AI (Artificial Intelligence) Has Its Challenges

As modern technology advances, we often hear about the promising outlook for AI (artificial intelligence). AI is said to have the potential for finding solutions to highly complex problems that humans could never solve on their own. As more and more data is generated in today’s increasingly digital world, AI can be viewed as the source of valuable answers in situations where the data is too voluminous for humans to go through looking for findings.

But, while AI does have promise, it is not a magic bullet. It does have its challenges.

A recent example is discussed in the February 22, 2021 Wall Street Journal article “IBM Retreat Highlights Hurdles for Health AI” by Daniela Hernandez and Asa Fitch. The article points out that IBM’s AI system had successfully played Jeopardy and beat humans at that game.   The article goes on to report that the Jeopardy success was supposed to signal the beginning of an era when “machines served up answers to questions big and small.” IBM planned to do that with Watson Health, which is the IBM unit that applies AI to healthcare. However, according to the article, Watson Health “didn’t live up to the hype,” it “struggled for market share,” and it “currently isn’t profitable.”

IBM is not the only company facing these AI challenges. According to the article, Google Deep Mind, an AI unit that “launched several healthcare-related initiatives” has also struggled with lack of profitability and with privacy issues. As the article points out, “the stumbles” by both IBM and Google “highlight the challenges of attempting to apply AI to treating complex medical conditions.”

In my view, as someone who held a position early in my career as that era’s version of what today is called a data scientist, health AI is challenging because it is far more complex than using AI to win games. Winning at Jeopardy primarily requires quickly searching through a voluminous, but finite, set of facts.  The Google Deep Mind AI won playing the game of Go. However, Go is a game and games generally have a finite set of rules, although Google’s AI is said to have identified a way to win at Go that humans never thought of.

In contrast to playing and winning a game, healthcare AI is far more complex. Successfully answering the world’s health care questions requires knowledge of all of the biology and chemistry of the human body, including so much that is still unknown today, plus understanding and navigating the unique industry characteristics of healthcare.

Furthermore, health AI has to deal with the challenges that typically affect all endeavors with AI, regardless of industry. These include dealing with AI biases and integrating human input with AI algorithms.

All of this makes it quite challenging for health AI machines to “serve up answers to questions big and small.” It is much easier to apply AI to situations like games, where although there are vast quantities of data, the problem being addressed is far more finite. Granted, AI may now work well in more finite subsets of healthcare. However, as IBM and Google are finding, there are major challenges with using AI to answer all the questions involved in something as complex as diagnosing and treating all human disease.

This entry was posted in Uncategorized. Bookmark the permalink.

Leave a Reply

Your email address will not be published. Required fields are marked *