EP 333: GenAI’s Impact on Science and Our Environment

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.

When we think of generative AI, I think a lot of people don't really think of science or biology, right? We just think of being more productive with that spreadsheet or writing a better sales cover letter. Right. Or, or something like that. But that's not what generative AI is, is capable of. That's not its floor, that's not its ceiling. That's just what a lot of us think. But there's actually ways that some of the world's most advanced bio companies, life science companies, are using generative AI in ways that really have the, the ability to impact life, to impact science, to impact biology. And that's what we're going to be talking about today on Everyday AI. What's going on, y'? All? Thanks for joining me. My name is Jordan Wilson. I'm the host and Everyday AI, it's for you, it is your guide for everyday people like you and me to learn generative AI, to learn really how it works and how other companies are putting it to work and how it's impacting our daily lives and society as we know it. And today is no different. So normally we go over the AI news. Technically this is pre recorded show, but we're debuting it live. So as always, you can just go to your everydayai.com and you can sign up for the free daily newsletter where we'll be recapping the news as we always do. If you have comments, go ahead, put them in, I'll be there, you know, answering and maybe our guests will be able to as well. So don't forget to sign up for that daily newsletter where we will also be recapping today's conversation. All right, enough about that. Let's go ahead and bring on our guest today and talk about how generative AI can actually maybe turn trash into treasures. All right, there we go. So James Daniel is the Vice President of Artificial Intelligence and Computational Biology at Lanzatech. James, thank you so much for joining the Everyday AI Show.

Thank you for having me.

All right, hey, can you tell us a little bit what you do in your role there and what Lanzatech is?

For those that aren't familiar at Lanzatec, we are making technology to help solve climate change. So it is time to move away from this practice of extracting fossil fuels, using them once and then polluting. We need to keep fossil carbon in the ground and so what makes this difficult is so many things that we use every day come from fossil carbon. So if you think of tennis shoes, hand sanitizer, laundry detergent, car tires, aviation fuel, what do they all have in common?

Carbon.

Exactly. Yeah, they all come from.

Yeah, I got one. Right. All right.

And so this is something that many people don't realize. So all of these types of material goods are made from carbon that's extracted from the ground. And so if we still need to make these things that we want to stop extracting from the ground, it sounds like an impossible situation, but it's not. Because the good news is we have enough carbon above ground to make everything we need. We just need to capture and recycle it instead of dumping it into the atmosphere. And so at Lanzatec, we are a technology company, and we're developing. Exactly that. We develop carbon recycling technology. And the way it works is we capture pollution and carbon emissions, we feed them to hungry microbes, and then the microbes eat the emissions and turn them into chemicals. They turn them into building block chemicals that are used to make fuels and everyday products. So you can go and buy things today that are made from carbon that we recycled. Things you can go and buy perfume from Gucci, athletic clothes and tennis shoes from Adidas made using recycled carbon.

Wow. So, I mean, let's just skip to the end. So how has generative AI helped in this process? Right, and I love the way that you simplified it there for, you know, people like me who aren't, you know, super, super smart into biology. I don't know if I've taken a biology course since I was, you know, like, 16 old. So how are you actually using generative AI to kind of capture this carbon above ground and recycle it and, you know, put it into use, into everyday products that we all, you know, use and love? How does generative. Generative AI play into that equation?

Yeah, AI is an enabling technology for us, so it enables many of the things we do, and it helps us make our carbon recycling technology more efficient. And so the main area where we apply generative AI is to engineer the biology at the center of our technology. And so I can talk a bit about that if you'd like.

Yeah, let's get into it. Tell us a little more about that.

Yeah. So before I talk about how we're using AI to engineer biology, I think it makes sense to kind of answer the question, why are we engineering biology? So we build these big refining plants, and so you can think of there as being two components to our technology, to our refining plants. So there's the bioreactor technology, which is like these big tanks, like steel in the ground, and then there's the microbes. So we put the microbes inside the tanks, we install the microbes inside the bioreactors, and then they eat waste gases and they make ethanol and other products. So the tanks and the hardware is fixed, but the microbes can be upgradable. And so we use AI to help us develop new and improved microbes. And so it's kind of like software upgrades. So you get a microbe version one, microbe version two, version three, that's even more efficient and at recycling carbon. Or you might put in a microbe that's customized to produce a particular chemical, like isopropanol. And so, and so, so we engineer biology to develop this technology. And this was, this was our first application of AI 10 years ago. And we started using it not because it was cool, but because we. We had this problem, right? We badly needed to solve this problem around engineering biology. And the problem was really what AI offers is it gives us this missing piece of the puzzle in understanding and predicting biology. We use biology, we love using biology. Biology is incredibly powerful. You can think of each one of our microbes as being like a little factory with really sophisticated machinery inside it. Every second, there are millions of chemical reactions that are happening. They're like these little assembly lines that are converting chemicals into. By harnessing this incredibly powerful biology, we can very efficiently recycle carbon. Biology is sophisticated, it's complex, and it's hard to fully understand how it works. But if we want to engineer biology using genetic engineering to get it to do all these amazing things, we need to be able to predict and understand. Otherwise you're just stumbling around in the dark. So you're doing tons of trial and error in the lab. You'll try something and it didn't work because maybe you have a knowledge gap, which meant you made a bad prediction. Sometimes you'll try something and it didn't work, and you just had no idea why it didn't work. So the key point is it turns out that machine learning is an excellent descriptive language of biology. So you can throw all of this biological data in AI system, It will learn patterns, it will figure out what's happening, and then it will allow you to make predictions. And so our scientists use this technology across our research stack. And a good example of generative AI is effectively a version of ChatGPT that we use that can speak biology.

