Science & Technology Q&A for Kids (and others) [Part 118]

Hi everyone. Welcome to another episode of Science and Technology. Q and A for kids and others. Let's see, I see a bunch of saved up questions here. I see one came in from Ollie. Do you think it's possible to sustain life on Mars? How far in the future do you see this happening? Well, you know, I think the, the first thing to understand is what does it mean to sustain life? I mean, people manage to hang out in the, you know, in the space station for, you know, a year at a time. It's, that's a very closed system where, where, you know, any, any food that was needed is, is, is sort of being delivered. And there's a, there's a, a certain, it's, it's not something where one is kind of where one feels that it is some sort of an independently sustaining life. On the other hand, at some level, you know, the Earth is an independent place where life is sustained in the sense that, you know, we, we breathe in oxygen, we breathe out carbon dioxide, the plants do kind of the opposite. We have in the end, a self sustaining system that can be independent of everything else and where we can have everything we need to sustain life. You know, people have tried to do experiments where you sort of have a closed object in which life can be sustained. You know, there are these nice sort of, what are they called, these sort of globes that you can get which have various kinds of critters inside them, plants and animals inside them, and they seem to be able to be sort of a closed glass enclosure where the only kind of external input is light from the sun or whatever else. And I think those, for a few years, those managed to maintain themselves as sort of a closed ecosystem. There was an experiment called Biosphere 2. It was done in Arizona or when was it? 20 something years ago? More than that actually now, 30 years ago, which was an attempt to kind of have a sort of a big greenhouse type thing where it would be possible to have humans hang out inside in a completely closed ecosystem. It did not work very well. All sorts of things went wrong. I happened to visit that place. It's now kind of open to visitors and things. It's no longer an attempted closed ecosystem. But in fact, I think when it was attempted to make this thing where it would have its own sort of its own plants and some, some animal life and humans and so on, it didn't succeed. I don't know how long it lasted. A couple of years maybe. And then all sorts of things started going wrong. Oxygen levels started getting out of the range where it was good for people and things like this. So not a great. Not a great sample of let's make a closed human habitable ecosystem that can just last forever. When it comes to Mars, the first thing one thinks about is does one make kind of a Biosphere 2 greenhouse on Mars where there's kind of sort of standard atmosphere inside and where one has plants and things going? I think the movie the Martian featured another kind of sample of kind of fictionalized attempt to do that. I think the other question is, I mean, Mars itself has a thin atmosphere that is certainly not sufficient for us humans. The surprising thing in the Martian atmosphere is that, yes, you can fly a drone there. The Mars helicopter successfully flew there. Although it's a pretty exotic object that has its rotors rotate at the speed of sound there and things like that. But it did successfully fly in the atmosphere like you could fly a drone on Earth. The question is whether, given Mars as it is, can one expect to somehow mine what's there and on the long term, kind of have something where you have sort of an independent human settlement there or something that doesn't rely on deliveries of stuff from Earth. And again, that relies on having, I think, this kind of greenhouse bubble type thing, probably people near the. There is ice on Mars, so you can in principle get water. And the thing that same with the Moon, actually, that in the south pole of the Moon, for example, there's supposedly ice there. And the thing that's exciting about that if you're going to use, for example, the Moon as a way station to get to other places, is that potentially you can take that water, which is H2O, with hydrogen and oxygen, and you can potentially use sunlight as an energy source to break that down so that you get separate hydrogen and oxygen, which you can then use as fuel and oxidant for a rocket so you can sort of make your own rocket fuel on the moon. And I think also on Mars, I think that the effort to make things, you know, make other things out of what's there on Mars, and I'm not sure how well that can be done. I mean, the other question would be, can you take something which is something like life on Earth and somehow have a version of it that works in the environment of Mars? I think the effort that people talk about to do that is something which I really don't know whether the physics of this works, which is take the ice caps of Mars and somehow blow them up with hydrogen bombs or something, and use that to put enough water vapor into the atmosphere of Mars. That you have a quite different kind of chemistry of the atmosphere that maybe is more able to support life. But I don't know if that really works. I think that's one of those things where you have to try to. Doing the calculation is hard, knowing sort of what's going to happen if you, if you do something like that. It's difficult. Just like on Earth. You know, people wonder, oh, you, you know, there are lots of nuclear explosions on Earth. Do you, if you, if you, if you had lots of nuclear explosions on Earth, would you put dust into the upper atmosphere that would make the surface cold for years and all sorts of other things like that. And nobody, I think, really quite knows what would, what would happen under those circumstances. It's one of these things where there's things that one can readily compute you using kind of what we know about physics. And there's things where, well, there's this effect, there's that effect. It's hard to know how things will come out. I think the other place that's interesting to look at in terms of this process of so called terraforming making a planet sort of more like Earth is Venus, which has thick clouds and a very high surface temperature, what, 350 degrees centigrade, I think. And people talk about the possibility of putting microorganisms in the clouds on Venus and having their biochemical processes be able to take what exists on Venus in terms of the. What is it? I think it has a bunch of methane and things on it and carbon dioxide and being able to kind of biochemically process that into something that's more suitable for life. I mean, the thing to understand is there's ordinary chemistry and there's biochemistry. What's the difference between them? In ordinary chemistry, the typical thing one's looking at is there are molecules, let's say in a gas, for example, there are molecules that are jiggling around and every so often two molecules will collide with each other and, and perhaps those molecules will be such that there'll be a chemical reaction in the sense that there'll be some forces that are associated with the electrons in the atoms that will make those atoms, for example, combine. So that you might have, I don't know, hydrogen and oxygen, you might combine them into water or something like this. This can happen in ordinary chemistry just because the molecules happen to run into each other then sort of, that's kind of plain ordinary chemistry. The thing that really became clear perhaps 150 years ago or something now is this idea of catalysts in Chemistry where instead of just having the molecules sort of randomly running around in a gas or in a liquid, you say, let's have a catalyst. Let's have a, in that case, a solid surface that has some kind of topography to it where the molecules will stick on the surface and they will