AI-Generated Genomes, Retinal Implants, and Palomar's Mystery Lights Explained

Intro

0:00Hello internet. This is your captain speaking Lester Narre joined as always by my co-host and our resident PhD Krishna Chowdery. We have three great stories lined up for you this week. Starting with a story out of the University of Toronto which is around generative AI for cancer detection which is published in cell genomics. This will be a good one. Our second story is a super story. A couple of institutions here involved. Science Corporation, Stanford University, UCSF, University of Pittsburgh, and University of Bond. They've created a wireless retinal implant that is helping blind people see. This is out of the New England Journal of Medicine. And our final third

0:42story actually is two papers around the UFO subject, which is now known as unidentified anomalous phenomenon. Interestingly, published in both Nature Scientific Reports and publications on the Astronomical Society of the Pacific. This is going to be really interesting. We're covering a lot of ground as always. We're going to have a great time. This is from first principles,

1:24>> my friend. >> How's it going? >> Quite well. Quite well. Back from our travels. >> Well, yeah. You know what happened. >> I I mean I I I do I am I You know what happened? >> The Doyers won the World Series. >> Uh, Los Angeles brought home another World Championship. >> Back to back, baby. >> Back to back. As predicted. >> As predicted. We We predicted the chemistry Nobel Prize. >> Yes. >> And now we predicted the World Series. >> Yes. >> I mean, what can we not do? >> Well, what what can we not do? Um, I I LA is obviously >> just the greatest place to be right now. >> Yeah. Whatever. I I'm still like re-watching

2:06um the they're putting out like World Series movies >> and I'm watching that. >> A great a great game seven. >> Yeah, it was an amazing game seven. >> Amazing series. Amazing game seven. You know, shout out to the the Jay's put out a good fight. >> Yeah. >> Um and we're going to we're going to start off with a little bit of, >> you know, difference >> uh to to Toronto, an incredible city.

Why start at University of Toronto (context + Hinton shout-out)

2:31>> An incredible city. an incredible university. The University of Toronto is, >> you know, one of the great universities in the world of science and technology. >> Yes. >> They're kind of like the Neils Boore Institute when it comes to AI. >> Yes. >> Because of Jeffrey Hinton. >> Shout out to Jeffrey Hinton. >> Yeah. I mean, they've done a bunch of other things. They discovered stem cells. They discovered insulin. So, it's an incredible storied university. And even though we defeated them, >> even though we won, >> even though we won, let's let's just say we won. Let's let's just say we won. And all these people being like, you know, I I'm We got a comment on TikTok saying that, oh, um, have fun have fun with the

3:14ring that you bought, you know, and I finally feel like the Yankees did in the 2000s. >> And now I get it because I don't care either.

“AI-generated genomes” story setup

3:24I don't I don't care. Yeah. Whatever. But you know the Toronto is a great city and the University of Toronto is a great university phenomenal research institution. >> Phenomenal institution and we're going to start off with with a story out of the University of Toronto which is this new study about how AI generated genomes promised to advance sort of precision cancer detection.

Generative AI beyond consumer apps; the research angle

3:52>> Yes. Uh and so that's sort of the story >> and we've we cannot avoid >> AI. We cannot avoid AI. >> It's it's just pre it's becoming much more prevalent. And it's interesting because a lot of times people think of generative AI as oh let me create an image about you know >> GPT and GPT generative. uh and so a lot of people think of the consumer use cases but in sort of this early re early stage research arenas there are some interesting things happening this being one of them so so how exactly are >> the researchers at the University of Toronto using generative AI for this detection process

The precision-oncology privacy bottleneck

4:32>> right well the problem is precision oncology okay it needs a lot of large genomic data sets >> yep >> but we also want to preserve the privacy of the individuals that are donating that data set, right? And so this creates a kind of bottleneck between the research and tool development for if we want to look at a genome, let's say, of a patient and try to figure out what is the specific cancer that this person has and how can we better mitigate that? What are the kinds of drugs that we need to actually do this? In order to actually train a model to do that job,

5:13we need a lot of data. >> But the data has to be, you know, without all of these privacy bottlenecks. >> The the sort of uh personal identifiable information.

What OncoGAN is; published in Cell Genomics

5:27>> Exactly. Yeah. And that's what's tough. Okay. And this encoan which is a generative AI pipeline. It was published in um

Synthetic genomes that mimic complex mutational landscapes

5:36I believe cell genomics cell genomics enco it creates a highly realistic synthetic cancer genome >> just completely within the computer. Okay. And it bypasses these privacy issues. >> That's actually really interesting. Okay. >> Right. And the key feature is it mimics all of these compu like computational complex mutational landscapes that are characteristic of cancers but it doesn't tie that data with the patient >> with an individual.

Goal: democratize high-fidelity data safely; accelerate tools

6:07>> Yeah. And the goal is basically to democratize access to this highfidelity genomic data, accelerate cancer diagnosis, treatment, prevention tools, all of the stuff that we want without actually compromising patient confidentiality. The idea being if you can provide this sort of substrate of the data set variety of research teams and institutions can do a work towards solving the problem uh without running you know >> into all of the regulations regulation issues >> which are there for a good reason >> which are there for a good reason. Um so let's go through what precision oncology means. Okay. >> Why it's actually hard to actually

6:48simulate these genomes and then what this current paper is doing. Okay. What this current paper is announcing. So the vision of precision oncology is we want to shift from things like chemotherapy and radiation which is kind of just a general sort of shotgun approach to this tailored treatment based on a patient's tumor, the molecular and genomic alterations in that tumor. That tumor. >> Okay. And we want to make it precision. That's that's where precision comes from, right? And so what we want to do is identify mutations and then administer targeted drugs. >> There's a data bottleneck where we have this paradox between sharing and privacy, right? Because the

7:29success of precision oncology and an AI to precisely identify what type of oncology is going on depends on a vast diverse data set, right? But every single genomic identifier is something that is unique to every individual, right? You've got three billion base pairs. And if I know you're 3 billion base pairs, I can figure out who you are, >> right? So reidentification is really possible. Like even if you deidentify data, like you took you took some patients genomic data and you took out some parts, it would actually be pretty easy to look up who you are based on genome genealogical websites, public

8:10records, the whole like 23 andme thing, um ancestry.com, like that that whole thing, right? So there's privacy risks and that can reveal dispositions to disease, potential discrimination, and you've got stuff like HIPPA that actually controls and creates barriers for that kind of data sharing. It's a very good thing that we're doing that, right? Because then without it, you can you can have potential discrimination on life insurance, long-term care, >> right? I was going to say the the big the big issue is you don't want insurance providers to be able to basically say, "Oh, because we know that you have this is the whole Obamacare pre-existing conditions issue." Yeah. >> If we know who you are and what your

Why “precision” is hard; data-sharing vs. privacy trade-offs

8:51pre-existing conditions are, we're going to charge you more because you're a higher risk and liability for us as the insurance provider. >> Exactly. Yeah. You don't want that. That's just blatant nonsense. Right. Yeah. >> The the other thing actually that it's tackling is um if you have a AI generated tool, whenever we make AI, you want to have a ground truth data set that tells you what is right or wrong, right? So that you can compare one AI model from another. Well, with with cancer data sets, it's really hard to do that kind of benchmarking because the the real patient samples have sequencing errors. you don't actually know what the tumor complexity is.

9:32There's no actual ground truth. So if we were to generate data sets that have a ground truth, then we can easily benchmark all of these other AI models. So when someone says, "Hey, I've got an AI model that I think does better. We've actually got a data set where we can test that." >> Yeah. >> Right. So So that's that's a separate issue from the privacy thing, but this thing is actually tackling both of them at the same time. And obviously there's a huge economic and human imperative, right? Like there's a global c cancer burden, 20 million new cases every year. Um 9.7 million deaths every year. There's a huge economic toll. Um drug development is long and expensive. Takes

10:13hundreds of millions of dollars. And if we have this kind of data set, we can train models that do it way better, right? We can we can limit all of that and democratize this approach. This is actually totally an organic anecdote, but literally I was in a work meeting earlier today and unfortunately one of my co-workers, you know, has a partner that um was just diagnosed with stage 4 >> and like you know I mean it is it it happens every day. Yeah. >> It it is present >> personally for millions of people. >> One two degrees of separation for so many people. you would know someone >> and and literally today I mean I I I I

10:55so this is clearly obviously >> uh a highly >> important and prevalent issue. >> Yes. Exactly. And so let's talk about like cancer itself. >> Okay. >> And why it's even like hard to do this. >> Okay. >> It's hard to like make these genomic data sets that are synthetic the way that we want. Okay. Cancer is a disease of the genome. It's caused by DNA errors. At the end of the day, that's what's happening. Okay, there's 3 billion DNA letters ATGC in our genome. And we're going to use this analogy of the genome being a library of cookbooks. Every single chromosome is a single

11:35cookbook that let's say one cookbook is for Indian food, one cookbook is for desserts and so on and so forth, right? So the genome is a library of these cookbooks. Each cookbook is a chromosome and then each recipe is a gene inside that cookbook. >> Okay. >> Okay. If we have errors in critical recipes, that's going to lead to cancer. >> Mhm. >> And there's really two types of big mutations that happen, these typos that happen in our recipe. Okay. There's germline mutations which is basically when a mutation is right when a letter gets substituted for another letter or some mistake happens in the genomic

12:18reading of the thing or I delete a bunch of letters so on and so forth right there there's two types of mutations that happen there's a germline mutation that's something that's inherited from your parents and that's present in every single one of your cells right because that's the you became you came from a zygote which is a single cell and so whatever mutation was in that zygote >> permeates >> that's all of you right then there's somatic mutations these are acquired during your lifetime it happens within a single cell and then the progeny of that single cell inherits that somatic mutation right cancer is actually primarily driven by accumulated somatic mutations line

12:59>> it's it's there are obviously cases where germline mutations create the cancer off the bat but usually It happens with age and there's these accumulated somatic mutations that cause your cells to do weird things, become tumorous, become carcinogenic and then create the cancer, right? And one key aspect is ankoan which is this generative AI it's trained on and it generates only somatic mutations. It doesn't get trained on the germline info. So it's it's already getting rid of that sort of privacy >> issue which is important to understand that there's two entry points

13:39>> one of which has less of a privacy concern >> by default. Um and that's where this uncle gan is focused on. >> Exactly. Exactly. Yeah. Now when it comes to the architects of cancers there's two types of cancers. There's a driver and a passenger. So the driver mutation these are critical alterations that give a growth advantage to some kind of cell. Okay. It can either be something like you turn on an ankco gene which is something that turns on the cancer part of the genome or it turns off a tumor repressor right a tumor suppressor. So then if if the cell wants to go become tumorous, this gene isn't being active telling it no, you don't

14:20want to do that. Not >> right. And then so these are the driver mutations. Then there's passenger mutations which are kind of like hitchhikers. They accumulate by chance. There's no growth advantage. It's kind of like just there, but it doesn't really affect the the cellular machinery in the way that it like turns it on and becomes cancerous. Right? tumors have very few driver mutations amidst thousands of passenger mutations. Okay. So, there's this distribution where there's a few that are targeted for these ankco genes or these tumor suppressor genes and then there's a bunch of noise. >> Yes. >> Right. And any kind of generative model

15:02like ankan needs to model that discrepancy where there's a few of the stuff that matters and there's a lot of noise, right? And Uncle Gan does exactly that, right? It replicates this tumor architecture by making a bunch of random passengers and not that many driver mutations. >> Okay. And the final thing I want to touch on is these mutations have fingerprints when it comes to cancer. Okay. There's characteristic patterns of mutations that give certain processes a leg up that becomes cancerous. For example, like UV light, right? UV light causes um a C to

15:43T mutation. Basically, you get this dmerization where like one leg of your DNA, you know, your DNA is a twisted ladder, but one part of your ladder is going to like staple itself. >> Like one leg of your ladder, it's got two legs. One leg of your ladder is going to staple itself onto itself and become a dimer. And then that's going to cause melanoma. >> Yep. >> For example, right? Tobacco smoke is linked to another singlebase substitution that'll cause lung cancer. >> So ankan which is this thing that people have made. >> It accurately reproduces that tissue specific signature right where it's like

16:23the lung cancer is going to do this and the skin cancer is going to do that so on and so forth. It's actually learned how to do all of that stuff. >> That's fascinating. >> Right. And then the the other thing, we're we're not done with all of the stuff that can cause cancer. There's also something called large scale vandalism, right? There's copy number alterations, CNAs. This means you've taken a whole chunk of your chromosome and you've duplicated it. Oh, okay. Yeah. Okay. So, now there's a duplication or even a deletion sometimes of large chromosomeal segments. So, you've got large parts like large chapters of your cookbook that are now been copied or completely deleted.

