112. The Baked-In Future of Science: Nick Edwards on Potato AI’s Quest to Become Your AI Scientist

I was like, "Okay, well, how would I do that?" Because I have to oxygenate these things. And so I used these Luawer locks and Jerry rigged this thing where I connected it to a balloon that had a stop valve and it had CO2 and oxygen, and I would bubble the solution while I was walking across the campus. (upbeat music) - Hey, you around, what's going on? - Oh, not much, you know, just take two of our commentary. - Yeah, I think we might have just say it. We're in the process of switching softwares up until now we've done all of our recordings in Zoom and we felt like it was time to evolve after 100 episodes. So now we're testing, and this is the second time we're recording this commentary. But what's on your mind besides the fact that we're trying something new? - Yeah, it's runnady now. So when it comes to software and the transition, the switching costs are high and it's just sanity. But no, I mean, I'm very excited for how things have been developing this year for the Gravving Podcast. I mean, wow, Chris Mason's interview, which was our last interview was mind bending. I mean, to be able to talk about genomics in outer space or in our efforts to have interseller colonization was so profound because Chris Mason, he's intellectual, he's a geneticist, he's worked with NASA, working with commercial space flight companies and it's just very exciting to be able to think about deep space and vast timescales. What an exercise, what a delight to be able to be human and think that way. - Yeah, I know Chris is fantastic and for anyone who hasn't listened to that episode, it was a last one we published and I think it's very much worth listening to. And one of the things you flagged that you want to talk a little bit more about was this collection of marine microbes that he mentioned. We've talked a lot about collections of microbes over the course of this podcast. But what was it about the marine microbe collection that you were interested in? - Actually, I was going down the rapid hole and I saw this headline that said, "The largest genetic database of marine microbes could aid in drug discovery and I was like, what? Ed's marine microbes." And so there is a consortium of organizations around the world, mostly academic institutions. So we're talking institutions from China, Germany, South Korea, Denmark, US, of course. We gotta get a piece of that. And they have a mass, this trove of more than 300 million gene groups from ocean bacteria, fungi, viruses, and they are making that information freely available online. So we're linked to in the show notes, what can we learn from 300 million gene groups, marine gene groups, it's just like what? And now that we have all these AI tools, which is the topic of our episode today, it's going to be interesting to see how fast certain information and insights can be acquired and used for human health and planetary health. - Yeah, you know what that reminds me of? I mean, that's fantastic. You are working on a multiple biome project, which I guess I'm working on too. - Oh yeah, we're in it together, Bra. - Yeah, what it reminds me of is there's this new visualization that shows the biomass versus the techno mass. The biomass is everything that grows on the planet. So that's plants, insects, every living thing. And techno mass is everything that we have built. So cars, buildings, concrete, asphalt, all that kind of stuff. And it turns out that in the last 20 years, the techno mass has exceeded the weight of the biomass. We're gonna put a link in the show notes to the bio-cube's visualization, 'cause I think it's worthwhile. We're actually gonna do a mini video on it, so that you guys can check it out. But I highly recommend you should check it out, because it just reminds you how much environment has been built on this planet. And what are we gonna do to reverse it? To grow more, which is what this podcast is all about. - Yeah, I mean, I saw that visualization. I'm like, wow, how creative it is, and how it changes your perspective, to be able to think on those large scales, but comparing those scales. I don't know whether to be amazed by the human capacity and capability over generations that we were able to convert biological materials into the built environment, but of course, that has come with a price. So we gotta get some equilibrium up in here. So this is one of those visualizations that are going to change the way you think about the world. - Yeah, yeah. And I think it's also worth mentioning. This episode will be published as everybody is leaving the JP Morgan Healthcare Conference. We hope that those of you who attended had a beautiful, productive, lucrative event. One of our friends reported back to me that his company just raised some money. So congratulations to them. I can't reveal that yet. But at the same time, right before the conference, A16Z and Eli Lilly made an announcement that they were gonna start a fund for $500 million to focus on the AI drug discovery. So we're gonna continue to pay attention to that. We do help clients who are using AI for drug discovery and other things, but it's a chunk of money. And clearly, it shows that people are interested and see that there's a lot of value to be created using AI for drug discovery. - Yeah, I'm excited to hear who they're talking to already and how many investments they're going to make with that $500 million. Is it going to be $10, $50 million checks or five? I mean, or 30? (laughs) You know, I mean, when you're thinking that a drug discovery, $500 million will only get you so far, but hey, look, we're talking about early stage development. You've got to kick start innovation. So that's very exciting. And we're very excited to our friends that we work with closely. There's a lot of startups that we know that we mentor, we're very blessed and fortunate to be able to mentor some amazing companies in the Bay of Pharmacy space. - Yeah, I mean, I think that's the perfect kickoff for our conversation with Nick Edwards of Potato AI. Potato AI is an early stage company. Nick is an interesting guy. He runs a podcast called Once a Scientist. And Potato AI is this company that focuses on really automating science using artificial intelligence. I think it's a fascinating conversation. - Yeah. - How do you know Nick? - I met Nick through the Mercdigel Sciences Studio program. So we are mentors there. So Potato is one of 10 companies this year that the Mercdigel Sciences Studio invested in. And we were just talking. It was amazing to hear that he is a fellow podcaster. And what he's working on is absolutely necessary. We can't wait for you guys to listen to this episode with Nick Edwards