AI Works Better with Human Intelligence, Too

Artificial intelligence applied to the right tasks can reveal insights that wouldn't otherwise be surfaced, and do it faster thn manual human efforts. But there are still some tasks that humans perform better than machines. Eric Williams, VP of Data Science & Analytics at Omada Health provided a view into how his digital therapeutics company is combining AI with human intelligence to help pre-diabetic patients lose weight.

Williams provided these insights during the Interop ITX session, Where Artificial Intelligence Meets Human Intelligence: Building Digital Healthcare for the Future. At the very beginning of his presentation, he noted that while we know that diet and exercise and weight loss can prevent Type 2 diabetes, motivating individuals to make the necessary changes to their lifestyles is difficult. Yet companies that provide health insurance to their employees, particularly those companies with self-insurance, have a vested interest in reducing healthcare costs, and diabetes is an expensive condition.

To help companies and insurers help patients make the behavioral changes necessary to prevent diabetes, Omada provides a "Symphony" of tools that include scales, wearables, and a food-tracking app. These devices send data back to the company. Omada also provides patients with behavioral coaching. Omada charges companies based on patients' actual weight loss tied to the program.

It's the "Netflix-ization" of healthcare, according to Williams, in which unprecedented amounts of behavioral data allows for a new era of behavioral science.

The company's data science team designed the program to be self-learning and constantly improving. As part of its evolution, Omada sought to complement its health coaches with  digital health coaches. Digital health coaches could interact with all the patients simultaneously, whereas human health coaches are limited to dealing with each patient one at a time. Plus, digital health coaches ensure that patients were following the prescribed program, eliminating the variability that comes with using many human coaches.

But there's also value to having humans in the machine learning loop, according to Williams. For instance, if a patient registers a 20-pound weight loss in a day or two, the machine may congratulate the patient on a job well done. A human is more likely to recognize the anomaly for what it is.

Humans can also help bridge the empathy gap, Williams said.

"It's still a safe assumption that human-to-human connection, empathy, understanding, and problem solving are valuable," he said.

And in some cases, a human may recognize something different is happening with a particular patient that is contributing to a lack of weight loss. In data science terminology that would be called "latent variables that we are not capturing," Williams said. But the human in the loop can identify and capture that data that the system doesn't yet recognize.