Latest / Deep Tech Germany – AI, Robotics & Frontier Innovation by Startuprad.io™ / Running AI on Biological Matter: Biocomputing with Fred Jordan

Transcript

Running AI on Biological Matter: Biocomputing with Fred Jordan · Deep Tech Germany – AI, Robotics & Frontier Innovation by Startuprad.io™

• WEBVTT 00:00:05.097 --> 00:00:12.737 Welcome to StartupRed.io, your podcast and YouTube blog covering the German 00:00:12.737 --> 00:00:17.657 startup scene with news, interviews and live events. 00:00:13.040 --> 00:00:19.920 Music. 00:00:19.737 --> 00:00:24.017 Hello and welcome everybody. This is Joe from StartupRed.io, 00:00:24.237 --> 00:00:28.537 your startup podcast and YouTube blog from Germany, Austria and Switzerland. 00:00:28.537 --> 00:00:35.257 And today, I would like to welcome Fred, who is an entrepreneur from Switzerland. Grüezi. Hello. 00:00:38.157 --> 00:00:49.637 You are the CEO and co-founder of two different companies, AlpVision and FinalSpark. 00:00:50.857 --> 00:00:56.217 To get this little bit sorted out and introduce yourself, I would just ask you 00:00:56.217 --> 00:01:01.177 to introduce yourself. and then we can work on the story why we're definitely 00:01:01.177 --> 00:01:04.077 here, a future technology called biocomputing. 00:01:04.157 --> 00:01:06.577 But first, let's tell your story. 00:01:07.677 --> 00:01:13.117 Well, you know, I'm a French physicist, actually, engineer. 00:01:13.757 --> 00:01:22.397 And I went to Switzerland about 30 years ago to make a PhD in signal processing. 00:01:23.617 --> 00:01:31.117 And then I met another guy doing PhD with me and then we create a first company called AlpVision. 00:01:31.437 --> 00:01:37.137 And this company specialized in applications of steganography, 00:01:37.717 --> 00:01:43.137 that means invisible hiding of information into digital image. 00:01:44.744 --> 00:01:52.224 Okay and we use this technology actually to detect counterfeit products using smartphones, 00:01:52.824 --> 00:01:57.004 so you take a shoe you want to know if it's authentic or 00:01:57.004 --> 00:02:01.484 fake you use our app take a picture it's going to tell you if it's fake sorry 00:02:01.484 --> 00:02:07.464 a stupid question if you're buying something on platforms that would be before 00:02:07.464 --> 00:02:12.844 ebay and now like all the competitors does it also work if you just have a picture of the product? 00:02:13.084 --> 00:02:15.144 Can you tell if it's fake or not from a picture? 00:02:15.924 --> 00:02:18.724 Well, theoretically, yes. 00:02:19.144 --> 00:02:26.464 But in practice, I would say 99% of the case, we ask our users to have the real product in hand. 00:02:28.444 --> 00:02:31.884 Okay, I see. So this was, 00:02:32.104 --> 00:02:37.284 you have to imagine, and when we started this first company with this friend 00:02:37.284 --> 00:02:41.644 and I, we were basically in a garage, okay, trying to use the theoretical findings 00:02:41.644 --> 00:02:44.224 that we made during the PhD and to make a living. 00:02:46.864 --> 00:02:51.304 Okay. And the company went and is still going very well, actually. 00:02:51.964 --> 00:02:56.464 We are protecting against counterfeiting more than 30 billion, 00:02:56.624 --> 00:03:01.004 30 billion of products each year. Okay. So it's really huge. 00:03:01.344 --> 00:03:07.824 You being in Switzerland, I would have an assumption that stuff you would protect 00:03:07.824 --> 00:03:12.164 would include, of course, expensive wristwatches. Yes, they may indeed. 00:03:12.684 --> 00:03:17.664 But, you know, confidentiality and discretion is a Swiss value. 00:03:17.924 --> 00:03:25.124 So we don't talk about what we protect, but the technology is provided to the 00:03:25.124 --> 00:03:29.164 consumers and nobody knows that it's us behind the scene. 00:03:30.359 --> 00:03:33.799 So we work, we make money with the brand owners that pay us actually. 00:03:35.039 --> 00:03:40.639 And so that was an interesting journey, okay? Trying to make money out of some 00:03:40.639 --> 00:03:42.479 theoretical mathematical findings, okay? 00:03:43.779 --> 00:03:47.599 And the company is still working well and we have offices now in Shanghai, 00:03:47.859 --> 00:03:49.299 in New York. So it's really cool. 00:03:49.679 --> 00:03:54.059 And in 2014, we said, okay, this company is going well. Let's make another one. 00:03:56.979 --> 00:04:00.239 When we run a company, it's not enough. It's not enough stress, right? 00:04:01.559 --> 00:04:06.619 Yeah, but the thing here is that we were lacking a little bit the R&D, 00:04:06.739 --> 00:04:12.639 the fundamental research part, which we love, the co-founder and myself. 00:04:12.879 --> 00:04:18.419 I said, OK, let's do something incredible in AI, artificial intelligence. 00:04:18.739 --> 00:04:25.219 Let's follow a path that nobody has followed before or seriously followed before. Okay. 00:04:25.719 --> 00:04:30.219 And that was in 2014. And we created FinalSpark. And this was, 00:04:30.519 --> 00:04:33.659 again, two of us in parallel with the first company. 00:04:34.099 --> 00:04:37.539 One question, Fred, doing a big shout out to your co-founder. 00:04:37.719 --> 00:04:41.459 What's his name? So he's Martin Kutter. 00:04:42.459 --> 00:04:48.559 And he's a Swiss German. And so we are actually fundamentally quite different 00:04:48.559 --> 00:04:50.579 because I'm French, he's Swiss German. 00:04:52.899 --> 00:04:57.899 But that makes actually a good team. So we've been working together for, 00:04:57.899 --> 00:04:59.439 I don't know, 25 years now. 00:04:59.719 --> 00:05:03.219 I totally assume you can both agree on fondue. 00:05:03.979 --> 00:05:07.679 Yes, yes. But this is mandatory. 00:05:07.979 --> 00:05:12.359 Otherwise, you know, federal state will take your passport. 00:05:17.099 --> 00:05:24.819 Okay. Okay. And so, basically, 2014, you found what is today FinalSpark. 00:05:24.999 --> 00:05:30.339 Can you give us a little bit introduction in the big picture and vision? 00:05:30.559 --> 00:05:37.039 How on earth did you bump into biocomputing? Because my understanding is you like physics. 00:05:37.259 --> 00:05:42.879 So, I would assume you fancy big machines, you like them. 00:05:43.139 --> 00:05:45.239 How did you end up at neurons? 00:05:46.079 --> 00:05:48.959 Honestly, this was not the plan. 00:05:51.199 --> 00:05:56.619 Really not. As you said, our expertise was in signal processing, 00:05:57.199 --> 00:06:02.799 physics, so mathematics, a lot of mathematics and programming and things like this. 00:06:03.279 --> 00:06:11.539 And so we started to go into AI by exploring new paths, but purely digital, 00:06:12.079 --> 00:06:14.079 like genetic programming. 00:06:14.079 --> 00:06:21.199 So genetic programming is that you are basically creating random source code, 00:06:21.807 --> 00:06:27.467 random source code, okay, that you compile and execute in the hope that it's 00:06:27.467 --> 00:06:28.907 going to do what you want. 00:06:29.047 --> 00:06:33.807 If you have any demand for that, I have a two-year-old here who's very enthusiastic 00:06:33.807 --> 00:06:35.467 about hitting the keyboard. 00:06:38.087 --> 00:06:44.067 For this to work, you have to create actually billions of programs of source 00:06:44.067 --> 00:06:47.607 code and execute them. As I said, very enthusiastic. 00:06:50.487 --> 00:06:56.187 And we used genetic algorithms approaches for this when you actually try to, 00:06:56.427 --> 00:07:00.427 you said that my source code is an individual, I'm going to make some crossover 00:07:00.427 --> 00:07:03.787 with another individual, which is another source code, and execute it. 00:07:04.167 --> 00:07:10.587 And so, okay. It's a very strange way of doing AI, and this is what we loved, okay? 00:07:10.767 --> 00:07:13.487 We were only looking at strange things. 