get aligned in some way. And then when another molecule comes along, that molecule will be arranged in such a way that yes, the second molecule fits in and you're more likely to have a chemical reaction. Whereas if they're in the gas, the molecules are kind of tumbling always around. And if it turns out to be the case that the molecules have to be aligned in a certain way for the reaction to be able to occur, the chance of that happening is much lower. So the big trick that biology has is it has things like enzymes, which are molecules, where they are arranging other molecules in just the right orientation to be able to undergo some chemical reaction. And there's a huge effect from that. I mean, there can be reactions that happen trillions of times faster or more because of the presence of enzymes, the presence of these sort of molecular catalysts that sort of do this organization of molecules in just the right way. And I think one of the things we learn more and more about biology is that you might think of biology as sort of an extension of chemistry, but really it's much more machine oriented than that. It's much more that molecules are being organized and you know, they're going in some filament or something, there will be molecules that are, that are, you know, it picks up a molecule and it has some process that actively moves it down the filament, rather than just this thing that happens in ordinary chemistry where things just sort of run around randomly. And so that's why it makes sense to say, oh, let's put microorganisms which are doing biochemistry, which are doing things that involve enzymes and other kinds of things that are sort of orchestrated organization of molecules to do something like terraforming Venus, rather than just saying, let's put some chemicals into the and hope for the best, so to speak. I mean, this is a general trend of the thing that we don't really know yet how to do is molecular scale, is doing kind of nanotechnology where we are organizing what molecules do, rather than just saying, oh, molecules are running around and we have chemical reactions happening. We already managed to do a huge number of important technological things just using ordinary chemistry. Although many of the reactions that are the most important industrially also use catalysts, solid surfaces where you have, where these Molecules get sort of macroscopic large scale solid surfaces where molecules get sort of organized in their orientation and so on. And as I say, biology has sort of gone one step further because it has molecular scale things like that. But I think the thing that one increasingly has to think about is how do we make molecular scale machinery. That is the kind of thing that we have in biology. And biology has some pretty nifty little devices like the ribosome, which is the thing that actually sort of starts the synthesis of protein, organizes from a piece of RNA that specifies what sequence of amino acid units should be put together to make a protein. It specifies what that. It is the machinery that has a stream of RNA going in and it progressively extrudes protein from that is assembled according to the pattern that's specified by the rna. That's a mechanism that's kind of molecular scale piece of machinery that biology has evolved. And it's been kind of the same for a really long time in biology. But that's kind of the thing to wonder about is how do we make kind of molecular scale machinery that does useful things. And kind of one trend that was around in the 1980s was, was this kind of, which led to the term nanotechnology was this idea that we would take things that we know about on a large scale, like, you know, like levers and gears and things like that, and find ways to make things like that out of molecules. That's one possibility. The other possibility is to have things that don't sort of mechanically work the same way that large scale machinery works, but is more suitable to be implemented in molecules. And that's kind of a different approach. And if you want to make things that, for example, do computation, we don't typically have mechanical computers. You don't need to make computation out of something that is, for example, mechanical. That was the plan in the 1830s, 1840s. Charles Babbage, ADA Lovelace that was kind of the idea of the analytical engine was let's make this thing out of cogs and gears and so on, and have it be able to do computations that way and be able to add numbers. People knew how to add numbers from the 1600s, actually, by using sort of cogs and things, but be able to do other kinds of mathematical operations and be able to sort of specify the controls purely mechanically. That definitely went out of fashion when electrical devices came in, because it's a lot easier to run electrons around wires than it is to have actual machinery where you're running, where you're moving cogs and so on. The big problem actually with cogs is very simple problem to explain. If you have an adding machine and it has all the digits represented by, each digit is represented by a cog, and you've got the number 999-9999 and then you're going to add one to it and you move that last cog, you have to somehow transmit the force to cause it to move, move that carry digit so that all the 9999 nines turn into 100000. You have to, you have to propagate that force all the way down that chain. And that's just a mechanically hard thing to do to both have it be working so that you can just be moving one cog at a time. And where you're moving all those, I don't know what, you know, 10 cogs or something like that all at a time. Much easier to do electronically with electronic switches where you're just moving electrons around rather than having to move material objects around. But so the way that our computers got developed is all based on electrons and electronics and so on. And sort of the key component that you need is some kind of switch, some kind of thing where whether electrons can go this way depends on whether there are electrons trying to go sort of 90 degrees the other way, so to speak. And originally that was done well in modern computers, the key pieces, so called field effect transistors, which have that feature that when you put kind of voltage across them, it prevents the electrons from flowing through the other way. And that kind of switching mechanism is sort of the key thing that you need to make logic and to make a computer. So also the thing you need to make amplification. Because when you're saying, oh, I've got the small effect and I'm using the small effect, that's kind of gating this big current, that's how you amplify things. Because you're saying small effect has an effect on this big thing that's happening. And that's how you do amplification. That was originally invented first for vacuum tubes. So in a vacuum tube, instead of, instead of having something where you have electrons flowing through a semiconductor, a solid semiconductor, you have electrons that were emitted from a hot wire and just streaming through a vacuum and being gated by the presence of other wires that produced electric fields or magnetic fields and so on, I think electric fields primarily in the case of vacuum tubes, that was a technology from around 19, 1900, 1910 kind of time frame where. You're having electrons flowing through Instead of flowing in a semiconductor. So, I mean, in a typical electric circuit, the electrons are flowing in wires that are made of metals like copper and so on. In a computer, sort of the innovation in a sense is having them flowing through semiconductors which don't conduct electricity like a copper wire does, where the electrons flow pretty freely through the wire, but they only sort of conditionally conduct electricity. Only when certain conditions are met does the electrons go through. And in the case of the earlier technology, one was using electrons going through a