Unlocking AI KPIs With Insilico

Audio:
Welcome to TD Cowen Insights, a space that brings leading thinkers together to share insights and ideas shaping the world around us. Join us as we converse with the top minds who are influencing our global sectors.
Brendan Smith:
Welcome back to another episode of Machine Medicine: AI & Health Care, TD Cowen's podcast series where we bring you the latest and most important takeaways from the state of AI and the healthcare sector today. I'm your host and TD Cowen healthcare analyst, Brendan Smith. And today, I am joined by Insilico Medicine's founder and chief executive officer, Alex Zhavoronkov.
Alex, it's great to have you. Welcome.
Alex Zhavoronkov:
Very happy to be with you, Brendan.
Brendan Smith:
So, for anyone new to our podcast series, Machine Medicine aims to break down the use of artificial intelligence across healthcare into bite-sized, digestible points, one episode at a time, highlight the biggest misconceptions, and then recontextualize each piece back into the bigger picture. And today, Alex and I will discuss some of the most important benchmarks and key performance indicators, or KPIs, that companies like Insilico are pioneering as an increasingly indispensable way to compare different AI platforms.
So, Alex, let's just dive right in. You and I have spoken before about some of the difficulties that investors often face when trying to assess the relative productivity and efficiency of different AI platforms within the healthcare space. I know you and your team have really led the charge here in terms of identifying and disclosing a range of preclinical KPIs, really arguing that more transparency, particularly within preclinical drug development, is not only helpful but, honestly, essential, given how far the field has come. So maybe let's just start with how you think about which metrics are most helpful to disclose and also why Insilico has decided to start putting some numbers around what essentially is an ROI for AI drug development.
Alex Zhavoronkov:
Thank you, Brendan. And I must say that TD Cowen was one of the first to spearhead benchmarking in our industry. I was very happy to see that in some of the reports and also at conferences. So, thank you.
When we started in this industry, and I would say the entire industry was kind of born around the time when deep learning started maturing and started showing the early signs of superintelligence and image recognition and voice recognition, text recognition. So I would say 2014, 2015. Then came generation. So that created a lot of hype in the industry. Actually, that was also around the time when we were born, and I saw many of the more classical biotechs repositioning themselves as AI-powered drug-discovery companies, or kind of old machine learning companies or computational-augmented drug-discovery companies repositioning themselves as AI. So it became universal.
And at that time, for investors, for employees, for the pharmaceutical companies more specifically, it became very difficult to understand who is who and how the progress is being made. In those early days, even myself, I didn't know how to benchmark. At that time, we were benchmarking by the number of research papers at top AI conferences, like NeurIPS, or the number of AI patents that you are going after, or some tangible demonstrations of AI performance.
And then, a few years later, maybe two, three, four years later, the investors and ourselves alike suddenly woke up and realized, "Oh, my God. Well, we've been at it for a while. Where are the drugs?" Because when you are fundraising, when you are publishing in AI, very often you are using the premise that AI is going to make drug discovery faster, cheaper, and with higher probability of success. That's kind of the motto. Where are the drugs? And around that time, investors started asking us, "Okay, where are the drugs?" It was like, "Oh, but we are doing this and that." But then the question stuck in my mind and we realized that "Okay, we can perform serious miracles at the preclinical side and also on the clinical side, solve very difficult problems for pharma. Why don't we actually go and do it?"
So, around 2018, we've demonstrated several proof of concept. For example, one of my important works, generative tensorial reinforcement learning, GENTRL, was published or was submitted for publication, where we showed that within 46 days, we can synthesize six molecules for a given target. Four of them worked, one went into mice and showed PK, 46 days, 21 days spent in computation. We thought, "Wow, great." And it was, of course, that paper got really hyped up and also criticized by more kind of traditional medicinal chemists because they were saying, "Oh, the target is not new. The molecule is reasonably easy to make and looks similar to others." But those were the early days of drug discovery, where I can compare it to kittens recognizing kittens in pictures. But then this, again, the question that investors and your employees started asking, "Where are the drugs?" But that paper allowed us to fundraise.
And 2019, 2020 was the time when we actually said, "Okay, well, now let's put AI to work. Let's see how quickly we can discover a drug. And let's now try to set up our own benchmarks in order to have experimental validation." And at that time, I don't think anybody has tried that. And even ourselves, we kind of decided to do that as an extension of that Nature Biotechnology paper from 2019, where we published GENTRL. We thought, "Okay, let's take another drug for as long as we can go. But now we're going to address this question of novelty. So take a novel target, go for a novel molecule, everything from scratch, let's time it and see how long."
