 Okay, we're back. We're live. It's the one o'clock clock. I'm here in likable science. Here's where I am. I'm Jake Fidel. That's Ethan Allen. He's the heart of likable science and he's likable, but he's also a scientist. In fact, he's our chief scientist here at ThinkDec. Thank you. Welcome to your show, Ethan. Glad to be here. So we're going to do something that's really interesting and it springs off an MIT newsletter article, and it's all about the genome, but it's also about artificial intelligence and how it has helped us, how artificial intelligence has helped us look at not only the human genome, but other species not too far from humans and look at that genome and tell the relationship. It looks into heredity and descendancy, if you will, the evolution of the human species. We know more about it now than ever. So what happened? What's the news story here? Well, the news story is a group applied very smart deep learning algorithms to examining a huge trove of data, modern human DNA data, ancestral human DNA data from old Homo specimens, Neanderthal data, denozovan data, which denozovans are another line of... Like Neanderthals. Like Neanderthals. Another place in the world and in the chain. Right. Australopithecine DNA. All kinds of different DNA. And these DNA records are incredibly sort of like giant code books. I mean, there's long strings of sort of meaningless strings of letters, which are coding for not only our 20,000 genes, with several thousand of these letters making up each gene, but huge stretches in between the genes of what used to be called junk DNA, but is now recognized as being very important. It's not so junk after all. It's very, very important. And reading the stuff and finding patterns within it and finding which patterns look like what other patterns that you're seeing in these other groups. It's a huge mind-bottomly complex task, and then matching that with what we know from bone structure, skull measurements, human migration pattern that's revealed by archaeological evidence. I mean, there's sort of multiple data sets stacked on one another. And the bottom line is what I find out is that we not only have sort of the lineage we knew, our basic Homo lineage. Some years ago we actually, we carry Neanderthal DNA, which had for years we thought we never did. I think I know some of those people here on Bishop Street. We all carry some of it, and some maybe a little more than others. And then we all carry this Denilovian DNA too. Again, different groups carry different amounts of it. But it turns out that's not enough. The AI found there is still missing something else we are carrying. So there's some other ancestral group that we have not yet identified. There is some thought that it's a, that the other group is a hybrid of the Neanderthal and Denilovians who went off and formed their own special group for a while and then came back and interbred with humans. But that's a little speculative still. It's mind-boggling. It's like a night in the museum because you get to see other species and you get to connect up, except it's not speculation anymore. It's not based on 19th century science. It's based on 21st century science. So now we can confirm exactly how evolution worked. This is different than before. Yeah, it's a great way to look and see a refinement of a story. We used to tell this very sort of simple story of human evolution. You know, one branch out of Africa, one migration, and now it turns out, no, it's much more complicated. There was an early exodus that spread widely and developed in these other groups. A later exodus of actual, our homo sapiens species that then for some while interbred with several of these other lines and various times and places and before they all went extinct. Well, it's thrilling. Thrilling. Not only our survival, but their demise, their extinction, as you said. So let's unpack some of that, though. You know, about six months ago, I went to a program offered by Kamehameha Schools and Ocean and across the street. And it was a sort of combined program in the in the facility that Kamehameha Schools has near the university. And there was a class and there were the people, a very diverse group of people, including some judges, believe it or not, who were in this class, what to learn artificial intelligence to get a smattering of it to see. And they had a, actually, a Japanese national gave the program and he got right into it. We all had computers and we all learned the basics of artificial intelligence. And, you know, if the one thing I carried away from that, it's a matter of comparing data, comparing images, mostly in that class and many, in many applications of AI. So you take, you know, you take a bunch of images you and you say that this is true about them or about one of them. And then you compare and you can see. It's like facial recognition. You can see the face because it knows the face. It's already done that and it requires a lot of computing power and it requires, you know, a fair amount of sophisticated programming. But fact is, you know, this is not all that complicated. It's the applications that are complicated. And that's what China is doing right now to apply it, you know, to various problems. So tell me how artificial intelligence works in the context you are describing as far as we can go anyway. Well, let me step back for a moment. Yeah, all human learning, virtually all human learning can really be said to be pattern recognition. I mean, that's sort of what we start out being able to do is, you know, as an infant, you're distinguishing one shape from another one color from another, you're beginning to be able to group orange objects against blue