Future of Science and Technology Q&A (May 10, 2024)

Foreign. Welcome to another episode of Q and A about future of science and technology. I gather we now have 100,000 subscribers. So thanks for joining us. I don't know whether that's a big number or small number. I don't pay attention too much attention to these things, but sounds like a decent number. So thank you all for joining us. Let's see, we have all kinds of questions here. There's a question from Kevin here. Have you used LLMs to analyze genetic language? I haven't done too much myself on that. I think the question is if you've got a piece of DNA, to what extent can you predict what's going to come next in that piece of DNA? Because the way an LLM is trained is there's a trillion words from the web and other places and what one's doing is saying to the neural net, here's a bit some text. Now let's mask out the next piece of text. Can you LLM learn to correctly predict what that masked piece of text will be like? So you can do the same thing for genomes and you can say given what's been there in the genome so far, what can you predict about what comes next? And that's an interesting question. To what extent? That's a thing that kind of, is kind of predictable. So some things are very easily predictable. For example, the codons, the groups of three base pairs that encode one of the 20amino acids. There are definite sequences of codons that occur and that don't occur. And there are things that sort of redundancy in the genome in that only certain kind of codons occur because there are 64 possible combinations of the four base pairs or triples of four base pairs. But there are only 20amino acids that should be encoded. And so some combinations of base pairs encode the same amino acid and so on. Some aren't used. Some represent things like stop codons when there is a transcription going on of a piece of the genome that says, okay, protein has ended now, or at least that section of the protein has ended now, stop transcribing with the RNA and cut it off. And that will be a piece of the protein that will then be assembled, but that will encode what then the ribosome will make into a piece of protein and so on. So an interesting question is to what extent the base pair sequence of our DNA, for example, is random and to what extent it isn't. So to what extent. If it was all completely random, then we wouldn't be able then the LLM Wouldn't be able to do any meaningful prediction. But so this really relates to. If we look at the proteins that exist in biological organisms, are they kind of just randomly chosen or are they chosen in a very special way to achieve all the things that biological organisms achieve? You know, about, oh gosh, how long was it? Maybe 25 years ago or something. I took a look at all the proteins in the then protein data bank and I asked the question, if you just looked kind of statistically at the amino acids and ultimately the base pairs that occur for those proteins, were they random or was there something very statistically special about the particular proteins that get used in biological organisms? At the time, I could find no statistical regularities. There were cases where some organisms have just a lot more A's and T's than C's and G's. Like I think crabs have more A's and T's than C's and G's because they kind of live in a place where it's easier to make the as and Ts than the Cs and GS, I think. But beyond those very simple regularities, I wasn't able to see any kind of statistical non randomness in the genome sequences that were actually used by biology versus the ones that were not. So there's sort of this question of, well, does any sequence of base pairs, or at least any sequence of amino acids, the proteins remember are these kind of chains of amino acids. So they're just, they're sort of. They're like polymers except that they have different elements. They're sort of knitting together on a backbone, these succession of different amino acids. And so a question would be, well, is any sequence of amino acids a reasonable protein? People don't completely know the answer to that. Some people say, oh, most proteins that you make with random amino acids won't fold. Because what happens with proteins is they start off with this sort of sequence of amino acids on them and then they fold up because of forces of attraction and repulsion between different parts of that chain. And they fold up in ways that we use for different purposes. Like for example, the. I think that the things. What is it? The proteins that are used in muscles make a kind of triple helix of elements in their backbones and so on. And the protein we use for like hemoglobin, the protein we use in our red blood cells makes this little cage that just fits an iron atom that we use to store oxygen as it goes around in our blood. And you can go through all the different. There are about 30 to 50,000 proteins that we use to make up us humans. And you can go through and each different one of them we use for a different purpose. And it sort of achieves that purpose by being a certain shape and having certain ways of binding to other proteins and so on. So there's the question of, well, if you just make a random protein, what, what will it do and will it be useful? So one question is, will it fold in a repeatable way? Because folding it's proteins, in order to achieve their function presumably need to fold in a definite way. There are some proteins that, well, there's some things that happen where there are misfoldings of proteins and actually there's a whole mechanism in cells to sort of get rid of misfolded proteins. Cells have some heuristics to basically be able to destroy misfolded proteins. And there's some thoughts that things like Alzheimer's disease may have some relationship to misfolding of proteins. And when there were, if you have these prion diseases which are the result of kind of forced misfolding of proteins. So when proteins misfold, they don't perform the functions. They're not the right shape and they don't perform the functions that they normally perform. But when you say, well, does a protein even quotes fold up? Well, any protein will form some configuration. Question of how rigid that configuration is is not clear. And knowing how rigid these configurations are is really hard because traditionally the way you know, well, what configuration does a protein form is to make a crystal out of that protein. That's a very difficult procedure, making crystals out of these very complicated molecules because they kind of have to all fit together in a regular way. And, and if they're weird shapes, that's a hard thing to do. Once you've got a crystal, you can use X ray