vacuum. Same technology has made cathode ray tubes, which were the kind of early approach to televisions, while the approach to televisions until pretty much the 1990s or or so. But vacuum tubes were, you know, a typical vacuum tube. You would have these. These things that were like a few inches tall and they were glass things with, with typically glowing wires inside. When I was a kid, for example, it was still the case that lots of, lots of devices had vacuum tubes in them. And you kind of open up them up and you could see these glowing, glowing things. One of the problems with vacuum tubes, whenever you have something like, well, it's like a light bulb, like an incandescent light bulb, where you actually have a hot wire inside something which is essentially a vacuum. Typically light bulbs have xenon gas in them. But. But you know that, that the rate of failure of the wire just breaking, for example, is quite high. Nothing like that happens in semiconductors. And semiconductors you can expect that that solid object will sort of last forever. But anyway, the basic approach that's used for computers is you've got lots of electrons flowing around in these kind of wire like things in a semiconductor, and you've got this elaborate pattern of kind of switches and so on between these wires. So it's really the story of computers is really the story of kind of navigating electrons through. Well, a modern microprocessor has the sort of density of wires and switches and gates and so on. That is more than you would get if you were to make a sort of street, a city that covered the whole earth and look at its street map, so to speak. Modern microprocessors have more circuitry in them than that, but it's that sort of kind of roads and, and junctions and so on that are happening. But it's all about moving electrons around. Biology is doing something a bit similar, but it's more about moving molecules around and atoms around rather than moving just electrons around. And it has kind of all this apparatus for arranging. Oh, you've got, you Know, the surface of a cell and it's transporting, you know, potassium into it and it's going through little, little microscopic tubes that were formed in the membrane around the cell. Things like this form those microscopic tubes formed by proteins that just have sort of holes in the middle of them and push the potassium through or whatever else it is. So, I mean, the kind of thing one thinks about is that's kind of the, that's the mechanism is, you know, how do you make kind of molecular scale machinery that does those kinds of things? You know, one thing that we know from biology, there is some machinery in biology that looks very much like kind of large scale gears and levers type machinery. Like for example, in bacteria, there are flagella, these, these things that kind of make the bacteria are able to swim, and they're these little tails that they have that rotate around like kind of like the screw of a ship or something. And the way those work is they actually have something that looks really like a gear that. And it really has, I think it even has 13, I think it has 13 teeth usually or something. And you know, you look at it under an electron microscope and you can kind of see it and it really looks just like the kind of, you know, gears that we might make out of metal or something like this. It's not made out of metal, it's made out of protein. And it's made out of a particular protein where when the protein sort of folds itself up, it forms that particular shape. One of the recent innovations, last few weeks, I suppose people have been talking about, is the idea of, well, how do you make a designer protein? How do you make a protein that's a certain shape? If you did want to make a protein that is, you know, some kind of gear shape, or if you did want to make a protein that has some particular form, how do you do it? It's hard to know. So a protein consists of these amino acids, which are clumps of atoms, and 20 roughly different kinds of amino acids. And a protein is formed from a linear sequence, a linear chain of those amino acids and all the different proteins we have. I know us humans, we have like, I know 30, 40,000 kinds of proteins that make us up, and those are specified by our DNA. But in the end, the proteins are sequences of amino acids. And they might be long sequences, you know, they might be hundreds of thousands of amino acids, or they might be short hundreds of amino acids or something like this. But what happens is there's this chain of amino acids and it happens to be the Case that those chains, they're pretty flexible and they fold themselves up. And the way they fold themselves up depends on the different amino acids, have different amounts of attraction for other amino acids and so on. Particularly there are two kinds, the hydrophobic and hydrophilic. Hydrophobic are ones that kind of push away from water. Hydrophilic are ones that kind of go towards water. And so those are, those are two features that if you put a protein in water, it'll tend to have its hydrophobic pieces sort of on the inside and hydrophilic pieces on the outside, I think. And that's one feature of how the thing folds up. But this problem of how does a protein fold up has been a hard problem because it's kind of like, well, there are many different configurations it could fold up in. You want to find the one that has the lowest energy, the one where the most proteins that amino acids that want to be together are together the most, where the amino acids that want to be apart are apart, those kinds of things. And that question of sort of how do you find that minimum energy configuration? A protein is flopping around, how does it, you know, it will find that in actual biology, if you actually make a protein, it will fold itself up usually into its minimum energy configuration fairly quickly. But the question is, if we're trying to just do that in a computer and work out what will happen, how do we do it? It's worth explaining that actually proteins don't always fold themselves up in exactly the right way. And in fact, there are some nasty diseases that happen from misfolding of proteins. And in actual biology, needless to say, biology has all kinds of mechanisms for making sure proteins fold up correctly. And there are so called chaperone molecules, for example, which are used to guide the folding of proteins and things like this. But in a first approximation, you have a protein, it's going to fold up in a particular shape. By the way, the protein may not be rigidly folded, it may flop around. It may depend on whether it's in water and things like this. But first approximation, it's going to fold in a certain shape. Often the way you figure out what shape it folded into is by making a crystal of the protein where you have many, many copies of that protein molecule all arranged in a regular array. And then you can use X ray crystallography to figure out what that regular array is and to deduce what the shape of the protein is that way. And so it's a bit of a different environment than a protein that's hanging out in Water, which is what it typically does in biology. But anyway, so this question of how does the protein fold up, it's a difficult thing to figure out because you could say, well, let's just try all the configurations and see which one has the lowest energy. But there are an immense number of possible configurations. If the protein is sort of a length N protein, there tend to be exponential of N roughly kinds of configurations to have to test. So people have been trying for ages to figure out sort of what are good ways to figure out how proteins will fold. Then a few years ago, kind of machine learning methods seem to seem to win the day. People have been trying that for a long time. It finally really worked. And what does it mean that machine learning methods figure out how proteins fold? Well, in