A pivotal event for AI happened when IBM’s Watson beat two all-time champions of Jeopardy! in 2011. This showed that the technology was far from being experimental.
IBM would soon go on to make Watson the centerpiece of its AI strategy. And a big part of this was to focus on healthcare. To this end, the company made several major acquisitions and boosted the headcount of data scientists.
But despite all this, the effort ultimately proved to be a disappointment. Keep in mind that IBM is now exploring the sale of the Watson healthcare business, according to a report in the Wall Street Journal.
The Difficulties With Healthcare And AI
When it comes to commercializing cutting-edge technology, it’s important to set forth concrete goals that are achievable and that have ROI targets. Trying to “boil the ocean” is often a recipe for failure.
In the case of IBM, it does look like it was overly ambitious, as the company was looking at making significant strides in fighting cancer and other chronic diseases.
“AI can work incredibly well when it’s applied to specific use cases,” said Nirav R. Shah, who is an MD and the Senior Scholar at Stanford University’s Clinical Excellence Research Center. “With regards to cancer, we’re talking about a constellation of thousands of diseases, even if the focus is on one type of cancer. What we call ‘breast cancer,’ for example, can be caused by many different underlying genetic mutations and shouldn’t really be lumped together under one heading. AI can work well when there is uniformity and large data sets around a simple correlation or association. By having many data points around a single question, neural networks can ‘learn.’ With cancer, we’re breaking several of these principles.”
The irony for IBM is that it likely would have been more successful by pursuing more mundane applications of AI, such as providing efficiency and better workflows for healthcare systems. After all, the company has a long history with such efforts.
The Data Challenge
Data is the fuel for AI. But in the context of healthcare, the data can be difficult to obtain—say because of privacy issues—and is often messy and complex. The “noise” can easily skew results.
But AI models for healthcare also require strong domain expertise. Advanced approaches like deep learning may not be enough.
“In general, medical applications are immensely complex and contain biological complexity and many compounding factors such as genetics, epigenetics and the environmental factors,” said Oliver Schacht, who is the CEO of OpGen. “This complexity and non-linearity which is often still only partially understood at all makes it inherently difficult to train an AI.”
The opportunity for AI in healthcare is certainly massive. In the years ahead, there will be major breakthroughs. And yes, they will impact millions of lives.
But to be successful, there must be a long-term approach and a focus on close partnerships. This will help to build trust.
“Today’s AI systems are great in beating you at chess or Jeopardy,” said Kumar Srinivas, who is the Health Plan Chief Technology Officer at NTT DATA Services. “But there are major challenges when addressing practical clinical issues that need a bit of explanation as to ‘why.’ Doctors aren’t going to defer to AI-decisions or respond clinically to a list of potential cancer cases if it’s generated from a black box.”
Tom (@ttaulli) is an advisor/board member to startups and the author of Artificial Intelligence Basics: A Non-Technical Introduction, The Robotic Process Automation Handbook: A Guide to Implementing RPA Systems and Implementing AI Systems: Transform Your Business in 6 Steps. He also has developed various online courses, such as for the COBOL and Python programming languages.
Credit: Source link