Yeah, a lot there. I Kind of want to dissect this one piece at a time. So one thing that you talked about, James, is your company, like a lot of companies, have been using AI for a long time. Companies in biology and life science have been using AI and deep learning, machine learning for decades. Specifically when it comes to the generative side, Right, when we talk about large language models, even for you personally, how has that changed? Not just, you know, the day to day work of people working in these, in these fields and related fields, but how does it also change what's possible?

Yeah, so there was a big algorithmic breakthrough in 2017 and that was the development of the transformer at Google. And so many, quite a few people know that the transformer has enabled generative AI technology like large language models and ChatGPT. Not as many people realize that that breakthrough has also enabled generative AI for biology. And so ChatGPT is trained mostly on text, right? On the Internet, on written language. Our biologists work with similar models that are trained on the language of life itself. So DNA is like a language, it stores biological knowledge. It's a sequence of nucleotides, like a sequence of letters. And so we use these big GPT style foundation models in the order of 15 or 20 billion parameters. So these generative pre trained transformers that are pre trained on a large number of biological sequences. So over weeks and over months, these models are pre training on biological sequences from all sorts of bacteria. Then during that process, the language model is learning something really useful from it all. It's learning some fundamental rules of biology in the same way that a large language model is learning rules of language as well as some understanding of the world. Then once we've pre trained one of these models, we can use it for tasks like ChatGPT, you could generate text, or we could use it to classify text. Biological language model, you can actually generate the language of life, which you can then engineer into a microbe. And so like I talked about how our microbes are like these little cell factories with all these little assembly lines. So you can imagine that proteins inside the microbes are like machinery on these assembly lines. And so we can generate biological sequences that allow us to create new or better proteins that we put into the microbes that will allow our microbe to make a particular building block chemical, for example. And so we've done this for like isopropanol or isoprene, which are important building block chemicals. And so really like if I look back over the last 10 years, that has been, I think the biggest breakthrough. Well There have been many breakthroughs, but I would say that is the biggest breakthrough that we've seen from that transformer model architecture.

Yeah, and you're right, right. Like I think as soon as, you know, the early GPT models, you know, started to, became or started to become commercially and you know, publicly available, different sectors. Right. Started to adopt him for their own uses. You know, I'm curious because you mentioned at Lens Attack that, you know, you almost have this version of Chat GPT internally. Right. And we've talked to some companies on the Everyday AI show before that have, you know, kind of built their own model or, you know, fine tuned their own model. So, you know, I'm curious, how has that process been been so far for your company and what has it allowed others to do? Because I'm sure not everyone, you know, is, is the vice president of AI and Computational biology. So there's probably people that have maybe a little bit more of a learning curve when it comes to working with AI. So what has having kind of your own internal model enabled your team to do?

Yeah, so there's, there's two aspects to my role. So I oversee our computational R and D and, and so we're applying these techniques across all stages of research and development there. But I'm also responsible for our AI strategy across the business. And so we've been adopting this technology across the business. One part of our broader AI strategy is, is quite simple. So like many businesses, we're just empowering all of our people with the skills and tools to use large language models because it makes them way more effective. And so I think all businesses are going through this right now, or if they're not, they should go through it if they want to stay competitive. And so I think we have many different use cases, but tactically, the first and the easiest thing that the team did was just giving all of our people across all functions access to a safe and secure large language model. So like a basic AI assistant to help with writing, summarization, brainstorming. And so we call it Lanza chat. It's like on 10 or chatgpt or Claude or Gemini. And one of the interesting things I've found is that I have yet to find a single person in the company who can't get value once they've been properly trained from using this tool. And so based on that, I can confidently say that I think all businesses can make their knowledge workers more productive by doing that. I'm sure you agree with that, Jordan.