a sense what's happening is there's a large amount of data on how proteins that have been studied fold up. And so what you're really trying to do when you say how is this protein going to fold? What actually happens in these machine Learning systems like AlphaFold and Rosetta fold and whatever, it's these various other different folding machine learning systems. What they're typically doing is step number one, compare the thing you just fed in the amino acid sequence you just fed in with all known proteins. And particularly when you look at the amino acid sequence that was fed in, you know, look at different sections of it and say, look at each section and say, how does each section compare with sections of proteins where we know how they folded up? So that's step number one. So that actually takes a lot of data. You have to have, I don't know, maybe a terabyte of data associated with kind of everything that's known about how all the known proteins fold up. That's step one. Step two is, okay, so you know that this particular region looks like a region of some other protein and it folded in this way. But how do you now glue together these little pieces where you say, well, this might fold this way because we saw it folded that way. In this particular example that was already studied. Now you have to kind of glue all those pieces together. That's the main place where kind of the machine learning kind of approach comes in, is that gluing together, step and, and sort of figuring out how do you, how do you arrange these pieces so they, so they make sense together. It's kind of, kind of similar when you look at an image generation machine learning AI system and you say, you know, you want, I don't know, I, I asked one of these systems to make A picture of a crocodile dancing on the moon yesterday. And so it's got to figure out, you know, it knows what the moon, surface of the moon looks like, it knows what a crocodile looks like. Maybe it knows what sort of a configuration of a, of a, of a person or a critter dancing looks like. It has to sort of merge together those concepts and make something which is a reasonable merger of those concepts. And did a surprisingly good job, actually. Kind of a funny picture, but that, that's sort of the. One of the things that machine learning is good at is kind of sort of take these pieces and kind of smoosh them together in a plausible way. And that's what has to be done in the case of proteins as well. Strangely enough, the same kind of smoosh things together and figure out how they fit. That is being used for language models for figuring out, oh, we started, you know, we start by saying, off saying something like, you know, the best thing about AI is how are you going to continue that? Well, you, what you want to do is, you might take, you might know that if you look on the web, there are, you know, 100 sentences that kind of start more or less that way, not quite that way. And you say, well, here are roughly their continuations. And then the big thing that you want to do with, with sort of machine learning is to figure out how do you, how do you correctly merge what you, what you've seen with, with what's already there? And that's what, you know, in a sense, that's how you can think about the large language models that make chatgpt and things work. That's sort of what they're doing, is this process of merging things together in a reasonable way. And one of the sort of interesting features is that these large language models, when applied to proteins, which are, after all, have nothing to do with human language and so on, they seem to have the same. They seem to have the ability to do this kind of merging process that you need and have it sort of be reasonable for proteins, just like it's reasonable for language. It's not completely clear why that works, but it could be what's happening in both cases is that there's some kind of regularity in how things work that we haven't yet discovered, that, you know, in biochemistry we haven't yet discovered that there is some very regular feature of how proteins are put together which isn't obvious to us, but which, which we can, which the machine learning system has detected, just like it's likely that there Are features of human language which are there but haven't really been, well, codified by us. Just like we know about things like the grammar of human language. You know, noun, verb, noun, and things is a typical thing we don't know about the kind of, the meaning, related, similar sort of ways that things get constructed in a general fashion. And that's perhaps what these machine learning systems have effectively found for us implicitly. And that's how they kind of how they manage to work. But one of the things I said, something that was just being talked about in the last few weeks has been maybe a month or so now has been the idea of can you make a designer protein by, for example, just giving as you do for these kind of generated images, you say, just make me a protein that will have a nickel atom inside it with a cage and will open itself up if you do this or that. Is it plausible to do that? Particularly if you read the biological literature, which has a bunch of text in it that says, oh, this is what happens, and here's a molecule that does that and so on. And I think one of the things that looks somewhat promising that you can essentially be able to sort of just like you can synthesize an image, you can imagine a crocodile dancing on the moon or whatever, you can synthesize that image and have the AI kind of put the. It knows what the sort of fundamental pieces are, puts them together in a plausible way. So something similar may be possible with molecules, with biological molecules, with proteins that you can say, well, you've got this one piece that, you know, fits nickel atoms inside it or something, another piece that attaches to this. Now this AI system will be able to merge those together to make you something where it says, okay, here's an amino acid sequence. Just make that amino acid sequence and you'll make a protein that has these particular properties. It's kind of an interesting direction for sort of thinking about nanotechnology and molecular scale machinery. I mean, the most important place for us where molecular scale machinery is obviously useful is an interface into biology, because we are essentially giant molecular scale machines which have all of these little mechanisms going on inside. And the question is, particularly when something goes wrong with some of those mechanisms, can you go in from the outside and put in technology that fixes those things? And yes, you can do that by having sort of large scale things that, you know, large scale objects that are at our sort of macroscopic machine scale. But it's much more interesting and potentially more than effective if you can have something which is Right down at the molecular scale. A machine that was sort of constructed by, you know, designed by us in an engineering kind of well designed for a particular purpose which turns out to be biologically useful. So anyway, I think that went quite far afield from the original question that was asked here. Let's see. Diana comments. Maybe biological switches would change everything for a biological machine. Yeah, you know, I actually don't know. It's a good question what the analog of a sort of biological switch really is. I mean, there are certainly molecules that will do things like they will have a protecting group that prevents some other molecule from getting at this piece that will cause a reaction to happen. I think maybe one can think about that a little bit like a switch. But it's a good question and it's a good question whether there are what the analog of an amplifier is. Well, actually, we know an example, pcr, the polymerase chain reaction, is an example of a biological or molecular amplifier where you