of Potato AI. (upbeat music) - Nick Edwards, Potato AI. Hello, welcome to the Grow Everything Podcast. - Thank you, Aaron and Carl. It's really a pleasure to come on. I'm big fan of what you guys are doing. I listened to a few of the recent episodes and some older ones and it's very cool. I was thinking, as we were starting, should I answer the first question everyone asks at the outset? - What's the first question everybody asks at the outset? - Yeah, please tell us. Tell me, this is great. You're professional. I mean, people, Nick has a podcast. So he could probably just do this whole thing, but tell us. Tell us what the first question. - Why did you name it Potato? - Yes. So I have kids and I think kids are just natural scientists. They have this innate curiosity. So it was thinking back to the magic of discovery when you're a kid and that's that experiment where you use a potato as a battery for an electrical circuit. It's fun, it's memorable. Nobody forgets the name Potato. - Yeah, and there's a lot of puns you can do. Your background has an amazing potatoes, but in the picture potato head, there's a lot you can do in there. And I saw the website, there's a lot of potato puns. (laughing) - Oh yeah, yeah. - It's fun with maybe a little too much, yeah. - No, it's great. It's great. And you don't have just one kid. You don't have two kids, but you have three. - I do. - Wow. - Yeah. - Well, I already with Carl, you guys are the trifecta or the three kid families. - The three kid trifecta. It's a lot of fun. We just had Thanksgiving recently and I was just mentioning who stayed in a pretty rustic place. So, but yeah, super excited to be talking to you guys about this and my kids are fascinated with science experiments. Sometimes I use some of these tools. I use potato every once in a while to actually get ideas for experiments. We made root beer one time, - Oh wow. - A root beer recipe. - Oh wow, it's really cool. We gave a little bit of description of what potato is in the intro, but tell us a little bit more about what it is. We know that you're tackling the reproducibility of science, but tell us. Tell us what it is for me from you. - The basic mission that we're going after is to organize the world's scientific information. To be able to enable autonomous science. So, the goal is really, how can we speed up scientific discovery, make more discoveries faster? There's a couple core problems and challenges that we're facing with that. Number one is there's way too much information. Literature is expanding exponentially, scientific literature. It's hard to keep on top of. It's incredibly overwhelming for researchers themselves. So, the exponential growth of literature is the first thing. The second thing is reproducibility, which is a big challenge. I don't know if you guys have ever seen this paper back in Nature in 2016, I think. I showed that something like 70% of research couldn't be replicated. More than 50% of the scientists they surveyed couldn't reproduce their own work. And so, to me, it's fundamentally a question of how do we know what to trust? And then how do we deal with this massive deluge of information? These are things I started thinking about a long time back when I was a graduate student at PhD at Brown, the strength program at the NIH. I was doing some. really complicated experiments where we'd inject a dental associated virus into mouse brain, a record from single neurons and slices, and it was this incredibly technically challenging, and most people give up on it after a few months, it's called patch complectrophysiology. So I'd spend hours digging through papers trying to figure out what pharmacology I needed to use, and then you start to realize there's just not enough information in a lot of the papers. And sometimes it's because it's just difficult to include all those details, but I just experienced it myself. I've seen that challenge throughout each of the companies that I worked with in biotech over the last bunch of years. I think the big unlock for me back when potato was still a dream, I started to realize that basically a lot of science is really communicated through language through this very unstructured text. So can we use AI and language models that are designed for language to actually unlock some of this information, make it easier to digest, be able to compare across papers, structures, and the scientific information so we can make it more accessible, easier to use, speed up discovery. That's the basic idea. That's pretty awesome. I mean, I know that there's people who envision a world where you've got a lot of AI agents that are possibly doing scientific work, but part of your vision is having AI research assistance to help you optimize protocols, reference literature, how far away are you from doing that? And what's the biggest challenge is to getting there? Yeah, so if you think about it, AI tools are constantly getting better. They're the worst they'll ever be. And you think of it almost like the progression of a scientist you get into the lab and you might start out as a research assistant. And I think that we're already there in some ways from that perspective because we've already got 600, 700 now-ish people using potato at many of the top institutions. So Harvard, Galtech, MIT, Stanford, they use it to review individual papers and come up with new ideas for hypotheses. We had somebody that came on recently and they put it in a nature paper. They found for next steps that pointed out some things that they had just submitted into a grant that they were working on. So it helps come up with some ideas, hypotheses. Then you can use it to generate literature-based protocols. You can use it to go back and forth and ask questions. We're constantly releasing these new features on a super regular basis. It just got incredible engineering with my co-founder Ryan Cosai. It already functions to that state where it can get me to a baseline, especially with a protocol, for example. It's not going to get you to something that you're going to run immediately in the lab. But basically every single person that we show it to, they're shocked at how detailed and how accurate it is. There's definitely room to improve and that's something that we aim to continue doing. And so the real goal here is not just an AI research assistant. It's an AI scientist. So for something like that, it becomes much more. You need closed-loop feedback. You need experiments that you can do. They can be done in the lab, but whether that's through automation or having people run those experiments. You need to analyze those experiments. You need to be able to run computational experiments as well and they connect to lab automation. So there's lots of those