How to make Health.AI mainstream?
A doctor who used big data to not only detect but beat cancer — her own.
FDA approving a company that detects whether you are having a stroke.
Predicting heart disease by looking at your eye health.
It seems like everyday we hear about news articles like this. But while there has been explosive progress in using AI to solve big health problems, we are still very far from mainstream adoption. Here are the three key things that need to happen.
1) Data Collectability
Arguably the most obvious tenet -- you can’t generate meaningful insights if you don’t have enough data. Ubiquitous phones with biometric sensors can provide some data, but it’s limited essentially to heart rate, steps and maybe temperature and sleep. How do you know really important biometrics encoded in your breathing or blood? To what extent are you willing to go to have this data to be measured and analyzed? Are chips embedded in your body the solution? We are moving progressively to a world where everything can be measured and monitored but we will have to balance them against privacy, safety, and ultimately even address the question of what makes us human.
2) Retrospective Repeatability
Many studies the media is quoting now are retrospective — you have diabetes, well let's see now if we can use your data to predict whether someone else today will have diabetes. There could have been many confounding factors, say people with diabetes were taking medications causing the body to have a certain response, so what you are seeing is correlation rather than causation. For a true prediction you need to do double-blind prospective studies ie here are two sets of people without diabetes, I predict the first set will get diabetes and another won't. And these studies need to be longitudinal ie follow someone for years. Not to mention that an algorithm will probably be judged much harsher than a human for false positives and false negatives. And finally these study results need to be repeated many times by different groups around the world to gain acceptance in the scientific / medical community.
3) Explainable Predictability
Medicine has to make very sure about the risks and that's not a bad thing. Take genetic therapy just a decade ago, there were deaths that could have been prevented and you can argue it actually set back the whole industry.
I find the biggest challenge of big data is it is like a black box we know there is prediction but we can't explain it. Getting doctors to accept advice without understanding that explanation is upending what modern medicine and in fact the scientific method has been preaching in the last 500 years. Until (If?) we get there, the very likely way that Health.AI will develop is to highlight issues -- if it detects an anomaly or a risk, it will encourage you to seek a doctor. Which is worth in itself, as any provider and payor will tell you that can save individuals their life and society billions of dollars.
These are purposely short articles focused on practical insights (I call it gl;dr -- good length; did read). I would be stoked if they get people interested enough in a topic to explore in further depth. I work for Samsung’s innovation unit called NEXT, focused on early-stage venture investments in software and services in deep tech, and all opinions expressed here are my own.