Over the last decade or 2 there’s been a bit of a renaissance on the hunt for the fountain of youth. A large part of this drive can be laid at the feet of a new paradigm on aging. We no longer look at aging as a strictly time-dependent process. As the science in to aging has progressed, we’re getting clear indications that time is a variable that leads to aging, but the relationship between time and aging is, itself, variable and nonlinear.
What does this mean? To understand this concept, you have to understand that there are 2 types of aging from a scientific perspective: chronological aging and biological aging. Your chronological age is the age most people are accustomed to. It refers to how much time has passed since the day you were born.
Biological age, on the other hand, is something completely different. In simple terms, biological age refers to the way that your body and the cells that comprise it function. The scientific explanation is a bit more complex and basically comes down to your risk of dying or contracting an aging related disease such as Alzheimer’s disease, heart disease, cancer, or Type 2 diabetes. As we age, our risk for these diseases increases, and that coupled with decreased fertility is the scientific definition of aging.
So what does this all mean?
Simply put, it means that time to me isn’t the same as it is to you: One of us more than likely ages faster than the other. For example, I’ve seen several classmates who were either younger than me or the same age pass away from aging-related diseases over the last few years. If you don’t get in to an accident, it’s likely that one of these very diseases is ultimately going to kill you.
If aging were simply the passage of time, we’d all pass away from these diseases in order based on our birth. But some of us have heart attacks at 35, while others have them at 85 or not at all. We all know this, it’s why many of us exercise, eat a sensible diet, don’t smoke, and try to get to bed early.
These factors simply slow down the rate at which we age biologically. There’s also a genetic component, some of our clocks tick faster than others, which is why it’s not a valid approach to compare yourself to someone else. Instead, it’s more sensible to look at how we age in relation to time. In other words, look at my age now in relation to what it was 5 years ago. This gives me my rate of aging and can be used to validate or eliminate lifestyle factors that are slowing or accelerating this rate, respectively.
Biomarkers of aging
One of the most difficult aspects of identifying the factors that affect our rate of aging is having valid biomarkers. While mouse studies are great, they typically look at how factors such as diet, lifestyle, and fasting affect aging in genetically identical mice. In case you didn’t know, aside from identical twins, no 2 humans are genetically identical.
Another problem is that we live a lot longer than mice, so figuring out if something makes people live longer takes decades. And going back to the genetic diversity I just mentioned, factors such as a ketogenic diet that can prolong life in some can actually shorten life in others with specific genetic polymorphisms. Do you really wanna wait 20 years to find out that the ketogenic diet you’ve been struggling to maintain has accelerated your rate of aging? Probably not.
This is why the research in to biomarkers of aging is so exciting. Now we can use data to determine if we really need to be spending $300 a month on supplements, restricting our food intake, or what the proper dose of exercise is that you need to perform to decelerate the aging process.
I’ve talked about a few of these before: DNA methylation, telomeres, and Young.AI. Today I want to focus on Young.AI, specifically how artificial intelligence can be used to dramatically improve access to biomarkers of aging. If you don’t know what the hell I’m talking about, go to www.young.ai
Pros of Young.AI for age prediction
Probably the biggest factor going for Young.AI is access. It’s free, it’s super easy to use, and it’s something you can use repetitively to track changes. All you need to do is either input data you already have or take a picture. You can’t get much easier than that.
There are 2 big fat problems with DNA methylation and telomere testing that restrict access. The first is that both are pretty pricey, $300 for DNA methylation and $100 for telomere testing. The second is that it takes 3-4 weeks to get the results back from telomere testing and 6-8 weeks for DNA methylation testing.
This last one may seem like a trivial issue, but people have terribly short attention spans. Most people, even when they’re most motivated to improve their health(aka New Years), fall off within 8 weeks. At that time they could care less what their DNA methylation age was in January when they tested it. This coupled with the price restrict the amount of data companies offering these services can collect.
This makes improving these tests more difficult as well. While DNA methylation is pretty well dialed in at this point from a validity perspective, telomere testing isn’t. And any sort of algorithm is only as good as the data set that drives it.