types of exciting things that we've started to make some early progress. And some of those things, but there's a lot of work to be done to scale it up into a full scale, AI scientist. But I firmly believe it. It already works as an AI research assistant. And people are finding a lot of value with it. Now that's amazing. Congratulations. That's great that you've gotten this far already that people are using it. 700 people, top institutions, things spread, word of mouth, and universities. So I can imagine you'll get more and more users in the near future. Very, very exciting. But I want to take a step back and think a little bit more about this relationship between the advancement and technology with the rate of how science is done. The rate of how biology is done. So with potato, you're aiming to make science faster, more reliable, which is necessary. But how do you balance the need for speed and research with maintaining the rigorous scientific standards that's required in doing experiments? Yeah, it's a huge question. That's the fundamental reason why we're taking the approach that we're taking, which is reproducibility and reliability has to be first because you can't make science faster if you can't trust it. That's the first and foremost thing that we're working on is focusing on rigor and making sure that it's accurate. We have a number of technologies one that we're using is retrieval augmented generation, RAG, and basically giving it the ability to understand language and some basic reasoning. But also only drawing from authoritative content that you're indexing. We are heavily indexed on peer-reviewed protocols right now, open access, but also just recently we're working with some of the major publishers. We recently announced a collaboration with Wiley. You want to be able to access some of this authoritative information. Make sure that information from protocols that you generate, you can cite back to individual papers and know where you're getting that information from. So that's one thing. Another thing that we're working on is we've built some tools to extract all the detailed information. For example, I give in paper. It could be any kind of unstructured scientific information. But I take for example, if I have this paper, I publish back in 2017 in Nature Neuroscience and I can put it into potato and it will pull out all the details and turn it into machine readable code. It puts it into a hierarchical structure. So you could be able to pull out all the methods from a given paper. There's a number of things that we can do with this technology. We're starting to work with large enterprise organizations and the applications are one I would say is comparing methods across similar papers to try and get consensus across them. The second thing is it's extracting this information and using it as context to existing data sets that companies already have to make them easier to use so that you can imagine if I have some sort of foundation model for biology, there's a bunch of data there. Can we attach metadata from papers or from experimental conditions so that information that can then become easier to ask questions about in plain language. The third thing that we're working on right now is if we can turn these papers into code then can we translate them into lab automation code so that you could actually run liquid handlers and lab automation with some of these protocols. That's really the focus is making it rigorous and then that enables the speed both now and in the long term. It's amazing because we start off by talking about the amount of scientific data that is generated and I also think about the amount of data that's generated in China for example and apparently we don't have full access or people say that people are disadvantaged because not everything is in English but I think probably most of it is but just being able to access all of that world of scientific knowledge whatever's behind paywalls you mentioned Wiley, Elsevier those big publishing companies they probably have so much that should be accessible and probably is not but let's switch gears a little bit because you also run a podcast called Once a Scientist and you've talked to people all over STEM some of which have been on our podcast as well. When you think about building potato and your experience with the Once a Scientist podcast what's the overlap what's the Venn diagram there what have you learned from interviewing all these scientists and what are some lessons we can take. Yeah that's interesting one of my guests SK Sharma he compared careers to a random walk which is a physics based concept things converge in ways and that's been a really interesting overlap for me so I started the podcast back in 2020 and to interview at maybe 90 or so scientists really about their careers because I just felt like when I was going through training I didn't have a lot of exposure to other things part of it was just because my head was down and I was in the lab doing work but part of it was I didn't find centralized sources where you could find out what does a scientist do whether you're at the bench or you're running something like messaging lab for example so I started interviewing people around that it's really interesting because I would start to get individual unique perspectives from a lot of scientists around the strategies they had taken some of it is post-hawk and so as people are putting together stories but so much of careers and progression is one stochastic I think but two it's an oral tradition and we have this weird apprenticeship model in science where information is passed from one person to the next sometimes it's on an appkin if it's scientific content sometimes it's just in the form of a discussion over lunch I think there's a lot of value in getting perspectives from lots of different sources pulling them together and then figuring out what makes most sense for you there's a lot of interesting parallels with what we're doing at potato because we're pulling data and scientific information from across a bunch of different publications people is language-based so there is some interesting inspiration there but I think one of the big things that I realized through the podcast was communication is really important and it's vital to anything you do within science whether you're a professor or whether you're starting a company the language is really important and I think to me that was one of the inspirations for actually starting potato is realizing that there's sciences not just numbers and data it is communicated through language and through stories yeah yeah well said yeah absolutely I wonder if you could take the data from once a scientist podcast all our recordings and connected