start with one piece of DNA and you end up with lots of them. And that's an example of where of sort of a molecular scale amplifier where in that particular case it's one molecule, you end up with lots of molecules. It's a little bit different from the way that electron type switches work, where it's not like you're getting more electrons, you're getting a different flow of electrons. Have to think about what the biological analog of that is. I'm not sure. Interesting, interesting thought. Let's see, there's a question here from universe within. Can you explain the fundamental principles of biology? I think this point about machinery at the molecular scale that really in some sense is the big idea of biology, is that you can have controlled things happening at a molecular scale. Now the actual mechanisms by which that works, we have this idea of DNA where we can kind of store the program of how we set things up. And then this idea of proteins, of molecules which can be of different shapes, where the shape is determined by this kind of computation like sequence that's on the molecule. I mean, that idea of make a molecule in a certain shape, have a sort of a standardized way to make molecules of different shapes. That's an interesting concept where kind of you've got, I mean, one kind of wonders what the analog of that concept is. For something like robotics. You say, oh, I want a robot that's going to be all kinds of different shapes. How do I make that? Can I have the snake that just has articulation that causes it to kind of snake around in all different ways? Kind of the analog of a protein, you know, that will be one way, I suppose, to imagine making a kind of an amusing way to imagine making a robot. I wonder whether that could be made to work where the thing is just one long snake and, you know, it articulates and you'd have, might have a, you know, a million elements in the thing. And if it articulates, well, you can make all kinds of shapes out of that. You could make kind of a space filling curve where the snake has kind of snaked around and filled up this ball. Even though it's just a one dimensional thing, just like a ball of wool or something like that, you're snaking it around to make this sort of three dimensional object. And then you might imagine, oh, you could also make a completely different shape. You could, I mean, in a sense, I suppose, you know, any old piece of knitting or crochet or something like that is going to be a thing where you're taking this one dimensional object and you're making it into a different shape. But now the question for robotics would be sort of, can you get there from here? That is, you've got this, this, this, you know, micro snake thing that makes this kind of physical object. And now you say, okay, I want it to walk, you know, how does it, how does it get those, how does it dynamically go from a configuration where the, where the feet are in this position to the configuration where the feet are in that position? Interesting question, whether that's possible. That's would be an interesting thing to work out, whether you can, you know, if you imagine something that is purely a one dimensional thing folded up, is there a way of changing the fold, changing the orientation of things so that it goes from the one configuration to another? I don't think, you know, a good question would be, does biology do that much? And the answer is not much. I think most, most molecules, most proteins in biology kind of, they, they go into a particular configuration and they mostly are fitting together like jigsaw puzzles with other proteins or other kinds of molecules, rather than saying, let me just deform myself to have a different structure. So I think that's one thing. I think that the question of sort of how biology organizes kind of large scale, achieves large scale things, it has all these molecules, it has all this machinery. How does it actually cause something biologically useful to happen? That's an interesting question. And I think less is understood about that than one might imagine. Because in a sense, one of the things that's happening is you've got trillions of molecules, or at least trillions of molecules that are all doing things that are organized together. And they're doing those things in parallel, these molecules. It's not like one molecule does something, then another one, then another one. They're all doing them in parallel. And how one organizes, how it even thinks about all these things happening in parallel, it's difficult. A lot of what we tend to think about as humans is very sequentialized. When we talk with language, there's a one dimensional chain of words that come out. It's not that we have lots of things happening at the same time. In our brains we have, you know, 100 billion neurons or whatever and they're kind of all firing. Not quite all at the same time, but they're all firing. You know, it's not like it's just first one set, then another set. Lots of neurons are firing at the same time. And, and that's it. That's one big difference. For example, when we look at machine learning systems today, the one big difference is that that whereas our memories are kind of, every part is active and doing something in our brains, in a computer as of today, the memory is mostly just sitting there remembering things and only the CPU is, you know, is going through and actually making changes. Whereas in our brains it's kind of like all cpu, all memory, all mixed together. But in biology in general, there's this idea of lots of things happening in parallel and kind of understanding at a, at a, at a sort of large scale level, what's happening when lots of stuff happens in parallel is something we're not particularly good at. I mean, our physics project actually in recent times has given sort of some new paradigms for thinking about what it means for lots of stuff to happen in parallel. One of the things that we are familiar with is in our normal experience of the world, we are used to the idea that sort of there is, we have space, we have kind of things happening at different places in space. And at some moment in time, things are happening at different places in space. There's like in parallel, across space, lots of things are happening at any given moment in time. And we kind of have some familiarity with that, partly because the speed of light is quite fast compared to our processes of thinking. So we're used to the idea that we can see all this stuff that's happening in different places in space all at the same time, so to speak, or all what's perceived as being at the same time by our brains because it takes our brains a long time to kind of say, oh yeah, I got that, and then go on to the next thing. And the speed of light is such that it's delivering information about things that are happening in different places in space, sort of all together as far as we're concerned. So we have some awareness of sort of that parallelism when it comes to things happening in different places in space. But, but quite how biology organizes these kinds of things is not really known. And in an example of something that is not well understood is when you have molecules that are bumping around and maybe they're being organized in a sort of machine like way. And you ask the question, how fast is such and such a chemical process going to happen? The usual way you work that out is you say, well, if it involves some amount of this chemical, some amount of that chemical, it's the, you know, you have concentrations of these two chemicals, you multiply them together. That's the number of combined sort of pairs of chemical A and chemical B that they'll be. And that's how you work out how fast the reaction happens. But that doesn't account for the fact that, well, actually after, you know, A and B interacted, they're in a sort of different configuration than before. And, and you can end up with