crystallography to find the structure of the protein. I think that was first done for myoglobin. Oh gosh, in the 1950s, I think, and famously it was done for DNA in 1953 of figuring out that it made a double helix by having essentially a crystal of DNA and using X ray crystallography, which, which is this thing that gives you kind of a holographic image more or less of sort of transform of the actual form of the way the atoms are arranged inside the crystal. And then you have to kind of decode that and that's non trivial to do. And that's a whole story of there are often many different solutions and you have to say, well, what's the plausible one? And what for Example Crick and Watson did in sort of figuring out the structure of DNA was to say, yes, there's a plausible solution which is this double helix, and then it has all this functionality and so on. But any case. So, you know, normally when a protein, when you know what shape is a protein, you're kind of completely sort of locking it into this crystal structure, which may not be the shape that the protein has when it's flopping around in water, for example, in a natural biological organism, there's a newer technique called cryo em, cryogenic electron microscopy, which kind of captures proteins that's more kind of like just freeze it in some state. But it's still not clear what the actual protein does when it's kind of in its natural habitat flopping around. It's certainly true that many float proteins have pieces that definitely move. So this question of does it fold? Is really a bit of an undefined question because every protein will form some configuration. Whether that configuration is very floppy, maybe that's a bad thing. If it's very floppy, maybe it's not. I don't think we really know yet what, how to think about that. But this question of if then if a random protein, sort of randomly chosen sequence of amino acids will make a protein that does something, people might say, people say, oh, it's only a few percent of proteins that fold. I'm not sure if that's true. I'm not sure we really know that. And it's not even clear what it means. They'll all do something. The question of whether they do something that sort of quotes rigid enough to be useful. We don't even know that rigid enough is an important criterion. But any case, the thing that has. So then the question is, well, you have these different proteins and what's presumably important is how these proteins interact with each other because that's how they're going to form all the different tissues that form an organism like us. And the question is, is it the case that it's like you could imagine you're just sort of throwing a bunch of random shaped Lego bricks and then the way you make an organism is by fitting together those LEGO bricks in some kind of creative way. Given the bricks you have, that's choice number one. Choice number two is given the organism you're trying to make, carefully figure out what bricks you should use to make that organism. I don't think we really know yet which of those things is the most important in biology. I would tend to think that there is a lot of kind of given that biology invented this protein. Then that protein gets recruited for a bunch of different purposes in the biological organism, more so than this protein is specifically set up to have this particular biological purpose. Now it's worth remembering that there are places where there is truly a random sequence, proteins or base pairs that get used, for example, in the immune system. One is routinely kind of the immune system, the adaptive immune system in mammals and so on, works by saying, let me make an antibody that is the right shape to sort of fit into the nasty antigens that are coming in to attack us. And what happens is there are, I don't know, 100 billion possibilities of this sequence of, I think it's 6, well, 6 to 12amino acids that form the so called hypervariable regions of antibodies. And they're kind of, that is forming little pieces of protein that is kind of random. And the whole point is that the randomness of that sequence is good enough that somehow it forms. As you look at all those possible random proteins, they are all kinds of different shapes. And as you scan across all those shapes, they form kind of the, they form sort of enough of, they fill up enough of shape space that they can somehow universally recognize the different antigens that we're exposed to. And I think they're like 100 billion combinatorial possibilities. And in the early stages of our life as humans, for example, before our sort of teenage years, particularly in the thymus gland that's somewhere right around here, the one is kind of processing these. Actually the thymus gland particularly deals with T cells rather than specifically antibodies. But it's the same general idea that one's sort of processing these immune system cells to say which ones of these are sort of shapes that will bind to things that are proteins that we already have in us, and which ones of these are not self proteins, are intruders, so to speak. And I think we whittle down those billions of possibilities to maybe 100 million kind of individual sort of individual sort of choices of antibodies and so on, and T cell receptors, which is another part of this mechanism that can recognize intruders. And so when an intruder comes in, it's like, okay, if the intruder matches one of those hundred million, then the one that it starts to match, it's starts to replicate very rapidly. And so you end up getting sort of a whole sort of collection of antibodies or T cells that are sort of going to attack the intruder based on its shape. So that's a case where the random sequence, where you truly get a random sequence of base pairs. Now, I mean, it's an interesting question. For example, of those 100 million sort of sequences that are sort of the result of the pruning from our early self, non self discrimination early in our lives, what is the distribution in shape space of those hundred million things that get left over? Different people will have different distributions of those things in shape space. How does one characterize that come and sort of make a science of how that works? And what can LLMs, for example, say about things like. Well, I don't know whether they can say anything about that. But this whole question about can you predict what will be in these sequences that exist in our actual kind of germline cells, in the actual cells that in the actual DNA that is sort of replicated across the cells in our body, can we predict that is really this question of how random are the proteins that get used by biology? I mean, there are some more detailed things like