AI Talk: Human compatible, Fitbit, telehealth

October 25th, 2019 / By V. “Juggy” Jagannathan, PhD

This week’s AI Talk…

Human Compatible: AI and the problem of control – Stuart Russell. Book review.

Stuart Russell is a well-known Professor and AI researcher. His book on AI with Peter Norvig is a staple for teaching AI in many classrooms across the U.S. I myself have used their book multiple times to teach graduate students about AI. His latest book, released this month, echoes themes that have been in the news recently. This book is about Artificial General Intelligence (AGI). What happens if we manage to create a super-intelligent machine? It could upend the world as we know it. Vast numbers of people will no longer need to work. In this case, perhaps everyone should be offered a Universal Basic Income (UBI)—a theme even taken up by a current Democratic candidate for President. Maybe we need to teach people the art of being happy and contented (this by the way is what a lot of spiritual disciplines attempt to do). 

Then there is the question of whether we will ever be able to create a machine that has superhuman intelligence. Skeptics believe that possibility is unrealistic, given the current state of affairs. Russell argues not so fast. We held similar views about nuclear fission and harnessing nuclear power. It was considered impossible until it suddenly became possible! Don’t bet against the ingenuity of humans. Whether AGI will be achieved in two decades or ten is not the question, but rather are we prepared for a world in which superhuman intelligence exists? What should we do in preparation? That is the genesis for this whole book. Don’t bet that we will not build such machines, but instead focus on how we will build our AI systems now, such that they are always subservient to our priorities, our preferences and always do our bidding. 

Russell echoes what many others have pointed out: Current deep learning solutions do not have the capacity to learn common sense and do abstract thinking. He suggests some of the abstract thinking embodied in first order logic formalism combined with deep learning may provide a building block for reasoning systems. Whatever the final approach, the thesis here is focus on building human-assistive systems from the start and we will not have to contend with such systems taking over the human civilization.

Fitbit venturing into monitoring space

We are all aware of Fitbit technology used to track workouts. Now, Fitbit is competing with Apple Watch to track atrial fibrillation (Afib). Fitbit is now a free covered benefit on over 59 Medicare Advantage plans and 100 different health plans. They have recently announced a partnership with Bristol-Myers Squibb-Pfizer alliance to track Afib. Fitbit is quickly trying to make its wearable go from “nice to have” to “must have”—tracking a variety of conditions from diabetes to heart conditions to sleep apnea! Its solutions are currently undergoing clinical trials and FDA approval processes.

App for Telehealth visits

UnitedHealthcare (the largest insurer in U.S. covering a third of the population), is going big with telehealth. It has launched a new app for customers that facilitates televisits virtually 24 hours a day. Just last month, Amazon launched a similar service for its employees—dubbed Amazon Care. The reason for the significant uptick in these types of visits is clear: It reduces utilization and decreases cost. Anthem is reducing the co-pay amount for these televisits compared to regular visits, explicitly encouraging the use of telehealth services. We are going to see more and more of these types of visits supplanting traditional clinic visits—it simply makes sense both from a care perspective and cost perspective.

I am always looking for feedback and if you would like me to cover a story, please let me know. “See something, say something”! Leave me a comment below or ask a question on my blogger profile page.

V. “Juggy” Jagannathan, PhD, is Director of Research for 3M M*Modal and is an AI Evangelist with four decades of experience in AI and Computer Science research.