17:03You've also got structural variance where you take a big chunk of your DNA and you flip it and now it's reversed. Now, it's just all sorts of chaos, right? >> Like there's so many different things that can happen >> that with cancer genomes, right? >> That are the driver and the cause of it, >> right? And the real and the real sort of challenge is how do we make a single architecture, a single framework that captures all of these different myriad effects >> right >> into a single pipeline >> that then can be used to train other AI to train other tools, right? And that's what this paper is doing. So

17:44>> they use something called um generative AI. We've seen this a lot, right? like uh a cat on a thing. This one, this particular architecture is called a GAN. It's a generative adversarial network. Okay, there's two competing neural networks. This is a very interesting way to actually train AI and to architecture it up. I think it's very cool. It's been used actually for image generation as well. >> Yep. >> But this time what they're doing is they're using it to generate genomic data. It's interesting because I've I've been watching this space for a while and I've I've seen GANs being used in the the image generation context

18:25now mapping it to this problem set is an interesting use of the architecture of >> general uh generative adversarial networks. That's interesting >> and and the basic idea behind general adversar generative adversarial networks is you've got two networks, okay? You've got one that's the forger and one that's the discriminator. Okay, the forger or the generator. This is the guy who's going to create forgeries from random noise and try to attempt realism. And then there's the discriminator, which is like the judge who's trying to tell whether the thing that was generated is a real thing or a fake thing. So both of these guys have

19:07the training set which is all the stuff that is actually real. And the the generator is creating fake versions of that. >> Yep. >> And over many many iterations both of these guys are going to get really good at their jobs. The generator is going to get really good at creating fake data and the discriminator is going to get really good at telling whether it's fake or not. And it's this competing adversarial. That's where we get adversarial from. This competing effect that actually causes the generator to get really really good at its job. So there's this substrate there's a substrate of truth that these two processes have access to. One's job is

19:49to create synthetic data based on that substrate. Yeah. >> The other is to say whether >> you're fake or not. And and because there's this sort of iterative back and forth between these two, you get to a place >> where you know the generator has now generated something that the discriminator views as being >> real because it's gone through that back and forth process. >> Exactly. And now it's like really good at making real looking >> real looking data data which goes back to our privacy issue and some of the other things we just >> Exactly. Yeah. Yeah. And so this cycle repeat repeats until the generator produces stuff that's indistinguishable

20:30from the real data. And this can be used for tabular data which is a lot of this genomic data, right? It's not like images and things like that. It's like tables of like how much of this is there, how much of this is there. Enko GAN uses something called Cabab GAN plus which is just a specific architecture that's used for tabular data. >> Got it. >> And it handles mixed data types, imbalanced distributions. So like the distribution is not like completely uniform. You have some stuff happening all the time. Some mutations happen all the time. Other mutations are rare. So on and so forth. So this takes care of that. >> It handles the complexity of of this kind of data set. >> Exactly. Exactly. Now the other thing that they're using is something called

21:11variational autoenccoders. These are simpler than GANs. And what they do is they basically take whatever input you have, they try to squish it down into a tiny little latent space, right? And that latent space has fewer dimensions than the original data. >> And then from that latent space, we have an decoder that >> tries to recreate >> what was originally compressed. >> Okay. Yeah. So this is used a lot. This is used actually a lot in like image generation. When we have let's say a 256

21:54x 256 image that has 256* 256 dimensions, right? Because each pixel has a value. And you can imagine if if there's only three pixels, you can imagine putting that on an xyz plane, right? where every single pixel's value has to do like the first pixel is how far it is in the x direction. The second pixel is how far it is in the y direction and so on and so forth. Well, now you have 256 times 256 dimensions. Every single image is highly dimensional. But if I were to take a bunch of photos of just a bunch of faces, right? The essence of the face is

22:36actually not that highly dimensional. There's always a nose in the middle. There's always a mouth. There's always eyes, right? Maybe maybe one of the directions could be how dark is the face, right? For us, it would be like maxed out. >> Very dark. >> Yeah, very dark. Like there would be one dimension that would tell the variational autoenccoder this is a very dark face, right? Then there would be one dimension that would tell it how far up in the nose is the nose and how how wide is the mouth and things like that. But the fact that there is a mouth is not something that needs to be encoded in the latent space because it's always there, right? The stuff that's varying

23:18is the stuff that's going to be encoded in that compressed middle part, right? And then you have a decoder that takes that middle part that says okay how much is the how much is where is this and where is that and puts it puts it into the actual image that it reconstructs. And this is something that we're seeing in emnest for example the emnest data set is a bunch of handwritten digits. >> Yeah. The key part about variational autoenccoders is that it's a continuous latent space. Meaning that all of my real data points are somewhere in my latent space, but I can pick any point in between and that'll result in something real. >> Okay? It's continuous. It's not discreet like

23:58>> this and this and there's nothing in between, right? There is actually meaning in between. And so you can see in that latent space of the emnest, it's going from the thing that looks like a nine, >> which is over here. >> Yeah. >> To a three to a six to a zero, right? And it's continuously deforming. >> Yeah. >> And all they're doing is reconstructing what happens if I move on that latent space like as a walk and I see what is the decoder reconstruct. >> Reconstruct. >> Does that sort of make sense? >> No. No. 100%. It's like it's like it's taking all of the information of the stuff that it's trained on and saying what is the essence >> of the thing.

24:39>> I'm going to put that into a landscape that I can traverse. >> You can tra Yeah, exactly. >> And then and then as I move through that landscape, it's going to create meaning with my decoder >> because there's there's different points on that landscape, for example, in the numbers that we find meaning in, which is like the nine, the three, the zero. But the there is this there is this this transformation continuous in between there where you can say oh this is like >> a 93ish >> thingy. >> Yeah. On my way from 9 to three. >> Exactly. Yeah. Exactly. >> Exactly. Yeah. So, so, so they're using both. And, and the key to using both is

25:20that you've got a GAN that's used for complex interdependent features like mutation counts, driver co-occurrence that that's there due to highfidelity matching. And then you have this variational autoenccoder that's there for continuous variables like genomic position. The position is a continuous thing, right? But the mutation count, that's a discrete thing. >> That's a discreet thing, >> right? So you're you're combining both into this hybrid approach to create this like realism in the data set. >> That's f that's actually really really interesting. >> It's they're not they're not limiting themselves to like one type of architecture right? >> They're really harnessing the true

26:02power. >> It's like there's this blended architecture which is using the the value of both of these like modalities >> and taking the best of both. >> Yeah. and and sort of using them in the context that they that they're good at >> in in in parallel or together. >> Yeah. Exactly. Exactly. Because the genome is is both continuous and discrete. >> Discrete, right? And so now we're >> we're getting like two soldiers here that that are doing the work, right? In order to train enco data from can panc cancer analysis of whole genomes PC AWG it's a data set that has 2,658

26:44whole cancer genomes like entire genomes. Okay. All 3 billion base pairs. >> Massive. >> Massive. Yeah. And this this data set is treated like sensitive information because it should be right because this has actual markers that can identify patients and things like that. But using that they've now created something that will generate data that doesn't have those limitations. >> They've created a derivative product from source data that is personal identifiable information that now has that degree of separation uh to to not run a foul of the practice. >> Exactly. Yeah. Yeah. Exactly. And from

27:24what I counted in the paper, there's five GAN models and one TVA model. So there's five GAN generative adversarial network models and then one um variational autoenccoder model. All of these guys are working together to create this giant genome right multiple giant genomes and obviously the question is to ask is like you know okay how real are you making this right the photo that you're making could it fool me? This becomes the with photos it's easy because we sort of have the eye test, right? This is the whole uncanny valley concept which is like, okay, I can look at something and be like, "Oh, the fingers >> Yeah. >> aren't right." >> I mean, now, dude, I'll be honest, dude.

28:06There's like videos on Instagram that I get fooled. I I do a double take and I'm like, "Oh, this is this is definitely >> there's a there's a Neil Degrass Neil Degrass Tyson just did a video of himself talking about how Earth is flat and he he held up an iPad to the camera. You didn't know like the video starts as just Neil Degrass Tyson talking about how the Earth is flat and then like he just moves the iPad and he's like that wasn't me." >> And it's >> it's like and it's the same it's like him in the same space. >> Oh my god. in the same outfit, in the same voice, >> and it's effectively indistinguishable. >> Yeah, dude. It's It's getting It's getting really good, right? So, at least we're harnessing it for like something

28:46>> something useful. >> Yeah. Um, so one of the things they looked at to see, okay, is the stuff that's being output by >> this neural network, is it something that can be real? Right? They looked at mutational density and type, they found a high similarity between the synthetic and the real data. Mhm. >> They looked at mutational signatures. They found that it's like pretty nicely correlated. Um, the real thing that got me was the genomic distribution. So, you can look at the entire genome of the human from, you know, chromosome 1 all the way to chromosome 23 with the sex chromosomes and you can follow where mutations are happening. Okay? And in

29:27cancer, mutations do not happen >> uniformly. >> Okay? There's going to be parts of the genome where it happens all the time >> and there's parts of the genomes where there's very few mutations. For example, if you're closer to the center of the chromosome, you're not going to get that many mutations. But if you're >> closer to the edge, you're going to get a lot more mutations just because of the physics of like, you know, you got a little bit more freedom towards the edge than towards the center, right? And so they they actually showed that mutations in their data set would follow nonuniform distributions. So on the on the top they have the distribution of mutations in a real sample.

30:08>> Yes. >> On the bottom you've got your generated. It looks pretty much the same. >> The eye test, you know, >> the eye test. Yeah. Exactly. And what they've done is like the axis on the x-axis is all the genes. Yeah, >> like all 23 chromosomes worth of genomes and they've binned it in like I think 1 kilobase >> bins and they've counted how many mutations are happening. Right. >> That's incredible. >> This is actually it's very it's very similar to what we would actually see. >> What we what we've been basically able to do is now create a a system that on the fly it's not fixed. Yeah. It's it's a it's a it's a generative model.