objects or triangles against squares. I mean, this kind of thing apples against oranges. You know, you're beginning to recognize there's people, there's men, there's women, there's boys, there's girls, right? There's bicycles, there's cars, there's motorcycles. These things all have their patterns you're recognizing. And so we learn the universe around. Right, exactly. And they've now taught computers to basically do that same kind of recognition, pattern recognition on very sophisticated levels. And they look around their universe is these monstrous data sets, these long strings of numbers, or letters, or gazillions of images. It's not only patterns in terms of graphics, it's patterns in terms of data in general. Yeah, it can be sequences of numbers are certainly patterings, right? Sequences of letters, I mean, words are patterns, you know, so it is it's all it's all sort of pattern recognition, and how much you can ask it to match things very precisely, or just sort of say your your goal is to get generally near near to it. And with that kind of thing, they now teach computers to teach themselves to learn all kinds of things. They now have computers that can play multiple games can play go and chess and other games and play them better than people play them. And they've taught themselves haven't been taught and programmed to play these they've actually played millions and millions of games against themselves and learn the patterns that every game they play, they learn right. And so what we have is recognition at a much higher level. We have learning, we have computers teaching each other. What we don't have is awareness. We're at the lip of that, I think it's not too far away. Right. But for now, it's it's very sophisticated learning about the world around us and teaching other machines to learn about the world. So what is machine learning? Is that what we're talking about? That is the other terms, machine learning, artificial intelligence are often you somewhat interchangeably people who are deeply in that field would probably argue those are two subtly different things. But to me, they're pretty much the same. Yeah, but yeah, it's it's a growing them and the applications as you say are enormous. We we just had people a few weeks ago from Ocean it on Ocean it has developed now a camera on a drone hooked up to artificial intelligence that can fly over a storm scene and tell whether telephone holes are upright or down and then report that back to the emergency crews. So they'll know where to move where to bring in how much equipment and how many crews so they can get communications restored quickly. Yeah. Yeah. And all it's based on I mean, fundamental level is a you have a you have a pattern of a telephone pole upright, you have a pattern of a telephone pole bound or partly down, and it's going to be able to tell that by comparing the patterns as they teach it initially with some clear examples and keep giving fuzzier examples and the computer teases them out until it understands what that difference is. And, you know, again, how to make a reasonable judgment on that. Yeah. So you may not get a perfect right result, but you get a reasonable judgment, which is no more or less than what you can expect from a human being on on the scene there. Right. And again, they use this now for radiological examining radiographs, x-rays and all I guess is that would be more accurate than a radiologist. Yes, because it's incredibly tedious work to look at image after image after image of these sort of gray scale images with little darker spots and lighter spots. Sure. It's mind bogglingly tedious. And but a machine doesn't get tired. And the machine says, does this match my image of a tumor? Yes, no, maybe. Yeah. Yeah. And run through it. Based on so many samples, more samples than the radiologists could ever have available to more samples and all the radiologists in the country could look at it in 100 years. Wow. Changes everything. Yeah, exactly. That's what this study did. So apply then the, you know, the notion of an artificial intelligence machine learning to what they did here to compare the human genome with the genomes of these other species. Well, okay, I haven't thought about this in real depth, but basically, it is like you had multiple sets of multi volume encyclopedias all stacked around and one huge set for current human beings, one huge set for Neanderthals and one huge set for Denizovians, huge up for Australopithecines, huge sets and then other sets of encyclopedias for migration routes, other sets for comparison of bone structure, other sets of encyclopedias for skull structure. And this machine essentially read every single word and every one of those sets addiction of encyclopedias and said, Oh, look at this, this piece of text here matches this piece of text here. I think these guys are pretty much more the same. And this doesn't quite match it, not quite as closely. So they're a little more distant from these guys. You come up with it, you come up with a judgment, right? It's the best thing all that they built a tree of life, basically a human evolutionary tree that that's probably considerably more accurate than anything we've had to date. Yes, let me go back for a minute, though. So what we're looking at is the bones, the composition of the bones. We're looking at the skull structure, because that's important in all this. And in fact, the MIT article had a picture of all the skulls of these related species, species related to human. And then you're looking, I mean, are we looking at DNA? Are they able to extract DNA from some of these, you know, prehistoric bones and skulls? Yes, the first Denizovian they found that