And we managed to fundraise for that. Thanks, God, at that time, it was great time to fundraise. Too bad we went kind of Spartan style and raised very little. Well, little, 37 million. Nowadays, people don't really consider it to be a serious series A for an AI drug-discovery company. At that time, it was a big boost for us, first big round. And we demonstrated that within 18 months, we can go from zero to what we call preclinical candidate or developmental candidate. For us, it's the same thing.
And for you and for all of your listeners, here is a crash course on drug discovery in the case that somebody might have missed that in graduate school or didn't go there. So, when you are discovering a drug, the process starts with the understanding of a mechanism of disease, why disease happens, identification of what we call protein targets. Usually, those are proteins that are driving the disease or heavily implicated on the disease. There are many of them, but you need to narrow it down to one or two or three, usually to one because the FDA wants you to have a very explainable mechanism. You validate the target experimentally. And then you identify either molecules or biologics or antibodies that disable that target. And then you try them in different experimental models. You try to do that in cells, in animals. And then you do toxicity studies and go into humans.
So, the preclinical candidates package or developmental candidate package includes everything up until you start formal toxicity studies that cost you a lot of money. Those are called IND-enabling studies, investigational new drug application studies. That require you to go for GLP toxicity studies, so good laboratory practices studies, in usually two species. So that would be, let's say, mouse and dog, dog and monkey, mouse and monkey, demonstrating the safety of the molecule. Then the FDA may or may not let you start a human clinical trial. Those safety studies are very formal and very expensive. Usually they take you 9 to 12 months. So, once you start doing IND-enabling studies, here is a trick. If you are doing it with AI, you may save a little bit of time, but most of the time savings happen before. So to the developmental candidate stage that happens before. And also, you cannot really modify the molecule or the target after you start those GLP toxicity studies. So that's when the bullet is fired, and that's it.
So, developmental candidate is where you make the bullet. And usually, traditional approach used to take about four-and-a-half years if the target is novel. So, we thought, "Well, how long is it going to take us?" And our first experiment where we took novel target, novel molecule, took 18 months. And we published it in Nature Biotechnology, explained how we did it, showed many, many experiments within this preclinical package. And we also realized that the quality of your developmental candidate package defines the success of the rest of the program. So you need to make it ultra-high quality because then you can actually sell it. There is a market for those things. And you can sell it at any stage, either a developmental candidate or later, in phase I, phase II, in IND enabling. So, there is a market. And the more novel your drug is, the more difficult it is to sell. You need to prove that it works extremely well. That's why a lot of companies are going after all targets.
So, we started benchmarking. So, okay, well, our first one was 18 months, but it was too novel. We had to progress it further. And we actually did not expect to progress it at all. We thought, "Okay, let us benchmark and see how long it takes." Once we nominated the preclinical candidate, I managed to fundraise more because investors immediately recognized, "My God, somebody actually has done that." Our entire team got extremely excited. We realized, "Okay, we actually have this capability." And at that time, I also got the co-CEO and chief science officer who knows how to discover drugs at scale. He worked for a contract research organization before, and before that, for Big Pharma, and before that, Harvard, graduate work. And we tried to extrapolate that first experience into a scalable model where we would go and try to develop many. So, in 2022, we nominated nine preclinical candidates. That was like, "Wow." That was my largest number of preclinical candidates nominated.
After that, we actually had to slow down, not because of our capacity, because of resources. Finances run biotech. Finance runs biotech. You cannot have 100 programs traveling in time unless you manage to sell a bunch to make it very sustainable. It's very difficult if you're going high novelty. So, once we nominated nine, we said, "Okay, well, now let's time it." And at that time, we set our first record, nine months to developmental candidate for a moderately novel target, QPCTL, that we actually nominated with a partner. And our longest time was still 18 months. And then our typical time was 13 months.
So, now we're in 2025. We nominated 22 developmental candidates, then reached human clinical stage, one, went into phase IIA and completed phase IIA. So safety and efficacy in human patients. And the study was designed for safety, but we unexpectedly saw efficacy. Well, traces of efficacy. So it's a very good trend. And we need to do a larger study, but it shows you that in five years, we managed to develop many, many good novel or moderately novel drugs that reached human clinical trials.
