My AI Doppelganger Tells Me What to Eat

This is The Optimizer, a weekly newsletter from Victoria Song, The Verge's senior critic, that explains and discusses the latest tools and potions people swear will change your life. Select Optimizer here. For about the past 100 days, my mornings have followed a new rhythm. When I wake up, the first […]

For the past 100 days or so, my mornings have taken on a new rhythm. When I wake up, the first thing I do is feed my digital twin. It consumes my metabolic data, and its first "meal" of the day is my body composition data, sent directly from a smart scale. After entering the program, I go to the kitchen to fill an Owala with lemon juice to take my medicine with. This is also included in the program. On Mondays, I follow it up with several calming breaths, take my blood pressure, and suggest it to my digital twin. Then it's time to consider breakfast. I consult my digital twin.

My digital twin isn't so much a robotic simulation as a color-coded food database with artificial intelligence that predicts how my glucose levels will react to whatever I eat. Too much red is not good, too much orange, green is ideal. This is based on the data I feed it, as well as data collected from a continuous glucose monitor.

A bowl of Cheerios protein and milk? My digital twin tells me it's orange if eaten alone, but green if I take a daily walk afterwards. So this is what I choose to eat and do. When that's all done, I enter it into the Twin Health app. Within hours, the app provides insight based on how my glucose is reacting and my "twin" becomes more accurate. Now, my twin is full.

Rinse and repeat for approximately another 99 days.

Digital twins are a relatively new concept in physical health, both in medicine and sports. It's a virtual model of a person built from various data sources—labs, wearable data, images, genetic profiles, etc.—that are constantly updated over time. Once built, doctors can use artificial intelligence to see how a patient might react to certain drugs or treatments and "test" them on their digital twin first. In theory, this could also help healthcare providers assess the risk of injury, predict disease progression, and implement preventive care while minimizing adverse effects for the actual patient. The idea is that this could one day help realize the dream of personal medicine. Metabolic coaching is perhaps the lowest-risk, highest-reward version of this technology that can now be tested in the real world.

[Image: The white scale is my twin's health scale. When I'm not making a digital twin, I sometimes use it to help control Pablo's weight.

My metabolic twin was "born" in June with a giant box of Twin Health delivered to my doorstep. Twin Health is a digital twin app that helps people "heal" their metabolism (eg lose weight, manage diabetes, sleep better, etc.). Not yet available for mass download. Digital twins are still experimental, and currently, the service is only available through corporate or private health plans. It is usually aimed at people with chronic metabolic problems. Twin Health provides everything you need to build a digital Twin Health needs 90-day Dexcom G7 CGM devices, a smart scale, a smart blood pressure cuff, three multivitamin and magnesium glycinate bottles, and my Apple Watch to build mine. This was followed by a trip to Quest Diagnostics, where the phlebotomist stuck a needle in my arm and sent vials of my blood for further analysis.

A few days later, I went on Zoom with Amanda's twin health coach—a human—and a doctor to review my lab results. Multiple numbers and formulas are shared on one page. As it turns out, my liver? Inflammatory, but understandable given that I have moderate non-alcoholic fatty liver disease. Insulin resistance - definitely present despite ideal HbA1c testing and fasting glucose. My pancreas seems to be working overtime, but my insulin producing beta cell count looks good. at the moment My triglycerides were also high, although the rest of my body fat seemed normal.

I knew I inherited blood metabolism. A year of CGM testing as a non-diabetic made this abundantly clear. But after the doctor signed off and Amanda asked me to define my goals, I began to discuss my vague desire to reduce the number of medications needed to keep my metabolism functioning well into my middle age. Can a digital twin help me do this?

maybe Now, Amanda recorded my goal, made a follow-up call in two weeks, and encouraged me to try to log as many meals as possible in the first 30-day period. When I reach the 90 day mark, I am scheduled for another blood draw.

In many ways, I'm relieved that my digital twin doesn't have his own avatar. At least, not that I can see. I'm happy for the abstract colored stickers. It's refreshing to see what color a cup of pineapple or a turkey hot dog with ketchup can be. (Orange and green, respectively.) Sometimes I'm reminded of Kazuo Ishiguro's dystopian novel Never Let Go, where clones are raised as organ donors. The scenario is not a one-to-one comparison. An AI digital twin cannot actually eat, die or feel pain. I'm probably anthropomorphizing my discomfort with the idea of ​​"testing" food on a virtual version of my metabolism, as if it were a food taster in a medieval court.

However, as the months passed, I couldn't help but notice my digital twin's hunger. And how much it took out of me, it was almost like I emptied out every time I "fed" my twin. Building this thing requires a huge amount of personal data - even for me, someone who is used to tracking and reviewing large amounts of health data. In my bi-monthly check-ins with Amanda, she noted that fatigue tracking was frequent feedback from Twin Health users.

My behaviors changed in both unexpected and expected ways. Whenever I have to prepare food, I end up avoiding junk food. I don't mind being blamed by apps. However, it did not lead to significant weight loss. And despite my stated goal of cutting back on meds, it didn't work out the way I'd hoped.

Nor did I learn anything amazing from my digital twin's metabolic predictions. Although it was valid to see that for me, a bowl of zoddobo jejige and white rice is a completely green or orange meal, depending on whether or not I'm active after eating. (Follow low carb diets. You can pry rice from my cold, dead hands.)

Twin Health's bi-monthly check-ins got me back on track. I've always been disciplined in my fitness routines, but being held more accountable than having to shout my progress out loud to another human being—one who isn't family—triggered my inner overachiever. After months of feeling too weak from medication side effects, I recently felt strong enough to run a 5K non-stop. I'm also back into strength training with some casual Amanda.

[Image: This is an orange treat because protein bars are still bars.

But most importantly, after 100 days, my follow-up labs came back with significant improvements. Amanda says my metabolic health score has increased by 9 points since joining.A year after my doctor's eyes fell on my liver enzymes, all but one came back normal. These numbers were slowly improving before I tried this whole digital twin business, but they fell even faster while trying this whole thing. My triglycerides dropped 70 mg/dL. There was a significant decrease in my hs-CRP, which is a biomarker for overall inflammation. I am still insulin resistant, but less so.

I really can't argue with results like that, can I?

I can, indeed. The digital twin concept is based on trust. As I said, metabolic use cases are high reward and low risk. If you give a twin lots of data on weight loss, diabetes management, or more vague goals like "better energy" or "better sleep," failure isn't immediately life-threatening. There are other drugs and ways to try.

Personalized animal medicine is different. Researchers are investigating digital twins for things beyond metabolism. For example, one area that is currently being explored is cancer care and research. There is huge potential there, especially when you consider how grueling chemotherapy can be, but also the huge challenges. The burden of trust for data privacy, for one. Digital twins are data hungry and their accuracy depends on a constant flow of quality and sensitive information about you. It then informs future decision making. If I felt overwhelmed three months into a low-risk environment, how would that translate to higher-risk scenarios? How does a digital twin exist after treatment? How does this data survive if treatment fails?

[Image: The CGM tests are still ongoing because my digital twin is hungry for that data.

As for me, I have decided that my digital twin will "live" for another three months. Despite my lack of success so far, I haven't given up on the idea of ​​cutting back on some medications. I could also use all the help I can get to get my liver labs back to normal. Amanda also says that now that I have a solid foundation, I don't have to be so aggressive with my twin data. Again, less stock means more flexibility.

So today, I'm asking my digital twin what he thinks about a protein waffle with a tablespoon of peanut butter and a drizzle of honey, along with a side of pineapple. Orange, but I can still make it green by walking.