30:50>> Yeah. So on the fly any team can now if you take this and say we want this kind of dynamics >> and they can generate now these cancer data sets that mimic real cancer genomes to a degree of accuracy or or comparable >> Yeah. or uh high fidelity that is usable Yeah. in in practical work in this space. >> Exactly. And that's that key that you just touched on, usable, right? What does it mean to be usable? Well, they actually tried to test it with something called deep tumor. Deep tumor is a

31:32>> again another AI that's been developed to identify cancers. Okay? They could fool deep tumor really well. But here's the key. Here's the key. >> Deep tumor is really bad at rare cancer subtypes. >> Okay? >> Okay. like lymph mlll there's only 35 samples in the training set okay and if you have low mutations you're going to struggle with like creating an AI that can identify that well they substituted generated samples into the training set and deep tumor did better once deep tumor trained on those the F1 scores improved >> that's fascinating so the the synthetic

32:13data from this GAN this ankco GAN when fed into an existing detection model >> that's like kind of established. People use it all the time. >> It's it's it's credible all these things. it has now improved its ability for the these edge cases >> these like edge niche cases because now I can I can generate genomes right for that specific thing that's not really prevalent >> in a lot of the real world data sets >> because the idea is you can take the deep tumor and then use it on like real world patient and like and and know that they do or do not have itactly and then if it comes out with a result that is >> exactly really powerful >> it's it's really it's really augmenting the data sets that we already have

32:55>> right >> with these synthetic data sets which by the way this is something that people do all the time in AI research right you do data augmentation where you take whatever data you have you like apply transforms to it when it comes to image data you'll like squish it you'll like turn it from black and white to red white and blue or vice versa you'll try to make the model more robust >> to the data >> quality >> but here they're actually able to create new data that can target these really rare cancer subtypes and establish a new paradigm for medical AI. Right. We can generate more data instead of we just need to collect more data.

33:36>> Right. Right. Right. Which which you know the there is a bottleneck on the collecting data piece. >> Exactly. And so you can much more quickly scale uh if you can uh create usable synthetic data sets that actually map to the real world uh use cases in a way that is demonstrabably true. >> Yes. >> Which is kind of the point part of the point of this. >> Yeah. Exactly. Exactly. And there's no one toone correspondence with the real patients right? >> You don't have to worry about HIPPA. You you do you solve the privacy issue while accelerating the ability to both create the synthetic data sets and then have those applied to existing tool

34:17sets or create new tool sets around all this process. And the team already they've got 800 synthetic genomes that are openly available. No ethics approval needed, right? So anyone can go and try to train >> a model based on that. if they've got a new architecture in their mind that really is specific to genomic data, they can test it on this. It's a new gold standard for benchmarking, right? Because now when I generate some kind of data set, I have a ground truth on what type of cancer it is, what are the mutations that should be flagged, so on and so forth. So, it's it's a catalyst for >> medical oncology. I think it's I think it's very cool. It's a huge new tool in

34:59the toolbox for oncology studies. >> Yeah. I mean, there's still limitations, right? Like the fact that we're doing this factorization where we first create the mutations and then we have another model that goes in and distributes it across the genome. It's not fully capturing this like complex interplay between features. Like for example, if you have a copy number alter alteration where you have these giant chunks of chromosome that get duplicated or deleted, that's going to influence local mutation rate. This is not something that captures that, right? It can't simulate complex events like um chromoth cis and like tumor subclones, but it's

35:41still a huge huge step forward, right? We we have to it's it's one step at a time. Yes, >> this has obviously been an issue that is not only affecting so many people, but it's been happening for so long and any any new step towards >> our ability to solve this more robustly is like hugely important. >> Yeah. Yeah. Yeah. Because like cancer is just like it's just such an insanely difficult problem, right? It's not a single cause, >> right? >> Um there's no single like panacea, >> right? Solution. >> Yeah. It's just yeah so so any any incremental and this this I think is actually a huge step because I think this opens up

36:23>> other people to do this kind of stuff >> to create large data sets to then have >> really AI going in full-fledged >> right right >> to try and help out >> right right in a in a way that again keeps in mind some of the aspects of data privacy that that are you know are very important when you talk about medical studies >> for all the reasons we talked talked about at the top of the story. >> Uh this is actually this is a really really again we said we're going to make sure we give Toronto >> Yeah. University of Toronto. I mean their AI is uh >> this is a big deal. >> Their their AI is always like on point. You know their AI research is is

37:03ridiculous. So >> this is a really big deal. >> This is a really big deal. I mean you Yeah, you guys lost the World Series, >> but >> we'll see you next year hopefully. You know, that would be great. >> You know, round three. >> Uh round three. you've given us an incredible new tool in the box ability to to do cancer detection um and expand the amount of people that can start really working at this issue in in earnest. Uh that's our story number one. >> Yep. >> Number two, the super story. The Avengers came together uh for for this wireless retinal chip >> that's helping solve for blindness. Yeah. >> This is in the New England Journal of Medicine. And again, we have Science

37:44Corporation Stanford UCSF University of Pittsburgh, and University of Bun that all collaborated on this story. You were really excited to talk to me about this story.

Story 2 starts — wireless retinal implant for AMD (PRIMA)

37:54>> I was really excited to talk to you about this because um my undergrad uh my undergrad PhD thesis, not PhD, my undergrad bachelor's thesis in uh Princeton had to do with the retina. And you know, we were doing fundamental science, but this is a really cool story because it's it's out in NBC News. Fox News even covered it. Um, with a tiny eye implant and a set of glasses, we can now cure age related macular degeneration, >> which is crazy. >> Which is crazy, right? >> Yeah. That that >> the solution is pretty ridiculous. Now this to to be mindful this age related

38:37macular degeneration which is basically age related blindness in the center of your of your visual field. This had no prior restorative therapies. Right. This is the first of its kind that is trying to tackle that. Okay. >> So basically if you had AMD you were just >> Yeah. You were just you were just blind in the middle of your in the middle of your field. You could see in the peripheral but in the middle you couldn't see anything. And now we h actually have something that is restorative right not just slowing it down >> restorative. >> It brings it >> brings it back. That's >> okay. >> The solution is something called a prima system. It's a photovoltaic retina

39:18implant microarray. It's a novel neuroprothetic. So it's wireless. It's subretinal. And we're going to get into what what all that means. Okay. This was published in the New England Journal of Medicine. The key trial is called Prima Vera. I don't know if the authors are fans of the music festival in Barcelona, but that's what they called it, the Prima Vera.

Trial headline numbers (81% report meaningful improvement; +5 lines)

39:42It's it's an incredible thing. Okay, 81% of participants 81% of participants said they got profound improvement, meaningful improvement. Okay. And remember those um those like eye exams that you had to do with like a bunch of big >> big like letters and then these guys the guys who had this over five additional lines on the eye chart >> really >> right >> that's >> that's that's a lot dude >> bad aigmatism and can barely see and so I >> and and and some patients regain the ability to read letters numbers and words

40:22>> that's crazy. Okay. >> So, let's talk about let's talk about how the eye works because that's going to tell you about how the mechanism of this thing works. Okay. The eye is basically a camera. >> Mhm. >> Okay. You've got a lens, you've got a

How the eye works (camera analogy)

40:38detector in the back, just like a camera. A camera has a lens and then a CCD in the back that the image falls on. And then the CCD captures how many photons are coming here versus there. What are the colors of those things? The eye is very very similar. Okay, the retina is an image sensor in the back. And this retina has photo receptors which are your rods and codes. You can think of that as your CCD. That's the thing that actually interacts with the light. And then if a little photon interacts with it, it's going to set off a little electrical signal. Then it goes through a bunch of different cells. They're called bipolar cells. Some of them are called amicron cells.

41:20And at the end of the day, you've got ganglen cells, which are the cells at the very end that take in all of that light information, convert it into electrical pulses, action potentials, and that goes through the oct optic nerve through your eye to your brain to the occipital lobe, and that's what's transmitting visual information to the brain's visual cortex. Okay? But it's really it's like mechanistically the eye is remarkably like a camera. You've got a lens and that lens you can squish. There's muscles on the on the end of your lens that you can squish and so on and so forth to let in more light, less light. That's your aperture, right?

42:02>> Um in in a normal DSLR, you've got the magnifying, right? And the magnification is like how much um like the the distance between your lens and the focal plane. >> Yes. Yes. >> But for us, we can I can focus on the microphone right in front of me and I can also focus on a mountain that's way way way far away, right? >> And the way that our eye can do that, we don't have the luxury of actually extending our lens. So what we do is we actually change the curvature of the lens >> in order to do >> in order to actually get different focal focal lengths, right? So it's remarkably

42:43like a camera. Okay? And in the back of your camera, in the back of >> the eye, you have the retina, which is this mini computer that takes in all of that information and then relays that back to the brain. >> It's sort of like the processing center. Once this comes once the raw data comes in, it kind of packages it up to send it upstairs. >> To send it upstairs. Exactly. And in the center of your retina, you have something called the macula and the phobia. Okay. Notice like I want you to just do this experiment where you're like looking at stuff. >> Mhm. >> Notice that you have a lot of detail in the center of your vision. >> Yeah. >> Compared to the periphery, >> right? >> Like if you really wanted to read

43:23something like that stuff that's on over over there, you would focus on it so that it's at the center of your vision. If you focus somewhere else, you wouldn't be able to read it. >> And the reason for that is you've got this structure called immacula and the phobia that has a high density of rods and cones, specifically cones actually that that that are the color receptors. So you you've got you've got like higher resolution there because the pixel density is higher. >> Not everything is not totally in focus all of the time. >> Yes. Exactly. And macular degeneration is actually a progressive disease that

44:05damages that macula. So >> the the macula is in the in the middle, right? And you can have two types of AMD, macular age related macular degeneration. There's the wet AMD where basically the abnormal blood vessel growth. There's a bunch of blood vessels in your eyes, right, that are keeping these cells alive, giving it oxygen, things like that. Those blood vessels can leak, rapid vision loss. And then there's dry atrophic AMD. Okay? And that's the gradual thinning of the macula. You get these extracellular deposits where the the cells are sort of just dumping their trash and then that kills everything. And all of that accumulation happens and it kills all of

44:46those cones and that are that are there in that center in that center. And so that center of your the center of your visual field >> Yes. >> is the thing that goes away >> which is like obviously

Dry AMD, geographic atrophy, and limits of drugs vs. restoration

45:00>> and that's the whole point. Yeah. Right. Right. >> Like that's that's where most of my information >> is coming from. >> Right. >> And that's the one that's more more common. Right. And then there's geographic atrophy. >> Um which it's it's at the end stage of AMD. you get these well demarcated patches of your retinal tissue where they die off. You get the photo receptors that are dying off. Um, and it results in like absolute scotoma. So, you get a blind spot right in the middle. >> And when you get the blind spot, you literally can't see in the middle of your visual field, >> which is so crazy, >> right? It's a complete loss of phototransuction. Mhm.