these kind of chains of causality of one molecule interacted with another and then it mattered that there was that history that determined that affected whether the molecule will interact again, and so on and so on and so on. Ordinary chemistry and the ordinary kind of mathematical methods used for chemistry do not take account of that. I mean, in the language of fancy physics, they're essentially doing mean field theory and they're ignoring correlations. And so the thing that one could certainly imagine is what does it look like when you take account of those causal relationships. One doesn't know how to do that because that requires kind of better understanding of how a parallel process of action happens. And that's something we've learned quite a bit about actually from our physics project, not yet applied well to biology. But those are sort of, those are important principles in biology. I suppose one of the, one of the meta principles of biology is just protein lets you do an awful lot of stuff. You know, protein made from, you know, you know, a small number of elements, small number of different kinds of elements, you know, these 20amino acids and just putting those units together in different configurations kind of gets you all the different creatures of the earth and et cetera, et cetera, et cetera. What biology usually seems to do is most things are made of Protein, but then it kind of recruits some other kinds of atoms to make calcium for bones or iron for red blood cells, or iodine for the thyroid gland or whatever else. It usually recruits other kinds of atoms that have particular useful chemical properties. And usually what it does is it makes a cage out of protein and just the right size to fit that kind of at. And enough atoms are sort of bumping around and come from your food and things like that that the ones that are the right size will get stuck in those cages and will do the right chemical thing that biology is sort of recruiting for. But I think that's the, it's kind of my, my view of sort of the, the mechanism of biology. And one of the things you see in biology is, is once you have some process, some molecule that interacts with some other molecule that does this and that and the other biology will tend to reuse that same idea over and over and over again. So for example, ideas like sort of natural selection, the evolution by a natural selection, that things are that if you want to kind of adapt to the environment, you can do that by saying, well, you know, the more adapted organism will have more children. And that means that eventually its genetics will take over the population. That idea, which is applied at a large scale for overall biology is applied at a small scale in the immune system. And it's kind of like biology reusing an idea that existed there. And you could say it, you know, we put it in a very human form of saying it's reusing the idea really. It's just that is a feature of sort of the fundamental formal computational feature of how things work. And that mechanism appears both in large scale biology and in sort of the small scale immune system. Let's see. I've been talking a whole bunch about biology here, but. And I mean, I think the William comments. Aren't neurons the biological analog of switches? Yeah, yeah, they're an example. I mean, but they are again, electrical. A neuron is. The, what's happening in a neuron is. It's like a little tiny battery, it's electrochemical device where you know, you, you have a chemical signal from one neuron and it, you know, the one neuron might, might connect with 10, 000 other neurons. And if it arranges, if the right electrical signals accumulate that the next neuron down the line will sort of, you know, will, will start getting excited and it'll start producing electrical pulses and then it will excite another neuron and so on. In the end those, the, you know, the battery that is the neuron that is making electrical pulses is eventually fueled by glucose and, you know, it's taking in food effectively. And that's how it's generating, it's producing, it has the energy to go and produce those electrical pulses. Let's see. Gosh. Lots of comments and questions here. Well, there's a slightly technical question here, but perhaps it's interesting to talk about. Question is, can machine learning help solving inverse scattering problems where the forward scattering problem is highly nonlinear? So what that's about is if you have. Something where, let's say you have, you have some kind of explosion inside the earth and it produces seismic waves, it produces deformations in the, in the rock that propagate. Like sound waves. Like sound waves propagate in the air. There are, there are similarly seismic waves that propagate through rock. And you've, and you're, you're, you know, it's propagating through the rock and eventually you pick up sort of the signal of what all those, what, what all the, the result of that whole process of the seismic waves propagating through the rock. And the question is, given that you picked up, you can work out if the rock had this particular form. You had granite here and you had limestone here and so on. You could work out from the, you know, knowing the explosion happened here, this is the rock that was here and there, you could work out, okay, so this would be the signal that I would get on the surface based on that kind of rock. So now the question is to solve the inverse problem of I've got these signals, what kind of rock is down there? And that's important if you're trying to prospect for minerals and things like this. And so the problem is that one of the problems with that is there may not be a unique answer. There may be many different answers which would lead to the same pattern of signals at the surface. And one of the things I think is being suggested here is, is, you know, can machine learning be used as a way to say what's the plausible reconstruction of kind of what rock there could be? Because it may be that there's a version of the rock that would produce those signals at the surface, but it's some weird formation that's never actually seen in geology. So, yes, I think that's a good use case for machine learning. I've wondered about some similar use cases. One famous use case is the, is the use case for x ray crystallography. So if you take a molecule like a protein, for example, make a crystal. I mentioned this before you're making a crystal where you just have a thing where you have this complicated shape protein and you've arranged it in a completely regular array, you've made a solid out of that protein where the protein is all, it's all lined up, it's all in this regular array. Then it turns out there's this technique of X ray crystallography. You scattered X rays from that system and the X rays probe certain features of that crystal structure. And the question is then, okay, what is the actual shape of protein based on the ways that the X rays scatter. So for example, when DNA, when the structure of DNA was decoded in 1953, the big innovation was, well, a person called Rosalind Franklin had, had found the X ray crystallography images of. This is what you get. If you've crystallized DNA, which is a hard thing to do, and then you've done X ray crystallography on it, these are the ways that X rays scatter from that crystal. So then the question is, okay, what shape of molecule will make X rays scatter that way? And there are many different solutions. There can be all kinds of weird shapes that would produce the same X ray crystallography pattern. And the thing that Jim Watson, Francis Crick did was to figure out that, well, actually if you have this shape, that is the stubble helix, that will produce the X ray crystallography pattern that was actually observed to be the X ray crystallography pattern of DNA. There would be many other possible solutions, but that was the one that is molecularly plausible