 So, my point is to provoke you to say something interesting, and I will listen, but I just my point to say a few introductory words. And one of the phenomena in biology which is mathematically very attractive is amplification of complicated objects according to their signature. And the simplest kind of used artificially is PCR, right? You have a signature and that's enough to amplify some molecules, exponential amplification and exponents, a big number certainly very attracts mathematicians when you only have ten to the twenty, it's very pleasant to see this, right? And my question would be, yes, what are the systems of the type is known and how we can artificially create new system of that kind. So the obvious thing beside PCR, of course in nature, amplification of survival, whatever it means, you install your bacteria and they have, say bacteria are the organism and not so big numbers, so they are less interesting. And you amplify them if they, for example, are resistant to some particular antibiotics and then another thing, even better, they have a strong amplification for viruses because they are small particles and how they amplify in the ambience of bacteria accept, which they accept. And another example, of course, the basic example is human or in general vertebrate immune, a quiet immune system, when you amplify proteins with some particular patterns corresponding to their being ability to dock, to have docking with pathogens. So there are these examples. And then one of them, this PCR, was kind of used by artificial system kind of by mathematicians or rather by computer science, by Edelman about twenty-five years ago, and this activity of DNA computing was kind of ignited and then it was continued, but I'm not certainly with my question to the audience, how reliable what you can learn? Say about fifteen years ago, there was a publication by, I think, by a name of Yahut Shapiro when they claimed they can make this kind of system going into some body and detecting prostate cancer, even curing it up to a point. Is it true? It was never confirmed. Is it true, by the way? Do you know something about it? But Yahut Shapiro... You can verify what... Again. There was this work by, in nature actually published by Yahut Shapiro, they can make this kind of almost universal computer, they make it on DNA, on amplification and with some enzyme, which they can implement on a cell. In particular, it can detect some prostate cancer and even do something. And then there was publication, but I don't think it was following, I haven't followed it myself. For example, what is, you know about that? Biologists. I probably... prostate cancer is very easy to detect. By TSA. Yeah. No, no, no, but it's... No, but that's the whole point. That's, I think, the difference of mathematical. It's not the question, easy or not, whether it's interesting or not. If you have a completely different approach, automatic, you don't look and say, you put my micro-machine inside and they do it and you don't look like this. Actually, you can do early detection of cancer. Yeah. By looking at the DNA in the blood. Yes. Okay? And you may find, like oncogene, like mutation in oncogene, you can predict that this patient may develop cancer. But this principally different approach. This is exactly the point. It's a principally different approach based on amplification. You put some cell in your body, some structure in your blood, and then multiplies if there is that. It's not like immune system. There's fundamental difference what you do immune system does, which is certainly by far more intelligent. All these tests are handmade, yeah? And they never go beyond certain level. Immune system goes beyond. So this is a point, imitate, in a way immune system, to use amplification and to use unspecified detection, which is selected then by some process. Right? And that's my question is how far you can go with that? For example, let me... So you don't know about this blood by your... What's up here now? No. This is my question. Of course, there are some other alternative ways, but this is kind of... It's a completely different story. Right? And the second point is how there was a discussion this by people trying to make minimal bacterium. The question is if we can make maximum bacterium. Namely, it will be bacterium and also probably collection of plasmids, which may be churned by simple stimuli, say, into viral particles. They go along together. And then there is a parameter which you want to select for. And to make it faster automatic. You have to create an artificial community of having a collection of plasmids, which are churnable automatically to the plasmid to viral particle, going to bacteria very fast developed because you can prepare bacterium to very fast evolution. And so it works like immune system and then does what you want without knowing how it works in several stages. So this is how it works, kind of parallel. There was recent development in machine learning. There was this in the 80s. There was competition between machine learning, between statistical approach, using statistical learning. On one hand, there were neural networks, and certain statistical learning went ahead because they analyzed more carefully. But then eventually in the last kind of five years, neural network took over where the point was not understanding what you're doing. You gain of not understanding. You do it by hand in a very primitive way because there is a power of amplification of some process. You iterate something. And the same you can imagine here. You may not know how a particular pathway works, but if you create a community of viral particles and bacteria, it can do it much better than you. If something has been done in this respect, and it's how feasible to create controllable bacteria, give a chemical signal, it starts mutating in