things about sort of where do genes begin and end on chromosomes or on the DNA. And that's been a long time thing that's sort of used sort of elementary machine learning to try and predict those kind of start and stop places and where transcription will start and stop and so on. I mean, there's a lot of other adventures there. It's clear that there's a lot of sort of post independent of the actual DNA. There are things about sort of modifications to the DNA that get made in different kinds of cells, modifications to the way that it's transcribed in different kinds of cells. You know, it's becoming increasingly clear that the different types of cells, maybe a thousand types of cells that we have in our bodies, that they all have the same DNA in terms of its original base pair sequence. But that base pair sequence can be annotated by additional atoms that are sort of hanging around the DNA and other kinds of sort of processes. And it's becoming increasingly clear that there's sort of a limited number of kind of transcription factors that can modify these things and will determine is it going to be a brain cell or a skin cell or a lung cell or whatever else. And that's kind of the sort of the big picture of what goes on is we all start from these kind of pluripotent, you know, can do anything, stem cells, that's what, you know, fertilized egg cells are that. And then those cells differentiate into all these different kinds of cells that exist and make up, for example, humans. I think there's like 14 levels of sort of differentiation to get to different kinds of cell types of that order. I don't think anybody's ever made the full lineage chart of how that works for humans. It's been made for this thing called C. Elegans, a nematode, a little tiny worm which only has about 2,000 cells in it altogether. The full lineage chart is known in that case and has been known for 25 years or so. But for us humans, I don't think it's known. But the thing that then is sort of the big new technology of last, I don't know, 15 years, 20 years, 15 years, actually, is this possibility of taking any cell and being able to revert it to being a stem cell, to use particular factors to reprogram the cell so that this cell that started as a skin cell, for example, can be reprogrammed back to being a stem cell from which other cells can be differentiated. And there's a bunch of therapies that are starting to come online, some starting to be successful for taking those cells that have been reverted to stem cells. So you can do that from. Okay, so you can take a generic line of cells that sort of is some standard lab collection of human cells, and you can make stem cells from those, and then you can take those stem cells and differentiate them into different kinds of cell types and. And you can use those for various kinds of therapies to replace cells that one isn't making oneself or whatever else. The problem with that is that if you take a generic cell line, the details of the proteins that will exist in that cell are not the same as the details of the proteins that you have identified in your early life as being self proteins for you. So your immune system will say, that's an intruder, let me attack it. And you have to use immunosuppressive drugs to prevent your immune system for going and attacking the things that are coming in to replace cells of yours. So the thing that's sort of the coming attraction in this area is being able to take your own stem cells that you get by taking a cell, just some cell from sort of almost any kind of cell from you, and then reverting it, reprogramming it back to being a stem cell, and then going forwards again and adding sort of transcription factors, either by letting the cells sort of replicate in an environment that corresponds to the environment that it would encounter in the body that would cause it to turn into this or that kind of cell, or maybe by just sort of externally injecting these kind of different factors to make it go into. To make it directly be a cell of a particular type. And Then using that cell of that type to replace cells that you want to replace for sort of therapy purposes. And that's kind of the story of, I think an increasingly successful probably direction in sort of medicine will be this kind of stem cell based therapy. And sort of, it's still a bit of a mystery when you do this reprogramming to get back from a, A, you know, some other cell in you to a stem cell of your stem cell, so to speak. Sometimes there seems to be some memory of what the previous type of cell that the thing was reverted from actually was, by the way. I mean, one of the places where people are sort of excited for purposes of longevity and things like this in dealing with stem cells is in making things like replace some of your immune system cells, which kind of as we get older, somehow the immune system cells get more tired out and sort of the young immune system is more energetic. And that's a good thing for fighting kind of infectious diseases and also for fighting cancer. Because often with cancer, the cells, particularly T cells, will be perfectly aware that there's something wrong and they'll all sort of go and start attacking the cancer. But there just aren't either enough of them or they're kind of too old and tired and maybe their mitochondria don't work well enough or something like this and they don't succeed in fighting it off. And if you could use sort of your own, sort of a new version, rebooted versions of your own immune system cells, you would have a better chance to fight these things off. So I kind of think that's one of the sort of coming attractions in, in this area. Let's see. I wonder if there are more questions about biology and then I can address those for a minute. And then. Okay, there's a, okay, there's a question comment from Jamie here. Will bacteria become more resistant and stronger against humans in the future or will humans become stronger? You know, one of the challenges is antibiotic resistant bacteria. As we make more use of antibiotics, we the, in order to survive, bacteria have to become resistant to antibiotics. And you can actually take, you can take a plate where you're sort of growing bacteria and you can literally see you put a little bit of antibiotic on the bacteria on the plate and the bacteria around there won't make it. And then sort of new and different bacteria will evolve. The random mutations that happen in the bacteria, the ones that can make it past the antibiotic will make it pass. The other ones, the bad mutations will die out. And so that is a place where you can sort of see biological evolution happening in sort of almost, well, real time. These plates, I