to potato AI is that easy and see what see what happens yeah I'm interested I thought at one point maybe I should do a chatbot based off of all these career conversations I've had with scientists and just make it so that researchers could go in and ask questions about what are the potential options what do people do and if I had time I would totally do that but yeah I wonder if you're interested I'm not sure what data would be the most relevant but there's so many applications for AI within science whether that's for something like basic discovery like we're doing or whether that's like career advice I guess yeah career bias and just digging through the data and synthesizing into something that's comprehensible to the reader even if it doesn't make actual sense but the way that the words are laid out triggers the idea triggers motivation to go down a certain pathway which you certainly have done I mean you were also a consultant at BCG, you were at startups. did everything. And then now you're here at Potato and as a podcaster. I think that's a great example of one career trajectory that scientists can go down. Yeah, again, I think that there's two key features of the phenotype of a scientist. One is curiosity and two is rigor. I'm pretty curious person. I started out as a basic researcher. I'm just the type of person I really like building things. There's this funny story where I was in my PhD and I was working at the National and Stone Drug Abuse. And I had to prepare these brain sections that I would use to record from single cells. And you had to make this very specific solution. That's called ringer solution. A lot of people in biology know about ringer solution. And we could keep the brain slices alive as long as we oxygenated them. And I had this specific solution. We could keep them alive for 10 hours and record from these cells. I had to do these experiments in another lab that was maybe a quarter mile away. So I had to walk over there. I remember one day I was doing it in the other lab and they didn't have all the same reagents. And it was inconvenient because I wasn't used to the instrumentation. So I was like, okay, well, can I just do this in our lab and then transfer it across campus? And I was like, okay, well, how would I do that? Because I have to oxygenate these things. And so I use these lure locks and Jerry rig this thing where I connected it to a balloon that had a stop valve and it had CO2 and oxygen. And I would bubble the solution while I was walking across the campus. And I've always liked to build things, whether it was like that or optical fibers. And so I got interested in entrepreneurship in my post doc because I was continuing to build things like microscopes. But I started to realize during that time that there were different ways to contribute to science. It doesn't have to be through the discoveries that you make at the bench. I started to have this realization that maybe that could happen through the business side of science. I worked really hard to got into BCG and was really interesting because I basically got a free MBA. Actually, I got paid to do an MBA. Yeah, yeah. They're focused mostly on the tech industry when I was there. I did some AI strategy and I thought that was really interesting. But I realized pretty quickly, I really liked the science and I just wanted to be closer to the science and building things. So I spent some time building software products at alumina for their internal sales teams and some startups within the last few years. I know I just have always taken this hypothesis testing approach to everything I do is what you learn in graduate school. You learn to iterate really quickly. You learn to be observant of the little details and the things that you might be able to capitalize on. That's exactly the types of skills that you use as an entrepreneur as well. So I think scientists from my experience make pretty good entrepreneurs because you learn to go through really hard things. You learn to iterate quickly. You learn to test hypotheses and that's what you're doing in business as well. Yeah, totally. Yeah, absolutely. Yeah, so when we talk about growing everything, one of the things we talk about a lot is how complicated biology is. It is like this mysterious black box that we keep unraveling and getting to know better all the time. So it's full of potential for innovation. How do you think AI is going to unlock the magical side of biology for researchers or entrepreneurs? Do you think it will do that? How long until we have got do the model of the cell? Yeah, interesting. Yeah, that idea is so cool. There's a lot of people working on it. It's a very, very challenging problem, of course. And a lot of it is a data challenge. Part of the way that we think about this is pull out the monotony, the mundane, the tedious things that we have to do because AI's can help with those types of things. So this massive volume of literature. I have to go through and read a single paper for, I don't know how many hours and try and extract like what did they do? What exactly did they do? Step by step. And then I have to compare that across multiple papers and that's hard work. And it's time consuming. That's low hanging fruit. Also the language models are good at being able to bring together those types of things that there's a mundane aspect to that. And so I think that it really allows for people to be more creative. And the idea is to really augment scientists. It's so that you take over some of those things. That's how I think about it. I think that it certainly doesn't take away the important aspects of coming up with hypotheses and really figuring out what the critical controls are for a given experiment. As it gets better, it's going to help more with that. But there's going to be this massive explosion of human AI collaboration that I think is honestly just going to propel science. It's going to be rocket fuel to scientific discovery. Yeah, I agree with that. Yeah, I'm truly exciting. I mean, we're talking about science writ large, but of course there's different flavors and I'm here at Massachusetts Gene Lab. We work with BiFarma, which I know you work with MS12. And then the other sciences, or what is it kind of a stepchild? Is the bio industrials world? No one really. We need more people to talk about. We think it's everywhere because this is the world we live in. But why we have this podcast, bio industrials, bio ingredients, climate biotech is here and it has been here. It's been silent actually for a really long time, but it's been cleaning our clothing. It's been giving us cheese and beer and all those wonderful things. But let's use it for other things. It'd be interesting to discover how much potato