45:40>> Okay. That is what this particular study is trying to handle. Okay. >> And I it's it's pretty common actually. Irreversible blindness among older place older patients globally. You've got something like 200 million people that that have that have this. In the US itself, it's a million people. It's it's it's an incredibly like big problem that we need to solve, right? Obviously quality of life goes away. There's a loss of independence. It's got a devastating impact on >> I mean >> all aspects of your life >> especially like when you know not that this is it's true it's been true forever

46:23but particularly when everything's so device driven and and you know >> Oh yeah I didn't even think of that. >> You know what I mean? So it's like you have to be able to like you know read and interact with screens and so it's impossible to do so. >> Exactly. Um especially in a device- driven kind of era. >> Yeah. Yeah. And you know, previously it was just like, oh, you're just you're you're totally not having a good time. Okay. There are a few drugs like there's a there's the first drug was approved in 2023, is um there's another drug called uh cyov. I guess the mechanism is you basically

47:03inhibit this cascade of stuff that's going wrong in the back back of your retina. It slows the lesion growth by 20%. But at the end of the day, it's not restorative. >> It's just stopping. >> It's just slowing it down. >> The degeneration, >> right? And that's that's great. >> Yes. >> But >> yes, >> for someone who's already had it, >> it's not really doing much. >> It's not Yeah. And so now we've got this

Why PRIMA is different: a restorative neuroprosthetic

47:28primo system. It's a new paradigm in vision restoration. It is restoring things. Okay, it's got two primary parts. There's an internal component. There's a tiny tiny photovoltaic implant. Okay, it's a little um cluster of neuro implants, right? It's it's it's a tiny little electronic device at the end of the day that goes in the back of your head, okay? And it surgically gets placed in the back of your retina, in the back of your eye. And then there's an external component which is specialized GA glasses with a video camera. And what that video camera is going to do is project light through your eye

48:09>> to that implant >> in the back of your retina >> and stimulate that implant to create electrical signals >> to trick your eye into thinking your photo receptors are still alive. >> That's that's okay. So the idea is you have this two-part system. >> Yeah. >> Which is effectively mimicking the real effect of photons hitting the back of the retina. But because all the the stuff in the macula is is now dead, we're not like regenerating the cells itself, but we are creating a synthetic version of the exact same input >> such that when it goes into your brain,

48:50it's receiving an identical signal to what you would in a natural sense, but in an artificial >> very it's like simple but at the same time extremely hard to do obviously because it hasn't been done yet. >> Yeah. Yeah. And it's only being done because of today's technology. So now let's get into let's get first let's tackle the implant. Okay. What is the thing that is going in the back of your eye? Okay. >> It's a 2 nanometer by 2 nanome little chip. Okay. 30 microns thick. It's a silicon chip. It's a pixel array of 378 different independent hexagonal chip pixels. 2 nanome by 2 nanome. It's way way smaller than a penny. It's like it's

49:31like, you know, uh smaller than even Lincoln's. It's like the size of Lincoln's eye >> on the on the penny in the in the Lincoln memorial >> in the Yeah. That's how small this thing is. >> Small. >> Each of the pixels, there's 378 pixels. Each of those pixels has a little mini solar panel and an electrode. Now, you're thinking, why solar panels? Well, solar panels create electricity out of light. Mhm. >> Now I've created electricity from the light that is going in into my eye. I don't need an external battery or anything like that. >> I'm so mad.

50:11>> That is I thought that was so cool, dude. >> So genius. >> I thought that was so cool. >> That which matters because a lot of times you can even use phones as an example. A lot of the weight that goes into the phone is the battery. >> This is the first one that is wireless. The reason why it's wireless is because they use this photovoltaic cell. >> That makes sense, >> right? >> That's genius. >> And I think that's so genius. >> Yeah, like there's an electrode and that electrode is what's actually creating the electric field to then tag all the other neurons in the retina to make it seem like >> right there is a photo receptor there. There's a rotten cone there. >> But the the energy for that electricity

50:53is coming from the photon itself. And it makes sense because your eyes are anyway, >> right? >> Yeah. That's smart. >> That's so smart. >> That's very smart. I like that. >> That's so smart. I love that. And and as soon as I saw it, I was like, "Oh, that's >> that's dope. >> That's dope. That's good." >> Yeah. And so what they do is they take this little chip 2 nanome by 2 nanome and they put it under the retina. Okay. This is deliberate where the electrodes are in proximity to the bipolar cells. So it's it's where the it's where the photo receptors would be. And there you can see on the left you've got a giant patch of where the photo receptors have died. You're putting that chip right underneath

51:34>> Mhm. >> where those photo receptors should be. >> Mhm. >> Okay. And that chip is now going to stimulate those bipolar cells which will then in turn stimulate the ganglen cells which will then go all the way to the back of your brain. Right? It's natural. It's efficiently processed signal and it's more sophisticated than these older retinal implants that were stimulating the ganglen cells that were directly going into the brain. These guys are mimicking the photo receptors >> itself. Yeah. Yeah. You see? >> Yeah. Yeah. Yeah. >> It's so it's so nice, dude. >> That I >> I love that. And what I what I love about this as well is that the the the pixel has a has a input. So that's where

52:17the source of the electric field like the the source of the current but it's also got a sync >> because if you don't have a sync then then you put a lot of current out that current is going to spread out and it's going to it's it's going to trigger all the rest of the cells everywhere else in the retina. But if you have a sync then it becomes localized. The current goes out it triggers the stuff in its neighborhood and it goes back in. >> It becomes for like more it becomes discreet. >> Yes. Exactly. >> Exactly. So now your pixel size, right, for like what I'm trying to see, that becomes smaller. >> They've thought of it. They've thought of everything, mate. It's really nice, right? Okay. So that's the implant. >> Now, let's get into the outside. Right.

52:58There needs to be something that actually goes and puts light to that chip that's in the back of your retina. Yes. Right. And that's where these glasses come in. So there's a user interface. There's a smart component. You put on these glasses. You also have a user interface. like it's kind of like a remote >> and that's your user interface that um you can like zoom in, >> zoom out. >> Yeah. Yeah. Yeah. Yeah. You can adjust the contrast and things like that. And then and then your your your glasses, they've got a little video camera, kind of like the meta glasses, you know, the meta glasses have a video camera. And those that those glasses then send in near infrared

53:39signals to the back of your >> retina, which is where the chip is. Why near infrared? Well, all the other natural photo receptors in your eye are not sensitive to near infrared. >> Uh, >> right? If you were sending invisible stuff, then you'd be blinded. by like a giant like flashlight that's like right next to your eye. But because it's near infrared, the only thing that is getting stimulated is the chip. >> The chip itself. That's that's actually that's really good. No, that's that's that's really good. >> It's so nice. >> That that's I'm I'm >> right. >> Yeah. Yeah.

54:19>> You're seeing why I was like excited about this. This is so cool. I I I just I I'm like I don't have anything to say cuz it's just it's it's so good. >> It's so good. Like they thought of it, right? And the fact that it doesn't have any cables. Like usually these things used to come with cables, but the fact that it doesn't have any cables simplifies the surgery. So you can go in to the corner of your eye. I mean, it's still a crazy surgery. I'll be honest. like you're going in with a needle in the the corner of your eye, but you're depositing it in the back of your eye. But it's a relatively simple surgery because all you do is you stick it in, you leave the little bit of chip where you want it to be, and then you take the you you take the the needle out, right? >> And it's it's it's it's actually

54:59incredible. The this chip efficiently absorbs 30 microns of silicon photo dodes, right? It's it's it's 30 microns of silicon photo dodes. You get this pulse light that's coming in at 30 Hz. And 30 Hz is you must know 30 Hz is about where um film >> Yeah. Yeah. Yeah. Yeah. Exactly. >> uh frame rate is. And that's because the ganglen cells that are going to the back of your head, right, that are relaying this information, >> anything faster than 30 Hz and it's really not discernible. Anything slower and it would look like stop motion. So, at 30 Hz, you're getting this flicker-free image. >> Yes. >> Of pulse light that's going in.

55:40>> This is really clever, >> bro. It's so cool. >> This is really really because they sort of because you have to both solve for the like technical imp. So, there's a variety of stages of this process, right? There's the actual like where in the eye do you make the intervention in order to get the restorative benefit. They made that sort of innovation by putting it in that like where the cones and rods are. >> Yeah. >> Then you have to think about how do you do how do you get power? >> Yeah. >> Then that's the solar piece. >> That's the Yeah. You made a solar cell >> which so brilliant. Then you have to think about like how are you actually going to turn it on? Meaning like how like how are you going to direct the light to make it actually work the way you want it to

56:20>> but not affect >> but not affect everything else. And so the near infrared as the methodology to do so is so genius. >> Yeah, dude. Oh man, that's so good. So good. That's quite That's quite nice. >> It's so nice. That's quite nice. Yeah, it's so nice. >> And let's talk about the study itself that came out in the New England Journal of Medicine, right? It's a multic-enter, so 17 sites across five European countries. You mentioned the University of Bond. I I mentioned them specifically because they kind of did a lot of the the groundwork, but a a bunch of other European institutions were part of the clinical trial, right? >> Sure. Sure. >> But you're increasing generalizability because of these 17 different places are

57:00trying it >> because people culturally genetically you want to have it spread across a variet. >> Exactly. Yeah. So 38 participants all of them over the age of 60 and all of them confirmed um atrophy in both eyes. Right. >> And what they looked at was something called the log mar which is the logarithm of minimal angle of resolution. It's effectively saying like in the center of your field of vision like how how big >> can you see things? It's it's this thing. It's the it's the letters and the driving test. You know, when you go to the DMV and you're like, "Okay, can you see or are you blind?" >> Those letters tell you how small of a letter you can discern. >> What line can you see on this one?

57:42>> Bro, I mean, the screen is pretty small, but I think I can see that. But but but in any case these guys after 12 month follow-up after after getting the after getting the implant >> 81% so 26 out of the 32 participants met the end point. >> Highly significant >> and it's equivalent to five lines worth of stuff. 20 59 more letters than they could have originally in that eye test. you know, I mean, this is this is such a huge I mean, we talk about this all the time, which is like why, you know, um this these studies are so

58:23important like the unlock that you get from something like this um for a huge population of people who otherwise we're just going to lose the ability to see >> Yeah. >> permanently. >> Yeah. >> Is now no longer like necessarily true. >> Yeah. Dude, there was one dude who went from legal blindness to like I can drive now. >> That's crazy. >> I mean, that's one dude, but still like >> and and this is like stage this is V1. >> Yeah. Right. >> Yeah. This is V1. >> It's it's the first iteration of this um in which you now have I think that the the innovation is in like all the kind

59:03of steps we just talked about because now you can iterate on each of those component parts for any number of different >> Exactly. computations or use cases etc. Exactly. That's a really really interesting. >> It's really cool. I mean there were detractors like if you looked at user satisfaction 69% so 22 out of the 32 reported medium to high overall satisfaction which it's still like if you weren't able to see and now you're able to see like that's 30% being like I'm not satisfied. And I looked into more of that. I think part of that is because of these things called serious adverse events that happen at the end of any surgery. Like you're look, you're

59:43you're sticking a needle into someone's eye and you're putting in a chip. Like there's going to be serious adverse events, right? And most of the stuff had to do with the trauma and healing from the surgery, >> not from the actual >> not from the actual device. Right. So if we can make that that process better, >> perhaps this can be a lot >> better. Yeah. satisfaction can be higher right? >> Which makes sense. That that does make sense. >> But at the end of the day, I think like this prima is is a quantum leap in terms of retinal implants. Like before we used to the first FDA approved um retinal imp implant was something called the Argus 2. >> Okay. >> Which had 60 electrodes. >> That was the that was this image I just

1:00:24pulled up earlier. >> Yeah. And that was that was wired though, >> right? You can see it coming out of the >> Yeah. Yeah. And and you could see that like in the retina itself there's a little like wire that like Right. This thing is just complete. It's just a chip naked on its own. >> Wireless. >> Wireless, dude. >> Wireless. No battery. >> No battery. >> It's just a photovoltaic cell. >> It's This is It's so clever. >> It's so clever, dude. And what what they're doing next, you talked about how this is just the beginning, right? These these pixels are about 100 microns in size. They're trying to get get it down to 220 micron. >> 20 micron pixels from 100 microns. So

