OR WAIT null SECS
Jennifer Grebow is editor-in-chief of Nutritional Outlook.
Life Extension’s Michael A. Smith, MD, gives Nutritional Outlook a glimpse of what the future—and present—holds in leveraging AI technology to create healthy-aging products.
Dietary supplements are born of research: research on health-promoting ingredients, on how ingredients should be combined, and on how a resulting supplement interacts with the body and ultimately benefits human health. Nutrition scientists are hard at work doing the research, but they face real-world limitations such as the time, resources, and cost that science requires. This is why nutritional research, like any research, happens gradually. But what if some of the research could be done quicker, more easily, and more cost-efficiently? Enter artificial intelligence.
AI technology is not new, but it is a newer prospect in the nutrition and supplements industry. Life Extension is one company that’s now using AI to help make discoveries that will yield the next generation of the company’s supplement products. The company’s goal is to develop supplements that support longevity and healthy aging. As it says, “For 40 years, Life Extension has pursued innovative advances in health, conducting rigorous clinical trials and setting some of the most demanding standards in the industry to offer a full range of quality vitamins and nutritional supplements and blood-testing services.”
Last year, Life Extension announced it had partnered with biotech firm Insilico Medicine Inc. (Rockville, MD), whose AI deep learning capabilities are helping Life Extension create cellular-health products, including those in its GeroProtect line.
Life Extension’s Director of Education and Spokesperson Michael A. Smith, MD, recently sat down with Nutritional Outlook to discuss why the company believes AI will not only help supplement companies create products faster, but create products that are better.
Nutritional Outlook: To start with, how does artificial intelligence work?
Michael A. Smith, MD: AI basically works by being able to analyze different data points at a level the human brain really couldn’t. It’s able to take a lot of information that you input and rearrange that information, find patterns, and do an analysis of that information that would honestly take a team of scientists years and years to do. The computer can do this very quickly, and it can start to make relationships between those data points…[T]he program learns from what we input, and it starts to understand that information at a deep level beyond even what you or I understand. We can start asking it questions, and it can give us quick answers that would take us years in a lab to get.
Nutritional Outlook: Can you explain the difference between machine learning and AI deep learning? From what I understand, machine learning is when you feed information into a computer, and the computer will generate some information, possibly with limits, based on that information. By contrast, AI deep learning somehow allows the computer to analyze data at a higher level and generate more meaningful information. Can you explain the difference to our readers and why the information AI deep learning produces is more complete or more meaningful?
Smith: This is an analogy that helps people. It’s based on online shopping. When you went to Amazon.com or to a lot of online retail stores early on, if you were to buy something, the computer would save that information. The computer would then say, “Well, this person bought product A, and a lot of other people who buy product A also buy product B and C.” And the next time you come on, or even when you’re at the cart level, it can offer up additional products to you. That’s machine learning. It’s very simple. “You’re buying this? Well, guess what? We now know through all of these other inputs that when you buy that, other people have also bought this and that.” That’s machine learning.
Artificial intelligence, the deep learning, however, is much more complicated than that. It’s almost like it knows things about you that you don’t even know about you. Again, this is an analogy, and analogies aren’t perfect. Analogies break down at some level. All of them do. So machine learning is: you bought that, and we know that you might also like these two things. That’s machine learning. Well, deep learning is, “Well, you not only bought that, but you’ve also been to this site and looked at this article. You’ve been over here and read something about the article again over here. We notice you also read a similar topic at another website over here. So based on all that, I’m going to make a different recommendation that has nothing to do with what other people have bought. It’s more based on what you’ve been doing and your behavior.”
That’s deep analysis. And that’s the difference.
So, machine learning at some level can be very powerful, but it’s not a deep look into who you are and what you’re doing. True artificial intelligence deep analysis actually looks at your behavior and what you’re doing. And, yes, I know this sounds scary, but the big online retail stores can do all that, everything I just said.
Taking that into what we [at Life Extension] are doing, I could use a machine learning application that says, “Okay, Dr. Mike, you want to make a product with grape seed extract. Well, a lot of people who do that also combine grape seed extract with pomegranate, and they often sell it as a blood pressure product.” That would be machine learning.
Deep learning would be something along the lines of, “Okay, you want to do grape seed? Well, here’s what we know. Grape seed can affect all these different pathways. It can actually work with pomegranate in these different ways, and we actually think that you could produce a product that’s more about nitric oxide production—and that’s how you should market it.” It’s another layer of analysis that machine learning cannot provide.
Nutritional Outlook: What lets AI analyze at that higher level?
Smith: Believe it or not, it’s based on your brain. Deep analysis is based on neuro networks. The best way for me to explain this is: think of the human brain and the human nervous system. You basically have a part that senses the environment. You can consider that the input. And then there's how the body responds to that, an output. And so they create these algorithms in the same way. There are parts of the algorithm that sense the input, what we’re telling it, what it’s bringing in from the environment. And then it can talk to different parts—different components, if you will—of this algorithm, almost like a nervous system. You know how nerves can go from your foot, up to your plexus, in your spine, and up into the brain? It’s the same thing.