plays in the other sciences too. Yeah, that's a great question. It's an area that I'm trying to explore and understand industrial biotech specifically. We're having a bunch of conversations with industrial biotechs right now because I think that there is massive potential from applications like use cases. We're starting to see some real interest from companies like that, whether it's from automation perspective or the things that we've already built are equally applicable. It's just an area that I think is super exciting and interesting. I'm a neuroscientist, I think about circuitry and emergent properties of the brain and that type of stuff. So I'm going to go off in a bit of attention here. This is a podcast. We are built for tangents. If you think about neuroscientists in an interesting field because there's all these different layers and levels of discovery, looking at individual receptors and ion channels and recording from them, then you can look at individual cells and action potential propagation across dendrites and across axons. Then you can look at synapses, connections between cells and you can look at plasticity and potential propagation. Then you can take a level up and you can start to look at circuitry and these large scale oscillations that are happening in synchrony. That's part of the reason that complexity and that layered aspect of neuroscience is just intellectual catalyst. I like thinking about it. I've started to realize synthetic biology has an industrial biotech, has some of those aspects of complex. I mean, you're thinking more about organismal and being able to look at its circuitry and there's different levels of analysis. I don't know. So I've just gotten more curious about it. I am definitely not an expert. It's something that I would like to learn about. So I'm curious. What are the areas that you guys think are most interesting right now from an industrial biotech perspective? Hi, a quick shout out to our sponsor, Messaging Lab, the force multiplier for biotech. Your biotech company is making the world a better place. You know that? We know that, but does the world? There's a big reason why some biotech companies attracted investors, sign up customers and get attention. That reason is strategic communications. At Messaging Lab, we translate complex science and economics into compelling narratives. And we have done that for the most successful biotech companies across healthcare, agriculture, personal care and beauty, materials and the list goes on. We're here to make sure your ideas don't only get heard, but resonate with your audience. So if it's time to amplify your company's voice, visit MessagingLap.com to explore how we can elevate your story and grow your business. Well, before I answer that, I just want to make a comment and get your reaction to this Drosophila Connectome and really the more recent news about E11 bio showing a brain map and being able to do it at a lower cost. I mean, I think AI tools fit right into that. But before I answer where I think AI should be applied to industrial biotech, just curious about your thoughts on those. So cool. Yeah, it's so cool. I talked to June Axe at E11 on my podcast a little while back and their vision, what they're doing in terms of our coding is very exciting, bringing down the cost of looking at whole brain connectivity or at least large scale connectivity. That stuff's cool, because then you can start to figure out what are some new architectures. It's understanding the human brain is understanding the model organisms, but they move up to human brain. But the applications for that are crazy, exciting for AI so that we can start to think of new architectures that can make these tools more efficient. So I think that's super exciting. Yeah. Just because we're on the topic and I had this thought in my waking state, it was about how we think and our brains are in our heads. But non-mile organisms like Octopi have decentralized brains. Yeah. And I was thinking, what would it be like to think from my hands? I know, right? It's unfathomable. As in grad school, I went into a lab at Woods Hole, a friend that they were just working with several pods and it was a cuddlefish. And we walked into the lab at one point and you could see they follow your movement and they're like tracking you with their eyes and they recognize certain people. And so they'll come out if they if it's somebody that they recognize they'll come out and they're friendly and they'll interact with them. But they're hiding behind the rock and like looking for you around rocks and I don't know, it's fascinating. But one of the things I was going to say I should mention is that thinking about the intersection with industrial biotech, I do I still want to get your opinion on where the interesting areas are. But I'm holding it. I'm holding it. We'll get back to it. We'll get back to hand brains later. Yeah. The exciting part about what we're doing with potato is that we're talking with all these companies around structuring their unstructured data and making it more valuable. So we don't even know all the use cases that there are for this. But we're seeing like tons of excitement from large enterprise organizations. So there's one interesting thing. about me for coming on this podcast is if we can start some conversations and if anybody this listening is interested in ways that we can structure some of your internal data and help with that, feel free to reach out to me and Nick at ReadySeptitator.com or you can reach out on LinkedIn whatever. But yeah, what are the areas that you think are exciting? Yeah, I mean, you mentioned one that we've done some work in when it comes to structuring all data, not us personally, but companies we've worked with because there is all this old data, the feeling or the approach has been new data is better and yet there's all this old data that can be interrogated. It just needs to be put into a format that AI or in the machine can read from. So I think that's one area that is more on the macro science level. But I think anything that has to do with process engineering on the industrial side and you mentioned being able to download protocols. We're in Brooklyn. One of the success stories of Brooklyn Biotech is open trunks, open trunks has like a open source liquid handler. They share protocols and that was just mind blowing to me. This ability to share protocols that you plug directly into a liquid handler and being able to share that with other people, not just for time saving, but also for reproducibility. So I think anything that has to do with process engineering is going to be a huge area for AI improvement. And then thinking about a format, everything that needs to be optimized on the