A decade back, the notion of keeping tabs on your steps or heartbeat seemed odd. The folks preaching about self-quantification were doing it in TED Talks, and journalists were covering this peculiar trend at conferences. Fast forward to today, and more than 40% of U.S. households have a wearable device, as per Statista. It’s now pretty normal to hear retirees chatting or showing off their daily step count. The trend of quantifying oneself is on the rise.
With artificial intelligence making relentless progress, researchers and tech experts are now exploring ways to go a step further—creating AI health coaches that go through health data and advise users on how to stay in top shape.
Plenty of proof indicates that wearables do bring some advantages. A 2022 review of scientific studies discovered that, among over 160,000 participants in all the studies, those assigned to wear activity trackers took about 1,800 extra steps daily, resulting in a weight loss of roughly two pounds.
Also Read: Claude AI is the new kid on the block; How is it different from ChatGPT?
AI can amplify precise measurements of metrics
According to Carol Maher, a professor of population and digital health at the University of South Australia and one of the review’s co-authors, wearables alter behavior in various ways. They encourage users to set goals, enable them to keep an eye on things that matter to them, and remind them when they’re veering off course from their objectives.
But, as Andrew Beam, an assistant professor in the Department of Epidemiology at the Harvard T.H. Chan School of Public Health and a researcher in medical artificial intelligence, points out, these effects tend to diminish over time.
To precisely figure out the metrics we’re interested in, like counting steps from a wrist-worn accelerometer, we need AI, but it’s a plain, unglamorous kind, explains Shwetak Patel, a computer science and engineering professor at the University of Washington and the director of health technologies at Google.
But, he adds, there’s a lot more it can already achieve: “AI can stretch the capability of that sensor to do things that we may not have thought were possible.”
This involves functions that are currently found in common wearable devices, like detecting falls and monitoring blood oxygen levels. Some researchers are attempting to utilize the fairly basic health data from wearables to identify diseases, including COVID-19, although usually not with the same level of accuracy as devices used in clinical settings. Up until now, AI has been a sidekick in the ascent of the quantified self. Now, researchers are aiming to leverage recent advancements to bring AI into the spotlight.
AI coaches in the offing
Patel recently teamed up on a paper where researchers inputted data from wearables into big language models like OpenAI’s GPT series. The models then provided insights about the data, which could be handy for clinicians making mental health diagnoses. For instance, if a participant in the study had inconsistent sleep duration data, the AI system would flag it and mention that irregular sleep patterns “can be an indicator of various issues, including stress, anxiety, or other disorders.”
According to Patel, the upcoming AI models can think for themselves, opening the possibility of using them for personalized health coaching. (However, some researchers contend that it’s still uncertain whether big language models can truly reason.)
“It’s one thing to say, ‘Your average heart rate is 70 beats per minute,’” he says. “But the thing that we’re focusing on is how to interpret that. The kind of modeling work we’re doing is—the model now knows what 70 beats per minute means in your context.”
The information from wearables could also enable AI “coaches” to grasp users’ health with much more depth than a human coach could, notes Patel. While a human coach might ask about your sleep, wearables can offer detailed, objective sleep data.
Maher has also been part of a team that reviewed the research on how effective AI chatbots are in influencing lifestyle behaviors. The review discovered that health coaches in the form of chatbots can assist individuals in boosting their physical activity and sleep, as well as enhancing their diets. However, the impact was somewhat less compared to what is usually observed for wearables.
These studies involved quite basic chatbots (created years ago, well before advanced models like OpenAI’s ChatGPT, for instance). Maher anticipates that more advanced AI health coaches would be more effective. However, she points out that there are still hurdles to overcome with large language models like ChatGPT, such as their tendency to fabricate information.
Also Read: AI can mimic human social skills in real-time, should we be scared?
Are there any downsides to this?
Beam highlights reasons to be doubtful about chatbot health coaches. Firstly, they experience a decline in effectiveness over time similar to wearables. Secondly, in the health domain, even human scientists, when provided with extensive data about an individual, still lack enough understanding to offer personalized advice.
Even if there isn’t solid evidence to give specific recommendations tailored to individuals from their health data, an AI health coach could keep tabs on whether a particular action seems beneficial and tweak its advice accordingly. For instance, heart rate data during a recommended workout could guide future exercise suggestions, as mentioned by Sandeep Waraich, product management lead for wearable devices at Google.
Google hasn’t officially revealed any intentions to introduce an AI health coach. However, it does intend to offer AI-driven insights to Fitbit users starting in early 2024. In August, The New York Times mentioned that Google DeepMind is developing an AI “life adviser.” On the other hand, Apple is said to be in the works on an AI health coach, known as Quartz, with plans for release next year.
It’s not only the major tech players trying to leverage data from wearables for ongoing, personalized health guidance. The health app Humanity asserts that it can pinpoint a user’s “biological age” within a three-year range using movement and heart-rate data. Their algorithm was crafted using data from the U.K. biobank, where 100,000 participants wore a wrist-worn accelerometer for a week.
The issue with monitoring metrics such as Humanity’s “biological age” is that there’s still no proof connecting these measures to real health outcomes, such as a decrease in overall mortality, according to Beam. He mentions that this is a broader problem with the use of AI in healthcare.