So, now, if an AI company is making a claim, faster, cheaper, higher probability of success, you can very easily ask them, "Okay, how many preclinical candidates? How long does it take you to a preclinical candidate? How many molecules do they synthesize? How much does it cost?" I can answer those questions. And if their AI is better than mine, I will actually try to use their AI. But the question of how to evaluate the output or AI platforms, or output of the AI platforms, it's very clear. We're all biotechs. You can call yourself tech bio all you want. But at the end of the day, the productivity of your team, of your platform, that's all what matters. And now we have benchmarks. In the past, traditional biotechs had four, five, six pipeline assets, and usually people were looking at the first one or maybe two, usually the first one. Now we can churn many, many of those programs at record speed with ultra-high quality because now we can also sell them.
Brendan Smith:
Yeah. Look, I think this is touching on a lot of really important points, particularly within, I would say, the understanding and maybe misunderstanding or short of understanding from a lot of the investment community in general. And I think one question we often get, or maybe just one point of contention, is that some investors are saying that AI healthcare companies, AI drug developers will just perpetually be disincentivized to disclose some of these KPIs. Just given the scrutiny, it can open them up to among peers, which, obviously, is kind of antithetical to the entire idea of trying to compare the progress of some of these different platforms.
So, I guess, from where you stand, I mean, net-net, do you think the field is now to a place where the blowback, if they can't compete on some of these KPIs, can actually outweigh the benefit of convincing potential investors that you're actually delivering? Or are we really to a space now where, "You need to show us and put us into a space where we can actually understand what are you doing, how are you differentiated." Are we there yet where companies are feeling that pressure to actually bring us back behind the curtain?
Alex Zhavoronkov:
Yeah. So, good question. I think that benchmarks are boring for journalists, for investors, for companies. So, people like flashy stories. And nowadays, even for SpaceX, if you look at... So, I love SpaceX in so many ways because this guy's actually set the benchmarks and they showed how you need to operate in AI drug-discovery space, because if you can do it well, you should take steps to launch your own rockets to orbit. And you can work on the next Starship, but you need to make sure that your Falcon is delivering payload all the time. And nowadays, believe it or not, media is not even writing about Falcons. So those launches became so routine that they don't make headlines. And same with us. So, in the past, when we were taking a drug into human clinical trials, the entire media would light up. One of our drugs went on the front page of Financial Times when we started the phase II. Because it's AI-generated drug, so people see that that is the first, and there are so much novelty there, it's never been done before.
But now when you start doing it routinely, people are like, "Whew, now what do I pay attention to?" And you pay attention to, let's say, big papers, the foundational model that somebody published and released a trillion, trillion, trillion new structures or molecules. Or now that foundational model can... you can talk to your genes, you can talk to your proteins. We can do that too. The question is that you need to be able to launch and deliver payload to orbit. And because of the hype and because of the very few investors, investment banks, analysts, reporters who actually are monitoring this niche, the overall general public and the kind of generalist community, they don't pay attention to benchmarks. They want to have flashy stories. So, benchmarks are boring. But I think that nowadays, if you look at the benchmarks for space, how many payloads did SpaceX deliver to orbit? So, criticize Elon all you want, but I think that he's outperforming many countries.
Right now, if you're actually looking at novel drug discovery, most countries have never discovered a novel drug and put it out. They have massive amounts of capital, they're investing in R&D. But even some of the countries that I'm not going to name now, very advanced, they help with the discovery of many drugs. But name one novel where there is novel target, novel molecule that reached human clinical trials. That is very difficult and also boring.
So, now what we are trying to do just for the sake of it and also to prove the point, since we're so geographically distributed, I go to new locations, for example, we set up in Abu Dhabi, we have 60 people, and now I'm trying to prove the point that with four people locally, I can try and go and discover a novel drug in a certain amount of time. So doing local benchmarking, trying to explain to people that now with AI, you actually don't need to invest massive amounts in infrastructure. You need four people. Let's see if it works. Global operations, so you can outsource a lot of validation. But the actual discovery, the actual design happens in the country.
So, if we manage to do that and show the benchmarks for a specific country that never aspired to do innovative drug discovery before because they thought it's too complex, too innovative, too difficult, if we show that it's possible, we might be able to empower nations. But in order to do that, in that region, you actually also need to be able to sell the drug. If you locally discover it, develop and sell, then it's a clear-cut way, and people will pay attention.
Now we need to also create the local benchmarks. And I think that my biggest problem in life is aging. So is yours. So is everybody's, right? Because we already have most of our needs satisfied. I don't know about you, but I'm very happy with what I have. All I want is to see more life and this beautiful future. And right now, just piggybacking on the US, a few European countries and China and Japan, it's not going to get us as far as if the entire world were come to the table and start contributing. So, AI will democratize drug discovery, and I think that we need to start setting local benchmarks as well.