Yeah, and I'm glad you said that. Because I still think that there's a common misconception out there in the business world that so many employees aren't going to find value from working with a large language model or you know, even a domain specific large language model is actually having a conversation just, just last night with someone who is in a similar position. They, they have a model for their team and they're like, well, you know, the, the rate of adoption is, is rather low right now and, and we really need to, you know, increase that. So, you know, I'm curious from, you know, someone who's kind of charged with making sure that, you know, a large company such, such, such as Lanzatech is what are some of your key findings or key learnings that maybe you can share with others who are in a similar position who maybe have their own model, but they're struggling to get more and more people to use it and to find value?

Yeah, so it's never just a technology thing, right? People quite often focus on the technology, but a big part of it is actually the people and the processes as you implement the technology. I mean like the, the data is there that shows that this stuff is making people more productive. Stanford just released their 2024 AI index. And I haven't read the whole thing yet because it's 500 pages. But in part of it, they, they summarize several studies from 2023 that assessed AI impact on labor. And kind of the consensus there is that it makes workers more productive, leads to higher quality work. But in terms of what we've learned, one of the interesting things is we had early adopters, so we had people who were already quite sophisticated and using these tools in their personal lives. And so we gave them a safe way to use us at work. We gave them something that was grounded on our organizational data and they did a great job, they created value. But what was interesting for me is the biggest wins that we had were with non early adopters. So people who had barely even heard of ChatGPT, or maybe they tried like, like the free one GPT 3.5, which, which isn't representative of what AI can do today. And so we, we gave them training. So again, it's not just giving them the technology, it's actually giving training upskilling. We gave them a little bit of a push and it was really interesting to see how much more efficient and productive they could be and some of the interesting use cases. So we had like many examples. One that I remember vividly is someone was gearing up to do a particularly well, they described it as a particularly painful and frustrating documentation task. And they had planned to do that in three to four months and they did it in six days in partnership with a language model. Wow. And so we've got some really great stories across the business like that. And so as people are kind of leading the way, they're talking to colleagues and then we're just, we have this transformation process that's happening. And so I think this will happen through all businesses, you know, and I.

Think that, you know, begs the question, you know, what happens, you know, what happens when you start to get those kind of efficiency gains across the board? Because it is a process, right? So, you know, having your own model is always, you know, much better than using a, you know, public model, right? When you can, you know, use rag and fine tuning and, you know, have your own domain, you know, knowledge in there and that helps get those, you know, efficiency gains like you talked about. You know, one thing that I always think about is, you know, the DeepMind, right? And they use a specialized model to, to, I think it was called Fun Search to kind of solve this unsolvable math problem that the world's smartest mathematicians had been working decades to, to solve. Right. So when you think about your end of the spectrum, when you think about, you know, you're talking about genetic engineering and solving these, these great biological problems, what do you think language models will help? Whether it's your team or just, you know, your industry, what kind of big problems do you think, you know, they could solve?

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 Gen AI. 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, their employees on how to use Gen AI. So whether you're looking for chat, GPT 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.

Yeah, so I'm, so I'm, I'm focused quite a lot on technology development. And so it's clear that AI accelerates technology development in general, which is, I mean, which is no different to how other general purpose technologies have accelerated technology development. So, like electricity or the computer, If I talk to the team and say, hey, for the next three weeks, do your work without electricity. It just sounds crazy, right? And so AI will be the same. And I think it makes people more efficient and also more effective. And so that's, I mean, that's exciting, right? Because if AI is a great accelerator of science, then that means we can do more science, we can do better science, we can develop technology more quickly. We need to develop technology more quickly to address many of the challenges that we face. And so I think that's, that's really the outcome, right? It allows people to do more and to be more effective.

Might it, might it be to the point? And again, I'm no, I'm no expert here. I'm, I'm learning along with the rest of our audience. Right. So we, we talked about in the beginning, James, you know, kind of this, this process of what you're trying to do at Lanzatech, very overly simplified, right? But, you know, trying to, you know, capture carbon out of existing products. Maybe they're, they're. Before they go to the landfill, you know, versus having to dig into new sources of carbon because the world needs it, right? Might there be a time when we don't need to be digging up new carbon and we'll be able to, everything will kind of be, you know, able to be recycled. Is, is that a possibility? You know, walk us through that?

Absolutely. I mean, that's our vision. I mean, by, like, I would like to see by let's say 2024, where every consumer can make a choice where they can buy products that have been made from recycled carbon as opposed to carbon that's come out of the ground. And as I said, there is enough carbon above ground to make everything we need. And so through technology, we can drive this change.

