AI For Runners And Athletes: We're Making Excellent Progress

Forbes

21-04-2025

BOSTON, MA - APRIL 21: Meb Keflezighi of the United States celebrates after winning the 118th ... More Boston Marathon on April 21, 2014 in Boston, Massachusetts. (Photo by)
In a recent panel conference on AI, three experts in the industry talked about breakthroughs in AI fitness wearables that will have enormous ramifications for runners and athletes.
A number of really good insights came out of this talk – pieces of the context around engineering that show us why AI is so good at supporting people in their athletic efforts.
I'm going to go over one fundamental idea from each of these contributors, to show three design aspects that are a huge part of how we're exploring AI use for athletics.
This idea about AI actually has two components – one related to the places where we put large language models, and another revolving around the difference between our human languagesand the language that our vital signs 'speak' when they are measured in real time.
Alexander Amini knows a lot about running - and tennis. But he's also involved in a company that's revolutionizing AI for our times.
In describing how AI can work for athletes, Amini first said it's about 'enabling AI to live where the data lives.'
He explained – early LLMs were put on the Internet, because edge devices didn't have the capacity to run them. Transformers were expensive and complex.
Now, the new models mean we can install smart LLMs on edge devices, and that will bring the cost down, while helping athletes who are in places without Internet access.
'It unlocks a completely new dimension of how we think about interacting with AI,' he said. 'AI doesn't need to necessarily be something (where) you send all of your data up to a third partyprovider anymore. It now can be something that is deeply embedded with your life, at every stage of your day.'
His other point was that by harness harnessing ourselves with these wearables, we're getting a robust set of data. It's far beyond what we get by talking.
'It really comes down to the way that we talk to AI,' he said.
He also broke this down further. Here's my interpretation: essentially, language is a set of symbols that we use to send ideas back and forth. It's 'compressed,' in his words. It's not raw. By contrast, if you think about all of the data that your body is giving off every microsecond, that's a different medium to communicate in. If AI is getting all of that data and using it,instead of listening to you talk, it can tell a lot more about you as a person.
'The data that we generate from our bodies is impossible to be compressed through language,' Amini said. 'It is highly multi-dimensional, and it's highly expressive signals, right? And it's constant. The amount of data is vast, and in many cases, it's not really possible to transmit the full complexity of our fitness in real time to these systems. So that's why, when we talk about bringing AI to ourselves … that's what we unlock.'
Another great take on this came from Emily Capodilupo, who has designed a tool called Whoop.
She explained the humans are not good at diagnosing whether they are dehydrated or not.
'We're actually horrible judges of being dehydrated,' she said. 'Humans tend to not realize that we're becoming dehydrated. In general … your body knows a lot … but there are incredible safety use cases, if you have algorithms running on your wrist, that can help you understand that ratio between how hard your body is working, and your pace: if they start to diverge too quickly, that could be a sign that you're on track to … hit a wall, or that you're sort of suffering from dehydration.'
This rings true to me as a runner: if you think about trying to self-diagnose as a human, it's really down to whether you feel thirsty, or hot, or you can tell by your skin that you're dehydrated. AI can tell down to a much more granular level what your body needs, and that's going to be an enormous advantage for any kind of athlete who's equipped with these tools.
Jamie is another person who spoke that day. He has done a lot of research on his own body, through a process called digital twinning.
A digital twin is simply a robust model of something in the real world. I often use the term 'doppelgänger' to talk about creating a system. That's a highly detailed simulation of something physical or outside of the digital world in the analog space.
The digital twin provides that playground of analysis that helps us to advance our measurements.
Jamie revealed that he has built 100 samples of himself, foraround $5000 each.
'I've spent a half a million dollars on my own biology in the last ten years measuring outcome changes,' he said. 'Those outcome changes add up into this digital twin that predicts my future, predicts all my disease risks, and we have blood tests now.'
In terms of access, he quoted William Gibson: 'The future is here - it's just not evenly distributed.'
'We have blood tests now for essentially every human disease,' he continued, 'that are somewhere between five(fold) and tenfold more accurate than the current predictions in the market.'
Going further, he explained that research in applying new technologies found that half of patients analyzed could benefit from a significant intervention based on collected data.
'In half of the people we've done these analysis on, we have found a life-changing piece of actionable information in them that is likely to add years to their life, and that's computed, predicted and analyzed,' he said. That's an amazing statistic.
Each of these three revelations illustrates the immense power of AI to help us to 'fine-tune' our lives, for fitness, for longevity, etc. In so many ways, we don't have to guess anymore.
Basically, a lot of what we previously considered science fiction is now real. Certain tools may or may not be on the market yet – we haven't seen all of this technology play out in our lives so far. But that's going to change soon.