they had one bone from the little finger and they used a piece of that bone to essentially get the DNA and show that she was of a different species. She wasn't Neanderthal, she was not, she was not Homo sapiens, she was something else. And that was from a little core they drilled out of a single little finger bone, basically. You know, I saw a something on PBS or CNN not too long ago on cable, about a teenage Mexican girl, or founded Mexico, because you can't say Mexican, that was, you know, from thousands of years ago, who was alone and being chased by either animals or humans into a cave. And she died in such a she fell and broke some bones when she fell and she died there on the spot. And luckily, you know, these scientists were able to get her bones relatively intact after all the years. And they went back and looked, and they did find DNA in old bones, they were able to do it. Right. And they were able to find out so much about this Mexican. I mean, this woman girl found in Mexico, her age, you know, her life history, what made her strong, what made her weak, how she spent her time. You know, that's the magic of science. You take the data, and you evaluate everything you have, and you come up with an analysis of all kinds of things that you wouldn't expect. Yes, I was just reading a fascinating analysis of a woman who studies Mayan skeletons and has developed this theory about the very sophisticated ritualistic sacrifice that Mayans did based on subtle markings on bones that she she has indicated a very highly sophisticated operator, basically slicing people's chests, you know, and being able to pull the still beating heart out out of the chest. Wow. But yeah, they can now find DNA or markers of DNA, very, very old stuff, and they're constantly pushing that boundary back further and further and further. Yeah, so I mean, but that's just one vector. So I go to the girl in Mexico, and I find some kind of DNA, you know, that much just a little bit. Who knows how it survived all these years, you know, because chemical processes of deterioration would would kill it normally, but you know, a confluence of events and we have it. And okay, then you look at it and you look at the you know, the biochemical composition of it at a molecular level, I guess. And now you now you have some data. And really, we stop there. And then we bring in the artificial intelligence, right? And you compare that with other DNA you found elsewhere. And now you start to get this is where the challenge really comes for the sciences. What How do you make something of this data? Right, you know, exactly. These are literally just strings of letters or strings of numbers, depending on how you want to look at it, just long, long, hundreds and hundreds and hundreds of thousands and millions of these strings letters together. And yeah, how do you start sorting through those? How do you break them up in meaningful patterns that sort of match our chromosomes? How do you then look at those and determine how much of it is how similar to another sample that you may have? What are you counting as similar? What are you counting as different? Yeah, it's really stunning stuff. And then of course, matching completely different kinds of data too. Sure. And one of the elements is carbon dating, which was so for the girl in Mexico. They want to find out when, when did she die? And it's not just looking at the physical specimens they were able to pull out of her. But also the carbon dating on how old those things were and lock it in. And then you can start really developing an evolutionary pattern, draw the chart. Right. Yeah. And now there's all kinds of other dating to they can look at decay of very relatively rare elements that decay in certain ways over even longer periods than carbon 14 decays and get extract time lines much further back now. Can we take a digression on that? And can you tell us how carbon dating works? So carbon dating works on the fact that there are a number of different isotopes of carbon. Most of the carbon around us is so called C 12 carbon 12 has 12 protons 12 neutrons or six six. There is so called carbon 13 and carbon 14 have extra neutrons in their nucleus. Prickly C 14 is a little bit unstable. It's it's a good molecule hangs to our good atom hangs together pretty well. But for a while it tends to like one of those neutrons pops off and disappears. And that happens very predictably statistically, you know, there's a huge variation. But you you look a lot of those atoms together. And what you find is, after about 5700 years, there's about half the amount of carbon 14 in any sample than there have been when you when you put it in there. Now, again, you're talking about, you know, when you're dealing with a little speck of materials, they're talking literally about trying to count. Yeah, count atoms, basically. So and so you can see carbon 14 can take you back, you know, some 1000s and 10s of 1000s of years. But after a while, there's not going to be enough of it left to do the counting. Yeah. But they're, let's say there are other radioactive elements that have much longer half lives on that that have half lives in the orders of 20,000 and 40,000 years. And so those now you can stretch things back hundreds and hundreds of 1000s of years in terms of doing timelines. So assuming you have enough material, and assuming you have, you know, the right equipment, which is expensive and major. How accurate a reading can you get on that? Can you get it to one year, 10 years, 100 years, what? I actually do not know off of using just one technology, but what what now happens is that typically science apply a lot of different analyses to a thing that