00:07:14.127 --> 00:07:17.467 And another strange thing, so we are testing a lot of things. 00:07:17.627 --> 00:07:21.067 And something else that we started to test was spiking neurons. 00:07:22.307 --> 00:07:29.107 So, I don't know, in the field of AI, normally people do not use so-called spiking neurons. 00:07:29.347 --> 00:07:34.447 Normally, in AI, you use artificial neurons, which are much simpler than this. 00:07:34.827 --> 00:07:41.027 You accumulate some values weighted by some other values, and you output another 00:07:41.027 --> 00:07:43.047 values given a threshold function. 00:07:43.487 --> 00:07:49.327 Basically, that's it. This is all, JGPD, everything is working with this. So simple. 00:07:52.527 --> 00:07:56.047 And we said, okay, we're not going to use this model. We are going to use a 00:07:56.047 --> 00:08:01.267 more sophisticated model, which is more realistic, still entirely digital. Okay. 00:08:02.227 --> 00:08:06.787 And that means it's a very well-known model known for 30 years ago. 00:08:07.007 --> 00:08:13.827 Okay. which basically is saying that this is a temple activation through time 00:08:13.827 --> 00:08:18.907 that characterizes the activity of a single neuron. It's a spike, okay? 00:08:19.647 --> 00:08:26.047 An action potential, actually. So you can also have some simulation of artificial 00:08:26.047 --> 00:08:28.547 neurons which are spiking through time. 00:08:28.727 --> 00:08:32.867 You can also create networks and do some learning. And we spent a number of 00:08:32.867 --> 00:08:38.847 years on this kind of model, also because one of the leaders of this field are 00:08:38.847 --> 00:08:42.207 actually in the Swiss Federal Institute of Technology, where we made our PhD. 00:08:42.467 --> 00:08:45.487 So there was some cultural connection. 00:08:46.247 --> 00:08:53.387 And after a few years, what we had was a few hundreds, hundreds of simulated 00:08:53.387 --> 00:08:55.907 wonderful spiking neurons, 00:08:57.107 --> 00:09:01.607 consuming about five kilowatts of power. 00:09:02.887 --> 00:09:08.987 So 5 kilowatts for 100 neurons I don't know if you put this in perspective with 00:09:08.987 --> 00:09:15.307 your brain it's 100 billion of neurons and it's not 5 kilowatts it's 20 watts. 00:09:19.006 --> 00:09:23.906 So then we really realized that, at this point, we could not scale this. 00:09:24.626 --> 00:09:28.806 Impossible, OK? If you already consume kilowatts with 100 neurons, 00:09:28.926 --> 00:09:32.266 you're basically hitting the limit of your model. 00:09:33.406 --> 00:09:37.146 Nearly no chance, OK? And then something else happened. 00:09:37.426 --> 00:09:41.566 We had a third guy who joined us at this point. 00:09:41.686 --> 00:09:46.686 And he was doing the military service, which you may know is mandatory in Switzerland. 00:09:47.426 --> 00:09:51.486 At the Swiss Federal Institute of Technology. Okay. 00:09:51.986 --> 00:09:58.346 And he was working in the lab of Professor Markram on the Human Brain Project. 00:09:59.606 --> 00:10:05.466 And this project, it's all about living neurons, not simulation, okay? 00:10:05.746 --> 00:10:09.886 But because it was out of our field, we were not looking at biology. 00:10:10.146 --> 00:10:15.086 It wasn't anything about biology, but he had to work there, okay? 00:10:15.086 --> 00:10:18.146 So he was working and measuring real living neurons. 00:10:18.966 --> 00:10:22.426 And of course, we were chatting together. And at some point, 00:10:22.846 --> 00:10:24.626 the idea came a bit spontaneously. 00:10:25.006 --> 00:10:29.206 Like he said, they are consuming almost nothing, these neurons. 00:10:29.806 --> 00:10:34.126 And on our side, it was the opposite. Our simulations are consuming way too much. 00:10:35.626 --> 00:10:41.146 The idea came at this point. We said, OK, instead of trying to simulate neurons 00:10:41.146 --> 00:10:46.446 on artificial networks, Next, let's try to use living neurons. 00:10:47.166 --> 00:10:53.506 When you just speak about living neurons, what do people have to picture? 00:10:54.026 --> 00:11:00.486 Because when you talk about living neurons as a chip, what I have in mind is 00:11:00.486 --> 00:11:04.486 like a silicon waver with some gray matter on it. 00:11:04.746 --> 00:11:08.786 Where do you get the neurons and how does it actually work? Well, 00:11:08.866 --> 00:11:11.926 what you have in mind is not so far from the reality, surprisingly. 00:11:16.746 --> 00:11:22.586 The thing is that most of the time, you cannot get neurons directly. 00:11:22.926 --> 00:11:27.646 Particularly, if you talk about human neurons, it's almost impossible to get 00:11:27.646 --> 00:11:29.786 what is called primary cells. 00:11:29.986 --> 00:11:34.006 Primary cells are really cells extracting from a living body. 00:11:34.206 --> 00:11:40.326 So you can get primary cells for mouse or rats, but for human, no. 00:11:41.166 --> 00:11:44.066 So there is something else. 00:11:44.146 --> 00:11:50.566 Another invention that came into play here is a Nobel Prize called Professor Yamanaka in Kyoto. 00:11:51.266 --> 00:11:59.686 He invented a way to take some cells of your skin and convert them into stem cells. 00:12:01.730 --> 00:12:05.690 Ah, and then from stem cells, you can go to neurons again? Yeah. 00:12:07.690 --> 00:12:13.070 Oh, yeah. Okay, I see. And this is what we do in the lab. 00:12:14.350 --> 00:12:21.370 Just five minutes before this interview, I was in the lab doing some experiments 00:12:21.370 --> 00:12:23.850 using these stem cells, actually. 00:12:24.210 --> 00:12:30.270 They are called induced pluripotent stem cells because we have induced the pluripotency 00:12:30.270 --> 00:12:34.330 because if you take skin cells that are not pre-repotent at all, 00:12:35.170 --> 00:12:37.330 the skin cell can only become a skin cell. 00:12:37.890 --> 00:12:41.470 It will never become a neuron. It's dead for them, okay? It's over. 00:12:41.650 --> 00:12:42.710 They have one thing to do. 00:12:43.010 --> 00:12:49.490 But if you induce the pre-repotency, then you can again create whatever cells 00:12:49.490 --> 00:12:50.630 you want, including neurons. 00:12:50.750 --> 00:12:53.750 And this is what I was working on five minutes ago. 00:12:55.490 --> 00:13:07.050 Ah, I see. And how can you put this biological matter into a working computer? 00:13:07.410 --> 00:13:15.270 How do you work with the connection? Because I do assume you have them arranged them in some matter. 00:13:15.430 --> 00:13:22.990 You cannot connect like every single neuron to its own very tiny electric circle or something. 00:13:23.770 --> 00:13:29.650 Yes just before i answer i will make a remark to you as a human being yes go 00:13:29.650 --> 00:13:34.510 ahead most of your neurons are not connected to your sensors, 00:13:35.833 --> 00:13:42.613 Okay. Just to be clear on this, which means in practice that what we do is that 00:13:42.613 --> 00:13:45.813 we are not first playing directly with neurons. 00:13:45.993 --> 00:13:50.553 We are creating something a bit more sophisticated, which is called organoid. 00:13:51.253 --> 00:13:57.853 So an organoid is a collection of living neurons connected together that creates like a small organ. 00:13:59.293 --> 00:14:03.493 That is going to be, in our case, a small ball of half a millimeter. 