EP 147: The AI Revolution in Biology - How it’s changing

This is the Everyday AI show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business and everyday life.

Everything is biological. So how can we use AI to shape the world that we live in? You know, maybe the, the medicines that we all need, the food we eat. How does biology and artificial intelligence shape that? We're going to be answering those questions and more today on Everyday AI. Welcome. My name is Jordan Wilson and I am your host. Everyday AI is a daily live stream podcast and free daily newsletter. Make sure to go check out that free daily newsletter. And we help everyday people like you and like me learn what's going on in the world of AI and how we can also leverage it. Right? And so many times we talk about, you know, software that we use, business growth strategy. So that's why I'm extremely excited because I'm going to learn so much today and I, and I know that you will too. But I'm extremely excited to talk about how, you know, AI is impacting the world that we live in and, you know, the biological makeup of just about everything. So stick around for that. We're going to get started, but before we do, as we do every single day, let's take a look at what's going on in the world of AI news. There's a lot, there's some interesting things today. So kind of speaking of biology. Well, AI is apparently like climate change. So that's the comparison that Google CEOs Sundar Puchai recently made. So during the APAC CEO Summit in San Francisco, Sundar Pichai spoke about the global responsibility to create frameworks for AI regulation, comparing it to the shared responsibility for addressing climate change. You know, I will continue to proliferate globally, making it necessary to create global frameworks for regulation. So Pichai said countries have a shared responsibility to build those global frameworks. So I was personally very, you know, when I read the headline for this story, I was like, wait, what does this all mean? But it kind of makes sense, right? It seems like individual countries right now are, you know, know, kind of going about AI regulation on their own accords. And this is probably the first time I've seen a real call for global regulation. So pretty interesting stuff there from the Google CEO. Speaking of big companies, Meta is now debuting some AI powered creator tools that seem to be a Runway competitor. So if you're in the creative content space, you probably know, Runway fantastic. Text to video, image to video tool. But the Facebook parent company just launched two new AI based features for video editing that can be used to post to Instagram or Facebook. One is called Emu Video, which generates four second long videos with just a prompt. And the other one is called Emu Edit, which allows you to edit short videos with a text prompt, you know, to say, hey, you know, erase this from this video and it does it. I'm extremely interested in this one. Meta Technically has been teaching, teasing this for probably more than nine months. And you know, Runway has really grown in popularity. So I'm excited to see what this new Emu will do. And I love that Emu commercial, you know, the life insurance one. All right, so last but not least, Google is delaying its new Keystone large language model to better catch up to OpenAI. So Gemini has reportedly been delayed. So right now Bard uses the model Palm 2, which is not really great, to tell you the truth, when compared to GPT4. So Google is facing difficulties in catching up to OpenAI as seen through this delay in releasing their new large language model, Gemini. And this also impacts the Google vs Microsoft Cloud Race with enterprise customers because Microsoft has seen some great success recently in this field due to their partnership with OpenAI. So Google is kind of taking a step back when they were reportedly going to be debuting Gemini this fall. Apparently it's going to be delayed a little bit longer. All right, so some big AI news and we always have more. So make sure to go to your everydayai.com Sign up for that free daily newsletter. We're going to have more news of what's happening in the world today in artificial intelligence and a lot more. But I'm excited now to talk about the AI revolution and biology and how it's changing. I don't think we've had a guest on the Everyday AI show that can help us look at the world in this way. So that's why I'm extremely excited for today's show. So let me help bring onto the show and please help me. Welcome. Here we go. I think. There we go. All right, so Anna Marie Wagner is the SVP head of AI at Ginkgo Bioworks. Anna Marie, thank you for joining us.

Thanks for having me. I'm really excited for this.

Oh, I am as well. Biology, right? Like it's a certain thing that I think we, we kind of take for granted because outside of, you know, biology and you know, your high school or college classroom, I think unless you're in the field, you kind of stop thinking about it. But maybe Annamarie, just tell us a little bit about what you do at Ginkgo Bio Works.