think they grow pretty slowly. It takes days for them to grow. But you can see kind of the way that evolution is happening as mutations allow some bacteria to make progress and others not. But it is sort of claimed that the use of antibiotics, the broad use of antibiotics for people has caused there to be more antibiotic resistant bacteria to have evolved in our environment. I kind of sometimes think that people go overboard in saying, well, even if you've obviously got some bacterial infection that comes after a viral infection that you've had, don't take antibiotics because it'll sort of increase the amount of antibiotic resistant bacteria. And that is an issue. But I'm not sure that it applies in all cases. And I suspect there are ways that you can make it be less true that the antibiotics that you take yourself, that, you know, the sort of a question of, do you infect people with the bacteria that have evolved to infect you after they were sort of squashed by the antibiotic and so on? And that's always, there's always this issue of, you know, you get an antibiotic and it's kind of like a 10 day course of antibiotics and it's like, keep taking it until the end, because otherwise there'll be these straggling bacteria that have evolved to avoid that antibiotic and that will be becoming antibiotic resistant bacteria that might bite you and they might also bite other people because they get sort of released into the environment. But anyway, so one of the challenges is that as time goes on, the bacteria evolve, they evolve to evade our antibiotics. We have to use new and different methods to attack the bacteria. I know one technique that people are experimenting with is to use bacteriophages to attack bacteria. So bacteriophages are viruses that are viruses for bacteria. So viruses are. They're little packaged up pieces of DNA or RNA and they get into cells and they basically take over cells, some parts of the machinery of the cell, and say instead of making more of the proteins that you should be making based on the type of cell you are and so on, you make more copies of the virus and then that's. And so it's been sort of taken over a cell. Well, there are viruses that will do that to bacteria, Bacteriophages. And at various times, people have used bacteriophages to do all kinds of cool things. For example, the CRISPR technique for gene editing came from a bacteriophage so bacteriophages have invented lots of different mechanisms over time, but particularly if you've got a virus that attacks and kills a particular kind of bacteria and you are infected with bacteria, one thing you can do is send in the bacteriophages. The challenge is to find which bacteriophage you want to attack a particular bacteria. And by the way, bacteriophages, you take a handful of soil, it'll be full with a zillion different kinds of bacteriophages, including probably ones that have never been specifically seen before. But the challenge is, can you find the one that will go against a particular bacterium that you're trying to attack? But that looks like something that may be a successful direction to go in. So. And there are other. I mean, the fundamental principle of antibiotics, like penicillin, for example, the first antibiotic, its principle is that the cell walls of bacteria have a different structure than the cell walls of organisms like us, the eukaryotic organisms like us, as opposed to prokaryotic organisms that are bacteria. And so what penicillin does is to sort of explode the cell walls of bacteria, but leave the cell walls of our cells intact. And different kinds of antibiotics work in different ways. I think the mycin family works in by affecting part of the cell replication mechanism of bacteria. And different, every so often, sort of a new scheme for distinguishing the way bacteria work from the ways that our human cells work gets found, and that provides sort of a new potential target for antibiotics. Um, so let's see, It's a question from pruts. Could you just build specialized cells from scratch using printed DNA? Well, human DNA is 6 billion base pairs right now. The kind of, the. The current sort of state of the art, if you just make a sequence of DNA, it's like 300 base pairs is what you can easily make. And there are new technologies coming online which will get to, like, 1,000 base pairs. So you can get to these oligonucleotide sequences, these short few nuclei, nucleotide sequences. You can just print them out of a machine, and you can also take those sequences, and by the time you have a reasonably long sequence, you can have its ends line up so that it will sort of assemble itself into a longer sequence. But it's a pretty long way to get to a sequence that's of the length that you would need in a human cell. Now, you can do other things. You can edit the DNA of an existing human cell. That's what the kind of CRISPR mechanism is all about. In its Original form that was just able to go in and find based on the sequences that existed around a particular place, around looking at the flanking sequences and then saying, okay, if it's a, G, C, T, D, ca, whatever it is on one side and something else on the other side, match that and then snip out one base pair in the middle. And now there are starting to be more sophisticated versions of that where you can insert little sequences and things like this. But it's kind of slow going. Editing a big existing base pair sequence using those techniques. It'll probably get faster over time, but that's sort of the approach there. Now, in terms of the rest of the cell, I think we're in pretty primitive state in terms of reproducing the other structures in the rest of the cell. I mean, for example, the cell membrane, famously the MRNA Covid vaccines, were embedded in lipid nanoparticles, which are kind of versions of sort of proto cells, the cell membrane of a cell in eukaryotic cells, lipid bilayer, which is particular molecules that sort of form this coating around the cell. And sort of the lipid nanoparticle is kind of a junior version of that, intended to encapsulate, well, pieces of messenger RNA in the case of these vaccines. But so, you know, finding something which sort of imitates the casing of a cell has been somewhat possible, but not quite there yet. But the full kind of structure of all the organelles in a cell, that's not something you can readily print yet. I mean, the state of the art in printing and actually the people I know who've been, who've been doing this, trying to print sort of human organs and particularly things like lungs and the sort of the critical lengths there are on the order of two microns, because that's kind of the length. Two microns, two millionths