particular locus, start taking plasmids, turned into viral particles, inject them when you give another signal, et cetera. And then you can create very fast, by far, kind of faster, because usually you make selection of bacteria for a particular purpose. You make some injection-based CRISPR, you make some gene-modified by selection, but it's a specific process. It can make a quasi-universal, not universal as a tuning machine, but universal like neural networks. General method, very fast. What was done about that, and what are possibilities, about the chances, what are constraints are there. And these are my questions to the audience, and certainly I just want to know what is none. And just... I think what is done a bit is using DNA to write information. Yes, that's of course what will be used, but this exactly means a point of amplification. This of course, yes, I agree, this has been done, but this is kind of... I don't think that we have... It's not DNA, you want to use bacteria and viral particle when they're already there in machinery for the production and doing complicated things, it's already ready for you. You just have to exploit it in some structure, and it's... I think it's quite... It seems feasible from generic principle. Of course, biologically I cannot judge what will be obstacles, how much it would cost, of course it would be a big thing to do, it's not what you do like that, yeah. It takes time and energy, whatever, but it looks very interesting at least to imagine, to know what the potential, what the possibilities are there to exploit and to guess, and mathematician may predict, guess how hard to do, and then of course try to do it. So what you can say about it? You don't mean that there is a theory of maximal size of bacteria or genome? No, they may be not very big, it may be a collection of bacteria, they have absorbed different plasmids, they have a collection of plasmids, a collection of bacteria not very big, but on the contrary they may have a rather small, they only survive in a particular environment, so they may be harmless, and then they absorb plasmids for particular function, then select them, they might be strong enough to be selectable, they don't have to be too big, they may incorporate some of them into... Where are you going to put the bacteria? Where are you going to put it? No, these bacteria will just be coming, you may just breed them in huge amounts, and then they, for example, you want to make bacteria producing... Put it in somebody's body? No, no, no, no, first it's the easiest thing to produce, you want to produce your antibiotics, so you make bacteria, it's very mutable, and there is some other bacteria you want to kill, and you put it, create a situation, and survive only if it kills, if it doesn't kill, it doesn't survive, and make very fast mutable semi-controller bacteria, so in a matter of days you produce antibiotics, days, not months or years, very fast, which you can do, of course, your patient is an unknown kind of bacteria, an unknown disease, infection, you combine this with his blood, and then in a matter of days it works like immune system, but much faster than under control, because viral particles produce faster than your immune system works, you can do it in much bigger quantities, not in a matter of a milliliter but a question of 100 liters, so you can amplify it by a factor of 1000, so you have the potential, if you know how to prepare, you have to make beforehand this universal bacterium, this universal system, easily adjustable, and that's the question is if you can do that, you don't have to be big, it's very much because it's very different from usual bacteria, but usually the known pathways, kind of, all partly unknown, of course. Why do you want to pack it all in one super bacteria? Why not just keep it to a very diverse population? You don't know what you're doing, it's a scheme, you must be adjustable, so you have to do it once and for all and have this in your reservoir and also a collection of these plus-mids or viruses, and then you guess which to throw in and just do it, it's just much like, why have universal computer? You don't make computer for each purpose, it's easier to have it, much more practical, do it once, once you create a hard work, you create a very massive work, and this is like, immune system works, yeah? It's one immune system, universal scheme, more or less, for all vertebrates, or at least for the world mammalian, this universal system, yeah. And how do you want to guarantee that it would only attack that virus or bacteria? No, exactly, by selection, if it does something wrong, it doesn't survive. Exactly, it makes selection mechanisms. What I understand is that you want to do this in vitro for that specific? Yes, in vitro if you want to produce some, you want to produce some antibiotic, you want to check for toxicity, of course, do the same, make selection on the next stage, of course, make several stages, yeah? But maybe it's mutated to attack that certain bacterial virus, but maybe also... Oh, maybe, many things may happen, maybe, of course, sure. You have to make proper provision for that, sure. Of course, there are many kinds of... The question if you do it, if you have a car and they say, oh, it can kill, pass the by, of course it can, yeah, but you do it so it wouldn't do, of course, make it automatic. I want to clarify again. You want to make a bacterium that can produce 100 antibiotics. No, no collection. No, I have to do the bacterium which you can control so it will go in particular mode of mutation and then become producing these antibiotics. You have computer program, it doesn't do just anything else, you have to program it every time. It programmable bacterium. So you want to have several buttons switch one to produce one antibiotic. No, no, no, no, no, no, not at all. I'm saying