00:14:03.493 --> 00:14:08.033 And to answer your specific question, we are going to put it, 00:14:08.273 --> 00:14:12.073 literally put it, deposit this ball on electrodes. 00:14:12.553 --> 00:14:16.853 And then we can use the electrodes to receive and send information. 00:14:17.393 --> 00:14:23.133 And as you correctly noticed, we are going to only discuss with a few neurons, 00:14:23.193 --> 00:14:26.373 which will be the interface with the rest of the brain organoid. 00:14:28.093 --> 00:14:34.973 I see. And the big advantage that you've already hinted in the beginning is 00:14:34.973 --> 00:14:43.313 that it uses really much less energy than a normal silicon waiver would do. 00:14:44.013 --> 00:14:46.733 Yes. Talk about a factor of one million. 00:14:47.773 --> 00:14:52.413 If you want to see some publications on this, or the publication for me is the 00:14:52.413 --> 00:14:57.933 one of Professor Artun, which I guess was done in the journal Fronties two years ago. 00:14:57.933 --> 00:15:04.253 You can google Hartung Frontiers what you can do you can provide the link to 00:15:04.253 --> 00:15:08.813 me after the recording and I'll link it in our blog post yeah perfect yeah because 00:15:08.813 --> 00:15:14.413 it's available for free so okay so you can get this so and indeed you know, 00:15:15.443 --> 00:15:19.923 Looking again at your brain, if I wanted to simulate your brain, 00:15:20.103 --> 00:15:24.203 I would need, by today's standard, a small nuclear power. 00:15:24.723 --> 00:15:30.523 Okay. Yeah, it sounds like a lot of energy. That explains why I eat such a lot of chocolate. 00:15:31.583 --> 00:15:36.503 No, because with all the chocolate you eat, you are still consuming 20 watts 00:15:36.503 --> 00:15:41.463 of power to run your incredible brain with 100 billion of neurons and 10,000 00:15:41.463 --> 00:15:42.763 connections per neuron. 00:15:42.763 --> 00:15:49.483 You know, nature has to be extremely efficient. 00:15:49.823 --> 00:15:57.183 And nature optimized nervous tissues for 300 millions of years to be energy efficient. 00:15:57.783 --> 00:16:03.943 So this is the benefit of and the good thing of using existing products of nature 00:16:03.943 --> 00:16:08.663 is that it's already there and it's already really, really optimized with a 00:16:08.663 --> 00:16:11.223 factor of 1 million better than what we can do in silicon. 00:16:11.223 --> 00:16:15.283 And let me mention that I'm a physicist, okay? 00:16:15.563 --> 00:16:19.923 So I love the integrated circuits and transistors and quantum mechanics. 00:16:19.943 --> 00:16:21.623 This is the things I know, okay? 00:16:21.923 --> 00:16:29.403 And I was not so happy to go in the biology direction because I didn't know anything about it. 00:16:29.483 --> 00:16:35.943 So if I could have avoided, I would have avoided it because I had to learn everything from scratch. 00:16:36.443 --> 00:16:43.183 So it's really as an engineer, you have, I would not say the responsibility. This is a too big word. 00:16:43.883 --> 00:16:50.523 You have to have the rationality as an engineer to take the best solution on the market, I would say. 00:16:50.943 --> 00:16:55.023 And the best solution here is not silicon, is living neurons, period. 00:16:56.243 --> 00:17:00.803 Extrapolating a little bit out into the future, what do you think could. 00:17:02.246 --> 00:17:09.026 Make, giving like all the big investments that we've seen just recently with 00:17:09.026 --> 00:17:11.866 a 500 billion into infrastructure for AI. 00:17:12.146 --> 00:17:16.506 What kind of difference could biocomputing make there? 00:17:17.166 --> 00:17:21.246 Biocomputing is going to open an entirely new industry. Okay. 00:17:22.006 --> 00:17:26.686 The most obvious application is that you talk about these investments, 00:17:27.666 --> 00:17:33.366 but also all investments in servers and the power conception that they represent. 00:17:33.646 --> 00:17:38.426 This power conception is converted into electricity, which is converted into $2. 00:17:39.786 --> 00:17:43.626 If you come up with a server like the Amazon Web Services, for instance, 00:17:44.166 --> 00:17:46.546 which consumes 100 times less, 00:17:48.246 --> 00:17:53.106 knowing that this is the primary source of recurring cost of these servers, 00:17:54.486 --> 00:17:58.086 the market is going to be huge. It's just obvious. 00:17:58.906 --> 00:18:07.366 And the question is, I would assume since it is not like serial production level 00:18:07.366 --> 00:18:09.086 you're working on right now, 00:18:09.246 --> 00:18:14.666 I would assume the production costs are currently pretty high and the energy 00:18:14.666 --> 00:18:17.226 consumption in the future is pretty low. 00:18:17.226 --> 00:18:22.626 And when you get to the point where also the production costs kind of matches 00:18:22.626 --> 00:18:27.886 the silicon costs, that's where it gets really, really competitive. 00:18:29.906 --> 00:18:36.646 Yes but you have to realize that the production cost is going to be incredibly 00:18:36.646 --> 00:18:44.086 quickly lower than any artificial device incredibly last week, 00:18:45.637 --> 00:18:52.777 I mean, in the lab, we have created, I guess, a few millions of neurons. Only last week. Okay. 00:18:53.897 --> 00:19:00.497 And the thing is that it's almost for free. I mean, when you have stem cells, they just multiply. 00:19:01.417 --> 00:19:06.157 24-7, you get more and more and more. And you talk about production. 00:19:06.957 --> 00:19:11.957 But it's natural. It's automatic. It's made for reproduction. production. 00:19:12.437 --> 00:19:15.737 So the problem that we have sometimes is that we have too much. 00:19:16.237 --> 00:19:21.437 So it's not going to be really a cost of production. 00:19:21.677 --> 00:19:24.777 There are going to be other costs, but not on this side. 00:19:26.977 --> 00:19:36.257 I see. So I do believe we now do have an understanding of what you guys are 00:19:36.257 --> 00:19:39.517 doing to replace silicon wavers. 00:19:39.717 --> 00:19:42.977 And just for my understanding and the understanding of the audience, 00:19:43.637 --> 00:19:52.697 would your biocomputing those balls there, which I picture as kind of very small bioreactor, 00:19:53.037 --> 00:20:01.257 would they in a computer completely replace the CPU or would you work alongside 00:20:01.257 --> 00:20:06.077 the CPU and all the other conventional pieces of a computer? 00:20:06.857 --> 00:20:10.197 Yes, you're right, alongside is the right keyword. 00:20:10.897 --> 00:20:17.377 As I guess, quantum computers are going to work one day alongside biocomputers. 00:20:17.797 --> 00:20:25.317 Like DSP, digital signal processors, work alongside CPU, and GPU work alongside CPU. 00:20:27.017 --> 00:20:30.157 So this is going to get a little bit more complex. 00:20:30.557 --> 00:20:35.157 But I can tell you what is not going to happen with biocomputers. 00:20:36.037 --> 00:20:38.557 Bioprocessors are not going to run Windows 11. 00:20:40.011 --> 00:20:41.951 They are not made for this. 00:20:44.531 --> 00:20:50.971 Biocomputers are very well done to run any task which is done for AI today. 00:20:52.351 --> 00:20:58.931 Because since AI is simulation of neurons, you can also use the real one. 00:20:59.391 --> 00:21:03.371 In many applications, this works. Not in all applications, but in many applications, 00:21:03.531 --> 00:21:05.491 it works. Particularly in generative AI. 00:21:06.571 --> 00:21:13.351 I see. I see. And that's basically with the big models talking about JetGPT, 00:21:14.111 --> 00:21:19.031 Gemini, DeepSeek, that is currently what the cutting edge of AI is. 00:21:19.191 --> 00:21:20.951 Yes, absolutely. It's a very good example. 