Yeah, sure. And I totally agree. I think we. I just think there's like this massively wasted opportunity in the way that we teach biology to kids. Like, it is. It is this incredible, like, alien technology that is capable of just such incredible applications. My favorite example, I use this one a lot, but it's like there's a. There's a protein in our bodies called ATP synthase, but it's like a 21,000 rpm motor that's like 10 nm wide. It's just like. It is bananas. And it all runs on code, right? It's like a C, T and G, not zeros and one, but it's like codable nanoscale physical technology. It's very cool. So what Ginkgo does in that sort of cool world of biology is recognizing that most of the stuff around us is biological, right? Like you mentioned it earlier, we are biological. Our food is biological. Honestly, like, most of the stuff we use has biological origins, right? Like, even the plastics come from petrochemicals, which came from fossils, which were animals and plants at one time, right? So most of our stuff can be made with biology. And our view was the way that the industry is organized right now to enable all of those applications is a bit broken because you're. You have all these companies that are focused on individual products. And what that does is it loses the opportunity to build the type of horizontal infrastructure that's been so enabling in, like the tech industry, for example, right? You've got these massive data centers that are allowing even tiny companies to access large amounts of compute, really cost efficiently. Like, we don't have that same infrastructure in the biological field today. And so that's really what Ginkgo is building, right? Like, we are building, it's a huge, huge wet lab run by robots and a lot of software that tries to lower the cost of doing like, physical biological experiments, like making DNA, sticking it in cells, testing those cells to see what they do. And then in the process of doing that, we generate a ton of data about biology, right? So what does this gene sequence do? Can we actually start understanding that? And that's put us in a really interesting position to start leveraging generative AI to try to get us closer to being able to answer these questions around, you know, if I wanted to design some new piece of biology that does something interesting and important, how would I do it on a first principles basis?

All right? And I'm going to maybe oversimplify this. And so we can put this Conversation, you know, frame it a little bit. So is it essentially that, you know, what you're doing at Ginkgo is, you know, collecting and organizing data from all different aspects of. Of the biological world and then creating models for everyone else to use. Is that kind of how it goes?

That's part of it, yeah. But importantly, we also do that sort of reinforcement learning stuff. So it's like the reality is that there's a lot of data out there, but there's a lot more we don't know about biology than what we do know about biology. And one interesting thing that I think I like to make people think about is if you think about large language models and human language, there's a really high bar because we invented human language. And so you're teaching a model to do something that we invented. Like, we've got high standards for that. We did not invent biology. Like, we are just students of biology. Like, we call it drug discovery, not drug engineering, for a reason. Like, we are out there discovering drugs. And what we want to do is we want to get to a point where we actually understand this space enough that we can start applying engineering principles to it. In the same way that we can build, you know, human language in the same way we can write computer code. All these things we invented, we can engineer with them. We're not really there yet with biology, but that's. That's what Ginkgo is trying to do. And so, yes, we have the model side, and then we have this kind of data generation side, because there's still a lot we need to go test and learn to make these models. Good.

And hey, as a reminder, everyone who's joining us live, thank you as always. Please get your questions in now. And it's always so sad when a great question comes in right at the end of the show. So what do you want to know about the AI revolution in biology? I think there's so much that we can learn here. Maybe. Anna Marie, Let's. Let's kind of start at the end. What does this look like? So, you know, everything that's going on in the world with, you know, AI and biology and what you're doing at Ginkgo? What does it look like in the end? Is it, you know, the ability for other companies to use these models that you create to help create, you know, better medicine, you know, healthier food, better crops? Like, what does it ultimately look like if this kind of marriage between AI and biology in the long term is successful?

