The Premises and Promises of Artificial Intelligence in Healthcare

Artificial Intelligence (AI) has started changing our world and our lives, including in healthcare — where, after initially slow progress, some of the early promises are becoming reality. But so are the many things we need to keep in mind in order to “do this right.”

The recently completed Partners World Medical Innovation Forum in Boston assembled some 2,000 participants over three days, including key policymakers, technology industry heads, clinical and research leaders from major provider organizations and others.

The gathering offered much excitement and inspirational gazing into the future of AI in healthcare, with Kyu Rhee, M.D., chief health officer of IBM Watson, observing (as a “Game of Thrones” fan) that “winter” for AI had come and gone. Indeed, the spring for the technology is now upon us — with momentum gathering from all corners of the healthcare universe.

The promise of AI is, of course, that the technology can, in its first iterations, make all of us in healthcare —clinicians, researchers, administrators, patients — more effective and efficient by taking over routine tasks and reminding us of what is known. For instance, AI can automate the routine reading and interpretation of medical images, such as X-rays and CT scans, exceeding the accuracy and reliability, and certainly the stamina, of even the most expert clinicians. It can also alert clinicians to newly available knowledge in considering their diagnostic and therapeutic decisions.

Consider the following: a physician may order a chest X-ray for a patient with respiratory symptoms to confirm or rule out pneumonia. AI can help accurately discriminate patterns in the image and help reach the right diagnosis. But it can also point out rare and/or incidental findings on that chest X-ray, such as in the imaged bone structures, which the clinician is not really looking for and may overlook.

Yet the even more exciting (but also more controversial) uses of AI and machine learning technologies are in generating new knowledge and asking questions humans would not think to pose. As several conference speakers noted, AI is changing the old adage ‘I don’t know what I don’t know.’  Jensen Huang, CEO of NVIDIA, pointed out: “AI is software that writes software, with the fundamental purpose of automating automation.” Consider the chest X-ray example: Once that image is read for a particular diagnostic purpose, it is relegated to archives. There is, however, still an entire treasure trove of information on such a medical image that automated machine-learning algorithms can continue to explore and find new associations in/from. AI allows us to move from “systems of records” to “systems of insights.”

Some premises of AI technology, however, should always be top of mind:

• AI requires a lot of data, something that the increasingly ubiquitous data sources, such as the exploding world of sensors that record our human condition and environment, make possible. But which data are sampled in order to construct and train the AI algorithms is key and should, therefore, be transparent. Add “bias in, bias out” to the well-worn “garbage in, garbage out” notion here, as the resulting AI application will have a different sensitivity and specificity for different populations, depending on how (un)represented they are in the data;
• AI algorithms should be continuously evaluated, validated and improved, with new evidence and additional data, as well as outcomes from their “hits and misses”;
• AI findings as the basis for any decisions must be readily justifiable, with their origins (rationale) continuously transparent;
• AI-enabled healthcare systems and decisions should, by design, have clear, embedded human oversight and accountability;
• AI systems can free us from routine, repetitive tasks and help us return humanism to all of our healthcare interactions, including patient care — where clinical care teams can provide empathetic, value-added advice and feedback to patients instead of being distracted by frantic data input into the electronic health record and other applications.

Indeed, for all the optimism and excitement resounding at the Partners World Medical Innovation Forum, there was also the sounding of many warnings about the dangers of AI further deepening inequities in society and healthcare (with applications narrowly sourced and built for the “haves” versus the “have nots”), the sliding into over-reliance on the technology without human checks and balances, potentially leading to catastrophic consequences.

In an extraordinarily somber address, Massachusetts Gov. Charlie Baker implored attendees to be guided by the notion that the opportunity of AI comes with tremendous responsibility, with all of us in healthcare held to a higher standard than most other industries, as we impact the lives and health of millions.

With technological advances such as those in natural language processing making so much possible, can we all resist the temptation to go too far, too fast?  The spring of AI in healthcare will require some growing from us all, as society and as individuals, so let’s work diligently toward a lot of beautiful blossoming.