00:21:21.471 --> 00:21:31.151 I see. I see. And now can you tell us a little bit more about what FinalSpark 00:21:31.151 --> 00:21:35.271 does in this whole future potential industry? 00:21:35.271 --> 00:21:38.311 Well, we want to be the first one to create a bioprocessor. 00:21:41.971 --> 00:21:46.191 And now you have to consider that this is fundamental research at this point. 00:21:47.691 --> 00:21:55.071 Let me tell you what is the main point of research. The main point of research is exactly the same. 00:21:56.737 --> 00:22:01.677 Not surprisingly. Then the main point of research was 30 years ago with AI. 00:22:03.117 --> 00:22:06.857 Do you know what was the main point of research 30 years ago in AI? 00:22:08.537 --> 00:22:15.417 It was about training. We were able, because I've been working in artificial 00:22:15.417 --> 00:22:21.797 neural networks 30 years ago, and we created networks of artificial neurons. 00:22:21.797 --> 00:22:28.037 And the problem was, how do you tune the connections between the neurons so 00:22:28.037 --> 00:22:31.517 that you get the output that you want for the given input? 00:22:31.817 --> 00:22:33.637 This is called training. 00:22:36.637 --> 00:22:45.537 If you have thousands or millions of connections to tune, it's impossible to do in practice. 00:22:47.037 --> 00:22:53.077 But someone came with a solution, which is called the backpropagation which 00:22:53.077 --> 00:22:58.477 is basically an algorithm or a mathematical approach which gives you the partial 00:22:58.477 --> 00:23:05.557 derivative of the weight of the connection in respect of the error the network creates. 00:23:06.337 --> 00:23:09.577 And with the help of this mathematical tool, 00:23:12.137 --> 00:23:18.077 we are able now to know exactly how to connect how to change the connections 00:23:18.077 --> 00:23:21.417 to have a learning platform. 00:23:21.637 --> 00:23:26.917 So when you talk, when you hear about machine learning, it is this learning. Okay. 00:23:27.617 --> 00:23:33.077 So this, finding the solution made a big difference, made the difference actually 00:23:33.077 --> 00:23:37.357 between non-working AI and working AI. It's very schematic, of course. 00:23:38.057 --> 00:23:44.397 But what is interesting is that it's the same problem today for us. We also have neurons. 00:23:45.177 --> 00:23:50.617 We also need to rewire them in a network so that when you input something, 00:23:50.937 --> 00:23:51.997 the output is meaningful. 00:23:53.797 --> 00:23:58.477 How do we do this? We don't know. Okay. 00:23:59.637 --> 00:24:03.057 No, no, but hey, let me tell you still a good news. 00:24:06.637 --> 00:24:10.157 One thing interesting, which was not the case with artificial neurons 30 years 00:24:10.157 --> 00:24:11.977 ago. We are sure it's possible. 00:24:14.454 --> 00:24:18.134 You know why? Because you are able to learn. 00:24:20.614 --> 00:24:27.074 So if a human brain can learn, the cells within the human brain should also be able to learn, right? 00:24:27.454 --> 00:24:30.194 Correct. Particularly knowing that it's not only the human brain, 00:24:30.374 --> 00:24:34.654 but bees and insects and very primitive animals are able to learn also. 00:24:34.854 --> 00:24:36.454 So we know it's possible. 00:24:36.714 --> 00:24:43.454 We just have to find how nature does it in order to reproduce it in vitro in a way which is reliable. 00:24:44.134 --> 00:24:50.034 Oh, I see, see, see. Okay. And that's basically the part where you are. 00:24:50.654 --> 00:24:57.534 My understanding is that there are only a handful of potential competitors out there. 00:24:57.694 --> 00:25:02.114 You've been talking about like three companies globally with FinalSpark being one of them. 00:25:02.414 --> 00:25:06.254 Yeah, yeah, it's great. It's an incredible era, actually. 00:25:06.734 --> 00:25:09.954 You know, we live in a world where it's very difficult. When you wake up in 00:25:09.954 --> 00:25:13.534 the morning as an engineer, I say, I'm going to work on something when nobody 00:25:13.534 --> 00:25:16.654 thinks about it and it's going to make a revolution. 00:25:17.574 --> 00:25:19.414 How many fields can you imagine this? 00:25:20.374 --> 00:25:25.014 And I'm going to start by myself. Yes, and I can do this. And biocompeting is 00:25:25.014 --> 00:25:29.034 one of these very, very rare fields that still exists today. 00:25:29.314 --> 00:25:31.954 And indeed, you're right. We are three companies in the world. 00:25:32.354 --> 00:25:37.654 And I'm not going to lie. I would prefer that there are more of them, not less of them. 00:25:37.654 --> 00:25:43.674 Because this is going to be a revolution but we need a critical mass of companies 00:25:43.674 --> 00:25:48.934 competing together so that venture capitalists and money starts to flow into this research. 00:25:50.834 --> 00:26:04.494 I see do you have like a kind of idea when you would or you guess you would be able to. 00:26:06.685 --> 00:26:13.105 Produce on an economic industrial scale biocomputing? 00:26:13.785 --> 00:26:19.105 Just for simplification, I know it's not correct, but biocomputing CPUs that 00:26:19.105 --> 00:26:22.505 we stick to a known frame of reference. 00:26:23.345 --> 00:26:29.285 Yeah. Our business plan states that we should be able to have a first a bioserver 00:26:29.285 --> 00:26:33.245 in about 12 to 13 years, one, three. 00:26:35.305 --> 00:26:42.585 So what we target is basically cloud biocomputing. So you know about cloud computing. 00:26:43.585 --> 00:26:47.825 And cloud biocomputing is the same, but powered by bioprocessors. 00:26:49.065 --> 00:26:55.545 I see. Basically, here around Frankfurt, we have D6, the world's largest internet node. 00:26:55.765 --> 00:26:59.945 So in the area I live around here, there's a lot of very big buildings with 00:26:59.945 --> 00:27:05.765 a lot of air conditioning on top and that's basically like all the places where 00:27:05.765 --> 00:27:11.325 all the calculations are done those data centers this may be replaced, 00:27:12.305 --> 00:27:13.845 maybe one time in the future, 00:27:15.425 --> 00:27:21.425 actually for me it's very clear biocomputing is not just a simple new technology 00:27:21.425 --> 00:27:26.065 it's a new industry that is going to stay and change everything in the coming 10 to 20 years, 00:27:27.385 --> 00:27:33.845 i see um before we get into challenges and ethics let us do a little ad break. 00:27:40.265 --> 00:27:46.145 Hey guys welcome back i'm talking to fred jordan co-founder and ceo of final 00:27:46.145 --> 00:27:50.645 spark one of just three companies in the world working on bio computing where 00:27:50.645 --> 00:27:54.565 they are going to replace or work alongside the CPU, 00:27:55.105 --> 00:27:58.205 a bio-CPU that we just talked about before. 00:27:58.885 --> 00:28:05.225 I would now like to go a little bit into the challenges and ethics of biocomputing. 00:28:06.225 --> 00:28:11.205 My question would be, what are the most immediate bottlenecks in biocomputing? 00:28:11.305 --> 00:28:12.985 Is it scaling? Is it stability? 00:28:13.265 --> 00:28:16.225 Is it reproductability? 00:28:16.865 --> 00:28:18.585 Or is it just teaching? 00:28:20.486 --> 00:28:24.966 I would say, teaching. First, teaching is the big thing, OK? 00:28:25.106 --> 00:28:28.526 Really, you know, it's incredible. 00:28:28.946 --> 00:28:32.906 Because in biology, you make a difference between in vivo and in vitro. 00:28:34.106 --> 00:28:42.366 Teaching and learning, in general, has been studied in vivo for the past 50 years. 