Yeah, yeah. So I think all those things that you mentioned. Those are things I hope happen actually in the like relatively near future. Like that, that should happen, you know, in my career, I hope. Right. And, and where I'd like to see that go is like imagine a chatgpt prompt. Like I as a researcher should be able to type in. I want to develop a. So like maybe I'm a doctor, here's my patient. I'm going to upload their genome and these are their symptoms. We've identified a cancer. Here's the pathology of the cancer and I would like to make a therapeutic for it. Please generate a therapeutic that kills the cancer and doesn't have any side effects for this patient. And you should have a model that is smart enough and has seen enough biology that it understands. Okay, well here's how I'm going to target that cancer cell and kill it. And here are all the other proteins that this person has that play important functions. And I don't want to touch those. I don't want to kill any of those healthy cells. I don't want to interfere with any of their other biology. And so I'm going to have a totally personalized therapy for that person. Like that would be. And as a doctor, I should be able to type that in in English and have some cool, you know, medicine, programmable medicine come out the other side. Like that is what I think this type of technology enables. But then remember, this is like every field, right? So my favorite sci fi example is I want to be able to sit down at a computer. Let's say I want to buy a new house. Instead of buying a new house or buying a plot of land and building it, I want to be able to go into a CAD system and design a house like the Sims or something and then have it like print out DNA for that house into like a seed that I can plant in the ground and like water and have it grow. Like, I mean biology is amazing. It can do these things. We just don't understand how to, how to manipulate that DNA in order to take advantage of all of those capabilities.

Are you still running in circles trying to figure out how to actually grow.

Your business with AI?

Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on Genai. Hey, this is Jordan Wilson, host of this very podcast. Companies like Adobe, Microsoft and Nvidia have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead. And some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use Gen AI. So whether you're looking for ChatGPT training for thousands or just need help building your front end AI strategy, you can partner with us too. Just like some of the biggest companies.

In the world do.

Go to your everydayai.com partner to get in contact with our team or you can just click on the partner section of our website. We'll help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on Gen AI.

I. Oh gosh. I mean the, like, obviously I'm, I'm not a scientist and I'm not, you know, I'm not one that even understands biology very well. I think, yeah, it's been 15 years, but the, the thought that that could actually happen. Right. Is amazing to me. But so in the long term, you know, or you said, hey, even in your career you'd like to think that these things are, are very obtainable, some of them, you know, being able to use kind of this, kind of this new wave of AI and biology to create better medicines, more personalized, you know, like in your example that doctor being able to sit down and type in real English. What are the challenges to getting there? Right, because obviously there's a lot of data. You know, making these models is not easy. But you know, specifically from your vantage point, what are the biggest challenges until we can get there in those various fields?

Yeah, yeah. So I think the challenge is twofold. One is related to what I mentioned earlier, which is we didn't invent biology, right? We're students of biology and therefore we don't fundamentally understand all of the rules of biology and therefore we rely on the study of real world data. And so then the question is, how easy is it to get the data? And there is a huge amount of data that is publicly available now, which we can, which we can and are using, but there's also a huge amount of data that we don't have. And to get biological data, remember these are physical experiments. Like you're printing DNA, you're moving liquids around, you're growing cells, you're measuring them. And so like one analogy I like to give is you can think about programming biology like programming a computer, right? It's ACTG, not 0 and 1. But you can think about it similarly. But a big difference is that flipping a bit is effectively free today in computers. Flipping a bit in biology, like switching an A to a T is not free. It is Very, very, very expensive. Even just printing that code, like writing the code is a few cents per base pair. Right. And so the issue we have, and one of the things Ginkgo is really focused on is that it's just really expensive to create high quality labeled training data for AI models. And so it's, it's, until we, you know, bring that cost down a few more orders of magnitude, I think that will be our biggest limiting factor.

Interesting. Yeah, it's, I, I guess unless you're in your position, you really don't know or understand how the whole process works. So thanks, thanks for letting us know a little bit about this. I'm going to go straight to questions because we already have some good questions and we'll get back because I have, I have more questions, but a good, a good one here from Dr. Harvey Castro. Thanks for joining us. Just asking, what are some top examples of kind of AI revolution or maybe some, some new AI advancements in biology? So yeah, some, some, some real examples. We talked some, some theoreticals and things you're working toward. What's kind of out in the wild, Maybe some things that you've, you've worked on.

