OR WAIT null SECS
Everyone in health IT seems to have a slightly different take on what the terms mean. Here's what Ayasdi CEO Gurjeet Singh thinks.
Everyone in health IT seems to have a slightly different take on what terms like “artificial intelligence,” “machine learning,” and even “data analytics” actually mean. No doubt, they’re all thrown around quite loosely (particularly when PR and marketing are involved).
Take Gurjeet Singh. He’s CEO of Ayasdi, an enterprise-scale AI company. Or analytics house, or machine learning firm, depending on your perspective.
“The problem with those 3 terms is that they aren’t very precise. Machine learning is the most precise of them,” he told Healthcare Analytics News in an interview at last month’s HIMSS meeting in Las Vegas, Nevada. His definition of machine learning—“a set of statistical algorithms in computer science whose primary function is to learn from data”—seems in line with a recent argument made by 2 informaticists in JAMA.
But that pair put everything from the Framingham Risk Score to deep learning systems on the same machine learning spectrum. Singh frames data analytics tools (which many consider risk scores to be) and AI (which others argue deep learning falls into) as distinct from the broader concept of machine learning—even if the reasons are nuanced.
“AI has been an old idea in the study of computer science,” he said. “The fundamental aim of AI is to mimic what people do using software. That’s in the broadest terms applicable. We’re not there yet. And a lot of what we have done in machine learning supports that aim. I can see how AI and machine learning are used interchangeably, because the difference is subtle.”
Singh thinks data analytics have the least stringent minimum threshold. “If you’re looking at a chart of your patient and it’s pulled from a database, that’s data analytics,” he said. “If you’re looking to figure out the top facility in your hospital system and you have a dashboard, that also qualifies as data analytics, right?”
But it’s very different from algorithms that, when shown examples, learn from data. That’s the way I would think about it.”