00:28:43.726 --> 00:28:46.646 Hundreds of thousands of publications on this field. 00:28:48.106 --> 00:28:52.506 However, comparatively to this, if you think about in vitro learning, 00:28:52.766 --> 00:28:56.406 that means not in a living animal, 00:28:57.366 --> 00:29:03.586 basically there is almost, I don't know, less than 10 publications. 00:29:04.466 --> 00:29:14.286 So that means in vitro learning would be you basically teaching a baby, 00:29:14.606 --> 00:29:18.026 an animal, something before it is born. 00:29:18.506 --> 00:29:22.566 It's even not a baby. It's even not an animal. 00:29:23.526 --> 00:29:29.446 When we create a brain organoid, it's just thousands of neurons connected together 00:29:29.446 --> 00:29:32.046 in a small ball. That's what it is. 00:29:33.446 --> 00:29:38.046 So it even doesn't look like a brain. It's a nervous tissue. 00:29:39.546 --> 00:29:46.986 But a nervous tissue, we believe it's reasonable that it's possible to make it learn something. 00:29:51.006 --> 00:30:01.846 I see, I see. Um, what is the upper limit to what biological neurons can compute efficiently? 00:30:02.026 --> 00:30:08.966 Or do you expect exponential progress, um, similar to something like Moore's law in silicon chips? 00:30:09.846 --> 00:30:15.846 Uh, yes and no. Um, in German, we do have a wonderful word for this. 00:30:16.066 --> 00:30:19.486 Uh, we say ja and nein, it's called ja, exactly. Exactly. 00:30:21.926 --> 00:30:28.366 No, because with computer chips, it's about artificial systems. 00:30:28.986 --> 00:30:34.886 That's so it's a question of light diffraction and how you can compensate this 00:30:34.886 --> 00:30:38.686 in order to engrave smaller and smaller circuits, OK? 00:30:40.446 --> 00:30:43.966 But here, you have to realize that we are not making anything. 00:30:45.557 --> 00:30:51.917 I am not controlling how well I am controlling very remotely, 00:30:51.917 --> 00:30:55.317 OK, how these neurons are growing. 00:30:55.777 --> 00:31:00.717 I cannot change the intrinsic, well, I can change to some extent, 00:31:00.857 --> 00:31:04.517 but I'm very limited in the number of change I can do to these neurons. 00:31:05.137 --> 00:31:10.997 For instance, the actual potential is going to propagate always more or less 00:31:10.997 --> 00:31:15.257 at the same speed in all the neurons I do, whatever I do. 00:31:17.517 --> 00:31:24.037 And I'm not going to be able to pack more neurons in the same space than they 00:31:24.037 --> 00:31:26.117 would naturally accept. 00:31:28.617 --> 00:31:33.177 A transistor is not accepting something. It's going to be a passive device, 00:31:33.177 --> 00:31:34.957 which is controlled by a human being. 00:31:35.177 --> 00:31:38.297 Here, you have to play with living things. So it's a bit different. 00:31:38.297 --> 00:31:42.017 You have to behave differently. 00:31:44.017 --> 00:31:48.637 But coming back to scalability, there is a big, big difference here. 00:31:49.557 --> 00:31:54.177 I can grow nervous tissue. Like I said at the beginning, it's half a millimeter, 00:31:54.397 --> 00:31:55.437 half a millimeter today. 00:31:56.277 --> 00:31:59.877 But tomorrow, theoretically, I could do one centimeter. 00:32:00.497 --> 00:32:05.477 I could do 10 centimeters. Actually, I could do 100 meters of nervous tissue. 00:32:05.617 --> 00:32:08.597 I could culture it. there is no limit to this, 00:32:09.957 --> 00:32:17.757 I mean a football field with a 5 cm nervous tissue is going to represent a serious 00:32:17.757 --> 00:32:21.457 amount of computational power but I will have, 00:32:22.317 --> 00:32:29.577 I will need to do nothing actually to get it's just more like when you do agriculture you know, 00:32:31.217 --> 00:32:35.937 so you grow things you grow plants and here you grow neurons So, 00:32:36.037 --> 00:32:43.277 it's more going to be in the scale of things, how big we can handle them. 00:32:43.957 --> 00:32:51.197 I was wondering at the moment you were talking about this, usually we only talk 00:32:51.197 --> 00:32:54.137 about electricity consumption. 00:32:54.217 --> 00:33:00.517 Do you also need to provide sustainment for the cells? 00:33:00.657 --> 00:33:05.337 I do assume they are not only living from electricity. Thank you. 00:33:06.470 --> 00:33:08.430 Well, they're not living at all from electricity. 00:33:11.070 --> 00:33:15.910 Not even remotely. OK? Electricity is not going to help them in any way. 00:33:16.030 --> 00:33:19.550 So we are not using electricity for this. 00:33:20.810 --> 00:33:27.490 For this, we use so-called culture medium. So these are medium is basically water, OK? 00:33:27.750 --> 00:33:29.930 With many things inside. OK? 00:33:30.310 --> 00:33:34.310 And the things that we put in the water to make cells live. 00:33:35.850 --> 00:33:39.970 I mean this mixture has been invented in the 60s, 00:33:40.670 --> 00:33:47.170 so it's not science fiction it's a very old school way of doing so we are not 00:33:47.170 --> 00:33:53.170 inventing anything here so you take water you have all the vitamins all the 00:33:53.170 --> 00:33:56.150 amino acids many salts, glucoses, 00:33:56.850 --> 00:33:59.930 and you can put your cells inside that 00:33:59.930 --> 00:34:05.810 are going to live for weeks and months mm-hmm talked 00:34:05.810 --> 00:34:09.830 about the ethics here how do you personally wrestle 00:34:09.830 --> 00:34:16.030 with the ethical implications of you using human derived neurons for computing 00:34:16.030 --> 00:34:23.070 even though they're their your own so here is the the big change okay um i was 00:34:23.070 --> 00:34:25.910 able to learn biology because this is still science okay, 00:34:26.490 --> 00:34:28.350 I'm an engineer. It's okay. 00:34:28.650 --> 00:34:36.650 But here, you talk about ethics. And now you hit the limit of what I can learn and my expertise, okay? 00:34:37.090 --> 00:34:40.130 And I don't know how to answer to these questions, honestly. 00:34:40.370 --> 00:34:44.730 It doesn't mean that these are not serious questions. I believe these are very 00:34:44.730 --> 00:34:45.850 serious questions, okay? 00:34:48.230 --> 00:34:54.010 And what we decided to do, actually, is to make connections with academics, okay? 00:34:54.750 --> 00:35:01.310 And because there are people working in ethics this is their job they are competent 00:35:01.310 --> 00:35:05.750 for this they are experts we are not I can tell you about many other things 00:35:05.750 --> 00:35:09.010 but not this one and many other things I cannot tell you anything, 00:35:10.570 --> 00:35:15.570 and for this academics what we did is that we are starting to come to philosophy 00:35:15.570 --> 00:35:21.050 conference where we explain what is biocomputers to a teacher and we say guys 00:35:21.050 --> 00:35:26.750 okay here is this thing We bring the science and bring the ethic. 00:35:29.050 --> 00:35:32.650 When I've been thinking about this interview here, 00:35:33.410 --> 00:35:40.930 I was curious, do you see a point where the conversation shifts from using neurons 00:35:40.930 --> 00:35:47.890 to do dumb, stupid computing to really collaborate with neurons in a fundamentally new way? 00:35:49.730 --> 00:35:53.510 First remark that strikes my mind when you ask this question, 00:35:54.776 --> 00:36:00.276 is that, don't you think that in the past few months, we're starting to collaborate 00:36:00.276 --> 00:36:05.716 with computers in an entirely new way, with ChatGPT and things like this? 00:36:06.296 --> 00:36:11.476 Yes, and I can see definitely human traits in there when they're getting lazy. 00:36:12.836 --> 00:36:20.196 So now the benefits and the consequence of the technology is always a vast debate, okay? 00:36:20.556 --> 00:36:25.136 What can I can already tell you, You're better off living today than 1,000 years ago. 00:36:25.996 --> 00:36:30.956 OK, so total benefit is clear. 