Yeah, so there are a couple of cool examples that we've worked on at Ginkgo. So one, one is, so there's a class of proteins called enzymes. And enzymes are proteins that basically catalyze chemical reactions. So it makes chemistry happen, it serves a function. And almost every program we work on for a customer at some point involves enzyme engineering or protein engineering. And so we've already incorporated many AI models and trained them on our data sets to help answer protein engineering questions. And so these might be things like help me engineer this protein to better catalyze that reaction. So I have to use less of it to get the impact I want or help make it more stable so it doesn't break down when it's sitting on my shelf or it doesn't break down when it's in a high temperature reaction or something. And so we have AI models that we deploy today that can help us take just a generic sequence from the wild or even from really no starting point and develop a new sequence that performs a better protein function. So that's one, Another really cool one is on the biosecurity side. So we obviously think a lot about the other side of the AI opportunity, which is the risk. So you know, there is like we are susceptible to biology. Right. So you know, if you hear about folks kind of doomsdaying the future of AI, it's Usually around the intersection of AI and biology. Like, what happens if you can make a bioweapon with AI? And so we've invested a lot in building biosecurity infrastructure. You know, how do you have, effectively, bioradar that is monitoring the world for new biological threats and then identifying. And this is where the AI piece comes in. If I find a new piece of biology I've never seen before, can I answer a few questions? Can I answer what it does and should I be worried about it? Can I answer, where did it come from? Like, is it engineered or not? And can I answer, what do I do about it? Like, how do I make a vaccine or therapy that protects us from this? And so we did some really cool work with IARPA, which is like the CIA's innovation agency, on one of those questions, which is, if you find a piece of DNA, can you tell whether it's engineered or not? So it's sort of the equivalent of like, if kids are cheating on a Test by using OpenAI, can you tell that AI wrote their exam? It's kind of the same thing for biology. Like, if I found a piece of biology, can I tell if somebody engineered it? Or was it just like Mother Nature threw something at us that we don't like? So it's a really, really neat application of AI today.

And I'm glad Tanya had this question just now. So saying, can you reiterate how you are going about retraining these models? Because, yeah, I think, you know, the example that you gave, it's, it's easy with human language or probably much easier, right, to train different AI models. So talk, talk maybe a little bit about how that works. So you said it's, it's expensive, it's, you know, complex. But yeah, maybe just, just take us a little bit more behind how you're actually training and retraining these models that deal with biology, which. Yeah, we didn't create.

Yeah, yeah. Well, we're able to leverage kind of the state of the art technologies and model architectures that are out there. And there are a lot of parallels between the architecture of human language and the architecture of biology. Right. So you can think about, you know, an amino acid or a nucleotide in biology as the equivalent of like a letter. Right. And, and then you see these tokens, right, which are combinations of letters that are reused frequently. Like, you see the same thing in biology. Like a token might be a sequence of amino acids that tends to show up in a lot of different places. Like it's a recurring Theme, it's a backbone or something. And then those are assembled to make things that have meaning. Right. Like an entire protein. And then you assemble proteins to make me. And so you can use a lot of that same model architecture. You're just feeding it a fundamentally different language, but it's still doing the same thing where it's recognizing. Okay, well, when I see this type of token in that context, well, the next token is usually that. I've got good odds that that token is that. And then the really tricky part comes in when you start wanting to build task specific applications on top of that with kind of multimodal data. Like what happens when I'm now importing. Okay, well, now I want to understand function, different kinds of functions, and I'm now importing different types of data and measurements. And then you get a little bit more complicated. But that's the basics of it.

No, I love it. And maybe it's also important to, you know, hit rewind on this a little bit because artificial intelligence is not new in biology. Right. I'm sure, you know, deep learning and, you know, neural networks have been widely used for many years.

But how, how, by the way, that I stole all their terminology from biology, because biology is amazing. Like a neural network is like brains. It's biology. It's all. It's coming through.

That's very true. Yeah, I'm glad. All right, hey, calling AI out, you know, hey, AI, whoever took that, give it back to biology. Right. But. But maybe talk a little bit about how with advancements in generative AI, right, Large language models even, how is this even changing AI in general in biology? And yeah, maybe give us a quick little history lesson and you know, what the new advancements in generative AI mean for your space.