of a meter is kind of the distance of tissue that is short enough that oxygen can diffuse, oxygen atoms can diffuse through that. And so you can get kind of from oxygen in your blood to oxygen that can sort of supply, kind of can get to different tissues so that they can generate energy and be alive and do all those good things. But the state of the art is like 2 micron printing size, and maybe it'll go down to 1 micron. But that's very big compared to the actual structure of organelles within a cell. That's sort of the matrix into which things like lung cells can be put. Kind of the framework, let's say collagen and so on. Framework that sort of makes the scaffolding to build organs. That's the length, scale of those kinds of things. And that is now just now becoming kind of 3D printable. And then the typical way you make one of these organs is you build the scaffolding and then you sort of pour into that a bunch of stem cells that you differentiate into the actual cell types that would populate that scaffold in the case of actual organs that we have. So there's a question from Kalka asking, can machines replace organs? Well, I think the first thing one will see and it'll be different for different organs. I mean, there's been lots of work done over the past 50 years on artificial hearts. The main problem with artificial hearts is that blood coagulates and it's really hard to get blood to flow. Blood is a complicated fluid that if it ever sort of gets stagnant, it will coagulate, it'll form a clot, which is sort of disastrous. So you have to kind of keep it steadily flowing. And our actual circulatory system has very flexible walls for our veins and arteries and so on and in the operation of the heart. And that is sort of set up to keep this really rather complicated fluid. Because a lot of our blood is in red blood cells, which are these kind of donut shaped. Well, they don't quite have a hole in the middle, but they have an indentation in the middle. Things that have a certain amount of flexibility. And as they move through our larger blood vessels and then going down to our capillaries, they arrange themselves in different ways relative to the flow. And it's pretty complicated. And. But if the flow ever stops, they clot. And that's, that's sort of one of the big challenges of artificial hearts. Even though the fundamental operation of the heart is really just pumping blood around. So in a sense that's a simpler case. When you look at other organs. There are kidneys, liver, things like this. There are sort of, those are things that have various kinds of chemical features. And yes, one can absolutely imagine having those things be set up so that they're manufactured, so to speak. And the best case is that you kind of make the scaffolding with 3D printing and then you feed in stem cells that come from the actual person that that organ is being used in. And then that person's immune system will say, oh yeah, these are all fine. These are cells that are just like me. And we'll be sort of happy with that. Now, in terms of making purely kind of machine like versions of different, different organs. I mean, most organs, you know, what do they do? Like, the heart is sort of a mechanical thing. It's just pumping blood around. But things like the kidneys are filtering blood, and that's a much more chemical kind of thing. The liver is a big source of lots of chemicals. The various glands we have are sources of chemicals and hormones mostly. And I think the things that typically happen are those glands, for example, are sensing chemical levels. So they have sort of chemical sensors and then they have chemical production capabilities. So the question of whether that can be done kind of machine like, whether that should be done kind of biology, like, is an interesting question. I kind of think that the coming sort of direction of nanotechnology is really about. Well, in electronics it's all about moving electrons around through wires and making them sort of do things that correspond to computations and so on. In nanotechnology, the interesting thing is moving atoms around and molecules around. And that's just what you need for biology. Biology is an example of sort of successful nanotechnology where you're doing things with atoms and molecules and so on. And so one can imagine a time when that's done sort of machine like, where it's engineered, how that stuff works. I think biological evolution. I just wrote something about this. Actually. Biological evolution tends to make very ornate solutions to problems of how to do things. They work, and sometimes you look at them and they're very clever in the way that they work, but they're not as direct as kind of what a sort of human engineered solution would look like. And so one can imagine there's sort of direct ways to achieve many of the things that different organs in the body do. I mean, the big one, I suppose, is the brain. And certainly in the end, there's kind of the possibility of kind of electrically connecting to the brain and doing things. And certainly sort of the first places where that's happening are various sort of in brain connections and various spinal cord kinds of things where you're taking the. The nerves in the spinal cord which produce electrical signals, and you're kind of decoding those electrical signals to run a machine that can help you walk if you've damaged your spinal cord or something like this. Let's see. Question from Doug. Several years ago, I read about research into DNA being used as logic case. Is this still an active area of research or has it been dismissed as not useful? The number one thing, I was sort of a little bit involved in the initiation of that direction a very, very, very long time ago. Although I didn't think of using DNA, I thought of using more proteins and things like this. But DNA is a very stable molecule. DNA hangs around for ages, millions of years, sometimes rna for example, and many proteins that we have are much less stable as molecules. So DNA is good, but because it's nice and stable and you don't have to worry about it sort of falling apart. But then the question is, well, what can you do with DNA? Well, the big thing that DNA does is single stranded DNA. DNA usually comes in this double helix with two strands, two complementary strands. So you know, A goes with T, C goes with G at every place on the DNA strand. But if you take it apart as a single stranded piece of DNA, it will be looking for its pair, so to speak. And what you can imagine doing is using that, I'm looking for my pair as a way to kind of match up things like tiles made of DNA. And so there was an effort in the 1990s, yeah. And then the early 2000s, to