One final thing that sets Young.AI apart from these other two types of testing is that you can use multiple data types for it. Currently, the 2 most useful sets of data that can be used are blood markers that are routinely collected by your doctor and facial pictures. In the future, you’ll also be able to use DNA methylation testing, telomere testing, and a bunch of other different types of tests of value. Futhermore, you’ll also be able to input your genetic data to help hash out individual differences in the response to different therapies.
This brings us to another con of telomere testing and DNA methylation testing: they measure 2 different variables that are likely affecting age through different pathways. Telomere testing is more or less assessing how many times your white blood cells have replicated on average and DNA methylation is assessing changes to your epigenome. They may track with one another, but probably not in a direct manner.
Using AI to drive things like biomarkers of aging is going to rapidly accelerate the science. This, above all else, is the most powerful benefit to this approach.
Cons of Young.AI for age prediction
Before I get in to the cons of Young.AI, I want to point out that the access variable more or less crushes any of the concerns. The architecture is there in the approach to make this far and away the best tool. Sure, DNA methylation correlates more strongly with chronological age currently and is therefore a better indicator of biological age. But if a tree falls in the woods and no one hears it…You get the point.
But there are some things in Young.AI that need to be improved. First, there need to be some standards to remove as many of the confounding variables from testing as possible. For example:
- Day of the week
- Time of day
- Hours, days, months, events leading up to the test you are going to input
- Inter-lab differences in testing procedures
- Angles, lighting, expression, and other variable in the pictures being taken
- Genetic differences that make us look older but may not have a direct impact on our health
What do I mean? Things like drinking coffee before the test, fasting for variable amounts of time, parties on the weekend, and circadian rhythms can all have pretty significant effects on the blood-based markers. Simply going to Labcorp instead of the Quest Diagnostics you typically go to can affect the reliability of the test, and people need to know that. Basically, variability in testing procedures is bad and this needs to be addressed to allow the neural networks of AI to do their job.
On the picture aspect of Young.AI, which will probably be the one most people use, variations in lighting, your hair cut, your facial hair, hair coloring, facial expression, skin moisture, distance from the camera and where you take the picture can all have a potential impact on your score.
Even just changing one of these variables can significantly change your predicted age, let alone if 3 or 4 of them vary. As an illustration, I decided to take 2 pictures today, both with resting bitch face. The first was after about a week of facial hair growth and the second was in the same room from the same distance about 10 minutes after shaving.
Note: That’s not drool in the 2nd picture, it’s my shaving mess 🙂
As you can see from the picture, I have some grays in my poorly growing facial hair. I also suck at taking consistent selfies, but there are greater tragedies. This minor difference along with some minor variations in the angle due to my sucky selfie technique caused Young.AI to predict my age differently. On the left, my age was predicted as 43.
In the picture to the right, sans facial hair, my age was predicted to be 39 years old. Not a huge difference in either direction over my actual age of 41, but I’m pretty sure I didn’t reduce my risk of dying of cancer by shaving. This is an issue with AI that needs to be addressed if we’re going to use it as a marker of biological age. And if we’re going to have robot overlords, don’t we want them to be smart robot overlords?
These are not huge problems, but certainly ones I feel need to be addressed. Fortunately, there may be a pretty easy fix to this, and this segues perfectly in to the direction I feel Young.AI should go.
Future directions for Young.AI
Before I progress down this road, I want to point out that I’m not a financial dude, AI dude, and I’m not at all affiliated with InSilico Medicine, the company that created Young.AI. So, most of the assumptions I’m making are based off of press I’ve read. I could be wrong about where I think they’re going but I’m probably not wrong about where they should go.
I suspect, based on what I’ve read, that Young.AI is going to be primarily rolled out to the “healthcare” and beauty systems. I put healthcare in quotations marks because I’m not quite sure how you can call a semi- to quarter-annual 15 minute relationship with someone you hardly know a system to optimize health, but I’m a cynic. Regardless, this is a mistake.
I see a much greater application of this technology, and a far more effective way to acquire data that will make it better, in the health and fitness industry. A healthy person sees their doctor 1-2 times a year while even only moderately healthy people go to the gym or fitness studio more than double that every week. And they almost always go on the same days of the week at the same times for most of the year.
