This venture capitalist sees plenty of promise in AI in healthcare. But challenges are ahead, too

“Garbage in, garbage out.” So goes the adage about computing that argues the results rendered by the “thinking” machines will be flawed if the information fed into the devices proves to be erroneous.

However, generative AI—the idea that a computer can self-correct—challenges the cliché, Dennis L. McWilliams, a partner at Santé Ventures, told Fierce Healthcare. Santé Ventures, an early-stage venture capital firm, specializes in healthtech, medtech and biotech investments. 

“One of the interesting things when you look at large language models like ChatGPT, they seem to have the ability to overcome that,” McWilliams said. “Not to get too technical, once they work through a certain threshold of training, the algorithms learn to sort out the garbage.”

For example, that’s not how it works with current computer algorithms that might evaluate X-rays, McWilliams said. “If your data set is messy, your algorithms are going to be messy. And if your data set has inherent biases, your algorithm is going to have biases," he said. "In generative AI—at least a preliminary view of these really large data sets—that, to some extent, gets easier to manage.”

McWilliams said that it’s still too early to determine just how AI will fit into the healthcare system. While the capabilities seem endless, McWilliams said that who will pay for the technology remains an open question.

For now, venture capital firms like Santé Ventures will back the growth of generative AI in this market. But they don't expect to lead that charge forever, McWilliams said.

“But at the end of the day, these companies are looking to build revenue themselves,” McWilliams said. “So, they’ve got to turn those investor dollars into revenue dollars and growth dollars. If you’re simply developing a new algorithm that you’re going to drop into the hospital system, it’s really difficult to convince the hospital to come out of their margin pocket to pay for that.”

Dennis McWilliams
Dennis McWilliams (Santé Ventures)

For patients, McWilliams sees AI as a next-generation WebMD, although that’s not on the market just yet. “People can have a much more immersive and complete evaluation done, as opposed to just going on the internet and typing into Google your symptoms," he said.

Providers will also be weighed by AI, he said. “As we all know, there’s a ton of variability in quality. I mean, if you go to an infectious disease expert in downtown Manhattan, that’s a lot different than going to one in the middle of rural New Mexico, right? For any physician who wants to access knowledge, AI is going to fundamentally change everything.”

And insurance companies? They must weigh the qualities intrinsic to healthcare, such as privacy issues.

“Insurers have fairly consistently denied paying anything meaningful to have artificial intelligence as part of a treatment algorithm,” McWilliams said. “I think that’s what a lot of the industry is trying to figure out right now: What to pay for. We’ve spent a ton of time really thinking about that and working through some of the nuances from a business perspective.”

On the other hand, he can see AI being a great help in claims processing. In addition, insurers may find AI useful in tracking fraud.

“They scan millions and millions of claims submissions,” said McWilliams. “They’re using AI to identify things that are out of whack.”

While the future is likely bright, there is still plenty of hesitation in healthcare about generative AI. For example, the Centers for Disease Control and Prevention and other experts have expressed concern that AI may not recognize racial and ethnic disparities in healthcare.

“I think that’s a definite concern,” McWilliams said. “You see that in some of the early AI training set. It’s a training set that includes data from all white males in their 50s. That algorithm is not going to be particularly useful for other demographics.”

McWilliams said emerging companies are working to address this issue.

“I think startup companies that we’re looking at make sure that their data sets are representative of the broader populations and the subpopulations that they’re trying to go after," he said.