1:01:04now we'll be able to fit 10,000 pixels instead of the previous 378. So now I'll get I'll be able to get as much fidelity as I do now, >> which is And again, now it's become an engineering problem, which is like much easier to deal with. >> Yes. Yes. >> Like like I'll take an engineering problem all day. >> Yes. Yes. 20 microns, right? That's that's about the size of the rods and cones themselves, right? You're now starting to to get to the limit the biology was getting to. Obviously, there's going to be challenges, right? You can't smaller pixel size means that the electric field penetration to the upper layers of that retina is going to be not as good. So, maybe you need these

1:01:443D electrode structures, right? Where you have like pillars, right, where the the the pixel is like this and you've got a pillar that injects electricity into the upper layers of the retina. But again, engineering problem >> engineering problem which which is infinitely more solvable >> than going from zero to one. >> Mhm. >> Um >> I mean this thing is like decades in the making, you know. >> But it's working and just it's so clever. I loved it. >> Yeah. No, this that that's really >> it took a while to get access to the the paper. It was in New England Journal of >> Medicine. I don't know why you guys Yeah. I I don't want to pay like I think it was like a hundred. I'll be honest. I

1:02:25think it was like a $100. So, I I asked a friend of mine who who is a doctor >> who does research and I need you to Yeah. I need you to give me a PDF. We're trying to We're trying to give you guys some traffic here, okay? We're trying to give some promo. >> Yeah. >> So, you know, free free advertising >> on the first principal pod. We'll we'll do deep dives on the stories and like honestly like give like good good understanding because a lot of times again like for the average person it's hard to really like parse and decipher what's actually happening in these stories and like what the implications are and what what's like the real world >> you know meaning >> and the real world meaning for this is >> like a big deal. Yeah, this is a big

1:03:06deal, right? Like now imagine in the future. Right now I'm thinking you can you can bring AI into it. >> Of course, >> right? You could have like the the the Iron Man thing where like the thing automatically zooms so you don't have to press a button to zoom. It'll just know what you're looking at and zoom in like >> Yep. No, I mean this it kind of like ties to the first story we had around the how AI is now being used for cancer detection. I mean there's going to be clever ways in which to implement it in this context. I mean you're you know there's all this stuff I mean these implants and these things you know we've talked about Neurolink before on the pod now we're having these retinal implants um you know there is a bottleneck around

1:03:46the surgery process itself like we're still you know meat >> you know we're still meat sacks and so that's still a bottleneck. >> Yep. >> It's still a bottleneck. Yeah. >> And it's going to be for a while. for a while. But we're we're getting we're getting into real interesting places. I mean, as someone who already has bad vision, you know, my you know, my wife has bad vision, too. She actually has I didn't do LASIC or anything. I just wear contacts and glasses, >> but her vision was so bad that at the time like LASIC wasn't good enough for her eyes. So, she got the implanted contact lens. >> Oh. >> Like like in her in her eye. >> Wow. >> Which is like And I was like, they could

1:04:27do that. >> That's crazy. you can do >> and this was like 15 years, you know, whatever ago. This was a while ago. >> Um, and at the time it was state-ofthe-art and this and that and yada yada, >> but I mean this is like orders of magnitude >> more difficult than just just just embedding a contact lens, you know, in the eye. >> Fascinating. Fascinating. Fascinating. >> Yeah, I thought that was such a >> That's a great story. That is a great great story. Um, it it gives me hope that if I start getting blind later in life, there's still a chance. >> Yeah, there's still a chance. There's still a chance. So, you're saying there's a chance. >> You're saying there's a chance. >> We're going to end with our story number three, which is our story about

1:05:08or related to at least unidentified anomalous phenomenon, which is more commonly known as UFOs. And before we kind of get into the two papers themselves, I kind of wanted to set the table a little bit because this is a subject and you know an issue that is extremely popular not only in the US but globally um and has in recent years gotten a boost because of a variety of events that have been happening particularly within the apparatus of the US government intelligence and military industrial complex structures. >> Yeah. >> And you know, this is a subject I've talked to you about a lot.

1:05:48>> Yeah. You're very passionate about it. >> Uh it's something that I I have I have interest in. I I work at a nonprofit that's working in this space, the Disclosure Foundation. And one of the things that people always get caught up on is origin. And what I would ask is that we we sort of keep origin to the side for a second in this conversation. um and just talk about what is without trying to ascribe an origin to because these are all things that are happening regardless of what you want to ascribe the origin to. >> Um and it's interesting regardless of what the origin is. >> And so I'm going to do a quick retrospective on kind of the recent history of the UAP subject.

1:06:28>> Um that is is the context in which these two papers are coming out. And so for many people, uh, they're probably familiar with the 2017 New York Times article that came out that basically talked about the fact that there's these US government programs that are stud studying UFOs that have existed for some time. >> Um, and that was sort of the catalyst for the modern era. This subject has a long history prior to that. Uh, that is outside of the scope of this episode. >> Yeah. And so we're not going to touch on sort of the, you know, the what I call the the legacy era of the subject. Um, but after that New York Times article

1:07:09came out in 2017, it contained three videos. This was the first time we're getting videos with chain of custody from the US government that were showing these anomalous objects. Right. >> Okay. But it was still only put out by the New York Times, which people will have different differing opinions about the quality of that institution as a newspaper. Uh that's changed since that time period. But the Pentagon actually confirmed that the what are cloally known as the Gimble, Fleer, and Go Fast videos are legitimate videos that have chain of custody to the US government. That happened between 2017 and 2019.

1:07:49What sort of proceeded from there was an escalation of the issue from a national security perspective. So in 2020, the then Senate Intel minority leader Marco Rubio, who is now the national security adviser Oh yeah. in the Trump administration uh was one of several senators who called for there to be an unclassified report to be created uh by the department of defense around >> what are what are these UAP what is this unidentified anomalous phenomena in 2021 the office of the director of national

1:08:30intelligence delivered that report uh in which they stated that there were 143 unexplained cases uh that you know again this was largely driven by Navy and other pilots basically saying look this is a safety of flight issue and we need to like be able to report and address this just for our day-to-day operations right >> again independent of origin or any of these other issues the DoD then created you know this UAP task force which was this sort of temporary task force that went out to both the uh intelligence agencies and the military uh sort of apparatus to try to understand are is

1:09:12are are you all seeing this? Where are you seeing this stuff? What is going on here? >> That eventually now rolled into what we now know as this permanent Pentagon office that's called the uh all domain anomaly resolution office that was established in 2022. Um, and that office was congressionally mandated to be created after that initial report came out because Congress is basically alleging that the executive branch, which runs the intelligence agencies and the military, are not providing them the information about what's happening as it relates to this subject specifically.

1:09:53And it's important to note that the term UAP is a specific term of art that does not include any other form of unmanned aerial systems. So this is drones, this is you know undersea unmanned like those are all discrete and defined and there's processes we have to deal with that. The reason why this became an issue within the government is that these platforms were exhibiting flight characteristics or behaviors that were beyond the scope of what was deemed as current or even next generation technologies. This has then been

1:10:33followed up by multiple hearings within the House and the Senate on this issue and has culminated with legislation that part of which was passed uh that was proposed in initially in 2023 but then passed in 2024 that mandated that executive agencies release information around the history of this subject. I mean, there's literally a a colloquy between Senator Schumer, who was the then Senate Majority Leader for the Democrats, and Senator Mike Rounds, who's a Republican, >> where they were sort of going back and forth, >> identifying that, you know, that this is something that has been an ongoing issue

1:11:15that Congress has been going after and the executive branch has been unwilling to provide information on. So, this is sort of a live issue. NASA had a UAP independent study group. They put out their own report around this issue. Um they are obviously struggling for funding. So that it's not something that they're spending money on because they're trying to just survive as it is. >> Yeah. >> There's been multiple whistleblower protections that have been passed. There's been defunding of UAP programs for the intelligence community. I mean there's this long list of indirect evidence around this subject but there's been a der of scientific research.

1:11:55>> Yes. Really the main primary thing that has existed in this space particularly in the last couple of years have been research papers that have been trying to define how to study this subject because of the reproducibility problem and the stigma both of which are big issues and the the reference point that has been addressed is like multime messenger astronomy studies has been the reference point for how to try to deal with this problem. But there are two papers that have just come out that are a first of its kind in terms of having actual data that again can be potentially either

1:12:36reproduced or replicated in other arenas from the Palomar Sky Survey and we're going to kind of go in from a first principles perspective to understand what these papers actually were talking about. >> Yeah. Yeah. So there's a lead author Dr. Beatrice Villa Royale. >> Vill Royale. Yes. >> Villa Royale. >> Yes. >> I hope I'm saying that right. >> Yes. >> So the lead author, Dr. Beatrice Villa Royale, um she actually covered this

Story 3 starts — Palomar POSS-I “multi-point transients” revisit

1:13:04issue before in her 2021 paper. >> Yes. >> Um that was published in Nature Scientific Reports. It was on nine simultaneous transients occurred in April 1950. Um, >> and it's it's important to note that like like research on this has been done >> has been done and this is this is kind of a follow-up paper but it's I believe a lot more thorough and I just thought that I just thought that the way that um the research was conducted you know the the the science sort of speaks for itself and then the interpretations come afterwards but >> you know on this podcast we always focus on from first principles what is the science that you did and um does it

1:13:47warrant the conclusions that you put out? >> Yes. >> Right. So, we're going to start with these papers. The papers focus on the Palomar Observatory Sky Survey. Okay. This is POSS1.

Samuel Oschin/Schmidt camera & huge field of view (13×13 moons)

1:14:03It was a sky survey that was taken up by the Samuel Austin telescope, which is a camera that is mounted on the Samuel Austin telescope at Valomar Observatory. There you can see the the telescope itself. It's got a Schmick camera and this thing is specialized for wide field photography. These cameras are actually pretty insane because so that camera the camera that is that is attached to it has a field of view of 6.6° by 6.6°. Okay. So 6.6 degrees by 6.6 degrees. The the full moon for comparison is half a degree. It's about like if you were to put your thumb right in front of your face, the the size of

1:14:44your thumb is about half a degree. This is 13 full moons by 13 full moons. So, it's 170 full moons. >> Mhm. >> Big is the field of view of this camera, right? And >> because it's so big, there's a specialized camera called a Schmidt camera. And what it does is you've got you've got a primary mirror, right? The mirror is about 1.2 m. That mirror then goes into the camera. You can imagine the the light is getting warped, right? As the light comes in, it gets warped. And so the edges of your field of view are going to get distorted. >> Yep. Yep. >> This specific camera is specialized for

1:15:26that kind of wide field photography. So it corrects for that distortion. And even then, the focal plane is actually not flat. You know how in like a camera your CCD is basically a flat piece of equipment because all of your light is coming onto a flat piece of equipment here because it's such a large field of view and it's so so big the focal the focal plane is actually curved. So in order to get a photograph your photograph the the the the plate in this case the photographic plate and these are all photographic plates that are used also has to be curved. And so you got to like warp it before you slide it

1:16:07in there and get this photograph. I thought that was really cool. Like the the amount of stuff that needed to happen in the 1950s in order to get these kinds of large field of view photographs right? >> Given the technology of this is purely photographic, which means that the other cool thing is that that that um the the telescope that we just saw that telescope is an equatorial mount. Okay? And what that means is that one of the axes is parallel to the Earth's axis of rotation. And so all you have to do is rotate along that axis and it'll follow the stars as the Earth rotates underneath. Right? The the the stars are all going to move in a circle around the North Star. And so in order to track a

