IBM Watson Flops For Cancer Treatment: Why Did AI Fail?

MD Anderson Cancer Center recently announced it was dropping IBM Watson’s cognitive-computing system, which was supposed to use artificial intelligence (AI) to expedite physicians’ decision making when treating malignant tumors.

Amid that move came a question: Was the problem with Watson – and artificial intelligence in general – or was it a human problem?

Goals of Arrangement

First, consider what Watson was supposed to do for MD Anderson, a major cancer-treatment center that’s part of the University of Texas system.

A 2015 Washington Post story, while mentioning some limitations of AI, took an optimistic view of the system and its medical applications. It called Watson as “a revolutionary approach to medicine and health care that is likely to have significant social, economic and political consequences.”

The story explained that Watson would help doctors find “personalized treatments for every cancer patient by comparing disease and treatment histories, genetic data, scans and symptoms against the vast universe of medical knowledge.”

In other words, Watson would digest piles of existing data and use it to produce a course of treatment. And according to IBM and some people at MD Anderson, it worked.

IBM defended the product, known at Anderson as the Oncology Expert Advisor (OEA), and the results of its application. In tests, a company spokesperson told Forbes the OEA’s recommendations were on target, agreeing with medical experts 90% of the time.

“The OEA … project was a success, and likely could have been deployed had MD Anderson chosen to take it forward,” the spokesperson added.

And one physician at MD Anderson told the Post, “I was surprised. Even if you work all night, it would be impossible to be able to put this much information together like that.”

What Went Wrong?

OK, then. IBM said the OEA worked and had some preliminary data to back that up. And the doctors liked it, too.  So what went wrong?

Part of the answer lies in a 48-page audit the University of Texas did on the project, which cost over $62 million. (The audit report has been taken down from the UT website.)

The audit points out that OEA had to use data from incompatible systems. It’s comparable to sticking a pie and a 20-pound turkey in your oven, and expecting them both to come out perfect at the same time.

Another part rests in one that’s as old as human intelligence itself: wanting to believe in miracle cures.

On top of that, the success of the early tests might have been exaggerated, mainly because the system was used to treat limited forms of cancer that may have been easier to diagnose and treat than other more virulent forms. For instance, the early version of Oncology Expert Advisor targeted just one group of cancer patients — those with a lower-risk form of leukemia, called myelodysplastic syndrome.

Medical Miracles

The media – witness the raves in the Post story – and even the medical community focused on those early high-flying successes as signs of medical miracles to come, building everyone’s hopes that Watson, and AI, could indeed “cure” cancer. After that, the sky would be the limit, right?

But high-flying successes can fall hard and fast. And they take down a lot of good ideas with them. In this case, the failure of Watson planted doubt about AI’s applications for medical treatment.

However, other cancer-treatment centers, such as Sloan-Kettering in New York, are still using forms of AI.

The lessons:

  • AI does and will have its applications in medicine, and elsewhere. But like any similar approaches, there will be some failures – many of them caused by the humans who use it.
  • We should be careful about crowning AI as a miracle cure for anything. It’s been shown to work in fields as diverse as psychotherapy and litigation. But touting it as the perfect answer, and particularly one that will overcome faulty data, only leads to skepticism about its applications.

Selecting The Right BI Vendor:
The Ultimate Guide

Choosing a BI vendor is all about finding the right fit. Our exclusive report will walk you through the process and help you select the perfect solution.Download Now