00:36:32.896 --> 00:36:38.436 Now, how can we interact with these nervous tissues? What is it going to change? 00:36:39.356 --> 00:36:45.316 I start to wonder, you know, with this large language model interactions that 00:36:45.316 --> 00:36:48.256 we are using more and more every day, like we have digital companion, 00:36:48.556 --> 00:36:53.616 like general AI is like it's their occur actually. 00:36:55.136 --> 00:37:00.576 It's actually less disruptive than we thought it would be so far. 00:37:01.216 --> 00:37:07.656 And the fact that this interaction would be done with nervous tissues instead 00:37:07.656 --> 00:37:09.796 of digital simulations of nervous 00:37:09.796 --> 00:37:15.016 tissue, what difference does that make actually? Is it that important? 00:37:17.196 --> 00:37:21.516 That is something that would take quite 00:37:21.516 --> 00:37:24.376 a lot of philosophers quite some time to work 00:37:24.376 --> 00:37:27.316 this out i see i see we're going to keep them busy for 00:37:27.316 --> 00:37:34.116 some time with these questions i see um talking about the applications and real 00:37:34.116 --> 00:37:42.416 world impact here um what do you think what industries will be the earliest 00:37:42.416 --> 00:37:44.636 adopters of biocomputing and why? 00:37:44.936 --> 00:37:47.196 So this one is quite simple, actually. 00:37:48.436 --> 00:37:52.296 All industries who are using AI, therefore all industries. 00:37:53.916 --> 00:38:01.216 And also those who use AI and care about their cost. All industries again. 00:38:02.356 --> 00:38:08.416 Because what we are going to develop first is a server when the rental price 00:38:08.416 --> 00:38:12.156 is going to be one-tenth to one-hundredth of of Amazon Web Services, period. 00:38:13.384 --> 00:38:17.264 This is what we're going to do. I see. I see. 00:38:19.944 --> 00:38:27.604 How soon could we see biocomputers solving real-world problems in fields like 00:38:27.604 --> 00:38:32.084 cryptography, optimization, or even drug discovery? No. 00:38:32.444 --> 00:38:35.444 For cryptography, I'm skeptical. 00:38:36.104 --> 00:38:37.764 This one in particular. 00:38:38.804 --> 00:38:41.824 It's not appropriate. I don't believe. 00:38:42.024 --> 00:38:46.484 I'm skeptical for this. I would really bet on quantum computers, 00:38:47.864 --> 00:38:52.904 not only because I love quantum effects and quantum mechanics, 00:38:53.784 --> 00:39:01.584 but for hardcore raw computation and speed, this is not the appropriate approach. 00:39:02.144 --> 00:39:08.644 Now, if you talk about drug discoveries, for instance, so here we can connect 00:39:08.644 --> 00:39:13.824 again with actually machine learning and with traditional AI, OK? 00:39:14.264 --> 00:39:18.824 And this can be used in the same way as AI for drug discoveries. 00:39:19.024 --> 00:39:26.184 For instance, protein folding and prediction of protein effects, this would work. 00:39:26.564 --> 00:39:28.624 Fundamentally, this could work the same way, actually. 00:39:30.544 --> 00:39:40.464 The main difference is always how much power you need to use to get this result. 00:39:41.864 --> 00:39:45.764 You know if you think about competitiveness in all countries, 00:39:45.804 --> 00:39:51.024 it's always about how much energy do you have and the cost of your energy so 00:39:51.024 --> 00:39:57.684 for the same energy you can have 100 times more it's like if you had 100 more energy, 00:39:59.089 --> 00:40:02.409 Or the energy was 100 times less expensive. 00:40:04.489 --> 00:40:09.529 I have to admit, I'm a big fan of the sci-fi author Peter Hamilton. 00:40:10.229 --> 00:40:15.089 And therefore, the question, could biocomputing have a direct application in 00:40:15.089 --> 00:40:20.429 human augmentation, such as interfacing with AI or even the human brain itself? 00:40:20.429 --> 00:40:25.009 Because there are already companies out there who are putting silicon chips 00:40:25.009 --> 00:40:32.009 into human brains. wouldn't it be easier to connect via a neuron by a computing-based chip? 00:40:33.369 --> 00:40:37.609 So first of all, science fiction and Peter Hamilton and all the other science 00:40:37.609 --> 00:40:43.069 fiction authors, I love them and I never go to bed without reading at least 00:40:43.069 --> 00:40:46.089 15 minutes of science fiction every day. This is a rule. 00:40:46.509 --> 00:40:49.869 Who's your favorite author before we get into the other stuff? 00:40:50.349 --> 00:40:54.229 Well, I have to say Arthur C. Clarke. Arthur C. 00:40:54.329 --> 00:41:00.829 Clarke, yeah. Yeah, so yes, you're right. 00:41:01.169 --> 00:41:04.149 In this discussion, we've been talking about biocomputing. 00:41:06.769 --> 00:41:10.589 But as soon as you start to look at neurons as small machines, 00:41:11.429 --> 00:41:14.329 you are changing the way you look at things. 00:41:14.429 --> 00:41:18.189 And when you change the way you look at things, this may lead, 00:41:18.189 --> 00:41:21.429 actually, to new applications. And you're right. 00:41:22.849 --> 00:41:25.489 Interfacing a human brain with a brain organoid. 00:41:27.689 --> 00:41:33.049 Is that challenging? A number of research has been published on, 00:41:33.589 --> 00:41:38.189 related fields, and you are first to consider that if I wanted to interface 00:41:38.189 --> 00:41:43.669 a brain organoid with your brain, what I would do first is I would take your skin, 00:41:44.609 --> 00:41:47.369 a different cells which have your DNA, 00:41:48.583 --> 00:41:52.463 So they will not be rejected by your immune system. They will be exactly your cells. 00:41:53.843 --> 00:41:59.163 And then I can tell you also, and you can sometimes see this online on our website, 00:41:59.463 --> 00:42:04.443 when we have brain organoids that we put together like this, 00:42:04.623 --> 00:42:08.403 we touch them, after two weeks, they're entirely fused. 00:42:09.483 --> 00:42:12.843 So neurons love each other. They love to connect. 00:42:13.203 --> 00:42:19.063 So it's absolutely not a challenge to connect. So the scenario that you described, of course, makes sense. 00:42:19.243 --> 00:42:23.383 So now, if you get into science fiction, you could say, OK, I train a brain 00:42:23.383 --> 00:42:24.803 organist to speak Chinese. 00:42:26.823 --> 00:42:31.383 I put this in my head. I'm going to speak Chinese. I was a bit skeptical that 00:42:31.383 --> 00:42:34.783 this could work as simple as this. 00:42:35.103 --> 00:42:41.083 But of course, things are going to change a lot in this direction. 00:42:42.083 --> 00:42:48.383 But not only this. I guess at some point you know today all the objects which 00:42:48.383 --> 00:42:50.483 are around you where you're talking. 00:42:52.503 --> 00:42:57.103 Well most of the objects I guess are not living you've got a mouse, 00:42:57.243 --> 00:43:02.183 a microphone these are not living maybe you have a plant or a flower these are 00:43:02.183 --> 00:43:06.643 living so you have two categories living, not living, 00:43:07.843 --> 00:43:13.803 and not living most of the time more and more these are things which are created by human beings. 00:43:14.863 --> 00:43:19.083 But in the future, you can imagine, if you talk about science fiction, 00:43:19.223 --> 00:43:25.203 you could have objects which are just in between, hybrid, partly living. 