they'll look at the kinds artifacts that are associated with a given skeleton. And there the archaeologists have built their own timelines of when these kinds artifacts were built. They'll look at pollen records. And again, the climate records will tell you that should be more or less on kinds of pollen at certain times. So they'll look at all this stuff together and they can sometimes get this stuff quite quite down to a relatively limited number of years. So you look at everything, right? And let me let me let me say that we're talking in the pronoun of a single person. But if you get machine learning to make this comparison of all these data vectors, so to speak, data can, you know, constituents, then you could probably do it better with with more data and get a better result, right? You couldn't do it without without machines now. I mean, realistically, even when I was in graduate school, a graduate student would spend their entire graduate career sequencing one part of one of the genes of 20,000 genes of the human genome. And now they can they can run off a whole genome of a person in a few minutes. And you don't need that student. No, no, you got machine that'll do it. The student has to be off doing something right more, more amazing. Yeah, at a higher level. But but this is the beauty of it. And you get all these multiple data sets and data sets that are very hard to compare. It is it's an apples and oranges kind of data. It's building the spreadsheet. It's building the fields. It's comparing the fields. Right. It's as they said in this Nature article, it's not just saying X and Y and here's a line. There's a thousand dimensions and you're making some squiggly pattern that connects these thousand dimensions in some reasonable way. And it's it's it's mind boggling complex. But the more we can get machines to do this, you know, the less mind boggling it is and the higher we can go on the thought change, but I want to unpack one more thing. You mentioned in that stream of vectors, geographical location, which I mean, you know, we know that humankind, at least the Homo sapiens species came out of the East Rift Valley in Africa and East Africa. And it went from there and they have maps you can open any any book and find not any book, but many books. Right. You can find maps of how, you know, the species traveled over, I don't know, 120,000 years or something like that all around the world and populated the same genome everywhere. And that's really interesting because because the evolution of the species depended in substantial part on the environment. humankind has a relationship with the environment. It tests us and the survival of the fittest is based against the survival of the fittest in that particular environment. Right. Absolutely. So tell me how they get the data, the geographical data you were talking about as an element of this analysis. So again, different groups of scientists have been looking at different aspects. So there are scientists who study the ecology of different regions of the earth and its paleo ecology is looking back at the ecology a long, long time ago. So they will, for instance, drill cores of mud out of lakes and they will look at the layers of mud and turns out layers of mud are very much like tree rings, particularly in temperate climates. Different things fall into different seasons. They pack down in very neat, orderly sequences. And again, by getting into one of those layers and starting to pull apart, what's in that layer? What are the shapes and tines and patterns of the pollen grains? Are these grass pollens? Are they tree pollens? You know, you can begin to infer, is it warmer? Was it colder? Was it wetter? Was it drier? Was there more thing, you know, more here? You get little parts of insect exoskeletons in that. And again, from those they can tell, was this a dry land or was it a swamp? So again, you've got that kind of data that can tell you a great deal about the environment. So once you know about the environment, you know about the challenges that the species had in that environment. Right. And you can get a beat on how that might have changed the evolution over thousands of years. Right. If you were stronger, taller, if you were capable of dealing with cold weather or warm or disease and all that, and so that would be a function of your environment. Exactly. And gee, God, you know, that actually opens the whole door to climate change and how climate change might affect evolution. But let's not go there yet. Let's have a break, okay? Okay. That's Ethan Allen. He's our likable scientist here on likable science. And he's our chief scientist and he knows a lot of stuff. And, you know, what I'd like to do when we come back from this break, Ethan, is examine what, what in fact we have learned about the greater processes of evolution from then till now, and what, if anything, we can infer and how, logically, scientifically, we can infer it going forward on evolution to follow. Oh, exciting. I'll see you right back. Hey, loha. My name is Andrew Lanning. I'm the host of Security Matters Hawaii, airing every Wednesday here on Think Tech Hawaii, live from the studios. I'll bring you guests. I'll bring you information about the things in security that matter to keeping you safe, your co-workers safe, your family safe, to keep our community safe. We want to teach you about those things in our industry that, you know, may be a little outside of your experience. So please join me because Security Matters, aloha. Okay, we're back with Ethan Allen, our likable scientist here on likable science. So Ethan, you know, trying to make sense out of this, it's a big question. I don't know if it was