kind of see whether you could sort of make compute things by having sort of tiles of DNA arrange themselves according to particular tiling rules. And you can think of many kinds of computational systems, whether it's a Turing machine, a cellular automaton, something like that, as saying, you've got this, let's say you've got a line of squares and then you're going to say, well, what's the next line of squares? Sort of going down the page going to be like, you can think about that as like a tiling problem where you're saying we're going to match up the correct sort of tiles with the things above to get a computation done. The problem is it's really slow and the other problem is it's not perfectly accurate. So now that's the bad news. The good news is you can have a trillion molecules all doing the same kind of thing. So potentially you have sort of an advantage in that direction. I don't think this has gone terribly far. And there was a sort of an enthusiasm about a decade ago for DNA storage where the idea would be, well, I said DNA was a very stable molecule. You know, can you take sort of your, your long term storage and instead of doing it in magnetic media or something like this, and discs, tapes, whatever else, magnetic tapes, magnetic disks, instead do it on DNA molecules. Could you basically just sort of put out your, your encyclopedia on a DNA molecule and just have it hang out on a DNA molecule? And by the way, you can have sort of distributed storage rather easily by getting that DNA Molecule replicated. And so you can have, you know. Well, first there was one copy of the encyclopedia, then there are a trillion copies of the encyclopedia because that DNA got replicated in lots of places. So it's kind of like a potentially good way of doing distributed storage. That's sort of the good news. The bad news is a pain to work with DNA, and it's kind of very slow to read things off. Now, maybe that will change because there are a bunch of new, both DNA synthesis technologies and DNA reading technologies that are coming online. There are several sort of methods right now for reading DNA that go at different speeds with different accuracies. And the bottom line is you see these machines that are sort of the size of refrigerators that just chug along over the course of a few hours, for example, and read big chunks of DNA. That's kind of the. The practical state of the art, so to speak. And there are increasingly. I always find it really bizarre and amusing that you see things that generate nucleotide sequences, pieces of DNA, and literally they will have these bottles that are labeled A G, C, T for the different base pairs, bases. And, you know, it's using those up just like you have a printer, a color printer that has, you know, cyan, magenta, yellow and black, you know, pieces of toner. You're basically building life straight from the molecules in these bottles, which I think is kind of, kind of notable. But I don't think the DNA storage and DNA logic neither has been booming in recent times. But maybe that will change as there's more streamlined methods for synthesizing and reading DNA. Let's see, Let's see. Elsie asks, to what extent do you think we'll attain some kind of universality in the reprogramming of live biological systems on par with current software systems running on silicon substrates? So the design problem has been traditionally much more difficult for biology than it is for programs. In a sense, we kind of have a fairly simple model of computation that ultimately we implement in the machine code of current computers. And based on that, one is sort of building up the programs one wants to run. So now, what's the analogous thing in biology? In a sense, the machine code of biology is significantly lower level than the machine code of typical computers. So it's an interesting question, and people have talked about this for decades, to build up kind of blocks which will be useful building blocks for sort of building the functions that you need in life, and then sort of putting these together in some sort of construction kit to be able to program life, to Program the creation of organisms in the same kind of way that we, we create programs in software that hasn't so far proved very successful. The idea of sort of let's just make the sequence for a particular organism to do a particular thing. There are cases where you can do that. You can say, let's add, oh, like somebody had just told me, you can buy these glow in the dark flowers or they're about to become available. So I just ordered some today which are kind of using the same proteins that exist in jellyfish that make them glow in the dark. But you can put those proteins into plants and so you can have glow in the dark plants. I don't know how well this is going to work, but it's just a cool thing that feels like a piece of biological programming and modernity, so to speak. But in any case, you can expect to, you can insert genes from other organisms. Like this is a thing that's done all the time in yeast, for example, and in E. Coli. You're inserting genes that make useful human proteins, like insulin, for example. And you can do that by having. It's really a bunch of bacteria that are making, that are just that think they're making proteins for themselves. But actually you inserted a gene that makes proteins that are useful for us humans. And I think that kind of sort of reprogramming of biological organisms has been done for 30, 40 years now. And it's kind of routine at this point. The question of can you from scratch sort of make an organism? There have been a bunch of efforts to find the minimal bacterium, the minimum organism that is capable of replicating itself and so on. And there's some evidence that, you know, one's close to at least one, one branch of the minimum organism. And how big is it? 15,000 base pairs, maybe I may be misremembering that, you know, the COVID virus is 29,000 base pairs. And. But I think sort of what size is the minimum organism? And that's something where that was mostly sort of its pieces were all got from existing organisms. It wasn't sort of designed from scratch. Now, you know, the idea of can we make an organism that wasn't, you know, that where we just sort of invent it from scratch. I mean, there's a, there's a sort of famous scene in the original Blade Runner movie where, you know, there's sort of a biology hacker who has all these little, little special purpose organisms that open doors and, and you know, do. Do particular functions. It's kind of like that's a. That's a vision of how things might work. I remember, oh, back probably 40 years ago, a person who many people saw as some kind of visionary. I