Yeah. So I think the big one for us. So historically, in. At the intersection of AI and biology, what we saw was that the focus was really on the architecture and there was sort of this acceptance that, well, because we're not going to have very much data, we need to have really good models that can deal with really small amounts of high quality data. And that was sort of the way that the industry thought about AI. Historically, I think what changed with the advent of the transformer architecture was this ability to take in large amounts of unstructured data and get useful insights out of it. Because again, historically, the biology field was very, very focused on small amounts of structured data. Tell us something useful and yeah, ML can kind of help us on the margin, but it was really around the quality of a specific experiment. That you would run to answer a specific question. Now, we can advance the state of the art by taking in these massive amounts of unstructured data sets that I can't understand. There's nothing I can really do with it. But these AI models can now start identifying the patterns and the functionalities that are present in that unstructured data. And that's become a much more useful foundation.

And if you are a little newer or not as AI geeky as maybe me. So kind of what we're talking about here is structured data is data that you can easily categorize. Right. And yeah, many different sectors have been using structured data for decades. But now with large language models and generative AI, it allows us to better use the unstructured data, which is, you know, data that maybe can't easily be categorized because it needs a human to interpret. So it's really cool. So maybe one thing. Actually another great question, because this was on my mind too, so asking. So, Taylor, thank you for the question. So saying, has ginkgo or other companies thought about using AI with crispr? I think that's how it's pronounced. I'm not sure for gene modification, but, yeah, I'm super interested in, you know, how biology, even biology and AI come together for gene modification. So, yeah, what's.

Yeah, no, I'm really excited about the gene editing space in general. So maybe just a quick science history for folks that have studied biology. So what is crispr? CRISPR is a protein complex. Like, it's a series, it's a biological complex that has emerged in microbes as they defend each other to defend themselves from viruses. And so what's interesting, like, if you think about the dirt in your backyard, there are billions, trillions of little bugs living in there that are constantly battling each other for space and territory and are getting infected by viruses. And so they're evolving very quickly, Interesting genetic tools to protect themselves. And so CRISPR is one of those tools. So we have a really large, a couple billion member, what we call a metagenomic database. So that is if you sequence all of these little microbes that live in random places and kind of understand what types of proteins they're making, we've got a very large proprietary collection of that. And so we've looked in that to see. Okay, well, what else is in there that looks like a gene editor? Because crispr, honestly, we discovered a little bit by accident, much like many of our medicines today. It was a bit of an accidental discovery, like penicillin, like bread got moldy and now we have antibiotics. It's great. That is the last couple of centuries of biology. And so when we now we have the ability to use AI to be much more targeted about, like, what type of a gene edit do we want to make? Because CRISPR is honestly a little bit crude. So instead of, like breaking all of the DNA apart and sticking stuff in and you've got all sorts of edits in places you don't want, can you find gene editors that are much more precise, much less disruptive, much safer, that can make bigger changes, maybe? So different types of genetic disorders that you might want to treat or engineering applications of gene editors have different requirements. And right now, the biological field tends to use the same hammer for every job. And what we want to do is sometimes you need a hammer, sometimes you need a screwdriver. And those are different tools. AI allows us to search the 3.7 billion years of evolution to find that type of diversity so that we can use better tools.

And what would this ultimately be used for? Right? Like, if we can, with the help of AI, you know, if AI and biology can cook up in the lab and, you know, make better DNA for us, like, what does that mean? Is that, you know, like. Like, what woozy is asking here about, you know, anti aging or what does that ultimately mean if we can use AI and biology to help alter DNA?