again, it's not how it works there. How it works with all the neural networks, yeah. It's not switch a button and get the result, right? You switch a button and start searching solution for a particular problem and then America goes for a cycle and then it finds a solution. So you make it mutable, right? You mutable particular locus and mutable when you control this mutation and then it becomes mutation and put it into some surroundings. If you have an idea what you want to produce divide it into stages. If you don't, you don't, right? So there may be possibility. And probably the most realistic not to change it. It's a hermosomal mutation but absorbing classmates which can be turned into viruses and this is the most mutable element so it make it the fastest, right? So it will be on the molecular level on the level of viral pathic evolution bacteria is supported, yeah? So the two stage process you can imagine, yeah? And this will be... There is a system which works by selection and learns to to kill bacteria, right? It's called the immune system. Yes, exactly. Wouldn't it be satisfactory if we just make kind of externalized immune system? I think it's more difficult. The immune system has many kind of different kind of cells involved, right? The amplification process what you want to use. But I think it's easier to have amplification with the viral particles and bacteria, right? It seems to me it's kind of more realistic to create artificial immune system outside of the body. I think this... We don't know if we want to see but it looks more... less flexible. Since you mentioned the immune system, etc. One major property of all these systems is the adaptation. Yeah. So it means the immune system in your body when we take it out and put it in something else in some other environment it's not an immune system anymore it's something completely senseless. Yes. So the structure that you want to define needs a priority. Some constraints of where it should survive and how it should survive. So like neural networks artificial networks whatever you want to have are priority conditions for how the system should look like before you put it somewhere because else it is not... No, absolutely. You have bacteria which are already sufficiently stable and then you add some features and keep selection. First it's a very slow process. You select the kind of bacteria with the properties you want and simultaneously you have to invent means probably they exist terming viruses to plasmids bacteria will be kind of symbiosis bacteria unless you want to get them out and infect other bacteria to make the process fast. The problem here is in your own body there are a lot of bacteria that are actually good and necessary for you. When you just develop it in vitro you don't have control for every variability that you might have. When you put it inside the living system an adaptable, mutable system No, but I don't speaking... Finding the most optimal material for example. But think about putting them in the human body it's another issue. If you can make them so well that you can go on to your body to do something it will be next level. I don't speak about that. I'm speaking but usually in biotechnology you make bacteria serving particular technical purpose, chemical eventually. Producing chemical either known chemical or consuming chemical or chemical serving simple chemical function like in the kind of bacteria. So you don't know exactly what it is and all bacteria are more or less similar for that. I don't know which I use usually. I think E. coli or some variation mutation of E. coli. Some very survival strong bacteria but you want them to absorb for example to carry plasmids with specific properties which you don't know beforehand. The properties you know what they want to do you don't know how they would do that and then to make it fast adaptable to this situation. Faster than normal because usually mutations are slow and they can make fast mutation in particular locus without disturbing basic machinery in the bacterium. It can go in principle by order of magnitude faster. And so with the bacteria much faster creating what you want. So not in years but in hours or days. Well considered instead of bacteria viruses because of course with plasmids and because they mutate faster, sure. There are similar things like that that are already being done to create new antibodies or they have different type of names dark pins or where it's not using the bacteria but it's actually using in vitro translation or sometimes it's... In vitro translation, I don't know exactly how we efficiently nowadays in vitro translation. Well, you can... I cannot ask him, he's an expert. So how massive and how stable it can be in vitro translation. Of course it's used artificial selection of proteins in vitro translation. There is a company, Moderna that's what they are based on in vitro translation. Yeah. They make RNA and then transfect it with ribosome. But who actually make translation? Have ribosomes separately from the organism in the cell or outside? Outside the cell, ribosome outside of the cell. On what scale? Make it also artificial. We just build it without the ribosome. Without the translate. Translating without ribosome? Synthesis. Synthesis. On life scale indeed, yeah? You tell me, how big the scale and how adaptable it is. If you want to... Evolve. It should be comfortable. But you say you can do it to evolve so it can be making excellent antibodies, yeah? Absolutely. So there are people who are doing molecular evolution. You can think of Don Hilbert at UTH in Zurich. Don Hilbert? Oh, Hilbert. Has been long working on this type of species how to make new... And as I said, quite some success in creating new proteins and new folds and new functions by different modes of selection. One of them is a ribosome display where the