Brendan Smith:
Yeah. I mean, I think you're touching on so many important points here. I mean, in particular, differentiating signal from noise, understanding not just geographic distribution of these benchmarks but the impact that actually disclosing some of this information can have, both on your own internal development and, frankly, on the space at large, too. And some of our conversations with investors, there almost seems to be this idea, I think particularly in capital-constrained environments, that using AI to move the needle on cost and time savings is a good thing regardless, though, obviously, some people are doing it a little bit more efficiently.
But from a valuation perspective, you're touching on a lot of points here. What do you think investors, in particular, really need to keep in mind when they're trying to compare some of these different benchmarks, different platforms. And I guess maybe better worded, what are they maybe underappreciating about where the field already is in this respect?
Alex Zhavoronkov:
Yeah. Good question. So, they are underappreciating how advanced AI is already. So, within this 18 months timeframe that takes me to go from zero to a pretty complex, innovative novel drug in developmental candidate, and I can do it faster if the level of novelty is shorter, the computational part takes maybe three weeks or maybe even less than that. And we usually synthesize 60 to 200 molecules per cycle now. And all of those molecules usually look good.
I previously made this analogy already just once, but imagine that you want to marry somebody, but all the candidates are coming from Miss Universe contest, or comparable for the opposite sex, or Mr. Olympia. So, all of them are great, but you need to choose one. So, that's why we synthesize so many with slightly different properties or significantly different properties, but they're very good. And nowadays, the selection from, instead of searching for a needle in a haystack, you generate a bunch of perfect needles. To choose from the bunch of perfect needle, the basic, the best of the best of the needles, that process takes longer than computation. So, computation is very, very quick. Yes, there are limitations. Yes, we cannot do many things and we don't understand biology extremely well yet. But on the chemistry side, on the antibody design side, we are approaching very high level of computational capabilities.
What I'm working on right now and what I'm kind of telling my team to motivate them, we're working on the pharmaceutical superintelligence. So I call it PSI, or PSI-1, where we are now. And for that, I usually give an example. You probably know Suno AI, where you can make your own music, very beautiful music with text, with lyrics. It will sing for you. Now you'll get many, many songs to choose from, and all of them are great. And it usually takes you about less than a minute to make a song. So, I want to approach drug discovery with the same level of efficiency on the computational side.
So I'll say, "Okay, well, we've got this cholangiocarcinoma, this specific mutation or this specific patient profile. Let's make a drug." And then within a few minutes or maybe a day, we're going to identify a target, make a drug, and after that, I see you in nine months with a developmental candidate because you still need to synthesize and test. And after that, you have to move with the speed of traffic because it's a very regulated area. And I really, really hope that the FDA, maybe somebody senior, would just look at us closer, and instead of looking at us in a fragmented way, would look at some of those programs in a very consistent, comprehensive, coherent manner end-to-end and say, "Oh, how about now we start looking at further acceleration where we can actually cut the regulatory burden and where we should increase the regulatory burden?" Because in some areas, it can actually increase, and you need to do that.
So, what investors and the general public alike don't understand is that in some areas of AI drug discovery, we are already very advanced. We're moving into the more advanced fields. But it's currently not very easily seen because of the need to validate over several years. So, basically, AI technology is advancing much more rapidly. We're now talking about from zero to product, a software product, in less than six months, something that people would be able to actually use. We can now do a prototype within a month and validate it to some extent. But if you are to make a drug, it's going to take you at least five years if you want to do phase II from zero, with your own chemistry. And even that might not be enough. We're also thinking about here that we're doing it in China, because there, it's super-efficient preclinical work, and then even early clinical can be done very, very quickly. In the developed... In other countries, it might take actually longer for many reasons. Also, because of the population is so small, if you're doing early-stage clinical trials.
Another thing that people have a misconception of is the value of the AI platform. So, the value of the AI platform gets diminished within a year. So, from one perspective, if it's validated, if you demonstrate that the model works extremely well, consistently, with many launches, with benchmarks, with preclinical benchmarks and clinical benchmarks, it is valuable. But at the same time, somebody else is working on something better already. So you need to make sure that this model that you're working on right now that you validated gets commercially viable. So you need to make sure that the cow that you spend so much effort growing and training produces some milk and preferably some other cows. And very rarely I see it happen now. People are just either too focused on the algorithm and are promoting their model or promoting the technology, or they're too focused on the drugs and on therapeutics, and they are completely disfocused on the platform. So, having this balance is very difficult.