These AI machines sense the environment. They can go, “Based on what we’re sensing, we should talk to this part of the algorithm, or maybe we should talk to that part of the algorithm.” And it starts to make these deep connections, just like your brain does, and then ultimately you output the best response…The AI technology gurus literally started with studying the nervous system, and that’s ultimately how they’ve layered this deep analysis that AI computers do.
It really is amazing, but at the end of the day, it’s simply a system that is able to sense an “environment” and decide how it wants to analyze that and produce a certain output that has the best optimal response, just like the human body.
Nutritional Outlook: Are there limits to the data we can feed an AI system today simply because there are limits to the knowledge we ourselves have today—such as a limited number of nutrients that have been discovered, or limited plant species that have been discovered, or a limited number known physiological pathways identified in the body? Do those knowledge gaps limit the output of an AI system?
Smith: That’s a great question. The limitation is not in what you just said. The limitation is more in the biases of the people designing the algorithms, and I’ll explain that. This is where the amazing AI technology comes in.
As a human being, I could study biochemistry all day long, and I could know all these different pathways. But you’re right. We don’t know all the pathways. There’s a limitation to what I know and what I can analyze. But what the AI technology can do is if it’s learned that a certain pathway does a certain thing under a certain condition, it can start changing the condition. It can start predicting how that pathway might interact and maybe what additional pathways you would need to get a certain result.
So the computer can actually say, “Here’s the current pathway you guys taught me, but in order for you to get an answer about cellular aging, for instance, I’m predicting that there are three or four other related pathways that you haven’t found yet.” And then scientists take that, and they start looking for those different pathways.
As a matter of fact, that’s how one pathway, mTOR, was in part discovered. mTOR is this antiaging pathway [whose discovery] came about through a lot of AI technology [that said] there must be another regulatory pathway, for a variety of reasons. And it was the computers that taught us that.
Nutritional Outlook: Interesting. So AI was involved, in part, in the discovery of the mTOR pathway?
Smith: Not directly, but it played a role. AI technology has played a role in helping to identify many different pathways, but it still took human study at that point—cellular research, lab research—to prove all that out. AI technology can’t go that far, but it can say, “Based on what you’ve been teaching me, if you want to talk about cellular senescence and how to control it, we’re missing something.” And the computer can say, “I think we’re missing a pathway that might be more connected this way. So look for something like that.” And so that can lead a lot of research, not just at the clinical level but the bench level, [done by] actual universities and scientists, microbiologists, who are studying this.
Nutritional Outlook: So the results AI generates are sometimes more predictive, and then they must be validated further. After that, what are the next steps? In vitro and animal studies, followed by clinical trials? Again, I’m thinking in the context of using AI data to create health-promoting supplements.
Smith: Let me explain how we do that…We teach the computer about cell aging. We can even teach it about what certain classes of nutrients, like polyphenols or flavonoids, do. Based on that information, the computer can say, “Based on the pathways and the nutrients you’ve been teaching me, you might want to check out these four nutrients. These four nutrients have potential for controlling cellular senescence.”
And we didn’t know that before. We never understood how myricetin, for instance—which is one of those polyphenols—actually could impact several different aging pathways in a positive way for the cell. We didn’t have that. There was a gap in our human knowledge. It was the computer that said, “Based on what you’ve been teaching me, and based on this deep analysis and these relationships I’m making, I think myricetin is one that could actually be really important for cellular senescence.” And we said, “Okay, wow. Let’s check it out.”
So then we take that information. We go back to the lab. Then it’s at the lab level where we confirm, “Sure enough…myricetin actually does help to get rid of senescent cells.” We never knew that before. It was the computer that suggested it or predicted it, and then we had to go back to the lab to validate it.
We always validate what the computer is predicting. As an example, let’s say the computer says, “Well, you know what? Based on everything you’ve taught me about cellular aging, Life Extension, I’m predicting…that you’re missing a couple of pathways that should do this.” So then we go back to the lab, and we try to figure that out and confirm that. And sure enough, in many cases, you’re like, “Oh, wow!” You end up finding these pathways in cells, and we only knew to look there because the computer predicted it.
Nutritional Outlook: So AI cuts out some of that explorative time that it would have taken researchers to arrive at some of those same conclusions. So it cuts out the time, the cost, the resources?
Smith: Yes. There’s a common poster most biochemists have [depicting] every known (to date) metabolic pathway inside a cell…Well, all of that information, all of that knowledge, has been put together over decades of research, with hundreds of thousands of biochemists each doing their little part in filling in this huge poster of all these metabolic pathways. Well, if we had AI technology way back in the 1920s and 1930s, we would have had that poster in the 1950s, not the 2000s. That’s the potential. AI could advance our knowledge of health, medicine, what have you, in such a rapid way.
It’s why we’re really interested in it as a company. I think this is an important point: Life Extension is a supplement company. We have product categories ranging from bone to joint health. But at our true heart, we’re a longevity company. And one of the big hurdles/obstacles that we run into is the fact that it’s almost impossible to study human longevity, even cellular longevity, because it takes so much time. It takes so much basic human knowledge. It takes so much input to be able to get an output—meaning, yes, you could live longer if you did something, [but] it would take so much money [to discover] that a lot of what we do in longevity research is based on hypothesis.