ribosomes presents a protein because they store all the translation and these proteins can be selected and then you can retrieve the ribosome and you have access directly to the nucleic acids that has included the protein you are interested in. So obviously, what you need is to have a way to link. Yes, absolutely. Of course, link proteins with RNA. Exactly. So that is done by ribosome display. But again, the question is how robust this mechanism is. You want to make it very simple and robust. For example, how fast you can make antibody, for example. Can make outside of the body, have infection and your immune system doesn't do that. What has been difficult is that you need a fold that is stable and to mutate only specific place in that fold. Yes. There are other groups, like a protein group who has been working on this problem and they have developed what they call a darking which is a protein that is stable in the cytoplasm. Antibodies are generally stable on the outside of the cell because they need disoptic regions. So a scaffold that is stable inside the cell and the difficulty is to mutate only specific positions. But then they make huge libraries that you can screen for. The question is can you imagine principle that can be used as an artificial immune system? This is currently being used to develop highly specific with very high affinity molecules protein peptides directed against whatever peptide of your interest. But just creating the scaffold and the libraries of mutants to 15 years. But now that you have it you can use it. You can use it. Maybe the simplest system is autonomous. You can go through chemical synthesis and selection using short oligonucleotides you can generate these short oligos that have higher affinity and selectivity for any protein. So that can be used for labelling proteins but also for targeting proteins. It's just all chemical. And I guess it can be very efficient. I think this is an immune system but if you learn something if you want to produce new antibiotics without knowing what you want to do it will be not so easy. And selection in bacteria in principle you know resistance to antibiotic development is fast. And the same way you can probably if you have well organized evolution make new antibiotics against a particular bacterium. This is what I'm saying. The immune system is very special of course it's still very shape very well kind of defined problem. You have to fit. But if you want to do something with some biological function you don't know how it's being chemically implemented. Then you may have the broad spectrum of you have collection of huge collection of viral particle of different kind of viral or plasmid mutation which will be on a really on a 10 to the 20 scale. Then what you need is a huge collection of enzymes because what you want is to create small molecules of antibiotic function. Yes. And for that you need to... But this exactly have a large collection of enzymes that you can combine in different ways. I don't think that this is being done. You can do synthetic biology and use a very small molecule that mimics the Zimadic activity. No, but this can be produced so easily on such a large scale and being modified so that how? In the same system how to do it? The bacteria is a machine which automatically can do that. The question is already bacteria is a machine which does many things if you properly control it. You don't have to think about that. If you work at every of them it will be very long. The question is to make it once and forever and then it will do very fast. The problem in evolution is that you need to have good results, right? Yes, exactly. And the question is how do you... Yes, exactly. That's the point. How to make good tricks to go to a desirable direction. Exactly. But sometimes it is, sometimes not. So it's easier to make resistance for antibiotics. Either you die or not. But this you have to do for other environments. That's my question. What are schemes to creating these kind of artificial rewards for bacteria or for viruses. And that's exactly the problem. Once you have it, you have tremendous possibilities. But again, the point is mathematically if you have little something, sufficient bias, you can amplify it logically with repetition and become very strong. The point is mathematically you don't need much. You're little bit in one direction and then you'll start working. The question is to have enough biological information to start thinking about mathematically and then hopefully implement it biologically. How far is your general idea away from the idea of molecular machines? Yeah, because they're biological, not molecular machines, they do machines. Yeah, and this is exactly what the Idleman and these people afterwards were trying to do. But they have obvious limitations because they're not as adaptable. You have to change them yourself. But here, by selection mechanism, they will be adaptable. You know, in a machine, you have to do it. And this thing will do it themselves. By viruses and bacteria, they are molecular machines. You only have to adjust them. That's the whole point. Of course, molecular machines, they're ready. The point is they're there in your hands. I mean, you have to do the subject of them. Very powerful, very versatile machines and just to learn enough points to make them do what they want in a controlled way. No, you just say, oh no, if you have some virus, if you go in but you can control and kill the cell, they inject the DNA and they become just a plasmid there in function of plasmid. And you may be controlled when you want to go outside if you don't like this bacteria or something. Of course, this can be done. I think many things just now we can do anything imaginable, locally in biology. The question is how to make out of them properly and build up a machine. I think all the ingredients