00:43:25.443 --> 00:43:31.123 For instance, a mouse that will have nervous tissue, that will recognize even 00:43:31.123 --> 00:43:33.223 before you press, that you want to press. 00:43:35.623 --> 00:43:39.343 So a way of interacting with objects which is totally different, 00:43:39.343 --> 00:43:43.343 because they are just the interface between objects and living things, 00:43:43.463 --> 00:43:45.203 that they integrate some living parts. 00:43:50.330 --> 00:43:53.750 I like this question if you had unlimited 00:43:53.750 --> 00:43:57.270 funding and no regulatory barriers 00:43:57.270 --> 00:44:00.970 what would be the first moonshot experiment 00:44:00.970 --> 00:44:07.290 you'd launch tomorrow so first this question is not as theoretical as you might 00:44:07.290 --> 00:44:17.810 think since i'm i'm looking to raise 50 million of euros at this point of course 00:44:17.810 --> 00:44:19.830 investor told exactly the same question. 00:44:21.770 --> 00:44:28.450 And the answer is clear. We are going to hire. First, we need researchers here 00:44:28.450 --> 00:44:30.690 because we have to test a number of things in part. 00:44:31.650 --> 00:44:36.370 We've gone quickly on a number of things, but there are a number of challenges that we have to tackle. 00:44:37.450 --> 00:44:42.870 And we have to create larger brain aganoids. We have to create a larger number 00:44:42.870 --> 00:44:44.630 of electrodes, increase lifetime. 00:44:45.110 --> 00:44:52.850 So the first thing I would do is that hire more people, increase the size of 00:44:52.850 --> 00:44:55.970 the labs to have more and more experiments running in parallel. 00:44:56.370 --> 00:45:03.250 I don't know what just sparked the idea, but just out of stupid curiosity. 00:45:06.550 --> 00:45:13.850 We're talking about living cells here. They won't be affected by computer virus, 00:45:13.990 --> 00:45:17.470 but could they be affected, meaning knocked 00:45:17.470 --> 00:45:20.390 out by a traditional virus or 00:45:20.390 --> 00:45:24.190 bacteria that gets somehow into the processor absolutely 00:45:24.190 --> 00:45:26.950 yeah absolutely and so 00:45:26.950 --> 00:45:31.890 what you describe is a very well known of anyone doing cell culture today and 00:45:31.890 --> 00:45:39.530 for and since the past 50 years so in the lab just below here again so i can 00:45:39.530 --> 00:45:44.790 tell you we are absolutely paranoid uh with this because uh they have no immune system, 00:45:45.510 --> 00:45:49.350 And we are not using antibiotics most of the time. 00:45:49.490 --> 00:45:52.930 So that means that if you have one single bacteria somewhere, 00:45:53.430 --> 00:45:56.050 everything is dead in 24 hours. 00:45:56.590 --> 00:46:01.630 So we have to be extremely careful about what we're doing. 00:46:02.753 --> 00:46:09.813 I see. When I was brainstorming a little bit on questions, I was wondering, 00:46:10.033 --> 00:46:16.673 traditionally, processors have been used to fulfill stupid, 00:46:17.113 --> 00:46:19.013 repetitive tasks. 00:46:19.013 --> 00:46:24.293 But could you see computers based on bioprocessors that can help in areas that 00:46:24.293 --> 00:46:29.413 haven't traditionally been associated with computing, creativity, 00:46:29.893 --> 00:46:33.293 art, helping handicapped people, talking about fusion of neurons, 00:46:33.493 --> 00:46:36.453 or even emotionally based AI? 00:46:37.253 --> 00:46:41.613 So your question covers a lot of things, okay? It does, yes. 00:46:43.393 --> 00:46:50.813 And your question also is based on, you have voluntarily put aside chat GPT 00:46:50.813 --> 00:46:52.793 in your question and LLM, okay? 00:46:53.173 --> 00:46:59.113 Because, of course, nowadays, computers are not what they used to be anymore. 00:46:59.513 --> 00:47:01.213 A little bit more. 00:47:02.673 --> 00:47:06.213 And yes, definitely we can. 00:47:07.553 --> 00:47:11.393 Well, the first thing is about the complexity of what you are generating. 00:47:11.613 --> 00:47:15.613 When you use real neurons, they are going to have way more connections. 00:47:15.853 --> 00:47:20.313 So it's one, okay? And they are in 3D. Our processors are not in 3D, 00:47:20.453 --> 00:47:22.353 okay? They are in 2D, okay? We should. 00:47:23.373 --> 00:47:25.933 Now, about interacting with you. 00:47:28.413 --> 00:47:35.513 It's very, not easy, but in principle, people do it currently in research. 00:47:35.773 --> 00:47:40.253 Easy to modify a neuron to detect molecules, for instance. 00:47:40.673 --> 00:47:44.533 Like if you breath, for instance, it could detect that you have a cancer. 00:47:45.784 --> 00:47:52.044 Just analyzing the molecules that you breathe. So talk about interaction between 00:47:52.044 --> 00:47:57.704 a living object and a human being. Could be also in that direction. 00:47:58.164 --> 00:48:04.504 So of course, the sensors are going to change a lot, because these are biosensors, 00:48:04.824 --> 00:48:11.624 totally good to detect complex molecules, which is extremely challenging in an artificial manner. 00:48:12.504 --> 00:48:17.244 So yes, the way we are going to interact, and the fact that it's living, 00:48:17.444 --> 00:48:19.664 and also it's living at the same speed as us. 00:48:20.444 --> 00:48:21.984 So it's interesting. 00:48:22.844 --> 00:48:27.124 A computer is performing, I don't know, 1 billion computation per second. 00:48:28.064 --> 00:48:32.744 This is not something that we can imagine. But the nervous tissue will work 00:48:32.744 --> 00:48:36.684 exactly at the same speed as the brain ergonomic that we have. 00:48:36.964 --> 00:48:43.804 OK, it's part of the, so everything is going to be in sync with us as human beings. 00:48:44.764 --> 00:48:48.464 It's a totally different way to approach computing. 00:48:48.604 --> 00:48:54.004 The last question of this section would be, what one unexpected but potentially 00:48:54.004 --> 00:49:00.144 game-changing application of bio-computing that nobody is talking about yet can you see? 00:49:00.364 --> 00:49:05.484 This is a very interesting question, you know, because the first thing I believe 00:49:05.484 --> 00:49:11.444 that the most important application is going to be absolutely game-changer, 00:49:11.644 --> 00:49:13.784 okay, of bio-computing. 00:49:15.464 --> 00:49:18.924 Is not only more efficient servers. 00:49:19.144 --> 00:49:24.364 This is already incredible, but it's not this. This is something else. 00:49:25.704 --> 00:49:26.984 But I don't know it. 00:49:30.304 --> 00:49:35.524 But, you know, there is a guy who was named Mr. Shockley. 00:49:37.024 --> 00:49:41.044 Mr. Shockley invented about 80 years ago the solid state transistor. store. 00:49:42.644 --> 00:49:48.844 I can promise and guarantee you, okay, that he had no clue that we would use 00:49:48.844 --> 00:49:51.484 his invention to create a smartphone. 00:49:53.944 --> 00:49:59.684 Okay, because that was 80 years ago. So if you ask me what are going to be used 00:49:59.684 --> 00:50:05.404 by your computers, I'm not able to answer. 00:50:05.564 --> 00:50:08.044 I know it's going to be big and way bigger than I think. 00:50:09.504 --> 00:50:15.164 Mm-hmm. Okay. I see. We'll be back after second short ad break. 00:50:24.251 --> 00:50:29.391 Okay, guys, we are back here with the last section interviewing Fred, 00:50:29.731 --> 00:50:36.571 co-founder and CEO of FinalSpark, one of just three companies in the world working on biocomputing. 00:50:37.711 --> 00:50:43.691 This is the very last section and I'm interested in like your personal take and legacy. 