And usually, I would say, if I were an investor and I were to look at the AI-powered drug-discovery company, I would give zero value for the platform. I would say, "Okay, platform is free. Show me that you can do it better than..." Well, not free. But, let's say, most of my software is commercially available. So you can actually buy it as a biotech. And you can buy it for... I'm not going to tell you exact numbers, but, let's say, a million bucks would take you a long way to acquire some of our capabilities. And you can basically say, "Okay, well, if I'm not better than Insilico, then maybe I should just spend a million bucks, and my platform is worth less. Or if I'm better than Insilico, then I need to demonstrate 22 developmental candidates, backed by peer-reviewed research papers. And also, I want to sell a few to demonstrate that they're of high quality. So somebody buys them. And preferably, I want to take them to mid-stage clinical trials and demonstrate safety and efficacy." So, benchmarking is extremely important on the drug-discovery side. And the ability to produce those drugs quickly, cheaply, with higher probability of success, that is the value of the platform.
Brendan Smith:
Yeah. And I think, look, again, you're touching on a lot of these lists of benchmarks at a given point in time. And I think you also mentioned this, which is, I think, really critical for people to understand a little bit better. These benchmarks are not static. If you're designing an AI platform that's working and doing what it should, to your point, over time, it should get better. It should evolve. So, what the benchmarks look like today should, in theory, be the worst they look moving forward. And you should be able to kind of track this over time and get a better sense of not just where you were on this day in 2025 but how is this actually evolving over time and can that actually speak to the capabilities of the work that you've put into and what it's ultimately going to be able to deliver out of it, right?
I've heard you kind of use this analogy. It's like, oftentimes, it's not just a snapshot, it's realistically a video, and in many ways, kind of actually a video game since we're living through it, right? And I think it's such a potent analogy for the evolution of this technology in so many ways, too. So, I guess maybe one last question for you. But just before we wrap up, I mean, given the number of different benchmarks that you can release, that you can disclose, that you can use to compare across different capabilities, is there, in any sense, a little bit of a hierarchy of which ones you think have a little bit more importance or that you watch for when you... When you look at maybe a new platform, like, "Oh, this has to be super competitive," maybe some of these other ones are, "It's nice to see, but there's one in two that I really actually need to see some very, very important points of differentiation, otherwise it's not worth my time," do you think that way? Or is it kind of an outdated pretext for looking at this?
Alex Zhavoronkov:
Absolutely. So, there are, I would say, baseline benchmarks. So, the baseline benchmark would be given a moderately novel or highly novel target. How long does it take you to developmental candidate, regardless of whether you're doing a small molecule or biologic? How many molecules do you need to synthesize? Or how much money do you need to spend on your biologics? What is the probability of success going into, let's say, IND? So once you develop the preclinical candidate, how many times did you reach the clinical trials? And how many times did you fail in IND enabling? So far, I had not had a single failure. That's the interesting part. Then you, of course, look at clinical benchmarks, but most companies don't reach that stage.
Then, second very important area that we are trying to now pioneer in terms of benchmarks is the level of complexity of biology and also the complexity of disease given very limited data that's available to you. So, previously, we were going mostly oncology because that's where you can get very quick proof of concept, lots of data is available. We were also going after complex chronic diseases where we can piggyback on longitudinal data. Now we are actually increasing the level of complexity, and I'm not sure how long it's going to take me right now, but we're going after pain. Pain is very difficult, especially if you're going after non-opioid stuff without a clear understanding of how the receptors work, how the pathways work.
Now I have some molecules that... And I've talked about that before once, so I'm not disclosing any undisclosed stuff here. But one of my molecules worked better than morphine. Currently, I'm optimizing that molecule. I'm not sure if I can deliver it, because currently, it works better than morphine if you administer it epidurally or intranasally. It's not brain penetrant. So now I'm trying to optimize the molecule so it can be brain penetrant and, at the same time, safe, because for chronic administration. But we've demonstrated internally... Currently, it's unpublished data, but I talked about it. In some experiments, we actually see the animals having less pain than when they're on morphine. Most likely, though, this is a non-addictive drug because it's not affecting any of the pathways that were previously implicated in addiction.
