe.
But the other, it's now sensitive to a new frequency band, this 10 to 30 hertz, right? Which was not accessible to us before. And this is exactly where you have intermediate black holes. intermediate mass black holes. And this is going to help us solve a huge... Below supermassive? So it's like smaller in size. Exactly. Smaller in size from supermassive, but larger in size from stellar mass. Ah, got it. There's been this contention in the black hole physics community about where the hell are the rest of the black holes.
We see a bunch of stellar mass black holes, right, with LIGO and things like that. We obviously see supermassive black holes at the center of galaxies. But... The black hole mass gap is what it's called. Yeah, yeah, yeah. This gap between supermassive and stellar mass, this intermediate mass black holes, where are they? Yeah, yeah. Well, now LIGO is going to be sensitive to those intermediate mass black hole mergers because those would happen slower. than the stellar mass. The stellar mass, they can get closer together.
And if you get closer together, you speed up, right? Neutron stars as well. The closer you get, the more you speed up. So with the neutron stars, you can actually start observing them earlier in their decay cycle, right? Where before you had to sort of wait till they got really close to create these big fast... gravitational waves now you can spy on them from the very beginning and see them speed up and get closer and closer together the idea and just to make sure i'm kind of understanding the idea is part of the spinning and the closeness is what it is what dictates the size of the gravitational wave that's being sent and this is where we talked about they have to be really really big when we started for us to see them because we couldn't detect at this range yeah by having this range
We can detect fainter gravitational waves, which one of the sources being these intermediate black hole. Yeah, and the other one being the neutron star mergers early on. Earlier, earlier in its... Yeah, so now, like, you can imagine, you can have an early warning system that something's going to happen, but minutes in advance. Right. Not, like, fraction of a second. Right. Because the gravitational wave itself lasts, the ones that we've seen, last, like, fractions of a second. Right. Because that's when they get close enough to actually get big enough. the waves get big enough for us to see.
Yes. But like now we can have a minutes long advance. Yes. Where now, even the James Webb with minutes, James Webb could drop what it's doing and point. And we could see it at the moment. And so this is a combination of being able to hear the universe and see the universe. And this sort of new level of accessibility allows us to combine both of these... tool sets in combination to be able to now not only hear it, have a window of time that's long enough that any number of different Earth-based or space-based detection platforms that are looking at infrared or optical could look in the direction of the source because we have four locations.
Yeah. Events. Oh, my goodness. It's like, yeah, we really are living in a golden age of... time domain astronomy right we had a golden age of astronomy in general but it's like we were looking at still pictures right right right and now and now the universe is dynamic and we have access to that extra dimension which is time passing and us then seeing the transition of these objects over time with really really high fidelity
Yeah. We have early warning systems now. We've got the Vera Rubin that's going to find random stuff. Yes. Yeah. And this is another. Now that we have this 10 to 30 hertz band that we're sensitive to, we got that unknown unknowns coming back again, right? This is, I think what's so interesting, this time domain astronomy point is it's like we're turning on the light for the first time. We've sort of been like having candlelight in this. We've been doing flashlights. Flash photographers. Flash photographers, right, right. In this like small window. And so not only do we have a larger window, we also now have this ability to see things in time.
Which again means our ability to understand what's actually happening in the universe around us is like right now, today. Like not 20 years ago, not 50 years into the future. We talk about Varouba. It's going to be putting out data every three days globally in real time. That's going to see things we've literally never seen before. Yeah. And lots of it. Like, lots of it. This is... I love these retrospectives because it... You know, science is a story. It's a human story. It transcends generations.
Yeah. We're building on the work, the blood, sweat, and tears of those before us. And we are... expanding the aperture by which we, again, can just understand so many pieces. And people always talk about AI in this, like, oh, in this purely generative LLM chat, GBT universe of, like, oh, these LLMs aren't that great. It's just this stochastic parrot that repeats back what it's already heard before. No, no, no, no. Like, this, like, one of the reasons, like, this.