wasn't. Wasn't my favorite person a physicist I knew who was fond of making sort of kind of outrageous statements. And his statement was, you know, oh, all this stuff with software, that's not going to last long. It's all going to be biological. Everything is going to be sort of biological. And it's like, let's. The, the. The sort of. The story ended up being, let's grow our houses. Instead of having a house where you have construction materials that are manufactured, you say, let's just have the right sort of organism to grow our tree house, so to speak. Literally a tree that makes a house kind of thing. That hasn't panned out, at least yet. I kind of think it's a little bit disappointing for us if that does pan out, because it's kind of like biology, which has evolved over the last 3 billion years or something, to be the way it is on this planet, to say, let's just leverage that rather than independently inventing sort of engineering technology to achieve the things we want to achieve. But, you know, it's been a tried and true things. We use wood as a material for lots of purposes, and wood is not something we invented. Wood is something that, you know, comes from biology, just like we eat lots of stuff that comes from biology. Now, we could eat sort of disembodied amino acids, disembodied components of proteins, conceivably, but we like eating the things that come from actual biological organisms, plants, animals, whatever else. And obviously there's a lot of effort these days to make various kinds of artificial meat. I'm not sure how successful that's being yet. Certainly the times I've tried it, I haven't been a big enthusiast. But that's kind of a. Let's just take the. Let's make these things to provide what we need, which is amino acids, which is what we use to build ourselves out of. And we could get them because some other organism could sort of set up those amino acids, or we could get them just as a sort of chemical process. So I mean, this question of sort of the programming of biological organisms to do useful things, it's tough to do that programming. It's tough to know what the organism is going to do. And it's kind of a story of my sort of favorite computational irreducibility phenomenon, that even though you set up particular rules for the organism knowing what will the shape of the ultimate organism be and so on is pretty difficult. So I think that's, I don't, you know, there are specific purposes for which we can kind of reprogram pieces of biological organisms. And that's very important for interfacing with us who are biological organisms, so to speak. But using it for sort of raw technology of the door opening creature or something like this, I don't see that as being such a near term thing. There's a question here from Natanan about how can cellular automata and systems like the game of life provide insights into evolutionary dynamics. I literally just wrote a big thing about that that came out a week ago today. So that's a good thing to look at. I think I might talk about that. It's kind of a minimal model for biological evolution that turned out to be really interesting actually. So let me not get into that in great detail here. It's better to kind of for me to do a session about that and show all the pictures and all that kind of thing. Yeah, there's comments here about artificial life simulations replicate aspects of biological evolution in a controlled digital environment. You know, those things always frustrated me a great deal. I interacted a lot with the kind of artificial life community, particularly in the 1980s when it was first coming online. And a bunch of the work that was done there was based on my work on cellular automata. But the thing that always frustrated me was what does it take to have sort of realistic biological evolution? It's like, oh, we need these populations of organisms and they have to interact in this way. And they have parasites and they do this and the environment has this characteristic and they compete and so on and so on and so on. And it's like, what is the essential thing that's going on? Because the fundamental question is, there are many fundamental questions, but one of them is why does biological evolution lead to complex organisms? It's not obvious that a complex organism is a good thing. Maybe just that little protozoan is the best organism we've had on our planet. And why do you need all these complex multicellular life forms that can do live streams and yak about life forms and things like this. Why does the planet need that? Maybe it's protozoa are sort of the single cell organisms are just the most successful organisms one can imagine by some measure of success in terms of replicating and so on. And Darwin in fact back, I quote this quite a lot and the last sentence of Origin of species from 1859 was sort of as the earth goes circling around the sun according to the law of gravity. So organisms more and more complex effectively and ultimately the higher animals, as Darwin called them, have been and are being evolved. So Darwin sort of thought there was this law of nature that would cause there to be complex organisms produced. But nobody ever found a reason for that until maybe this little thing that I just wrote where I found this very minimal model which starts to get a handle on sort of the essential phenomena of biological evolution. By the way, those essential phenomena, I think are deeply related to the essential phenomena of machine learning. But that's a different story. Let's see, Let's see. Proz is asking a baseline bacterium that we fully understand the effect of every gene on its own and of all genes as a whole. Yeah, that's sort of the idea of a minimum bacterium. I'm not sure how well that's been achieved and I'm not sure how, whether, I mean, I'm not sure how well factored how modular kind of the, the operation of these organisms actually is. You know, I don't know to what extent you can say. I mean, clearly modularity is a thing in biology. That's why there are separate cells. And it's not just everything in a big vat. You know, modularizing into cells is clearly important. Modularizing into separate organisms is clearly important. Modularizing into organs within an organism is important. There's a lot of modularization, there's a lot of treat this thing as kind of a unit on its own, interacting only somewhat loosely with other units. But I don't know to what extent there is, you know, that, I don't know to what extent you can really point to mean that's sort of the, to what extent are the different cell types and so on doing some very complicated dance with complicated inter orchestration. And to what extent can we just take one out