What these AI computers or algorithms are allowing us to do—helping us predict new pathways, predicting how those pathways might work clinically in people—is speeding up the longevity research time frame. Instead of decades and decades, it’s bringing it down to months and months. And that’s why we’re so excited, because we’re learning about things we can do right now with nutrients to improve cellular health and cellular aging—and ultimately your and my aging. And that’s the real power: speeding up what we’re learning about human longevity.
Nutritional Outlook: AI is especially useful in the longevity work that Life Extension is doing. Would it be as useful in other health-supplement markets like joint health, brain health, etc.?
Smith: It has applications for all of that. I mean, even just take a new nutrient that, based on some current research, some clinicians think might be good for arthritis. Now, it’s much cheaper, and it takes a lot less time and a lot less people, to do a joint-health study. You could set all that up—but it’s still costly. It’s still time-consuming. Where AI technology could help—and, by the way, the human medical centers, pharmaceutical companies, are starting to look at AI technology for this—is that you can almost simulate a clinical research study within the AI environment, within the AI algorithm, and it could predict whether that nutrient is really going to be good for arthritis, before you even spend the money on a clinical trial.
So it has application across all product categories in finding potential new nutrients and helping you decide if that nutrient’s even worth spending time and money on researching. That can be quite powerful for the health industry across the board.
Nutritional Outlook: What role is AI now playing in Life Extension’s work and the supplements it’s introduced, like GeroProtect. How is Life Extension leveraging AI today?
Smith: Right now for us it’s about identifying and validating geroprotectors. That’s just a fancy word for protecting age—basically, helping us age better. And right now, a big focus for us is on a handful of aging pathways—a handful of senescent pathways—ultimately using AI to bring to market faster products that could affect overall human longevity by improving cellular health and aging. That’s where we’re at with it right now.
But we’re very interested in the application in other product categories, helping us predict laboratory research and clinical research before we even do it, to help direct our research monies and time. So, we’re keeping an eye on AI technology, watching how it’s developed and how it could help us in the long-run as almost a partner in research and product development.
But right now, for us the big focus is the geroprotectors—nutrients that impact cellular aging, which then ultimately impact my aging.
Nutritional Outlook: How is artificial intelligence currently being used throughout the nutrition industry? Is the rest of the industry using AI to any great extent? Is it still very new in this space (versus in other industries)?
Smith: I don’t think too many other companies, to my knowledge, are using AI technology at the product-development level like we are. There are a lot of research centers and universities that are increasing their nutritional research departments, and that’s probably where AI technology is being utilized the most—in helping nutrition scientists, biochemists, understand aging pathways, antiaging pathways, metabolic processes. Using AI technology to fill in that poster even more (the poster that I mentioned before). That’s where most of the medical industry in general is—and when I say medical, I’m including supplement, nutrition, what have you. It’s really at the lab level right now. But I think it’s a pretty good guess that over the next few years you will see more companies using AI technology in helping them develop products. I think it’s just a matter of time.
Nutritional Outlook: Is AI an affordable technology for companies in the nutrition and supplements market?
Smith: AI isn’t that new, but in terms of developing products with it, it is pretty new. And like any new technology, it’s really expensive at the beginning. But as things get better, as algorithms get better and there’s more competition to use certain AI algorithms, price will come down. It will become more cost-effective. I mean, look how much money your first computer cost versus today. You can get a great laptop with more memory than you could ever want for $800-$900, whereas 20 years ago the basic home computers didn’t do anything and they were very expensive. I would expect that as competition increases and companies become more interested in it, that cost will come down. And that is when you’ll see an explosion of AI technology.
Nutritional Outlook: We’ve talked a lot about how AI gives us answers faster, answers that we might not have had for years to come. It gives us those answers today. But are nutrition and supplement products created using AI actually better in terms of being more effective? Why might they not have been as good had they not been created using AI?
Smith: There are two ways to answer that. At some level you could argue—and the computer scientist guys who do this, they would argue—that, yes, the AI technology does produce more efficacious products because it’s telling you about nutrients, it’s telling you about doses, it’s telling you about combinations that you would maybe never think of yourself. So, at some level, it could produce something that’s more efficacious.
But I don’t know if that’s really the endgame for us, because we’re going to validate everything anyway. It’s really more about increasing our own knowledge of longevity and cellular longevity, introducing us to nutrients and pathways today that we wouldn’t know of until decades later. It’s really about advancing our knowledge and then being able to use that knowledge to develop formulas that are efficacious.
So, on some level, you can make the argument of more effectiveness from AI formulas. But on another level, it’s really just about advancing our knowledge at a pace that we could never achieve by ourselves.
Bonus: Hear Michael A. Smith, MD, and other experts discuss opportunities in the healthy-aging supplements market in this recent webcast, “Cellular Health: The Next Big Market Opportunity”