IBM (NYSE:IBM) Watson Health, and other machine-learning company’s immediate impact on healthcare may have been significantly overhyped, promoting a timeline that it may not be able to reach, according to a new article from MIT’s Technology Review.
The fault lies not with the company, however, but the complexity and relative unavailability of data that it and others machine-learning systems needs to “learn,” according to the article.
Machine learning processes require constant adjustments to internal processing routines to produce the highest possible percentage of correct answers to a set of problems, according to the article. This process works well with cases where it’s easy to identify problems or solutions, such as malignancies in X-rays, but becomes exponentially harder with cases in which the relationship is unknown.
“But for potentially groundbreaking puzzles that go well beyond what humans already do, like detecting the relationships between gene variations and disease, Watson has a chicken-and-egg problem: how does it train on data that no experts have already sifted through and properly organized?” article author David Freedman wrote.
When teaching a self-driving car and it’s machine-learning systems, anyone can label trees or signs to be recognized, but with medical data, it takes specialized training and deep knowledge to make such decisions, the article argues.
Another issue with machine-learning, AI-driven solutions in healthcare is a lack of data on patients outside health records, including eating habits, drug use, air quality and other factors, the author argues. Part of the reason that data is hard to find is because hospitals have been slow to take up modern data-driven practices, according to the article.
IBM Watson Health actually has a significant leg up on the competition, according to the author, as it has the ability to acquire data that other machine-learning plays can’t. The company has acquired Truven Health Analytics, Exployrs and Phytel, all of whom are dealing with large data sets. The company also has partnerships with other companies which provide it with more data, and therefore the ability to train its system more efficiently.
While machine-learning has been overhyped, the author argues, there’s still a chance that Watson and other companies like it will make a significant difference in health outcomes, just not as quickly as they had promised.