to make. And so this idea of using machine learning as the way to do that inversion, it's a good idea. There are cases where that inversion is done which don't need machine learning. So famous case is tomography for CT scanning and MRI and so on. What's done there is, that's sort of a simpler case of inverse scattering where what you have is, okay, you've got X rays, they're going through a person, the X rays, when they hit bone, for example, the X rays don't go through as much as when they're going through soft tissue. And so if you, if you kind of, you know, if you X ray your hand, this was the early thing that roentgen did around 1900, was you x ray your hand. You see the, the bones are white. Well, because the way the film was being used, that's the place where the X rays didn't go through. And. But that's where you shine the X rays from one side and you see what what comes through? And the answer is, where you have soft tissue, the X rays come through. Where you have bone, the X rays don't come through. So the basic operation is you're seeing. What do you see on the other side when you shoot X rays through something? I mean, so that's the basic operation. Now the question is, how do you reconstruct the sort of what's inside you by just, you can X ray from one side, so to speak, and you can just see kind of the shadow of what's inside you, the shadow of your ribs or whatever else on the X ray film. The idea of tomography is to go and look at, from all angles, to just go around the circle and just say for every different angle, how much are X rays absorbed? So you're going in this orientation, you're absorbing this much. You go in this orientation, you absorb this much. And essentially what you can think about doing is imagine that for every orientation, you knew how much, how many X rays were absorbed. And imagine you were cutting out little, I don't know, trans pieces of transparency or something where you, where you make them darker if more X rays were absorbed. Well, actually lighter, because the X ray tends to. X ray film has this inversion thing, but, but let's say it's lighter when more X rays were absorbed and darker where fewer X rays were absorbed. And you just lay these strips down and you arrange them around a circle. And knowing that in this orientation you've got this amount absorbed, in this orientation you've got this amount absorbed. If you do that, then the picture you'll get will basically be a picture of what's inside. And that's how tomography works. When a CT scanner does its thing, it's going to tick around. It has a X ray source, X ray receiver, and it's just finding what's happening at all different angles. MRI is a little more complicated because there happens to be a. It happens that the, the way that works, the, the, it's, the Signal is at 90 degrees to where the, where the, where the, where the magnetic field is and so on. So it's just a slight more geometrical trickiness. But it's the same basic idea. But that's a case where you can do inverse scattering very definitely in a very mathematical way. In many cases of inverse scattering, that's because, actually, because it is a linear problem. It's just the amount of X ray absorption just kind of adds up. In many of these other cases, nothing like that happens. And so, yes, I think it is a good Use case for machine learning. And I'm actually surprised. I mean, I've certainly asked people about the case of x ray crystallography, and people sort of said, yeah, that's a good idea, but I'm not sure anybody's done it. Perhaps they have. But yes, it's a good use case because it's one of these things where it's kind of the reconstructing what's plausible, particularly based on many examples of what really is there is a good kind of thing to do. Okay, Kylie is asking, how do X rays work to see only bones and ignore skin? Okay, so the basic point is that it's a question of what. What absorbs X rays more. So, for example, if you. Well, if you shine a light flashlight through your hand, you'll see. You actually do see something, which I think you can see Bones, if you do that. The, you know, for visible light. Well, the problem is for visible light that the photons of light, as they go through your hand, they keep on getting scattered. They keep on getting, you know, going in a slightly different direction and so on. They don't kind of go straight. They don't most of the time just go straight through. And that means that even if there was a nice, you know, bone there, the photons will keep on getting kicked around in sort of a random walk. And so it'll fuzz out any image of the bone. For X rays. X rays are much higher energy photons. They're usually produced. Typical X ray source of X rays is you accelerate electrons, which you can do with an electric field because they're charged particles. You accelerate them, you crash them into a target of lead or something else, actually nodes. Some other lighter materials I think are usually used. And that will produce what's called bremsstrahlung breaking radiation. And that is what will happen is the electrons will basically excite atoms in the material and those atoms will emit X rays. And those X rays are. Then you can then collimate those X rays to only keep the ones that are going in a particular direction. And that's how you get your X ray source. But anyway, the X rays are just high energy photons. Same kind of thing that's in light, but higher energy. And those higher energy photons, most of the time when they're going, things are mostly quite transparent to X rays. Things are mostly, you know, our hand, for example, mostly the X rays just go straight through. But the X rays get some atoms, some types of atoms will stop those X rays. And the basic rule is that the. An atom. So if you look on the periodic table, you go hydrogen, helium, etc. Etc. Etc. As you go through successive atoms, you'll, you'll find the, the. There are more protons in the nucleus of those atoms. And if I'm remembering correctly, the, the rate when you sort of shoot electric X rays through roughly the, the scattering of X rays, I think, goes like the square of the atomic number. So it's kind of like, it's. As you, as you make. As you look at heavier elements, they tend to have more effect on X rays. And let me see, why is that? That is because. Yeah, that's because the, the. Well, okay. It's because the, the electrons, when there's a larger charge on the nucleus, it's kind of pulling electrons closer to the nucleus. And when you pull them closer to the nucleus, they effectively have higher energy. They're kind of spinning around more rapidly. You can think of it that way. They're kind of spinning around more rapidly. They're pulled in more by this higher charge nucleus. And so that, that causes them to be associated with higher energies. And that means that when you have they. When the X rays can be absorbed there, because there's this place where the atom can have kind of. The electrons in the atom are kind of able to absorb X rays that are of higher energy than in the case where there's a lower charge on the nucleus and the electrons are kind of sort of have kind of more, more lazily going around the atom. And they just don't have the ability to absorb photons that, that say, I'm going to kick you pretty hard and that those electrons just, they can't absorb that kick in the same way. So in, in what, what happens is as the, as the atomic number goes up, as you go further in the periodic table, you absorb X rays significantly more. And because our bones have calcium in them as opposed to having carbon, oxygen, hydrogen, which is what most of our sort of soft tissue is made of, calcium is further along in the periodic table, so it absorbs X rays more. So we see X rays, we see the X rays being stopped there. And because originally X ray stuff was done with all ordinary photographic film where the places where it wasn't exposed, it