UKRAINE – 2021/02/19: This photo image shows an IBM logo on a smartphone screen. (Photo … [+]
SOPA Images / LightRocket via Getty Images
A defining event for AI came when IBM’s Watson beat two Jeopardy champions of all time! This showed that the technology was far from experimental.
IBM would soon make Watson the centerpiece of its AI strategy. A big part of that was the focus on healthcare. To do this, the company made several major acquisitions and increased the number of data scientists.
Despite all of this, the effort ultimately turned out to be a disappointment. Note that IBM is currently investigating the sale of the Watson healthcare business, according to a report in the Wall Street Journal.
The difficulties with healthcare and AI
In commercializing cutting edge technology, it is important to set specific goals that are achievable and have ROI goals. Trying to “cook the ocean” is often a recipe for failure.
In the case of IBM, it looks like it is too ambitious as the company has made significant strides in fighting cancer and other chronic diseases.
“AI can work incredibly well when applied to specific use cases,” said Nirav R. Shah, MD and senior scholar at Stanford University’s Clinical Excellence Research Center. “When it comes to cancer, we’re talking about a constellation of thousands of diseases, even if the focus is on one type of cancer. For example, what we call “breast cancer” can be caused by many different underlying genetic mutations and should not really be grouped under one heading. AI can work well when there is consistency and large amounts of data around a simple correlation or association. With many data points around a single question, neural networks can “learn”. With cancer, we are breaking some of these principles. “
The irony for IBM is that it would likely have been more successful had it pursued more mundane AI applications like efficiency and better workflows for health systems. After all, the company has a long history of such endeavors.
The data challenge
Data is the fuel for AI. However, in the healthcare context, data can be difficult to obtain – due to privacy issues, for example – and it is often cluttered and complex. The “noise” can slightly distort the results.
However, AI models for healthcare also require strong expertise. Advanced approaches like deep learning may not be enough.
“In general, medical applications are immensely complex, including biological complexity and many compound factors such as genetics, epigenetics and environmental factors,” said Oliver Schacht, CEO of OpGen. “This complexity and non-linearity, which is often only partially understood, makes it inherently difficult to train an AI.”
The opportunity for AI in healthcare is certainly huge. There will be major breakthroughs in the years to come. And yes, they will affect millions of lives.
However, to be successful there has to be a long-term approach and focus on close partnerships. This will help build trust.
“Today’s AI systems are great at beating you at chess or at risk,” said Kumar Srinivas, chief technology officer of the health plan at NTT DATA Services. “However, there are major challenges in dealing with practical clinical problems that need to be explained ‘why’. Doctors won’t limit themselves to AI decisions or respond clinically to a list of potential cancer cases when generated from a black box. “
Tom (@ttaulli) is a Startups Consultant / Board Member and author of Artificial Intelligence Fundamentals: A Non-Technical Introduction, The Robot Process Automation Handbook: A Guide to Implementing RPA Systems and Implementing AI Systems: Transform Your Business in 6 steps. He has also developed various online courses, for example for the programming languages COBOL and Python.