It makes far more sense to take pictures of the same person under nearly identical conditions throughout the year to iron out the minor wrinkles caused by changes in facial hair, hair color, tan, etc. Of course, you can effectively couple the data from blood marker tests with the picture tests for a more complete picture.
I could see this as more of a system to not only make Young.AI better, but also make the gyms, studios, and fitness professionals that use it better. Fitness professionals can be trained to take pictures in a consistent manner and in a set room to remove variability. They can also explain that minor differences in facial hair can confound the results but are ultimately needed to perfect their personal data. Acute lifestyle factors that alter the data can be collected by the same professional and this can be entered in to Young.AI so it can learn. (Go on a bender this weekend? Poor sleep last night?)
Programs can be tailored around the data and the rejuvenation effects of specific classes or programs such as yoga can be used to market the programs once facial pictures are validated for age prediction. Even better, programs can be individualized based off the Young.AI data to circumvent that genetic individuality problem. As a witness in many gyms over the last 20 years…
- Most everyone is doing something random
- Some people exercise too much
- Some people exercise too little
- Some people center their fitness program around things that don’t serve them very well
- And most have an absolutely terrible approach to fitness from an aging perspective (Super-restrictive diet and obsessive bootcamps much?)
Another great application is corporate wellness programs. Any company willing to shell out the cash to promote wellness in their company most certainly wants to validate that the wellness company they’re using is living up to their end of the deal. At the end of the day, in business, all that matters is the bottom line. Why shell out hundreds of thousands of dollars to a wellness company that doesn’t reduce your employees’ risk of the chronic diseases of aging that cost a fortune to treat in medical and productivity costs? Seems like a great use of AI right there.
It’s a bit surprising that the fitness field isn’t an avenue that InSilico is pursuing. You have a huge range of different types of people from fit to fat, trying different diets from Keto to time-restricted feeding to low fat, exercising in many different ways, and that are all interested enough in their health to spend at least $50/month on their health for a gym membership to accomplish something they could probably do at home. Certainly a more ideal population to train a machine to learn how to predict biological age than blood work that’s probably disproportionately coming from sick people who go to see their doctor.
Another one of their novel uses of AI in longevity is in the development of supplements that target pathways similar to the pathways researchers in the pharmaceutical field are pursuing for longevity. Again, fitness facilities are ideal in this regard because members of these clubs likely account for a big percentage of the billion dollar supplement industry.
Add to this the potential validation a tool like Young.AI can provide, particularly the bloodmarker measure, and you have a homerun. Ultimately, developing programs in fitness clubs and corporate wellness programs can drive the supplement end of InSilico’s AI program through validation with the Young.AI tool, while driving up the amount of data they collect for machine learning to progress. That data is extremely valuable in improving the biomarker aspect of InSilico’s AI program.
But more than anything else, from a health and wellness perspective, using something like Young.AI in this way can expose people to this new concept of aging where time isn’t the only number you need to worry about. Maybe, just maybe, this will get people thinking that therapies that can slow or reverse the aging process are both possible and worthwhile to do.
Artificial intelligence has the potential to drive progress in aging research and age regeneration in humans light years ahead of where they are now. Of course, the use of AI for this purpose is in its infancy, so the quicker we can get more and better data the quicker things will progress.
While it makes sense to form collaborations focused on big pharma, beauty, cosmetics, and supplement companies, I think the more immediate direction for the use of AI in aging and age rejuvenation is in the health and wellness industry. Let’s face it, AI is only as good as the data it’s exposed to, and the more data the better.
Furthermore, while I’m quite bullish on the notion that we’ll have viable age regeneration pharmaceuticals, therapies, and nutritional products in a decade or 2, we’re not there yet. The best we have for age rejuvenation currently is exercise, fasting, calorie restriction, stress management, and sleep. And until we solve that whole cancer thing, I’ll probably shy away from activating single aging pathways such as mTOR, AMPK, sirtuin/NAD, and telomerase with gene therapies and pharmaceuticals. For now I’ll just work on hitting them all by addressing my lifestyle.