The Limitations of an ‘AI Health Coach’ in Addressing Global Chronic Disease Challenges

Last week, two big names in the artificial intelligence (AI) and wellness industries announced a collaboration to develop a “customised, hyper-personalised AI health coach that will be available as a mobile app” to “reverse the trend lines on chronic diseases”.

Sam Altman (head of OpenAI, maker of ChatGPT) and Arianna Huffington (a former media executive who runs a high-tech wellness company called Thrive Global) announced their new company, Thrive AI Health, in a Time magazine advertorial.

Health is an appealing direction for an AI industry that has promised to transform civilisation, but whose huge growth of the past couple of years is beginning to look like it’s stalling. Companies and investors have pumped billions into the technology, but it is still often a solution looking for problems.

Meanwhile, venture capitalists Sequoia and the investment bank Goldman Sachs are wondering out loud whether enough revenue and consumer demand will ever emerge to make this bubble feel more solid.

Enter the next big thing: AI that will change our behaviour, for our own good.

Personalised nudges and real-time recommendations

Altman and Huffington say Thrive AI Health will use the “best peer-reviewed science” and users’ “personal biometric, lab and other medical data” to “learn your preferences and patterns across the five behaviours” that are key to improving health and treating chronic diseases: sleep, food, movement, stress management and social connection.

Whether you are “a busy professional with diabetes” or somebody without “access to trainers, chefs and life coaches” — the only two user profiles the pair mention — the Thrive AI Health coach aims to use behavioural data to create “personalised nudges and real-time recommendations” to change your daily habits.

Soon, supposedly, everybody will have access to the “life-saving benefits” of a mobile app that tells you — in a precisely targeted way — to sleep more, eat better, exercise regularly, be less stressed and go touch grass with friends. These “superhuman” technologies, combined with the “superpowers” of incentives, will change the world by changing our “tiny daily acts”.

Despite claims that AI has unlocked yet another innovation, when I read Altman and Huffington’s announcement I was struck by a sense of déjà vu.

Insurance that manages your life

Why did Thrive AI Health and the logic behind it sound so familiar? Because it’s a kind of thinking we are seeing more and more in the insurance industry.

In fact, in an article published last year I suggested we might soon see “total life insurance” bundled with “a personalised AI life coach”, which would combine data from various sources in our daily lives to target us with prompts for how to behave in healthier, less risky ways. It would of course take notes and report back to our insurers and doctors when we do not follow these recommendations.

In a related article, my colleagues Kelly Lewis and Zofia Bednarz and I took a close look at the theories of behavioural risk that might power such products. A model of insurance based on managing people’s lives via digital technology is on the rise.

We examined a company called Vitality, which makes behavioural change platforms for health and life insurance. Vitality frames itself as an “active life partner with […] customers”, using targeted interventions to improve customer well-being and its own bottom line.

Similar projects in the past have had questionable results. A 2019 World Health Organization report on digital health intervention said:

The enthusiasm for digital health has also driven a proliferation of short-lived implementations and an overwhelming diversity of digital tools, with a limited understanding of their impact on health systems and people’s wellbeing.

Hyper-personalisation

Altman and Huffington say AI-enabled “hyper-personalisation” means this time will be different.

Are they right? I don’t think so.

The first problem is there is no guarantee the AI will work as promised. There is no reason to think it won’t be plagued by the problems of bias, hallucination and errors we see in cutting-edge AI models like ChatGPT.

However, even if it does, it will still miss the mark because the idea of hyper-personalisation is based on a flawed theory of how change happens.

An individualised “AI health coach” is a way to address widespread chronic health problems only if you envision a world in which there is no society – just individuals making choices. Those choices turn into habits. Those habits, over time, create problems. Those problems can be rooted out by individuals making better choices. Those better choices come from an AI guardian nudging you in the right direction.

And why do people make bad choices, in this vision? Perhaps, like middle-class professionals, they are too busy. They need reminders to eat a salad and stretch in the sunshine during their 12-hour workday.

Or – again from the AI health coach perspective – perhaps, like disadvantaged people, they make bad choices out of ignorance. They need to be informed that eating fast food is wrong, and they should instead cook a healthy meal at home.

The social determinants of healthcare apps

But individual lifestyle choices aren’t everything. In fact, the “social determinants of health” can be far more important. These are the social conditions that determine a person’s access to health care, quality food, free time and all the things needed to have a good life.

Technologies like Thrive AI Health are not interested in fundamental social conditions. Their “personalisation” is a short-sighted view that stops at the individual.

The only place society enters Altman and Huffington’s vision is as something that must help their product succeed:

Policymakers need to create a regulatory environment that fosters AI innovation […] Health care providers need to integrate AI into their practices […] And individuals need to be fully empowered through AI coaching to better manage their daily health […]

And if we don’t bend society to fit the AI models? Presumably we will only have ourselves to blame.