was white. That's. Our bones happen to show up white on sort of typical X rays because that's the place where the X rays don't get through. The film wasn't exposed back in the day when people were using photographic film for X rays. And so it ended up white. The Places where the film was exposed, it went, it went dark. And those are the places that correspond to our soft tissue where the X rays were just successfully going through rather than getting absorbed. Mri, by the way, sees different things. So MRI is all about hydrogen and it's all about seeing essentially where the hydrogen atoms are in, in one's body and things like other kinds of imaging technologies. Like there are ones like PET scanning, positron emission tomography that is seeing, I think glucose. Let's see now, I think, let me think. That certainly sees where metabolism is happening. But I'm now confusing myself a little bit because in, well, in functional MRI imaging for looking at where in the brain is active, you are seeing glucose metabolism and it still must be primarily seeing what's happening in the presence of hydrogen atoms. I have to think about that a little bit more to see how that works. The Prutz is commenting that that bone opacity comes down to water contents. There's more water and tissue. Yeah, I think the point is that, that, that, I mean, what's happening is the X rays are getting, I mean the, the, yeah, water doesn't stop X rays or it does eventually stop X rays, but at a much slower rate than things like bone. I, I think it's the calcium content of bone that is the primary thing that stops X rays. And for example, if you have, you know, if you have a, a metal plate because you, you know, broke some bone and it got repaired with a metal plate that will show up. The metal will be higher atomic number and it'll show up. It'll completely stop the X rays. Okay. Yolo. YOLO is asking, can those X rays be used for automatic detection of changes within the molecular structure of crystallized solids? Well, X ray crystallography is, people don't do that on sort of a real time basis. It's kind of like it's a very delicate thing to make these crystals and it takes a lot of effort. You know, people spend a year trying to crystallize this kind of protein and it's a big kind of complicated art. Just how quickly do you cool it down so that the protein molecules have arranged themselves to make into a solid and so on? It doesn't tend to be something where you just say let's deform it and see what happens or something. It's not a real time kind of activity. It's very, very slow activity actually. So I think that the, there are techniques. Let's see, I'm trying to remember how this works. There's an exafs xf, what is that? X ray absorption. Oh boy. There are various techniques that are not doing so. Ester crystallography is based on the fact that there's a regular array and you're effectively detecting the periodicity of the array. You're detecting different periodicities in the array. Mathematically, you're doing a Fourier transform. What you're saying is the atoms in this orientation, the atoms are arranged in this regular kind of sequence in this orientation, they're arranged in this sequence. That's roughly what you're detecting in X ray crystallography. I don't think, gosh, there are techniques for looking at very fast molecular processes and I think there's one that involves X rays, but I'm not remembering how that works right now. Let's see, Universe within asks, do you think ecology will play an important role in understanding molecular computation? Interesting. I mean, the question is, what do we know from ecology and is there sort of cross pollination of things we know from ecology about, Oh, I don't know, you know, what's the invasive species analog in molecular computing or at the level of molecular biology? I'm not sure. I think that ecology is still a comparatively blunt science. I mean, we know the sort of mathematical aspects of ecology are mostly things about game theory, mostly things about saying, well, we've got this number of this species and that number of that species, then this will happen. I mean, there are very fundamental problems in ecology which are completely unsolved. Like, you know, for example, why isn't there always one winning species? How do you estimate how many different species there will be within a given area? I mean, there are weird rules like these claims that on islands, the number of species depending. If you have a small island, you have different number of species and there's some kind of power law distribution. I'm not sure if that data is correct. It's probably rather hard data to get that on different sort of sizes of islands and things, you'll get different numbers of species. And there are all these kinds of things that are observed about the number of species that exist. Like, for example, there are perhaps 10 million species of critters on the earth. Estimate that 10 million number. You know, the Earth is a certain size, it's a certain age, et cetera, et cetera, et cetera. Why is it 10 million? Why is it not a trillion? Why is it not a thousand? You know, why are there different numbers of species? What is the relationship between. Well, we see speciation happening, we see the formation of distinct species, and you Know exactly what, what, what determines kind of this whole sort of interaction between different species. What determines, you know, can we have a, you know, in a sense, a new species often involves an innovation in some niche. It figures out, you know, the, the new kind of bird figures out how to, how to peck its way into this kind of nut or something. And it's sort of an idea and it's kind of like, how do we explain the space of ideas? How do we describe sort of how these different species, each of which sort of has an idea. What is that space of ideas? And how do we think about how that fits together? You know, we have a more microscopic version of this in microorganisms. For example, when you have a bacterial colony, you know, if you look at, I don't know, the, the gut microbiome or something, you're looking at all these different bacterial species there and how they interact with each other. And you know, if you eat a lot of this kind of food, they'll, it'll make an environment which is more rich in whatever and then you'll, you'll end up, you'll end up changing that microbiome. And how that works and what the ecology of that is is sort of a perhaps more immediately experimentally accessible version of ecology. I mean, there's also ecological issues in things like cancer and so on with different types of cells being produced and such. Like, I actually realized that I have to disappear here because I have to go do some day job things. And I noticed Zayden commenting that ecology is more complicated than physics. I wonder how true that is. You see, what happens in physics is like all these gas molecules are bouncing around in the room. They're doing really complicated things. It's just that the particular observation we make of them about their overall pressure or overall flow is one where that complexity is hidden. So in ecology, the issue is what do we care about? And if the things we care about include how many humans survive on this part of the planet? You know, that may be something that, which rubs our noses perhaps quite literally into kind of the computational irreducibility and all this kind of complexity of what's going on. But there may be some questions we can ask in ecology if we could figure out what those questions are. Well, we could say, yeah, we got a science answer for that, just like we can have a sort of simple science answer for these kind of large scale features of gases. Anyway, lots of interesting questions. Thank you for asking those. I see many more about different topics and I look forward to chatting with you all again next week. So thanks. Bye for now.