00:50:45.591 --> 00:50:52.711 I'm curious, what motivates you to work on this every day and what keeps you awake at night? 00:50:52.711 --> 00:50:58.591 You know, I co-created a first company that makes money, which is very good 00:50:58.591 --> 00:51:00.651 and hires people. We're very good. 00:51:03.111 --> 00:51:14.431 So now look at Earth from 100 million of kilometers, which you will see a small ball like this. Okay. 00:51:17.171 --> 00:51:21.751 Imagine you are standing in space alone looking at Earth like this. 00:51:23.031 --> 00:51:32.171 What can you do on Earth that is still meaningful when you look at this distance from Earth? 00:51:36.611 --> 00:51:39.991 And making more money is not one of them. OK. 00:51:42.051 --> 00:51:50.411 However, creating a new form of intelligence, yes, it makes sense. 00:51:52.791 --> 00:52:03.991 I see. If somebody writes the definite history of biocomputing 50 years from now, 00:52:04.451 --> 00:52:08.831 where do you hope your name appears in that story? 00:52:10.171 --> 00:52:18.311 You know, if it appears only somewhere, I would be so delighted. 00:52:18.771 --> 00:52:25.931 You know, first you have to imagine that 12 months ago, we were basically totally unknown. Okay. 00:52:26.431 --> 00:52:30.431 And then we became viral with the publication we made in Frontiers in May. 00:52:31.871 --> 00:52:38.571 Two weeks ago, I found there was a Wikipedia article, which is titled Science in 2025. 00:52:39.231 --> 00:52:42.971 Where they list all the important things that happened in science in 2024. 00:52:43.811 --> 00:52:45.371 And we are listed here. 00:52:46.351 --> 00:52:50.231 That was talking about science fiction. This was really science fiction, 00:52:50.271 --> 00:52:52.931 because I could not believe my eyes. 00:52:53.291 --> 00:52:54.991 So I don't know what. 00:52:56.191 --> 00:53:00.331 But right now, I'm not thinking about this. I'm just thinking about the next 00:53:00.331 --> 00:53:02.371 experiment and what we can do better right now. 00:53:04.476 --> 00:53:08.836 I see. Another theoretical here. 00:53:09.396 --> 00:53:14.416 If you could have a conversation with Alan Turing, John von Neumann, 00:53:14.496 --> 00:53:21.836 or any other computing pioneer, what questions would you ask them about biocomputing? 00:53:22.576 --> 00:53:28.176 Well, von Neumann would love the idea, of course, because it is a von Neumann machine. Yes, yes, yes. 00:53:29.156 --> 00:53:31.496 So we will have a good friend, okay? 00:53:33.556 --> 00:53:37.836 There are many things about information theory, about spatiality, about movement. 00:53:39.716 --> 00:53:44.136 I definitely would need to prepare this meeting. 00:53:47.676 --> 00:53:54.736 And arrive with like a physical catalog of many, many pages just full of questions? 00:53:54.736 --> 00:54:03.096 Maybe not only questions, but I would actually ask, what would you do if you 00:54:03.096 --> 00:54:05.996 could program living neurons? 00:54:06.756 --> 00:54:11.156 I would actually, I think, I'm sorry to say, or maybe it's obvious, 00:54:11.376 --> 00:54:14.776 I think I would ask the same question that you asked me since the beginning of this. 00:54:18.076 --> 00:54:25.036 Sorry to disappoint, but... That's totally fine. That's a big brace for me here. Totally good here. 00:54:25.516 --> 00:54:29.956 What do you hope your biggest contribution to biocomputing will be, 00:54:30.136 --> 00:54:34.076 not just scientifically, but also maybe philosophically? 00:54:34.456 --> 00:54:43.116 I think there is something with technology, is that there are human beings and 00:54:43.116 --> 00:54:47.536 technology, and more and more people start to say it's not good. 00:54:48.856 --> 00:54:51.756 Technology is too far from human beings. 00:54:52.476 --> 00:54:57.436 Actually, technology is almost against human beings, OK? 00:54:57.776 --> 00:55:00.916 And as an engineer, I don't like it, OK? 00:55:01.096 --> 00:55:05.776 I want to reunite technology and human beings, OK? 00:55:06.056 --> 00:55:14.656 They should live in harmony, OK? And I believe biocomputers and synthetic biology can achieve this, OK? 00:55:14.796 --> 00:55:21.376 We have done very good artifacts, which are called machines, artificial things, 00:55:22.659 --> 00:55:26.559 We still need to have a good interface with us as human being. 00:55:27.459 --> 00:55:33.099 This is the last piece that we should build. And we are going to do it, I'm sure. 00:55:35.559 --> 00:55:40.039 Talking about going to build it, going to do it. You already talked about fundraising, 00:55:40.039 --> 00:55:42.439 going to the very mundane pieces here again. 00:55:43.139 --> 00:55:50.159 You guys are currently on the outlook for raising funds, right? Yes. 00:55:50.979 --> 00:55:56.819 Yeah, indeed. So our mission is to raise 50 million of euros. 00:55:58.479 --> 00:56:03.699 And so we are talking with a number of people worldwide, actually. 00:56:04.019 --> 00:56:09.679 So we hope we can gather this maybe in three series, ABC, 10, 2020. 00:56:10.659 --> 00:56:15.139 Honestly, I would prefer to avoid this because I don't want to be running after 00:56:15.139 --> 00:56:18.859 money constantly. I want to work on the science here. 00:56:19.079 --> 00:56:22.379 And this is the big challenge and the fact that we are standing in Europe. 00:56:22.659 --> 00:56:27.999 Which doesn't help at all because I would already have this money in the US probably. 00:56:30.559 --> 00:56:35.299 If people made it until here, that's almost an hour of listening. 00:56:35.499 --> 00:56:38.939 Thank you very much. Clearly appreciate it. But those people are definitely 00:56:38.939 --> 00:56:41.199 interested in you and biocomputing. 00:56:41.459 --> 00:56:44.939 Are you open to talk to new people looking for talent? 00:56:45.259 --> 00:56:55.979 Yes. So we handle actually a constant stream of people who want to work with us, several per week. 00:56:56.799 --> 00:57:00.999 And these are only people who want to be hired. In addition to this, 00:57:01.099 --> 00:57:03.959 we have internships who want to work here in the lab. 00:57:04.319 --> 00:57:12.759 So we answer to everyone and we have some process to get people working with us. 00:57:13.449 --> 00:57:18.469 Great. So everybody who's interested now, we link down here in the show notes 00:57:18.469 --> 00:57:23.429 your website and the people can just basically reach out to you directly. 00:57:24.109 --> 00:57:28.209 Yes, we can also do is that we have a Discord server. 00:57:28.609 --> 00:57:35.409 So they can go to the Discord server and we will answer directly to the questions of everyone. 00:57:35.709 --> 00:57:40.649 So first they can see all the questions which have already been asked and then 00:57:40.649 --> 00:57:42.009 we are happy to exchange. 00:57:43.709 --> 00:57:47.409 Great. So, Fred, thank you very much. 00:57:47.689 --> 00:57:54.889 It was a pleasure to talk to you. It was more fun and much more extensive than I expected. 00:57:55.549 --> 00:57:57.769 Thank you very much. It was a great pleasure to talk to you. 00:57:58.009 --> 00:58:01.269 Hopefully, to have you back in a few years and talk about your successes there. 00:58:02.009 --> 00:58:03.289 Yes, I would be happy to do so. 00:58:04.529 --> 00:58:07.009 Great. Thank you very much. Have a good day. Bye-bye. 00:58:12.009 --> 00:58:21.089 That's all, folks. Find more news, streams, events, and interviews at www.startuprad.io. 00:58:21.429 --> 00:58:23.589 Remember, sharing is caring. 00:58:24.240 --> 00:58:37.120 Music.