covered, I really wonder that it was covered in the MIT article, but what, what have we learned? What can we learn? Why should we care, really? Because, you know, these sound like profound scientific experiments and, and findings, but at a very scientific database level, what, what larger picture can we get about the evolution of the species or all species? The evolution of all species on the planet from learning about the connection of the homo sapien and these various other contributing species? Well, I mean, for one, you can, you can look at this and say, and ask a simple question. So why is it that we have retained certain neanderthal genes? What, what value have those genes brought to our species? What good did they do to people who possess them that enabled those people to reproduce and keep, keep that gene in, in the germ line, you know, and what, what other genes didn't make it? What did those genes do? And yes, they, they are now actually just beginning that line of investigation, trying to find out what that neanderthal DNA does when those genes get turned on in our development. And some of it has to do with aging processes, they already know that. Some of it may have to do with, with some brain development. This is there just sort of scratching the surface of this whole, this whole area now. But it's the kind of thing that, yes, you can learn a whole lot about because genes are sort of expensive. They're biochemically costly to make and to maintain, and if they're not good, they rapidly get dumped out, that the people who possess them don't reproduce. That's, that's ultimate marker of a bad gene, right? Yeah, yeah. And conversely, the marker of a good gene is, you've produced a lot, lots of descendants, right? Yeah, yeah. And so, assumedly, the genome that we have today is, represents pretty good genes, you know, that it's, it's scavenged from other species as it were, which is really an interesting idea if you think about it. We don't often think of ourselves as stealing genes or borrowing genes from other species, but we, we now know we have done this in the past. Yeah. The second part of your question, of course, is much tougher, is where, where does this take us? Because we have now a, you know, biological evolution that is proceeding along and has always proceeded to perfect the species within the environment. Right. And if the environment is changing, that makes it all the more complicated. So it's biological evolution is never going to have, is never going to hit perfection, right? Because the environment is changing, but it's always going to track the environment in some sense and try to match it. Yeah. But now we have an environment, a physical environment that's changing much more rapidly than it ever has before. We have a cultural environment and we have a technological environment that all of which can influence the genes. And we're now able to go back and start actually manipulating the genes due to our technology and our culture, right? And literally physically manipulating the genes. We talked a while ago about the CRISPR babies that have been born, that the genetically modified human beings that have been born. Yeah. So, and so your question about where this might take us, I mean that. It's really unanswerable. Yeah, yeah. There are too many variables even from machine learning, I think. Because one is we, you know, we don't know the effect of climate change and that's a study all in itself. Right. So you can't throw any assumptions in. You can only say it's going to change and dramatically and more quickly than before. Secondly, the species itself has the technology to change the genes and not only the genes in one generation, but the genes going forward. We can modify ourselves and our progeny. That's pretty scary. We can do a good job or maybe not. And that's unpredictable also. Right. So I was just reading a fascinating article about honey bees. And honey bees in hot weather have to stand in the entrance to the hive and fan to get cool air going through. But it turns out they're not doing this sort of randomly. They are standing in the hottest places of the entry of a narrow entryway and fanning there and not in the cooler places and purposely sort of setting up a good draft flow. Furthermore, different bees have different thermostats for that that behavior so that as a colony they can even cooler whether you'll still get some bees doing it hotter whether you'll get different sets and maybe more bees doing it in a very hot weather maybe a whole bunch of bees will be doing this. So yes in some sense as the temperature global temperature rises will we decide at some point we better start manipulating our genes to be more thoroughly tolerant or tougher to be able to withstand flooding or you know or whatever you know traits we want to pick. And that gets to be a very interesting kind of scenario right. Challenging and scary. Yeah yeah I mean it's one thing to go and fix a genetic problem. Somebody has a genetic defect that you can go in and fix early on and let the live a relatively normal life. That's grand and virtually no one will argue that's that's allowable. But it's not that simple. Right now once yeah and who gets to have this technology how would we decide who to use it with. Maybe you have machine learning artificial intelligence can look into the future and crank all these variables together somehow and make some speculative you know expectations and guesses and whatnot and tell us about the future of the of the human race. I don't think I want to know right now. It's too scary. It is it's a little frightening to look at. Thank you Ethan. Yes indeed. Great to talk to you as always. As always.