now with CRISPR, you can control things fantastically. The question is how to use it. And amplification is the key word. You can amplify things which you cannot do otherwise. You can beat your intuition in any experience. You have an expansion, it's smarter than we can do anything. What you may want to think about is naturally systems evolved because of our positive reinforcement but also because of negative reinforcement which means stress. As a matter of fact, inducing a mutation due to stress is a less harmful way maybe later on for applications than by reward. Because the major reward... No, reward is survival. The only reward is survival. With certain ways you make certain group survival faster and then move it in a desirable direction. And of course you have to do it with ways. You don't kill everybody but just create some gradient of survival and then search it exactly as you do in a neural network cycle after cycle, there are other ways. And this is a mathematical... Again, we don't know by the way how it works. It's extremely efficient but for no unknown reason. But also in evolution we don't know why it works so efficiently. We don't care in a way. We just want to exploit it. That's the point. So we came up with one suggestion which is to create... and there are two of them libraries either of small proteins or of small DNAs synthetically and then just test them and come up with a good... antibiotics which we already do it. You want to have the bacteria do it. No, no, we can use this library. This library can be incorporated into plasmids and then use. I don't think that you heard that you need a reward in order to... if you want to go through a bacteria reward system. And the reward system that you can do is by the environment. For example, it depends on what you want. So if you want antibiotics you surround your library of random... No, no, but you want to kill... No, no, but it's very easy. But look, look, look. You want to kill particular bacteria. You control the situation. When the concentration doesn't go down you kill them. The only reward is death. The only reward is life or death. It's very easy. Certainly. I understand it. No, no, but how do they do it in practice? Of course it is trivial. Of course, if it's an environment to add an agent which kills, doesn't kill the bacteria. That's easy. Of course you can do that. This already not so bad. The question is still how to make the system of this movable plasmids easily controllable. So here they're active. Here they're not active. How to make bacteria mutate in a particular locus and mutate in a nice way. Not just by noise, but inserting certain interesting sequences in your proteins. You can induce, just increase mutagenesis. No, but mutagenesis. What kind of mutagenesis? You don't want, like this polymerase do, just making a noise. You can use it at some point. You want to insert some pieces of proteins which may be useful. You may have library of those and insert them there. You have library of millions of plasmids, in which you can be brought on random or random, because if you have 10 to the 16 of bacteria, you can easily insert the billions of different plasmids, which you can prepare beforehand, of various properties and in different combinations. Experimentally, you have to do experiments. You cannot do it partly prepared and do experiments. The question is if you can create such a system, such a project. I think it's a big thing to try to do, but I think it's great fun to do it. I think you can start with small numbers, like small number of plasmids or whatever, and provide a perfect principle. I think for a small number like 3, 5, it wouldn't be a big deal, because you can have different agents that elevate, activate one, and all of that. Now, if this works, then it will take a long time to build up some kind of huge amount of agents and be huge library, but it may take a long time. Yeah, of course it will take. The first stage will take a long time. The whole point will take me 10, 15 years to create the system, but then it will work very fast. Like computer you have to invent the right system. I think to start with, you need to do... Yeah, but I'm not well, I don't know how to start and what to do. Yeah, this is the whole point that I'm asking you. How can you imagine such a project? As a mathematician, you can only kind of listen, say it words, you don't know how to do that. There are so many things you have to know. So you know the bacteria which are used to bleed cancer, BCC Yeah. Because they use glycoses, yeah, because... They're about 50 years old, they still do it. They try to attack the cell which use glycoses, right, I guess. Right, and of course, they say there's a lot of possibilities to make it more efficient. To breathe this bacteria to kill all cell which use glycoses but then select them to kill all the cancer cells. But this is the most sophisticated level. I don't want to go on this one kind of science fiction. So I'm trying to make it more or less realistic, yeah. But this, of course, ultimately what you want to do, yeah. So you start with something that is already useful. Yes, absolutely. And then modify it, modify it in this way, exactly. And make it much more better tuned. That's sure. That would be very close. This, of course, one of the inspirations, there are bacteria which select like some of the glycoses that they invite actually immune system, yeah. They don't kill the cells directly, I guess, yeah. But that's certainly fun to pursue this, yeah. But this again, from a mathematician point of view, there are this really mathematical kind of idea there. So it may be. Perhaps where your idea could what would be important in your idea what we are doing currently a lot is simply trying to develop individual lineages. Right. That, for example, are resistant to an antibiotic. Right. And