Habit and a GPT

If we live and eat healthy, take time away from the computer, and take our medicines. we should live healthier lives. This is what we want to happen. But many of us don’t adhere to the doctor’s advice and don’t walk every day or we forget to take our medicines every day. We fail despite our desire to make lifestyle changes the doctor prescribed. This is as true for diet and exercise as it is for clinical treatment programs.

Thomas H. Davenport from MIT and Ravi Kalakota from Deloitte writing about The Potential for Artificial Intelligence in 2019 found that out of 300 US-based healthcare practitioners, nearly half admitted that 75% of their patients were not fully engaged with their treatment plans. In other words, a significant share of Americans are not changing their diet or exercising more when their doctor prescribes it, or are not taking their medicines as prescribed.

I’d be willing to bet patients would prefer their treatment plans be successful, especially when it means not having to come back to the hospital. There are a host of challenges related to treatment plan adherence (inadequate insurance) that are beyond the scope of this short post (and my expertise) but to put it simply, habit science has some pretty good answers.

43 percent of what people do every day is repeated in the same context, usually while thinking about something else. Ok, that sounds easy. I don’t often think about the steps involved in brewing coffee in the morning, even though there’s at least 6 steps involved in making it for my wife and I.

It also takes 60-63 days - frequency-dependent - to achieve habit change. I’ve built products for health care practitioners using a human-in-the-loop design (via App/Whatsapp), but this was before OpenAI and ChatGPT.

What I wanted to know was how did AI affect behaviour? We’re experimenting with a GPT for learning and service delivery use-cases, but we did find some promising evidence from the pharmaceutical industry, a subclinical Trial from AbbVie in Chicago, Illinois.

The 2018 subclinical Trial by Shafner Bain EE and team demonstrated an improvement in adherence to treatment programs for schizophrenics of 25%. They demonstrated that AI platforms could increase adherence, rapidly detect non-adherence, and predict future non-adherence. This means, there is evidence that generative AI can improve patient outcomes when carefully designed with physicians.

Use of a Novel Artificial Intelligence Platform on Mobile Devices to Assess Dosing Compliance in a Phase 2 Clinical Trial in Subjects With Schizophrenia

Essentially, Generative AI equips an attending physician (or medical team) with new tools that can be used in treatment plans. Combined with medical knowledge and personal medical records, generative AI can:

• Predict behaviour based on behavioural insights data
• Trigger/Receive prompts via API
• Provide timely prompts at key decision points with contextual-awareness.

Here is what we know. It takes 60-63 days to gain or change a habit, and habits cover about 43% of what we do everyday. Seems like if we can get those new good habits up and running, we would have healthy behaviours happen automatically, something I think sounds amazing (as I consider going downstairs for a snack). But don’t take it from me, read how Wendy describes habit:

“A habit happens when a context cue is sufficiently associated with a rewarded response to become automatic, to fade into that hardworking, quiet second self. That’s it. Cue and response. Notice that there’s no room in that mechanism for, well, you. You’re not a part of it, not as you probably think of yourself. You—your goals, your will, your wishes—don’t have any part to play in habits. Goals can orient you to build a habit, but your desires don’t make habits work. Actually, your habit self would benefit if “you” just got out of the way.”

― Wendy Wood (1)

Here is a scenario where generative AI can be helpful. A newly diagnosed diabetic will often be described treatment plans that include lifestyle change if they wish to avoid future medical treatment. At holiday family gatherings, that newly diagnosed diabetic may eat a small piece of cake (say at a sister’s house party), and afterwards may become tempted to throw caution to the wind and indulge in all the sweets they want for the rest of the holiday.

We call this the “What-the-hell-effect.” Obviously, from a health and happiness perspective, the newly diagnosed diabetic would have preferred a better outcome, perhaps one where they estopped after the first piece of cake, or didn’t eat the cake at all.

It is for these moments, when a doctor can’t be there, that generative AI can be a big help.

By combining a doctor’s treatment plan, and a patient’s own goals, generative AI has the potential to improve patient success by 25%, when used to help us establish new habits.

References

(1) Good Habits, Bad Habits: The Science of Making Positive Changes That Stick

‍