It's doing things that we really wanted it to do. And it's doing it way better than we could do it. Right. Or that we necessarily even thought it could do at this stage in its development cycle. And I think it's why it's so important to talk about these AI stories in the context of the experimental design, of the historical storyline of these, like, instruments. Yeah. Right? Because at the end of the day, science is so much about instruments. Yeah. And when we talk about experimental stuff, and this is why we spend hundreds of millions of dollars on this stuff.
Yeah. I just, this was so, so fast. We started off first with our CRISPR story, how we understand what CRISPR is, what is Cas9. The 2020 Nobel Prize. The 2020 Nobel Prize. The legal battle between two of our elite research institutions around who owns the human domain aspect of this, and how AI is making that gene editing process better. And the same kind of ideas, different implementations, different details, but...
using AI as a system to augment and make our existing tools more advanced with LIGO for gravitational wave detection. And what we're going to be able to see with that is so crazy. Yeah. And how good it worked. And also the timescales of iteration with this AI stuff is much smaller than the hardware timescales. And so we have these long arcs of time to get these new hardware instruments online. And then we have these much shorter timescales of where we can now use these new AI models to continue to –
polish the diamond. Another great American exceptionalism story with LIGO, too. We need to fund our elite research universities and institutions. Yeah, and we need to do basic science research because, you know, the technologies that are coming out of LIGO, like these control systems, this is going to be useful in all sorts of technology. Right, right, right. But, like, it was pushed to its very limit on LIGO, right? Because things like LIGO are what...
make us go all the way to the very end. I mean, think about it. We're sensing to the quantum limit. Yes. There's going to be insane amounts of quantum sensing technology that comes out of this. Yes. Control system technology that comes out of this. The mirrors, the vacuum technology that's going to come out of this. Yes. You know, it's... All of those had to be new fundamentally... New fundamental engineering solutions that have applications well beyond gravitational wave detection.
Yeah. Your new iPhone is going to have... You know, these AI-driven optimization models for capturing your 32x zoom. Yeah, exactly. Selfie. It's going to be crazy. Unbelievable. This is a great lead up to our Nobel Prize week. Again, tune in. We're going to be doing – there's no channel where you're going to get better coverage of what's happening in this year's Nobel Prizes, understanding the scientists, understanding the history behind their discoveries, how that human story turns.
What does it mean moving forward? I'm so excited to have you on that. It's going to be great. It's really good. Again, we're going to be doing all three days. Monday, Tuesday, and Wednesday. Monday is when the Medicine Prize comes out, Tuesday the Physics, and Wednesday the Chemistry. And Princeton is almost certainly going to be on the podium for one of those, at least we hope. At least we hope, yeah. At least we hope. My name, again, Lester Nare, your host, joined as always by my co-host and our resident PhD, and one of the greatest science communicators of a generation, Krishna Chowdhury.
Thank you, sir. This is from First Principles. We'll see you guys for Nobel Prize week.
Do pigeons navigate with their livers? A surprising study proposes that iron-rich immune cells help homing pigeons sense Earth's magnetic field. But other researchers argue that the mechanism does not hold up. Lester...
How can a weather satellite see Ramadan, Christmas and a blackout? Dr. Miguel Román joins Krishna Choudhary to explain NASA's Black Marble: nightly maps of the lights on Earth, from a single band on the VIIRS...
Can sound destroy a tumor without an incision? Dr. Zhen Xu joins Krishna Choudhary to explain histotripsy: focused ultrasound that creates tiny bubbles to mechanically break apart tissue, without relying on heat. From a...
What happens when AI moves beyond the chatbot and into the laboratory? From Claude-designed proteins that bind scorpion toxins to robots that assemble optics experiments, we examine what AI can do for science and what...
Can light restore aspects of sight? Can recalling a cold experience change an animal's metabolism? And what can ghost particles and mirror-image molecules tell us about the universe? Lester Nare and Krishna Choudhary...