the reward there is very clear because the antibiotic attacks the bacteria directly. But there we are having a slow evolution because we are actually selecting on individual clones and not on a population that has a population that's better. Yeah. And if you had a way to have a system that has and ends for a change of DNA, so this is where with recombination where the different bacteria can exchange. Yes. Among themselves. And where the reward is not at the level of the individual clone but is at the reward of the population. Then you would probably be able to evolve things much faster. Exactly. This is exactly to find this fast evolution scheme. Yeah. That would be the closest parallel to the deep learning where it's level. Right. Right. But then you have to and computers have to weigh simple chemical control directing in this way. It was prepared for that. There was a big key to stop this process or start this process start to exchange faster, smaller, whatever or slower. But this you have to make already some kind of organisms controllable in this way and then using their populations so we can do it really play as if it's a computer. And then we can make experiments and just do it. We'll make experiments with different algorithms with a computer first. It's a very primitive thing. But one way is to measure the environment so you want to try to find a bacterium that is able to change the environment. For example the great the molecule that you have put in. Absolutely. This was the first problem. Yeah. And then what you need to know is to measure the environment and to respond to the changes in environment by adding things. For example, for bacteria we'll be giving glucose only when your concentration starts to go down. Right. We can do this. Of course it's done by technology. People do this a lot. And you need to make sure that the reward is not at the level of the clothe but it's at the level of the population. Level population, right. No, but the point again, this was done a lot in technology. People do this all the time. The question is to create particular kind of bacterium who will do it much faster than usual bacteria. The whole population organize it and control. So you have to make some big preparation. We'll design, engineer some kind of DNA some bacteria first. And then start doing that. Well, the best is you did not have to. And the best way to perhaps do not have to do that is to have it change have a system where you have you have a number of plasmids that can exchange between bacteria. Yes. And so if you have two populations of bacteria one that is highly metagenic and the other one that is good at expressing the plasmid. Right. And have them exchange. Right. The highly metagenic will be unstable, right. Eventually. It may very well die. So we want to make it control at least whether it's metagenic or not. You could do that by having two populations. You have to make it in response to the selection and the other one that does the metagenics. Yeah, this. Okay. No, just I cannot say but I imagine because they exchange the DNA they may exchange by the wrong genus that all metagenic. Right. So you want to still to have some the control on the chemical control properties of this bacteria. What you say is you allow too much freedom. My idea to make it feel like computer. So they do what you want to do. They're not do exactly they don't have this freedom being or not being metagenic. You see. Then we really will be much better control what you do. But what you say is kind of more close to real life. I want a more artificial situation. No, you'll be far away from real life because what we will do is that we will have a decision of labor. Yes. Of course, I understand the idea. No, no. I think it's maybe first step but then you have to more sophisticated and more control. But this of course how it can work. We have population of different kind of bacteria and they exchange at least plasmids and this metagenic is not but this plasmid must be mutated here then go there and they don't mutate anymore. But then you have to know how to make balance from one to another when you stop it. The trick would be to make sure that your selection is on the population of the bacteria. On one part of the population of the bacteria. Certainly. But then you have to get some of this and this eat. You give the lactose and this glucose and you can balance this something like that. Of course, but this is what you have to imagine that and just then try to make experiments. At the end result you want to be kind of the simple for use and very efficient, right? Something. Why is this a mathematical idea? I don't understand why. Well, because you think mathematically what is a system is amplification. If you try to make a measurement dynamical system, there is amplification. You have no way to speak about that. And if you describe it in general terms then you can imagine how it will behave in this situation. Of course, mathematical in the way it's a structure which can be potentially described in general terms. PCI is a mathematical phenomenon. It's a shame, but in addition it didn't invent it. I think absolutely. Someone who never published a paper invented it. That's for sure. But even then when Malish came up with that he was very unhappy exactly because it was mathematical. It was not biologists or simple. I think it was very annoyed. It was mathematical in nature. No, no, just the whole idea of evolution of Darwin, I think he was a mathematician he couldn't make multiplication table but he was not really a great biologist but he had mathematical intuition. Mendel. Mendel, yeah, another story. He was biologist, he used mathematics but Darwin was a philosopher who used mathematics and was so successful despite the fact he was a very poet biologist actually. Very nice experiments in biology. Darwin? No, he