With machine learning and AI in healthcare, can you speak the language?
As artificial intelligence and machine learning start to make their mark on healthcare in a big way, there's no shortage of hype. But there’s also no small amount of uncertainty about just what it all means – literally.
"We haven't settled on how to talk about this yet, and it's creating confusion in the market," said Leonard D'Avolio, assistant professor in the Brigham and Women’s Division of General Internal Medicine and Primary Care (part of Harvard Medical School), and CEO of machine learning company Cyft.
"If I describe what I do as cognitive computing, but a competitor describes what they do as AI or machine learning or data mining, it's hard to even understand what problems we are trying to solve."
That's not ideal. Because the problems that can be solved in healthcare with AI are numerous and notable, said Zeeshan Syed, director of the clinical inference and algorithms program at Stanford Health Care – whether it's better decision support at the bedside, better business intelligence for the C-suite or big-picture challenges such as managing care "across complex networks of providers for complex populations and complex diseases."
There are important distinctions between AI and machine learning, said Syed, and further differentiations with regard to how machine learning can be put to work.
"AI is basically getting computers to behave in a smart manner," he said. "You can do that either through curated knowledge, or through machine learning."
Curated knowledge means you could hardwire in data that, for instance, "if a temperature reading is above 102, someone has a fever. That's getting the computer to behave in an intelligent manner, but it's using existing knowledge embedded in the system," he said.
With machine learning, on the other hand, a computer "derives knowledge from the data, going beyond just the basic notion of AI, to uncover new insights."
In healthcare, machine learning has a big role to play, "coming up with new biomarkers and new understanding of how different parameters correlate with disease and disease onset," said Syed. "Predicting diseases. Preventing diseases. Treating diseases. Along that entire spectrum there is a role for machine learning."
That said, there are specific types of machine learning, and each has different approaches and applications.
"At a high level, there's supervised machine learning, where you have data about a problem, and information about certain outcomes," Syed explained. "You're essentially trying to predict or classify or diagnose the outcome from the data you have access to: You're trying to create knowledge or learn knowledge that's of the form of the relationship between the data you have and a known outcome. That's why it's called supervised: You're learning with the knowledge of what the outcome is."
Then there's unsupervised learning, "where you basically just have a bunch of data, and the goal is to find interesting structure in that data," he said. "It's not necessarily related to any particular outcome, but it's just what are the interesting characteristics of it. What are the anomalous records in a set of data you have."
Semi-supervised learning, meanwhile, "is sort of in the middle," said Syed. "You're trying to learn an outcome and understand what the relationship is between different parameters and data on that outcome, but in addition to having small amounts of labeled data, where, for example, for certain patients you might know what their outcomes are, you have a large volume of unlabeled data, which still helps you improve the process of learning from the small amount of labeled data you have."
Finally, "there's an area called reinforcement learning, which typically focuses on being able to sequentially interact and learn from things, and then factor that in to iteratively improve your decision-making over time," he said.
While those are the four major types of ML, "there are lots of other different areas that are coming around and growing," he added. "There's a real opportunity to derive new knowledge and insights."
All of these distinctions are important, of course, but not quite as critical as a larger point, said Harvard's D'Avolio.
"I'm less concerned about about the subtle differences between machine learning and AI," he said. "For healthcare, it's more important to start with: What problems are we solving?"
There's a temptation for many to think of AI as superhuman – even imbue it with humanizing characteristics: "We like to give it names – literally give it names – like Alexa, Siri or Watson," said D'Avolio.
But AI is not superhuman, and it's not going to replace physicians. It's a set of tools to help them do their jobs. And irrespective of the terminology, "if you attempt to implement these tools without best practices and milestones and ROI, it doesn't matter what you call it," he said. "You'll most likely end in a wrestling match with the vendor about how to get out of the contract."