1:16:48single spot in the sky, the telescope has to rotate with the Earth so that it points in one direction in the celestial sphere. And in order to actually do that back then, they had a manual astronomer that would look through a 10-in smaller telescope to a guide star and then and then control the telescope manually to keep that guide star in the exact same position. So there was a lot of like manual stuff going on back in the day. And I think this is really important to see to to understand >> how how the data was collected, right? It's not with this modern technology where now everything is automated and

1:17:29nothing is actually in equatorial because it's it's easy to it's easier to make bulkier telescopes in the azimuthal mount >> where I can control both axes with a computer and I I just have to like do the calculation and control both axes and I'm fine. Right. Um, so this guy survey produced a thousand photographic plates, about a thousand, one in red, one in blue. So then you'd have about

Curved focal plane & photographic plates: why the hardware matters

1:17:542,000. Um, this is an example of one of the photographic plates. Um, it's a onederee by one degree patch of the plate that is targeting Pletes, the seven sisters. Um, and the the haziness that you see, that's actually not a defect. Those are actual clouds around those stars, those dust clouds around those stars that that are creating that. It's a beautiful beautiful example of one of the photographic plates that was actually taken during that time. The typical exposure was about 45 to 50 minutes. Um what we're focusing on here are the red sensitive plates. They used to take red sensitive and blue sensitive. Here we're focusing on the

1:18:34red sensitive because that's what was used in the study here. Um, and the limiting magnitude um is 22 in terms of apparent brightness. The higher the magnitude, the the dimmer the thingy is. And this is about 1 million times fainter than what the human eye can see. Okay. So that's what's limiting the faintest object that this telescope can see. >> Can see. That's like the the lower bound. >> Yes, that is the lower bound. So that's the telescope that we're using and that's how we're getting these images. Now, the photographic plates. We got to talk about the photograph. >> We got to talk about the plates. >> What does that mean? >> Yes. >> So, photographic plates are 14x4 in. And

1:19:14these 14x4 in are placed at the back of the camera where all of the light is coming in. It's 1 mm thick. These are incredibly thin for how big they are. They're like the size of a record, you know, a vinyl record. Um they've got a light they they've got a photographic emulsion on them. Okay. A light sensitive silver halli crystal layer. Usually it's silver broomemide and that's that's suspended in gelatin. Okay. Bromium is the h hallogen that they're using here. Halogen is like that column on the periodic table that's second to last on the right. These plates are bent as I was saying so that they can match the the the focal plane

1:19:56which I thought was just really cool to think about. Um and then the process is a photon comes in >> from your telescope and it liberates an electron in this silver broomemide and that electron that that liberation of the electron forms a latent image spec. You've got this free silver atom. Okay, that creates a silver atom cluster. Then during development that latent image spec catalyzes and the silver broomemide that's that's that's around there goes and turns into black metallic silver. Okay. And then finally you have a washer a kind of fixer that removes any unexposed silver broomemide. This is

1:20:37like the old school you know where you like wash the thingy and then that removes any >> silver broomemide and what you're left with is a permanent negative image. And that's why in the in the previous image that you saw, if you actually go back to photo two, you'll see in the previous image, the dark spots are actually where the light is. >> Okay? >> Right? >> Cuz that's where the silver is getting deposited. Okay? So you get this negative image, >> right? >> Of of the dark is actually where the light is coming in. >> Okay? And the white is where there was no photo reaction that happened. Okay? And so that's how we actually get these photographic plates. >> That's what a photographic plate

1:21:17>> that's what a photo photographic plate is. And that is what the chemical reaction of what a photographic plate means. Okay. Now what this what what what this sky survey is doing and and the the data that this particular group used is the digitized version of those photographic plates. Okay. So the digitized version in the 1980s and 1990s, we did a digitization process where those glass plates used um micro densitometers that basically went through and digitized these photographic plates. You've got a light beam that measures the opacity of the photographic

1:21:58plate and that converts that opacity into a digital pixel value. So the more opaque it is, that means that there was more dark, >> right? There was more darkness. And then if there's more darkness that means that the there's more light over there. And so and so now you've converted from >> photographic plates to a digital sort of pixel values, right? Yes. >> And there were two >> digitization processes that happened. Okay. There was the digital sky survey. Yes. >> That was um done with a pixel size of 25 microns. And then there was a

1:22:38supercosmos sky survey >> and that was done at the Royal Observatory of Edinburg >> with uh higher resolution of 0.7 arcseconds per pixel. >> Got it. Okay. So the first one was 1.7 arcsec per pixel and then this one's 0.7 arcsecond. So it's just like more gra like grainier detail, you know, sort of think about >> a slightly higher fidelity. Yes, exactly. And now you've got a vetting step, right? Where now we can create computer algorithms that look at these digital photographic

1:23:18versions and then try to find transients, little spots in the sky that aren't there in both aren't there in current photographs of that patch of the sky, >> but that were there >> that were there back then. Okay. >> Right. And what we can also do is we can rule out any scanning artifacts because we've got two different digital versions >> of the photographic plate. So if something happened in the scanning where I got a little speck that shouldn't be there in both at the same spot, >> right? The probability of that is insane. >> Is insanely high. The the point is the the the digital scans there's two there's two versions. >> And so if someone was at asleep at the

1:23:59wheel Yeah. when they scan version one, >> it should not be there in version two at the exact same spot >> in in such a precise >> exactly >> way. And so what what you're sort of there's a layer of of of validation. >> Yeah, there's like a redundancy here. >> Redundancy here because we have two digital >> Yeah, we have two digital versions of this. Okay. And so now we get into the Vasco project, right? Which is the vanishing and appearing sources during a century of observations. This is led by Beatatric Villa Roel. And the the point of this project is to have a systematic comparison between the

1:24:40digitized versions and modern sky surveys. Right? You look at what the Palomar Observatory saw back then and you look at what we're looking at right now. Yes. And you've got an automated software that flags these vanishing the present present then missing now, right? And there's also appearing. So not present then but appearing now, right? You've got an automated software that flags these things. And initially you've got a catalog of 100,000 transient candidates. And here we're seeing two different images. One from pass one, which is the first sky survey and pass two. And the circles represent two

1:25:21little spots that you can see are exactly not there in the second one. Right? It's like pretty obvious that those two dots were there before and now they're not there now. Right? And so that's a transient. You've got a point-like source that's there in some plates, but in the later plates it's not there. >> And the idea is, you know, we have two points in time and we should be seeing identical things in both those points in time. >> Yes. If it's a star then yeah >> because of >> and it's not identical. >> It's not identical. Right. So >> people are going to be skeptical about this. Right. For example, it could just be plate defects. Sure. Right. Both this

1:26:04this point of source of light or whatever this thing that we're seeing in the photographic plate, it's there in both the digital versions. Well, that means it could just be in the plate, right, itself. What is a hallmark of something that is in the plate? And this is something that comes from Nigel Hamley at the University of Edinburgh. He's one of the critics. He looks at something called the full width at half maximum of these points of light. Okay. So, the full width at half max is basically imagine you've got a gausian. Yeah. >> What you're what you're looking for is what is the width of that gausian when the height is half the the peak height. Okay. And what this tells you is

1:26:46something about the noise in the data gathering process. Now I told you this this um this telescope, right? It's traversing the night sky. >> Yes. >> And it's being controlled by a human being. Well, there's going to be a little bit of noise in how I control that telescope as it tracks, right? So, even a stationary point of light is going to have a little bit of jitter, right? As I move this telescope, the telescope is going to not totally track it completely. And even a point of light like a star that is definitely stationary over the 50 minutes at least, you know, it might be moving over like several years, but like over 50 minutes, it's stationary. that's going to have a

1:27:28little spread in my photographic plate. And what he noticed was these transients, the ones that the ones that sort of appear and disappear, they've got uh a spread that's way smaller >> than the stars. >> Mhm. >> Okay. >> Mhm. Yeah. Yeah. And that to him was a telltale sign that this is a defect because if it was a if it was a light source, then the light source should have the same sort of error that the stars do >> because they're all light, >> right? Does that make sense? >> It does. It does. >> Right. Like all of the other things have the same sort of spread because it's from the same jitter of the telescope. >> It's uniform across the system, >> right? >> Yeah. Yeah. Yeah. >> That's it's a it's a key thing. Yeah. >> Now, the contention is the the sort of

1:28:10counterargument is well, if it's truly transient, meaning like it only like lit up for a minute or a fraction of a second and it had a bright signature, then only one part of that 50-minute exposure is going to have that light come on and then it's going to go away. So then of course the full width at half max is going to be way shorter because >> it's not going to have time to accumulate that noise, right? But now we're here. We're at a degeneracy problem, right? Because a genuinely short-lived astronomical event, right? Something that happens for a fraction of a second is going to look identical to a

1:28:51plate artifact. >> Got it. Right. you know. Yeah. >> It's like you're looking for you're looking for a shortlived astronomical event, but that's going to look exact, you know. >> Yeah. Yeah. How do you distinguish between what is being argued as a plate artifact versus a true >> Yeah. >> uh short-lived transient event? Because it's hard to distinguish between those two. >> Yeah. Yeah. Yeah. The the the measure that we're using is degenerate to um discerning between one or the other. Does that make sense? >> It's it's it's one of the two. >> Yeah. But >> but like the measure that we have which is this full with a half max is like it's going to be the same for both. >> Got it.

1:29:32>> Right. So So which one is it? And that's why there was a necessity for an independent line of evidence. >> Okay. >> And this is where we're getting into the two papers that we have now. >> Okay. >> Okay. So it's it's this it's this challenge of well it could be a defect or it could be a truly transient object. Which one is it? And the the the interesting note here just as a a a caveat or a point of order is there there was a emphasis on this being you know the polymer sky survey which happened in the ' 50s was pre-sputnik yes uh sky survey and nowadays when we talk about transient

1:30:13light artifacts it's starlink it's other satellites, other whatever you Chinese classified platforms, but there's a lot of stuff >> Yeah. >> that we have now put up there. >> And so it is much easier nowadays to say, oh, it's X, Y, or Z. >> Yep. >> One of the interesting aspects of why this distinguishing between whether it's a plate artifact or not is if it's not, we're seeing these transient light anomalies at a time period where we

1:30:54>> we at least didn't put anything up there. Right. On purpose. >> On purpose. >> Right. It's this is from November 1949 to December 1958. This is before >> Sputnik, which could be interesting. >> Yes. >> Because of the fact that it can't just be pointed to being >> Starlink passed by. >> Exly. Exactly. Exactly. And so now that is where we're getting into the two papers that have come out from the Vasco project. Okay. What they've done is with the first paper, they're doing a spatial arrangement on individual plates and they're arguing about um the transients, how they look on individual plates, and where they are in the night sky in terms of

1:31:37when they were when these photographic plates were taken, where was the Earth's shadow, and if we can do some kind of statistical comparison and really show that these are stuff that's in the night sky and not photo photographic plate defects. And then there's a second paper that talks about temporal correlations in time between the stuff that we see in the photographic plates and stuff like nuclear tests and UAP sightings which we'll get to. So study one, this is the alignment hypothesis. This is Villa Royale um and others in 2025 in PASP. Yes, the paper

1:32:18is aligned multiple transient events in the first Palomar Sky Survey. The hypothesis is that these multiple pointlike transients are aligned along a straight line on the same plate, which is very unlikely by chance, and they could be real physical objects. And that is the argument that they're making. Okay. So, they've got a catalog of about 300,000 short duration transients from this Palomar survey that we just went into. >> Yes. >> Okay. And the first thing that they do is they ask how many of these things are aligned. They're in a line. Okay. And here we've got a photo um of plate five. This is

1:32:58candidate five. >> Mhm. >> There's five different thingies that are all in a line. Okay. And they use this thing the the a statistical framework from Edmunds and George published in 1985. and it calculates how often you would expect these things to happen by chance. >> Okay. >> Okay. And they find a sigma of 3.9 that this is not chance. >> Okay. >> Okay. 3.9 is it's pretty high. Okay. It's it's not like five sigma which we've talked about in our previous

Alignment significance (~3.9σ) and why it’s not decisive alone

1:33:31episode about the uh strongest evidence yet for life on an exoplanet which was uh three sigma >> which was three sigma. This is 3.9. It's higher, but it's still not five sigma. Sure. Right. But it's kind of interesting. Okay. And this is just candidate five. There's other candidates that also have around three sigma, right? And they they go through this in their paper. Um that's actually not personally what I found the most interesting. The alignment is like I don't know by chance. I don't really know how to reason about that in my head about how often to do things by chance. And also the the framework the the statistical framework by Edmunds and

1:34:12George it relies on a bunch of different things like the number of objects in the sky the assumption about the density of objects how high they are like all this other kind of stuff right the earth's shadow test is I think a decisive diagnostic so here's the logic you've got the sun you've got the earth and the earth casts a shadow on the rest of space right when the moon goes in in the earth shadow that's how you get a lunar eclipse.