had a very absurd idea about inheritance and the blood insisting on that. He'd made experiments later on. No, of course he was not an Egypt. But his experiments on plants on plants. It was later in life actually and life was in plants. And this was good, yeah. In plants I think it was very interesting. But in evolution what he thought about inheritance was biologically I think not very good. But mathematics was a fantastic idea. You can scheme everything by big numbers. It is an exponential function and this was why it's a mathematics. It is an exponential function. When you have these numbers like 10 to the 18 or something of viruses in the environment I mean it's a big number. Everything can happen there. You cannot predict it. All your intuition goes to hell. It's big numbers. And this is a which you can want to exploit them without understanding them. You're asking a question which Darwin didn't solve, right? Darwin taught us about variation and selection. We thought a lot about selection but the structure of variation is still a mystery. And your question is exactly... But now we have to control variation. Yeah, you have to some control of variation. It will be not noise, not a point of substitution, but actually big chunks of DNA put in and out which is certainly by far more efficient, right? It certainly may be lethal with higher probability which is good so we don't have misfeeds. But when it relaxes, it relaxes very well. And then, of course, there may be point adjustment. This was actually a mistake by Darwin. He believed only such how to speak in point mutation. He never believed in a large mutation which actually was probably crucial in evolution. That's the mystery that some call evolvability, right? How come variation happens on some sort of a low-dimensional manifold which allows for selection? Yeah, but this actually... I think it's very much mathematical issue which we don't understand there. It is we don't understand how it works. And this we will not understand with bacteria as we don't understand neural networks. It's not the issue to understand. It's mathematical structure and you can be guided by some intuition, mathematical. But we don't understand rigorously what's okay. We have to live with this. It's a metaphoric answer with some poetic license. It's modularity. We understand that... Yes, modularity is a factor. Modularity will appear in life everywhere, even in the universe. You have classes of a matter up to some point where gravitation acts. And then it disappears. It's not universal. So everywhere it works for some reason. Otherwise, on a structure you have rather chaotic things where only kind of... not everybody can make life out of them. We have homogeneous structures. It's heterogeneity. It's kind of organized by modularity, clusterization. Otherwise it wouldn't work. That's why it's so. But exactly how it works we don't know. That's probably why your imaginary immune evolvable bacterial immune system doesn't exist. We know how to create... how to introduce mutability in the future. No, no, with the point it will have your purpose, not their purpose. That's the whole point. Actually it was that when we were saying, show me an organism which serves somebody else's purpose and I said, of course you can do it. It was easier to say. But this exactly came to all. Because it will serve your purpose not good for the system, good for you. Because you adjust the final result and modify it all the time. But machinery all the way there we don't use it enough. All the bacterial machinery. As a mathematician it's exciting to have it without even understanding detail, understand how it can combine them. Understand and again only experimentally. You cannot imagine it. You can only roughly imagine it. And the schemes of use it may be rather general. How you make certain algorithm how you make computers which actually what people were trying to do in molecular machine they imitated the scheme of computers which I think is not right. Because they're not adaptive to the system. You have to really go along with what you know what happens in biology. And then come with the result. So I have nothing to say. I was only asking. So we can make this I think artificial evolution what you said we are pursuing and talking to the right people. So if you want to continue with this subject one lesson from success if you perceive this as a success of deep learning is that it's not really about deep learning it's about deep teaching. The secret if you ask me is in the sort of incremental construction of the system. So then in designing the kind of system that you're asking for I guess the secret would be not to have selection of the particular but incremental. Cascade, absolutely. Cascade with selection. Cascade with selection. Right, on population between cascade and which is organized in cascade is exactly it has an amplifying power of structuralization of what happens. But this again I have to think about that and just try to make it. And it's not easy for biologists to invent simple indeed model problem. Model situation which actually can be done and see how it works. But you have to... You may want to talk to Diethard Hautz on this. He's the director of Max Planck Institute for Evolution Biology and his research is exactly what you're asking. Sure, exactly why I'm asking the audience where it happens and how to bring this together and bring these people together and just discuss. It's an exciting issue. Because it works on molecular evolution genetic basis of adapting adaptation processes I think it's the direction. One point you studied but I think it's fun to make it really in lab to make it physically. Of course mathematicians just only can desire that we cannot do it. Again I have nothing to say. There are some names. These names must be incorporated in the future.