You know, I'm curious, even for yourself, you have a very unique vantage point, you know, working for a large company such as Lanzate. So you are both looking forward into new innovations and new biological problems that you're trying to solve for the world. But then you're also looking internally right at your team and looking for ways to use generative AI to help make the company more efficient and more impactful in its work. So I'm curious, even for you, what is whether you want to talk day to day, week to week, month to month, but what is the way that you are excited by the, the prospects of using generative AI in your own work? Because I think that'll help open up the eyes for a lot of others.

It's, and it's a difficult question to answer because you can see how it can touch all aspects of knowledge work. Right. One, so one thing that's quite exciting for me, so I, I have a background in science, and I mentioned that AI accelerates the scientific process. And so you think, what is the scientific process? So you do background research, you might design experiments, you'll analyze data, you'll interpret results. And so one thing that I find quite exciting and one thing that I wish we had like 15, 20 years ago was, was the ability to, to automate a lot of this process of background research. And so I'll give an example of what we are doing. So before someone does an experiment, the first thing you want to do is you want to understand what's been done before. Right? You want to. And that often requires reading a lot of scientific literature. Inside our company in Lanzatech, we have an internal scientific knowledge base that We've built over 15 years of science knowledge, and it has over 30,000 pages that our experts have written. And so you imagine if someone's about to do some work, they need to find out what's been done before. And that can take a while if you're having to do it yourself. I can see examples of where it might have taken five days of going back and reading and figuring out what's been done before. We built a large language model on that knowledge base to help our scientists do that. And now you can do that in a day. The language model can build those connections, extract the insight. That type of thing is really exciting for me just because of its ability to speed up the process of doing science.

Wow. So you, you know, I'm curious, what's, what's next? What's next for whether Lanza Tech and, and using AI, Maybe can you talk about any, you know, big projects that are coming down the pipeline or just for your industry in general? Right. And, and as we look at, you know, kind of the, the topic of today about how we can, you know, turn trash into treasures? What's, what's next, you know, what should we be looking at, you know, as we hope to, you know, solve this problem or solve more with generative AI.

So one interesting area for us in science is using AI to allow us to do really efficient and effective experiments. And so that's an area of development. There's a lot of hype around this idea of an automated scientist. I, I don't think that's the right framing of the problem. We still have our scientists who are working in partnership with, with these techniques, but that, that's quite exciting for me because if you can really accelerate your ability to do experiments or really decrease the need for the number of costly and time consuming experiments, I think that'll really transform things. But what's, I mean, from my perspective, and this is actually something that we communicate with the teams during our training when they're using this. I think it's really important when using this technology that the human is always in the driver's seat. And so that's the expectation that we're setting across the board when we're using this technology. The human, the person is always checking results or is always owning the work output instead of saying, oh, the AI did this. That's why there was a mistake. We never want to see that. One of our values as a company is we own our decisions. And so that's the way that we're framing the use of these tools. You own the output, you always stay in the driver's seat. And I think with that, that gives us the best of this human AI partnership.

Yeah, I'd really like to use that. If, if I have a bad show, you know, I'll just use that crutch. The AI. The AI made me make this bad show. But I mean, you bring up a good point. Right? Because in the end, as generative, you know, even generative AI becomes more and more commonplace in our daily lives and in society, in our daily workflows, you do have to find that right balance. And I love what you said there, James, about you human is always in the driver's seat. Right. Even if it's an autonomous car, you know, with AI, like, you still have to put a human somehow, either literally or figuratively, in the driver's seat to make sure that they have accountability for those decisions.

So.

So James, we've talked about a lot in today's conversation, but, you know, I'm wondering what is the one main takeaway that you hope that people can take away, you know, specifically as it comes to, you know, kind of the work that you all are doing and how you are using generative AI to get it done?

Yeah. So I personally think we're entering this kind of golden age of technology. Development. And from what I see at lanstec, so across the world, right, we have incredibly smart scientists and engineers and innovators who are dedicating a lot of effort to solving these big challenges, challenges like climate change. And so I think AI is just really allowing people to massively increase their impact. I mean, we're excited about AI, right? But what I see at the moment really, is IA intelligence amplification, and that's what we're seeing right now. And we're really excited about that.

Love that. I just jotted that down, you know, flip the script. I love that intelligence amplification. So another gem for us. So thank you so much, James, for joining the Everyday AI Show. We really appreciate your time and your insight.

Thank you for having me, Jordan.

All right. Hey, a lot that we covered in today's episode. Maybe you were driving, maybe you were out walking your dog, or maybe, like me, sometimes science and biology can be a little bit confusing. So don't worry. We're going to be recapping everything in today's highlights from the show. So make sure to go to your everyday AI dot com, check out the daily newsletter and also check us out again. We hope to see you back tomorrow and every day.

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 youreverydayai.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.