We're also going after biological processes like muscle wasting or muscle gain. Imagine exercise in a pill. I would love to do that, right? I'm not a big fan of the gym, but I need to do that because that's one of the very few weapons we have against aging right now. And that's very unfortunate after all of those decades that the humanity tried to go after this biological process. Thankfully, and thanks to GLP-1s, there is a path to approval and also there is huge demand because GLP-1s induce muscle wasting. So you actually lose muscle on GLP-1s. And GLP-1s are not going to significantly help you if you are already pretty fit. But if you want to get a little bit more gain, you might want to go with muscle-wasting drugs that will prevent sarcopenia, especially for the elderly, and improve muscle everywhere and improve your cardiovascular function, probably, as well.
So, we are going after higher level of complexity, new targets, new molecules. We now need to develop absolutely new and extreme safety standards. So, when you're going after chronic diseases and aging, you need to make sure that the molecule is not only beneficial but also it should have zero side effects. All the side effects that it may have should be beneficial to the patient. It cannot have any damage. So, now I'm also working with organoids, with new AI modeling to achieve new levels of safety. And basically, we need to simulate the entire world in order to be able to do that, unfortunately. But with the current resources we've got, we're trying to simulate to the extent where we can and start setting benchmarks for chronic diseases, for chronic conditions that are significantly beneficial.
So, now that I've established my Falcon 9, I'm building my Starship. And the Starship should take us to the next level of biology and novelty, because it's very difficult to outperform nine months to development candidate, right? So, we've demonstrated it's possible. If I get it to five months or four months or three months, it's actually not going to make a lot of difference because it takes me usually much longer to sell this drug to a pharmaceutical company than it takes me to discover and develop to a certain level. So, nowadays, we need to go for just higher level of novelty and try to decipher the complexity of human biology a little bit better because that's your key to aging.
Brendan Smith:
No, I think this is fantastic. Again, we've covered so much great ground today. And it's a conversation, I'm sure, we'll continue to have for the foreseeable future. But before I let you go, there is one thing I like to just ask all of our guests, and I think it's especially poignant today, given how much we've talked about. If everything we've discussed today really goes over someone's head, but they've made it with us this far, what is the one really critical point that you would want everyone listening in to remember and take away from our conversation?
Alex Zhavoronkov:
Well, if you've made it this far, so here is a punchline, and don't freak out. You need to realize very... So, as humans, we try to self-deceive ourselves to believe that something else is important. We find ways to distract ourselves from this thought, "We are all going to die." And we don't like to think about it. It's not a pleasant thought. And before you are going to die, if you're not going to have some extreme event and accelerated event, you are going to lose everything and die. And this loss of everything is never pleasant. You may think that you're healthy, but you might be operating at, like, 50% of your capacity from your peak. And you are going to be losing. And at the end, there's going to be some disease or some extreme event that's going to kill you. And it's never pleasant unless you are overdosing or something. Don't recommend doing that either.
So, aging is your greatest enemy. And right now, it's not China. It is not the Middle East. It is not somebody who you just don't like in your own country. It is aging. And we need to unite. We need to fight together. We need to pull resources. We need to collaborate. We need to globalize. We need to enable countries around the world to join this fight. We need to tell this story to our loved ones and the people we hate in order to maximize quality-adjusted life years for everybody on the planet. I think that if you are looking to do ultimate good, you need to maximize the number of quality-adjusted life years on the planet.
If anybody develops a drug that makes everybody on the planet live one year longer, we're talking about 8.2 billion quality-adjusted life years. That is more than 170 million lifespans. So you're going to be super-altruist if you do that. But it's never a one-man journey. So it's a team. And unfortunately, right now, from what I see, the world is becoming more fractured. It's becoming more emotional. It's driven by sensationalist news. Even though what we need to do is just put our head down and work, work, work, collaborate, because we might be the first generation to live an extremely long and healthy and beautiful lifespan, or might be the last one.
Brendan Smith:
A clear and unequivocal call to action, if I ever saw one. So, thank you very much for hopping on and talking us through really this pretty critical point in the evolution of AI. I'm sure, again, we'll have plenty more to discuss over the weeks and months ahead as it pertains to all these measures. So, really, thanks again for joining us, Alex.
Alex Zhavoronkov:
Well, thanks for having me. TD Cowen is amazing. And we would love to work together and read your beautiful analysis. It's something that I love to read.
Brendan Smith:
Always love to hear it. Thanks very much.
Alex Zhavoronkov:
Thank you.
Audio:
Thanks for joining us. Stay tuned for the next episode of TD Cowen Insights.