Artificial intelligence won't solve all of medicine's great problems

Almost 20 years ago, the medical and scientific communities were overjoyed. With the Human Genome Project finished, there was an air of inevitability that the causes of some of the most common and destructive diseases would soon be pinpointed and eradicated. It’d be simple: one gene, one problem, one solution. We even heard Francis Collins, at the time, say “over the longer term, perhaps in another 15 or 20 years, you will see a complete transformation in therapeutic medicine.” Unfortunately, it was never going to be that easy.

As a result of the Human Genome Project, we do have a much clearer picture of what lives inside of us and why. But the other side of that coin is that, as a result, we also know disease rarely ever follows tidy rules, like a single gene as the root problem. Multiple genes play multiple roles, and a quick, targeted solution is most often not possible. Even for conditions identified as being caused by a single gene, like Huntington’s Disease, we’re still working on a first treatment, much less the be-all, end-all solution.

Read more commentary:

Artificial intelligence could identify you and your health history from your step tracker

I am not a robot. I'm a doctor and my patients need the real me.

Medicine alone can't lengthen US lives. We need to invest outside the health care system.

The confidence in the Human Genome Project as a silver bullet mirrors the enthusiasm many now have for artificial intelligence. When it’s not driving our cars, selecting your baseball team’s starting lineup, or even creating art, AI has been slated by many to make huge strides in disease treatment. Using it, we will be able to map the root causes of disease, then run simulations to determine the exact treatments we need to slow and stop them. With just a little more time, our use of AI will solve some of the most treacherous illnesses we face.

Sound familiar?

The truth is, there is a lot of promise in AI. The limitation of many Human Genome Project-guided solutions was that each problem was tackled looking for one or two sources, virtually by hand, using the “maps.” Artificial intelligence can do that path-finding at an exponential rate.

AI helps, but nothing in medicine comes easy

And while the Human Genome Project developed a look at our bodies in totality, the humans tasked with working with this data are physically unable to factor in the complexity of the human conditions. Programs equipped with AI have the benefit of considering data from entire populations when working on our genome, quickly recognizing patterns and outliers that might have taken us years to root out — if we ever did.

But it might be time to temper our expectations. The reality is, AI isn’t exactly new: It’s been a part of medicine, to varying degrees, since the 1970s, even before the Human Genome Project was tackled. And AI has been wrestling with the data from that project and its derivatives for decades now, with limited success. If we’re being honest with ourselves, there’s a chance that the best we can ever do is create a more accurate map of our risks for disease using measures of family history.

Clearly, AI has opened doors to a lot of new and exciting things. Learning from the past, if we can just meter our excitement and allow for this technology to be developed at a reasonable pace, with realistic goals, it’s possible that we could reach some of the heights we’d always hoped. We need to realize, once again, that nothing in medicine ever comes easy, and all of the intelligence in the world, artificial or not, won’t change that. 

Jason Moore is the director of the Institute for Biomedical Informatics at the Perelman School of Medicine, University of Pennsylvania. Follow him on Twitter: @moorejh