The Earth’s-shadow test: a cleaner diagnostic idea

1:34:38So if we were to think that all of these things are plate defects, all of these transients are plate defects and not actual objects in the sky that are reflecting sun sunlight, then if they are plate defects, the number of things that are happening in Earth's shadow should be the same as outside Earth's shadow because it's not dependent on whether the thingy whether the patch of sky I am looking at is in earth's shadow or not. >> Right? >> If on the other hand these objects are reflecting the light from the sun then there should be fewer transients when

1:35:21I'm looking in earth's shadow >> right >> than when I'm not. >> Right. >> Okay. And I know the coordinate of the spot in the sky where I'm looking, right? Because each of these photographic plates has a right ascension and declination, which is our latitude and longitude in the sky. We have our time of day. So from that, I can calculate which patch of the photographic plate is in Earth's shadow and which patch is not. I do have to do an assumption based on like how high I think these objects are, but they make an argument which I actually don't quite fully understand, but they make an argument that um these objects are

1:36:01probably at geocynchronous orbit, which is 42,000 km. Um and at that altitude, I can calculate which object which part of my photographic plate is going to be in it Earth shadow and which part is not. Right? And then from that I get an expected number of 12 thou 1,200 objects but I only observe 349. >> So that's a pretty statistically significant sample. And this is where I remember when you first came to me with the story you said 22 sigma significance right the statistical significance. I

1:36:41was like what are you talking about? This is what they're talking about. They're talking about that like the 349 is so far away from 1223. It's 22 sigma away. They actually did a more realistic comparison where what they did was um they they looked at the actual sky because this was actually kind of cool. So the the photographic plates in Palomar didn't uniformly sample the entire night sky. The 22 sigma comes from the assumption that the photographic plates are completely uniformly sampling the night sky. On the other hand, some patches were looked at more than others, right? And some patches were completely missed. So what they did was they did a more realistic

1:37:23simulation where they said, "Okay, suppose I only looked at those patches and I randomly generated objects. How often would I see defects versus not defects and all that other kind of stuff?" And there they get a lower significance, but it's still a lot. 7.6. six sigma. That sits better with me. Okay. 7.6 >> seems more reasonable. And and the benchmarking that they did kind of made more sense to me, right? Because they're they're taking into account this prior distribution of like where the telescope was looking rather than an assumption that oh, it just looked everywhere. Well, it didn't. Right. >> It's it's not the Vera Rubin Observatory, >> right? Right. It's not just like Yeah.

1:38:03Yeah. This is the 50s, right? like it took us 80 years to get to get to the Vera Rubin. So, so, so that's the first um that's the first story. The second story which is study number two, this is published in um nature scientific reports and this is correlations with the atomic age. Okay. Transients in the Palomar Observatory Sky Survey may be associated with nuclear testing and reports of unidentified anomalous phenomenon. This is it's got UAP in the title. >> It's got UAP in the title in in in in a nature subjournal >> in a nature subjournal. Springer Springer publication, >> which which is a a significant thing in

1:38:44and of itself given the uh aversion to doing anything related to this subject from most peerreview journals. >> Yeah. I mean, I was pretty surprised myself. >> Yeah. When you were like cuz I brought up like well what what journal is it? I was like and you're like wait really >> wait nature. Yeah. Fine. I guess if it's Springer, we can do it. >> We'll cover it if it's >> Yeah, we'll cover it. Yeah. So, here what they're doing is they're doing a more of a time sensitive approach. Okay. >> Mhm. >> They're looking at the rate of the transients. >> Yep. >> That appear in those photographic plates that I talked about. Yes. >> And they're trying to correlate it with dates of above ground nuclear weapons tests. >> Yes.

1:39:24>> And daily counts of UAP reports. Okay. So the the timeline is again 1949 to 1957. They get a bunch of transient count. So for every single day they've got photographic plates and they count how many transients were found on each day. Each day also gets a binary flag on whether it's within a day of a nuclear test or not. >> Right? And back then I think only three countries were doing nuclear tests. The Soviets, the US, and Great Britain. Um, and then there's also daily UAP reports from the UFO CAT database. And these three things are what they're

1:40:05trying to relate. >> Yeah. Y >> they use something called a generalized linear model. I'm very familiar with something called a generalized linear model because my PhD had a lot of that. Basically, these kinds of models are what you want to do when things are not just simply linearly correlated to one another. In my PhD specifically, I was trying to correlate the spiking of neurons in terms of many different behavioral motifs like which way he's looking, how fast he's running, where he is. All of these things can have varied responses that are not necessarily like, oh, if I'm looking this way, I just have more. Maybe it's like a weird response, right?

1:40:46Where only if he's looking this way, there's going to be more. So all of these like different multimodal effects can be captured by a generalized linear model and that's what they did. It can also simulate weird distributions that aren't like just like uniformly distributed or gausian distributed. If you've got like a binomial distribution which is like like coin flips. >> Yep. >> If if you've got you know did I see something or did I not things like that it's very good at modeling those kinds of things. Right. So there was a correlation to nuclear testing. So the transients occurrence in these photographic plates was 45% more likely within a 3-day nuclear test window.

1:41:27Okay. And the main thing was it was 68% higher likelihood on a day after the test. >> Right. >> Mhm. >> It's pretty significant. >> It's pretty significant. There's also a correlation plot that they do with UAP reports where using that generalized linear model analysis 8.5% increase in transients for every additional independent UAP report on the same day. I don't really care for that as much. Like UAP reports to me are like I don't know humans are fooled very easily and UAP reports are just humans being like I saw something that was crazy. Um there's I I can't think of whether there are tests that I could do to to see if this

1:42:08would work. Like for example, what if you like assumed that like on average half of the UAP reports were nonsense? If you took out half, would the correlation still be there? If you did a bunch of samples where you took out half and you looked for correlation, would there be some samples where the correlation would be like way higher, you know? >> Yeah. Yeah. Like that's something that we used to do in my neuroscience lab in PhD is like >> how sure are we that this variable is correlated to this variable where if I were to just take out a subset of the data if the correlation still persists then >> it's it's a more robust finding right and given given the shape of that plot

1:42:49I'm not sure if I believe it but the the the nuclear testing thing is just like that's just there right so yeah So, so those are the two reports. There's one report in PASP that's physical evidence. You've got the Earth's shadow test that a substantial fraction of the transients are sunlit objects, right? And then and then you've got the scientific reports that circumstantial behavioral evidence that says that the timing of these transients is affected by nuclear tests, right? Um there are people who obviously

1:43:30um are skeptics, right? >> As as we also saw with the same paper I just brought up earlier about the strongest evidence yet, that came immediately with follow-ups from multiple institutions that pointed to why their modeling was not sufficiently robust. >> Yeah. Like as of now, I haven't seen any follow-up scientific papers. Um because all of this all of this data is publicly available and I actually saw on the >> on the two papers it's it's readily available for people to look at. So I encourage anyone to to do their own analysis. Actually >> I I want to make a funny note on the paper that I keep bringing up within two

1:44:11weeks cuz it was was it >> it was a week dude. >> It was a week within a week cuz it was out of Cambridge offered one of the >> Yeah. Yeah. Cambridge had the original paper and then the guy at Oxford was like no than a week after came out with the rebuttal and then multiple came out. It's been more than a week. >> Yeah, >> it is getting this paper is getting quite a bit of just obviously a lot of the science media is has is doing variety of different coverage, but there's not yet necessarily been false as quickly. Yeah, >> there's a number of reasons that could be true. >> Yeah, I mean one one of them I think the big one is probably it's not an active uh scientific field like exoplanets and exoplanet researches. There's everyone's trying to make a name for themselves,

1:44:52right? They've got a career to look after. This one is not like >> something that, you know, okay, so you prove this wrong. It's not going to like give you a leg up in astrophysical career. >> There's no there's no incentive for people who are looking to climb a ladder >> to follow up on. >> Yeah. Yeah. Yeah. Yeah. So, so that could be one of the reasons. Um but in any case um the scientific American actually had an article about it that I was reading and they showed some rebuttals right so there's the um Hamley argument which is from the University of Edinburgh he's an expert Nigel Hamley he's an expert in photographic plate analysis and his main contention was that full width at half

1:45:32max right the fact that these transients are so pinpointy and precise compared to the stars that are kind of spread out because of all the noise from the detectors. And there's skepticism that, you know, this is a classic signature of an emulsion or a flaw or a plate defect. And it's not a celestial object because a celestial object would have this kind of noise. >> Um, and one of the crucial things that he says is the ultimate orbiter is actually just the physical evidence. The plates are still around. We can take those original glass plates and examine them under a microscope and not examine the digital scans. And under a microscope, it's going to be very

1:46:12obvious whether this is a defect of the silver just like being kind of weird at that point or if it's a genuine astrophysical source. So, I think I think that's, you know, he's he's proposing a way to >> to make it go away >> to make it go away. I I think that's good. So if you're interested in doing a microscopic study of the digital plates and you would like to collaborate just in the comments below, >> yeah, let us know. >> Let us know. But and we already talked just previously about >> even in the paper itself a means by which that is not necessarily the likely

1:46:52outcome. However, you can remove it. Yes, the earth shadow point. you can still remove it entirely from being the cudgel that you're seeing being put out because a lot of times in the coverage of it, people say, "Oh, it's probably plate defects." Without actually explaining >> Yeah. >> how the paper points to why it's not plate defects. >> Yeah. Yeah. Yeah. >> Which is the kind of the issue. >> The Earth Shadow thing is like pretty pretty substantive, I think. And so, however, it can the the the beauty the beauty of us still having the original plates is that it it's falsifiable. >> Yes, exactly. Um, there's another one that's proposed by uh Sean Kirkpatre,

1:47:33who's the >> Dr. Jean Kirkpatre, >> former Pentagon AO head. >> We we like to call it arrow. >> Arrow. Okay. Arrow head. Um, I guess that's like a anomaly office or something. the the that's the all domain anomaly resolution office. >> Okay, got it. >> That was created in 2022 as a result of Congress pushing on the executive branch to like give them more information about again because >> military and intelligence officers were coming to Congress saying this is happening and we we we need help in dealing with this and we're not getting the support we need. >> I see. That was that was the impetus for why the office was established and its mandate was twofold which was