clinical trial side of the world, there's just so much there. I know there's a couple companies that are tackling it. There's probably dozens of companies that should be tackling it because you don't get enough people participating in clinical trials or often the wrong people. The number of people that drop out is astounding. And the whole drug development process, I mean, we could say on one level is broken. It just takes too long, costs too much money to get a drug to the market. So anything that we can do with AI to optimize that process to identify patients, I think is going to be huge. And not to mention that the front end of that on the drug discovery side, there's $60 billion, I think is a number of investment that's gone into AI drug discovery. But that's just one part of this whole thing. You've got to optimize molecules. You have to be able to synthesize them. You've got to do digital testing on them. There's so much that can be done before you actually hit animal models and then the clinic. Carva, what about industrial biotech? You know it's all about this discovery. You said I'm working digital sciences program. I did say bioprocessing. That's awesome. I'm excited to explore it. That's the exciting thing about broad scientific platform technologies that we can reach across different areas. We are definitely working within traditional bio-farma. Eram Knows as an advisor to the Merc digital sciences studio program, by the way, appreciate all your help with that. You're doing a great job on it. Well, thanks. But I think many of these areas that I just don't know about are super exciting. So yeah, processing. Yeah, bioprocessing is a big thing. I would say like we've been talking about a lot is microbiome dynamics because on the one hand, yes, you do have the gut microbiome which is being studied by the farmer world and the nutrition world. But then you have the environmental microbiome side. So the soil microbiomes and what's going on in the oceans and the trees and how they relate to the health of the planet and how they can influence different solutions to help, like for example, clean up PFAS or forever chemicals that are toxic chemicals that are pervasive at this point in soil and in water. I mean, I think one of the fascinating things that Carl and I being at this intersection between bio-farma and bio-industrial is to see where's the crossover, what's the overlap? Because biology is biology is very complex. But what can we learn? Say, for example, from the variety of different immune cells and then the variety of different microbiome cells that exist and they all have different types of interaction. So you're talking about connectum. What are those microbiome dynamics that cause a process to then create a different function or product? Whether it's a drug or some type of molecule that your microbiome can create versus you taking a pill and it's made in a process outside of your body, but how do you trigger or influence or reorganize different cells or influence them to produce different molecules for your health or for other things like producing more collagen to look younger. I just before this call, I went down to Brian Johnson Rabbit Hole. He released a new video about making his face look younger and it's better. Like before it was kind of like a skeleton, but now it's like, okay, okay, you're doing it. Good job. He's using different techniques. And he's share what those are. And he's like, no, you don't just use fillers because fillers can end up in different parts of your body. It's not that great. But there's another methodology to boost your own collagen production. He was doing clork restriction, but then he added a little bit more calories, which maintained his biomarkers, but then helps his face get a little fuller naturally. So he is looking very young for you guys look at him recently. But anyway, this is to be said, the intersection. So potato is like science or biology, agnostic, it'd be interesting to see how potato can influence both sides of the biology spectrum. For sure, for sure. I wonder if he was like crushing up the Doritos. There's tissue clearing. Yes. A little esoteric, but yeah, yeah, just so our audience, yeah, there was the Doritos. We talked about this before, but the Doritos, there was the molecule, the tartar zine. Tartar zine, okay. Yeah, tartar zine. Yeah, it can be used for tissue clearing. Yeah, so you can see through your skin. Yeah, that oh my gosh, I don't think he was doing that, but that is a thought and you can't unthink seeing through skin. Yeah, for sure. I think that there's going to be some really interesting applications. Some of these computational models like cell modeling and looking at interaction networks, I'm really excited for those types of applications. I think that there's a lot of power and new computational tools as well. We've gone down that path a bit and there's David Baker and Demis Sassabis and John Jepre just won the Nobel Prize for some of these types of tools. We think there's a lot of promise. They're hard to get started with. It's hard to build these models and so that's really why we're focused on this unstructured data to help augment and fill in some gaps for some of these types of things. Yeah, yeah. So you talk about data, which is really important. There's this whole struggle or there's this need for validated accurate data. How does potato play into that? You said earlier at the top of this episode that 70% of studies are not reproducible. Some scientists can't reproduce their own studies. What does that mean for the data that has come from those studies? And how do you balance the fact that maybe some people there's unstructured hypotheses that aren't very good and again, the data not being very good, but how do you balance that? How do you think about that? Yeah, it's a challenging problem is how do you know what to trust? So one of the ways that we have really focused on that early on is indexing mostly just protocols, which are the step-by-step boring instructions of how to do an experiment. Add this reagent, add that reagent, mix it, centrifugiate, and those are the types of things that people publish because they want people to reproduce them. That's one of the ways in which we're really focusing from a data perspective is focused on those lower level instructions. This is also where you have to have this really strong human AI collaboration and human oversight to have the expert come in and review it and know what are the things that I can change, what are the things that I might not be able to. And so that's part of the way that we're thinking about this with comparing and doing this consensus across multiple papers. So if you can pull out all the insights from an