Yeah, I mean, I think we are only limited by our imagination. Like, this is. Biology follows rules, but it feels like magic to us because we don't understand the rules. And so I think what AI does is it, you know, shows a little bit more of the how and the why behind the magic. And so, yeah, I mean, anti aging seems like it's totally within the realm of what we could choose to do with biology. I do think you shift pretty quickly into the world of what should we do with biology. Right. I think that that will actually, to me, be the much more interesting conversation that we're having 10 years from now. It's not what can we do? It's what are we as a society comfortable doing, doing. But I do think this intersect we are at that inflection point to the vertical part of the exponential curve on a number of dimensions. Like, it is getting exponentially cheaper to read DNA, exponentially cheaper to write DNA, exponentially cheaper to do the types of experiments we are doing in our labs that help us train AI models, which are getting exponentially cheaper to run and more advanced. And so that's the intersection of a lot of exponential technologies. This stuff is going to move really fast. And I think we're going to have a hard time keeping up with that speed. And the more interesting questions will be how do we choose to, how do we choose to use it?

Yeah, because then I, yeah, and another great question here that, that leads to how is AI being regulated in biology? Because yeah, it does seem like once you achieve, you know, certain breakthroughs, there's probably some ethical choices to be made or some deep conversations as a society. But at least, you know, right now and in the near future. How is AI being regulated in biology?

Yeah, fits and starts. So there's, there's like the basic infrastructure of like drugs still have to go to clinical trials and get FDA approval before patients get treated with it. Right. So there's like basic stuff of before things hit the market, there is a regulatory process on AI Specifically. What you see in like the executive order that came out a couple weeks ago is like, the government is certainly very interested in how AI is being applied to biology. And so in the same way that they're saying, hey, please check in with us if you're training a really, really big human language model, they're also saying, check in with us if you're training a model on biological data. And so we are collaborating very closely, obviously with our partners across both the private sector and the government. There's also a lot of self regulation that's happened in this space. So for many years we've been part of a consortium of companies that have the capability of doing DNA synthesis. So this is like writing biological code where we self regulate. And we've said, all right, we need to make sure any piece of DNA we're printing is not a known pathogen. Like you can't call us up and make anthrax. Like that should not be possible. Right. And so there's some self regulation that's happened. But I think, again, I do think this is one of the hardest questions because this technology is going to move so fast, it's going to be really hard for regulation to keep up. And I think there's a delicate balance here of sort of recognizing both the benefits and the potential risks of this technology, but also recognizing that it will be this technology that allows us, it allows us to respond to the risks as well. And so making sure that responsible parties are able to advance the state of the art and be in a position where we can respond to bad actors or accidents like that is, that is really important.

And you know, as, as we wrap up here because Anna Marie, we've, we, we've talked about Everything. I'm glad we got to the ethical piece, but we've talked about how AI is being used, how it could be used, some of the ethical concerns. But you know, maybe what's that one takeaway that you hope, you know, the everyday person can, can learn from today's episode and, and how, you know, AI is, is being used in, in the biology field and how it impacts their lives. What's that one thing that you want people to know that will help them understand this field? That it's kind of hard to understand?

Yeah, I mean, I think like my main shtick right now is like, I think biology is the future kind of computer science. Like for the last 20 years, everyone was like, oh man, I wish I studied computer science because that's the future. Like today that thing is biology. Biology is programmable. Like it is the next frontier of major scientific discipline that we will be able to proactively forward engineer. And so making, helping kids fall in love with biology, helping adults who are interested in shifting gears care about biology and recognizing the impact it can have, both for good and for bad, is my top mission. And so if I can get a couple more people interested in the field of biology through this, mission accomplished.

All right, well, hey, there's so much to be interested in. So much we couldn't even get to. But hey, that's why every single day we put out a daily newsletter. So there's going to be a lot more there. But Anna Marie, thank you so much for joining the Everyday AI show and helping us better understand this AI revolution that's going on in biology. Thank you so much for joining us.

Thanks for having me.

Hey, and as a reminder, like I said, please do go to your everydayai.com. there's going to be a lot more. We're going to share a little bit more on what Ginkgo is working on, as well as maybe some topics we didn't get as long to dive into. So make sure you go check that out and we hope to see you back for more Everyday AI. Thanks y'.

All.

Appreciate it.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going for a little more AI magic. Visit you your everydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.