1:48:15unfortunate because it was basically a mandate that that had conflating uh outcomes. One was we want you Arrow to take data from the Department of Defense, all of the combatant commands and the intelligence community intake it and then analyze what we're seeing as these UAP reports and if it's conventional send it to the office that exists to deal with it and if it's unconventional figure out what we need to do to figure out what what it is. Right? That was one of its mandates. The other mandate was a public communications mandate to help to demystify this issue to the public by creating public education

1:48:56materials and this then the third. The problem with that is you're asking you know the department of defense which is not a public communications office to do public communications. There's a reason why NASA >> is its own agency. >> Yeah. And you don't have NASA running out of the DoD because we would never get anything out of >> Yeah. >> that structure. So there's sort of a confl there's there's a friction between the two things the two things Congress asked it to do which is why a lot of the public is frustrated because they are not very forthcoming with all they have thousands of these reports. We've basically seen none of it. >> Yeah. Okay. Well, he he he put his uh Sean Kirkpatre he he gave his two cents

1:49:38which I thought was actually quite good which was he said you know there's nuclear detonation and clearly this thing is correlated with nuclear detonation. Now when you have these high altitude nuclear detonations, you know, you can have radiation particulate stuff that goes away and it can create transient luminous events like churnov radiation or even reflective metallic junk like the the stuff that made the nuclear like you know there's going to be stuff that gets thrown out and that junk could reflect light in the way that we're seeing in these Palomar sky surveys. This is interesting because I was actually I was just at the soul conference at in Lake Missouri, Italy this past week where Dr. Bishers Vill

1:50:21presented these two papers. Um, and one of the there's a Q&A section where one of the people actually asked about the churnov radiation >> uh as a solution to the problem set. Someone can clip this and put the response. I don't think it's public yet. I didn't I didn't write it down, but that she did have a a response to that issue. >> Yeah. >> Um which I I unfortunately can't replicate in this context, but that was that was brought up by someone in the audience as the like obvious next step in the progression as as an answer to it. >> Yeah, I can imagine like I mean for churnov radiation it would probably be like streaky rather than single point sources. Um but you know if

1:51:04you have like metallic junk >> then it's that >> then that that could like just like you know as it's turning like create that reflective. So uh to me the metallic junk is is a a better explanation than chirkoff radiation or ionization because those have very characteristic >> signatures um on photographic plates. But one of the challenges that Sean Kirkpatrick said was, you know, we've got GPS satellites now. Why don't we try to reproduce these results with the GPS satellites? Um, I think it would I think it's harder said than done because GPS there's so many satellites now, right? It's not just like you're

1:51:45gonna have so many streaks. So, >> for a 50minute exposure, you're just gonna get completely wrecked by Starlink, right? Um there's another proposition by Elliot Gillum at the um SETI Institute search for extraterrestrial intelligence. Um he said you know meteors traveling directly towards the telescope could appear stationary dots not streaks. Um it's viable but like now you got it's traveling directly at the telescope like okay >> and that that volume level and simultaneous simultaneous multiple aligned. >> It's a little challenging. Yeah, it's a little challenging given the numbers. Um,

1:52:26there's a Princeton astrophysicist Robert Lupton who has this look elsewhere effect. He's basically saying that, you know, in a massive data set with thousands of plates, millions of images, finding a few apparent alignments by chance is not that surprising. And you can't actually use the Edmonds and George formula to do that kind of statistical significance to figure out whether stuff is in a line because there's so many things that are in a line. You know, you're you you can easily fool yourself into thinking that the thing that is in a line is there for some reason, right? And the probabilities need to be corrected for the vast number of stuff that's out there.

1:53:06>> This goes back to your not looking at the alignment as the interesting part. >> Yeah. Yeah. Yeah. And I I think there's a I need a little bit more convincing on the alignment being something that's interesting. Okay. >> Right. Um but I mean the debate is there. >> You know 7.6 sigma deficit of transients in the umbra. That's not something that I think we can ignore. Um, I think it forces a shift from are any of these transients real to okay, given a significant fraction of them are probably real sunlit objects, what are they? >> Okay. Um, it's it's it's certainly like giving me

1:53:48pause. I'll be honest, >> which I want to say for the audience, I've been working this guy for a long time. So to be giving him pause >> Yeah. is is >> it's like that's statistically significant. >> Yeah. Yeah. Yeah. Yeah. Yeah. Yeah. Yeah. You probably have like the number of times you've tried like nah. >> I've tried so hard. But but I I I thought you'd find this interesting even just from the experimental design perspective, the clever approach of trying to do a study like this in a way that's compelling. >> Yeah. I I mean I mean clearly like the team the team in the PASP paper at least the the one in the proceedings of the

1:54:29Pacific Astronomical Society that one that one has done its job of like you know going through the proper science and things like that. The the time stuff I'm a little less excited about. Uh the the one with the the earth shadow thing, that's just like solid calculation, you know, >> good math. >> Yeah, that's just like good math. Um there is like there was one that was kind of interesting. So candidate five, the five transients um systematically in alignment that happened on July 22nd, 1952. >> July 27th. 27th. >> Oh yeah, July 27th, 1952. And they mentioned that this coincides with the

1:55:11um most intense weekend of 1952 Washington DC UFO flap, which I hadn't heard about, but I looked it up. And we've got a photo showing like the the the recorded number of of incidents that were reported. And apparently like there were visual sightings, fighter jets were scrambled, there was a press conference from the Pentagon. So, I didn't even know that this thing happened. So, this is one of the things weird, bro. >> So, this is one of the things that's so interesting. Why would they do that? >> It's it's so interesting because there's again this is why when I start at the beginning of this there's this whole

1:55:53legacy of this subject that exists um that's very dense and there's a lot of information there um and you know unless you have an entry point you won't know it's there but you know we've UFO flaps are consistent and persistent throughout about the history not only of the US but global society and it it it is interesting to see it how it cyclally comes back up. Um but that was candidate 5. Yeah. And that was the one that was like super aligned. I mean um I still think it's pro it's probably coincidence but it's it's an interesting coincidence. I'll say that.

1:56:33Um, the other the other thing they they do at the end of the at the end of the PASP paper is they look at 3D simulations in Blender and show how these reflective objects, if they're like slowly processing something that has a shape like that, could create the kind of signature that we see where like randomly it's at exactly the right angle where it's reflecting the sun into the Palomar telescope and then it goes away real quick. So then you get a really sharp, you know, image spec in your photographic plate. It's speculative at best. I think I

Where we land: interesting, speculative; next steps & Rubin data

1:57:13think um I think it's certainly interesting. I think I think the Earth's shadow thing is the one that like is really like sticking with me, right? That calculation is is quite nice. And the way that they do the the way the way that they do the controls where they simulate a random subset in that photographic plate and say, "Okay, how many are going to be in Earth shadow versus not like Yeah, that's see seems to be the way that I used to do like you know these bootstrap >> Yep. >> things to try to figure out what the significance of my result was." So yeah, I I thought it was I thought it was pretty cool. I think next steps would be to just look at the photographic plate, make sure that it's not a defect. take take a look at the plate not a defect and then from there

1:57:53>> if proven to be the case >> now we can have the next stage of the conversation. >> Yeah. Which is like okay and like you know with modern surveys perhaps there's a way to get an idea of how often this is happening in Vera Rubin. Vera Rubin's um completely public all of the data is public like immediately after it's >> and it's up and running right now. >> Yeah. Yeah. So we have already three months of data from the Vera Rubin. If you've not already listened, take a look back to our previous episode where we covered the Ver Rubin Observatory, why it matters. It's another sky survey, but it is >> just light years in terms of the structure and substance of how it works and what we talked about the data, the

1:58:35immediate data availability, no embargo, da da da da. Again, some people be like, well, you know, the government's going to like try to like wipe away their classified projects that are >> No, but this is just I mean, and it's also it's been I think let's say 3 months. >> Yeah. >> 3 months meaning 100 days, meaning there's 30 photos of the entire night sky already. Right. >> Right. Palomar took years for one photo of the entire night sky. This already has 30. >> It's an incredible The largest camera. >> Yeah. The largest camera in the world. >> In the world. um creating the biggest amount of data and one of the most incredible data processing and

1:59:15distribution platforms. Yeah. >> Uh for it's it's just an incredible. So the the point here being >> there are things that can be drawn from this study in terms of now looking at us having this larger much more robust data set and applying some of those thoughts and learnings to it. in addition to continuing to validate and replicate using the old the the the plates from the Palomar sky survey as well. Like there's multiple avenues of exploration here um around this idea of these transient light artifacts that are occurring. I I really was excited to just get your

1:59:56perspective on this. >> Yeah, I thought I thought it was I thought it was a cool set of papers. Yeah. and and >> definitely the PSP one I I liked >> and just also talking about it from first principles. Um again, >> yeah, I think it's important to understand because because this is historical data, we're not used to analyzing that, >> right? And the analog nature of it. It's not digital, right? You're not sticking a CCD in the back of a telescope. This is at the end of the day a chemical reaction that's giving us something. So, it's important to understand the whole chain of custody, >> right? from photon to pixel. >> Right. Right. And how how do we even get to pixel from photon? >> Um an incredible set of stories this

2:00:38week. Uh we did our lucky number three. Uh we started with our generative AI for cancer detection out of University of Toronto. Again, sorry again about the loss, but you're going to save the lives. going to save the lives of millions of people. >> Yeah, >> with with that that great great story. Followed up by our super story on the retinal wireless retinal chip. We're making blind people see again. >> Yeah. >> Notless restoring. >> Yeah. Restoring. >> Not stopping, not slowing down. Restoring. >> Yeah. and and we ended with our uh interesting possibility of data related to what have now been renamed UAP,

2:01:19formerly known as UFOs and these transient uh presputnik transients in the Palymer sky survey. I'm very interested to see people apply some kind of similar methodology to the Ver Rubin data sets. They're there. They're publicly available. You're interested in working on that, please shoot us a DM. uh we can make the connections. Uh episode 15, we're continuing to cook. >> Uh unbelievable engagement on just I I still can't believe people care about science as much as they do. >> Yeah. >> Um >> it's I love to see it. >> It's you you do love to see it. Science is important. We are not covering I know people are begging us to cover three

2:02:01Atlas. >> Yeah. >> There's no data. The government shut down and there's no data for us to cover. Uh, however, Europa Europa Clipper is in the tail. >> Yeah. >> Right now, >> as we speak. >> Right now. >> Yeah. >> But we can't do anything. >> Yeah. >> Because there's nothing >> the ESA and um the the guys around Mars have taken photographs of gray atlas as it went across. It's it's reached it's gone through the closest point of approach from the sun and now it's on the other side of the sun. So, we can actually see it again from the Earth. It's on its way to Jupiter. But we have no data that we can talk about. >> So until we get the data, we won't cover it. Um I know that Avi Loe has talked

2:02:44about the uh non-gravitational acceleration which can have a prosaic explanation, but we can't know whether it is that prosaic commentary tail >> or not thingy or not until we get the data. >> Yeah. >> So we we hear you guys. You want us to cover 3II Atlas. uh open up the government, give us the data and we will talk about it. >> Yeah. >> Uh I am your host Lassinari joined as always by my co-host and our resident PhD Krishna Chowdery. This is from first principles. See y'all next week.