individual one that you can do it across them and you can say, okay, what's the consensus here? What is this paper doing that might be a little bit different or that's unique so you can start to pull those things out and make them visible to people? That's really where we're going next with some of this extraction technology that we build. Yeah, that's pretty cool. I'm just curious thinking about your career, podcaster, academic, now entrepreneur, with extended BCGs. But what are some of the unique challenges or even insights that you've been offered to the entrepreneurs that are listening to us? What things do you take from your academic time or your BCG time that are helping you out now is the CEO of Potato? Yeah, I'd say it's a couple things. One is the ability to wander into the unknown and push forward on something despite a lot of ambiguity because in science, that's what it is. You're on the frontier of knowledge. You're grasping around the dark and trying to figure out which way we're going and testing things as you go. And you learn a lot through that process that teaches you to be resilient and to have a certain level of grit that I think is really important in every aspect of a career, but particularly as a founder of a company. I think that's something that has really left an impression on me and shaped my approach to career. There's that. I think there's, again, that curiosity that I have some ideas and I'm just going to test them. I'm going to do fast iterations. I think the third thing is the communication. One of the things that I noticed as a scientist was that many of the labs that had the most funding that all the PIs were speaking at all the conferences and they had high profile papers. They were all good communicators for the most part. And it's not something that just comes naturally. You have to put time into it and you have to figure out there's a reason why it takes time to write a really good paper, for example, because it's challenging. So you have to put in the work that academic training is a really good proving ground for any type of career that you go into, but I think particularly for finding a company working to start a, the more I've gotten into the business side of things, the more the communication has been even more and more important. That's been distilled into me. Like BCG, you had to figure out how to address things at a very high level so you could speak to an executive and say, like, here's your answer. And here's the three supporting bullet points underneath that. And then go into details as necessary. This is like very different from the bottom-up style of communication that we typically have within science. That was a big learning for me. And that's That's something that I still-- I have to focus on and figure out ways to address things to different types of audiences because when you're out in the world talking to customers and our customers, our scientists in some cases, but some of them are not or talking to investors, you have to zoom out, think a little bit meta, like how do I position this? And I think that's really important. Those are the things I learned. No, that's great. I do have a couple of questions. You mentioned your users. So 700 users, there's Harvard, the biggest schools, but you said it's not just for scientists. I mean, other citizens scientists using it. What are your thoughts about making potato accessible to people that don't have this very expensive scientific degree, but want to be a scientist? Yeah, that's an interesting question. I mean, we do have people come on and use it for different types of use cases. Anyone that is interested and curious, you can go to readysetpotato.com and sign up and use at least certain aspects of it for free. I think the word an interesting point in the history of science, when it's going to be more and more accessible and people are going to be able to pick up some of the background knowledge using computers that will open the door. I'm super excited about those types of things. Some of the more advanced tools that we've got for data extraction or more for enterprise types of organizations because they're the ones that are dealing with large massive amounts of scientific information. The free tools we started as an ability to start the conversation, but we do see people using them for growing fungi or different types of citizen science applications. So I'm curious to see how that evolves. Yeah, that's really cool. And I'm really curious to another question because this is another world that we live in is the world of DSI, which has been really interesting to us. So we mentioned people adding protocols or using protocols. How does potato address attribution to people who upload certain protocols? If a big organization uploads a protocol, can a citizen scientist use it? Will they see the logo widely on there? What does that look like? So there's a couple things in there. One is that all the information that people upload is private to their own individual instance. We don't train on it. And so whether that's an organization that's using it and they're uploading some protocols or some papers, it can't be viewed or accessed by others. So I think that's a really important component because privacy has to be a critical piece of this. I think the second thing is a question of attribution right now with the retrieval augmented generation, RAG, it cites the individual papers and where information's coming from within a protocol. The question of how we manage attribution from a publishing perspective, whether it's putting it on some sort of DSI platform or publishing it journal, those are things that we have not come across yet and we'll have to explore and figure that out. Yeah, yeah, we'll introduce you to our friend Michael Fisher, this is his world of decentralized science. So I'm super excited to see where it goes. Yeah, yeah, sure. He has a platform too, but for DSI and what is about creating the ledger, putting everything on the blockchain, being able to identify core sources and then track the sources across usage. It's interesting. I mean, it's also about challenging the current scientific establishment, the way that it's being done in academia, the way that the publishers have a stranglehold on scientists and scientists pay to be published. It's crazy. I know you interviewed C. May Chow. She talks a lot about that as well. It's interesting. It's going to be exciting. There's all these new organizations that are popping up different ways to structure science. And that's I think that's just going to continue to happen. And DSI is exciting because you have to have some sort of version control for science. Looking at the ways that institutions are going about, there's