and say this is modularly doing this. This is essentially the subroutine, the function that does this. And it's connected only very loosely in a very organized way to these other things, I think. And in fact, the experiments I was just doing on biological evolution suggest that often the solutions are much less cleanly separable like that. There are things where when you look at it, it's like, well, yeah, okay, you can see it has certain mechanisms in it and in the end it produces some great results. But to say, can you take it apart and say this cog interacts with that cog and that one and so on to put all these things together, it's not clear you can do that. A question from Swiss here. Do you think gene regulatory networks function as observers like us? Well, so let me not get too deeply into that right here, but there's a whole story that I've told in different times here about kind of our view of physics and our view of the fact that we perceive the laws of physics that we perceive because we are observers of the kind that we are. And mostly I've thought about brain based observers, that is about kind of what in human brains, one of the key things we do is to try and sort of concentrate all the different activities of all those different neurons, those 100 billion neurons in our brains. We try and pull out of that a single thread of experience, a single kind of experience over the course of time. And I mostly thought about how brains perceive things because that's the main thing I have available to sort of internally consider that. But you could also say, how does for example, the immune system experience kind of the world? How does a colony of ants, where there's less independence on the individual ants, how does it experience the world? How does a human society experience the world? Human society, we all think we're doing our own separate things, but we are all sort of linked together by sort of human society. And there are certain kinds of collective things that happen for better or worse sometimes in human society. And that's kind of like a brain taking all those neurons and making a definite decision to do this or that. And there's sort of a question about to what extent are the ways that, let's say the immune system or generally regulatory networks and gene regulatory networks sort of reacting to the world and to what extent is the sort of conclusion of how they interact with the world kind of brain like it's an interesting question. I think in the immune system, for instance, we can kind of ask things about how does the immune system make a decision? Make a decision means we're just going to generate gazillions of clones of this, of this particular, you know, this particular cell and so on, copies of that. And to what extent, you know, how does it make the decision to do that? Might have some complicated network of different interactions that eventually say, zoom off in that direction. And I suppose the question is to what extent can we see that as being like our brain like decision about oh yes, we saw a cat or we saw a dog type thing? I'm not sure. Interesting question. And I think it's something where it's very possible that sort of the things that we've seen in Physics will replicate in biology because in some sense the way decision making works in something like the immune system or in other biological systems is sort of similar to the way it works in brains. But I haven't thought that fully through and I'm not sure. Let's see. Oh, gosh, many questions and I have to go soon here. Jonathan asks, might it be possible in the future where if there's a crime, we could just put a DNA sequence into a computer and it would show a picture of that person? The answer is, I'm sure the answer to that is yes. And you know, when people have said sometimes, oh, it's just putting out DNA and it's all very, you know, it's all very privacy preserving, I am fairly certain that one will be able in time to sort of reproduce things like facial features from, you know, from DNA. Now, having said that, you know, even identical twins, if they're kind of, you know, live in different environments, they may look fairly different and certainly they're, you know, their, their hairstyles might be different. They might wear, you know, they might do all sorts of different things. They might have had different nutrition in their lives which causes them to have, you know, different, you know, a different appearance for that reason. So it's not the case that everything about us humans is, and our appearance is determined genetically. I mean, there are even things like kind of the details of fingerprints which are, which are associated with folding in the, in fetal development. The fingerprints generally have the same general characteristics between identical twins, but they're not identical between identical twins. So there are aspects of kind of the way that we look that is determined by things beyond sort of genetic aspects. But there's clearly a lot of genetics in the way that people look. Now what's been difficult is it isn't just one gene. I mean, there are a few things like eye color that are determined by a small number of genes. And so there's very definite brown eyes, brown plus blue makes brown and things like this roughly, except when it doesn't quite make brown, it makes green instead. But let's ignore that. The simple kind of Mendelian inheritance idea is that there are sort of a very simple set of kind of discrete combinations for things like eye color, but other kinds of things like, you know, what's the, I don't know, the shape of the nose or something like this that's not determined by, presumably by a small number of genes. It's presumably determined by a large number of genes. And exactly how those work and how they interact has been quite unclear. I mean, a classic example of this is human height, which is clearly genetically related. The best predictor of a child's height is kind of the average of their parents heights, assuming that they all have the same nutrition and so on. But knowing kind of which genes are associated with determining height has been difficult. And there's probably a lot of genes. And in fact, there's some evidence for that because heights follow a normal distribution, and that's kind of what you typically get. If you just have lots of different contributors all being smooshed together and averaged, you'll typically get a normal distribution out of that. All right, I think I unfortunately have to leave you guys and go back to my day job here. But a lot of lovely questions which I would look forward to addressing another time. So thanks very much for joining me and see you another time. Bye. Sa.