At the 2016 AI World Conference in San Francisco, Shivon Zilis and James Cham presented their newly minted Machine Intelligence 3.0 Landscape. Among the notable features of the infographic was the dedicated healthcare section, separate from the rest of the industries. In addition to the 25 healthcare specific companies the Bloomberg Beta partners highlighted, there are also a number of cross-industry companies listed, like CognitiveScale, that are working to tackle challenges of health management with cognitive computing.
It might be autonomous cars and Go champion computers that receive the biggest hype, but as Zilis and Cham explained, “the techniques used to conquer the gaming world [are] moving to the real world.” The choice to give healthcare its own section in the 3.0 landscape is indicative of the impressive and bold strides machine intelligence is making in the industry.
It may seem surprising that healthcare is a pioneer in artificial intelligence, given the industry’s reputation for resisting change. We tend to think of hospital administrators as luddites, burdened with a complex bureaucracy. Day two of the conference featured a panel titled “AI and Cognitive Computing in Healthcare,” and the speakers were not shy about the colossal challenges they face as tech innovators in healthcare. Overwhelmingly, trust emerged as the chief characteristic needed to keep cognitive computing thriving with hospitals.
“The challenges are really on the regulatory side,” John Axerio-Cilies said. John is the CTO and Founder of Arterys, a startup building cloud-based medical imaging software. “Each country has a different policy around privacy and a different set of regulations.” Getting the appropriate permits and having the agility to work within so many distinct environments is particularly costly.
Corey Kidd, CEO of Catalia Health, added that even when working with a single hospital, “There are lots of layers of bureaucracy and people to convince, even when you’ve got many supporters within the company.”
“At CognitiveScale, we think of these regulatory and privacy challenges in terms of trust and interpretability,” CTO Matt Sanchez explained. “The black box does not fly for regulated environments,” referring to the often obfuscated nature of machine learning models. The gap between your input to a model, and the output the model produces, commonly referred to as the black box, can make your machine learning models seem like flawless magic, or irrelevant noise — neither of which make a particularly impressive impression on care managers.
Inability to properly explain where a decision came from has saturated healthcare with apps promising to revolutionize hospital workflow. However, when you’re dealing with the intricate aspects of a person’s wellness, you can’t simply serve up a piece of advice. You must be able to trace that insight back to the source and figure out why it was given. “There’s too much hype out there,” Corey said, “and that detracts from what is real.”
“AI has to be thought of as trusted and transparent, particularly with regulated industries,” Matt said. “We have to prove that the algorithms we develop do what they intended, and a regulator has to be okay with it.”
Catilia Health’s approach to building that trust is through a “cute robot that sits on the user’s desk or bedside.” The company markets Mabu, a robot health assistant that uses psychology behavioral models to encourage positive behavioral change. The reasoning behind the cute robot? “It’s not about the technology,” Corey explained. “It’s for the psychology. The benefits of a face-to-face interaction.”
Building trust with patients stems from understanding them as more than their EHR. Matt said, “In our work with Intermountain Healthcare, we found that when you provide patients with technology that helps them understand the environmental and behavioral effects on their health, it has a dramatic impact on how they self-manage.”
David Ledbetter, a Data Scientist at the Children’s Hospital Los Angeles, explained that his team approaches trust by building close relationships between the technologists and the clinicians. “We go on rounds and shadow the nurses. I‘ve even had the chance to sit in on some surgeries.” By watching the decision making process in real time, data scientists can better understand where the data came from and create actionable and useful solutions.
“Ultimately, we are trying to augment each other, and not replace each another,” David said. By working within the hospital, David and his team are able to create visible, effective change with their care manager co-workers.
But when you don’t have the benefit of working within the hospital, as is the case for most of the healthcare innovators, then trust must be forged in a different way. “I’m not going to walk into a retailer and tell them — rip out your E-commerce system that you’ve spent millions on and implement this AI one,” Matt explained. Similarly for healthcare, “its about incremental changes. Working within their system one insight, one process at a time. It’s not about waving a magic wand and fixing everything all at once.”
The ultimate hope of these cognitive technologies is that patients will receive more personalized care. “Yes there’s Epic, yes there’s Cerner, but I’ve never met a physician that loves their EHR system,” Matt said. These systems eat away at time that physicians could be spending with their patients. With the help of artificial intelligence, the goal is to scale physicians’ ability to use the technology, opening up more time to forge meaningful relationships with their patients.
Applying AI to healthcare is a bold vision with great promise and there are already tangible results to prove the viability. Building a culture of transparency and trust is integral to keeping the pace. Machine intelligence companies must operate with agility to find those spots for change — process by process, insight by insight. If companies cannot prove the integrity of their mission, AI in healthcare will vanish into another winter. The successful ventures will be from those who build out this trust infrastructure, breaking down the divide of the technologist and the physician, to formulate mutually beneficial solutions for a future of sustainable, quality care.