new types of nonprofit organizations. There's just different ways that people are going about things. And I think we're going to see massive shift in how science is done in general, writ large. And I personally feel like we're at the beginning of the next scientific revolution. I don't know exactly how it's going to play out, but I think this is a turning point for science. Yeah. A lot of science is about observation. And you talked about working on microscopes. I had worked at a company that used machine vision to be able to look through LiDAR to help people with our fall prevention. So it was a very advanced, maybe over an engineered solution to prevent people from falling out of their beds in the hospital, but it worked really well. But anyway, I always think about like, can we just see what happens? There's my microscopy. You could see things happening, but even on a more nano angstrom level, that's crazy. Okay. Could you be able to see that just have the machine vision being able to see real time and have real time representation of what's going on, which it would happen instantaneously because things happening in a very small scale happen like quantum seconds. I don't know if that's the term. Yeah. Multi-modality is going to be very interesting. When we start to combine computer vision applications with the language applications and lots of other modalities as well. Audio. Yeah. That's the modality. Everyone's getting to enjoy at this current moment. That was my world too. We can detect different biomarkers through your voice. You can detect your heartbeat through your voice because your voice and audio is waveform. So what is your heartbeat? So you're able to learn a lot just through voice. But yeah, the multimodality. Then you got your AI scientist right there if they can see, listen, taste, smell, digital nose, digital tongue. Those are things. Yeah. I mean, I don't want to get too far out there. That's me. Sorry. Welcome to my universe. No, no. I think that it is going to get really interesting, especially when we have humanoid robots. They can be doing experiments in the lab. You need to be able to figure out what's being done in the lab right now. So I don't know what that future is going to look like, but I think it's going to be exciting. Yeah. But it's going to be touching a potato, which is what the humanoid hand is doing behind in your back. Oh my god. Yeah. Guys have to see this potato picture. It's like, what is that? Piece of art called? I can't remember this. It's a team chapel. Yeah. That's my collage. Low. Yeah. That's right. It's hilarious. So instead of a lighting bolt, it's from the potato. That's funny. Okay. All right. We're going to, before the potato pun start, we have a fully baked episode right here. Well, slice it up, dice it, dip it in ketchup. I mean, okay. I lost it. Yeah. It is the creation of Adam, by the way. Oh. BDAM. Yeah. Yeah. That's why I said. Yeah. Oh, yes. Yeah. That's right. That's right. But this has been awesome. I mean, it's so incredible to just speak to you about how to do science in new ways, leveraging AI. Everyone's talking about different ways of leveraging it, but to do reproducible scientists is so important. So we're very excited for you. And we're like minded podcast friends. So that's incredible. I'm going to have to have you all podcast at some point as well. That's super kind of you. We were once a scientist. Yeah. So we were. Yeah. But yeah, we'll have everything in the show notes. ReadySetPotato.com. ReadySetPotato.com. That's right. That's right. Awesome. Yeah. Cool. Well, thanks so much for joining us next. This was so much fun. Let's do it again soon. Carl, Eram, thanks so much. This is a lot of fun. So I appreciate you taking the time. Yeah. See you later. Eram, what did you think of that episode? Wow. It's funny because we were thinking of what we should title this podcast. Since the title of this podcast is The Baked in Future of Science. Yes, potato puns abound, but it's interesting because what we do a little behind the scenes look is that sometimes we use Shatch B.T. to generate podcast titles. And we're like, The Baked in Future of Science. I'm like, is this AI mocking us? Right. Yeah. They're right. You're in the matrix. There's a destiny. Everything's calculated out. The Baked in Future of Science is an AI scientist. So I was like, okay, that's hilarious. But yeah, it's amazing to see how much progress he's making. And this is just the start. We got to make sure we get the right hypotheses and look at all of the papers and the published papers that are of authority and the new papers as well. Make sure we're looking at the right information and it's the start. And then to be able to take the information and put it into lab automation is just, wow, maybe I'll be a scientist in a few years because it's going to be a lot easier. We are all scientists, Eram. I think what's interesting is that Nick presents one part of the equation when it comes to the science that is done in labs where they're commercial or academic. Last week I'd mentioned this biotoken and they talked about being able to sit down and having an AI agent design a drug and really optimize that drug, iterate on it and then even take it through preclinical testing, which really made me feel like I'm not prepared for this agentic AI era that we're about to enter into tomorrow. But it really makes me wonder what else our AI agent's going to enable and what Nick is doing is really a key part of that. I'm very excited about it. I would say the biggest thing too is we talked about this a little bit in the episode but it's going to be interesting to see how these AI agents are going to make decisions on different tasks through the, I guess, information pipeline. You have, again, maybe 1,500 papers on one topic and how do you score each paper? How do you know what's a good one? What's a bad one? And of course we're already dealing with inherent bias in AI depending on what training data it learns from. So it's a tough challenge but that's what entrepreneurs do, that's what engineers do, as a scientist do, they try to find a solution to a problem. So it's going to be exciting. I think a lot of us are excited to see how these tools are going to be used from research to actual physical products in the world. Agreed. That was a great episode. Thank you Nick and thank you to all of you, our listeners. We really appreciate you. Let us know your questions. We're very easy to get a hold of. There's our contact information in the show notes. The grow of the thing hotline is out there our number. We do get calls from people asking us things so feel free to get in touch. We love you and we love our community. That's the pod. (gentle music)