“These are life-or-death scenarios”: Meet the scientist bringing AI to Toronto hospitals

“These are life-or-death scenarios”: Meet the scientist bringing AI to Toronto hospitals

Bo Wang, the new chief AI scientist at UHN—Canada’s largest hospital network—on embracing tech, cutting down wait times and whether robots are coming for health care workers’ jobs

Bo Wang, who just became the Chief AI Scientist at UHN
Photo courtesy of University Health Network

In the near future, patients at Toronto’s hospitals will be greeted by the stiff metallic smile of a scalpel-wielding robot. Okay, just kidding—but it’s safe to say that the field of medicine will soon become a little more Jetsonesque. Case in point: the University Health Network (UHN), Canada’s largest network of research hospitals, just appointed its very first chief AI scientist, Bo Wang, who has been tasked with overseeing the implementation of AI tools in hospitals across the city.

Both Wang and UHN have some experience in this field. Earlier this year, UHN launched its AI Hub, co-led by Wang, which aims to bring scientists and clinicians together to develop new AI technology. Wang’s latest gig makes him the very first chief AI scientist in the country. Here, he breaks down what the job entails, why hospitals have been slow to adopt machine learning and why he doesn’t expect physicians to be as resistant to the tech as, say, Hollywood screenwriters.

You have a PhD from Stanford. Is it safe to assume that you’re a PhD-style doctor rather than an MD?
Ha—correct, I’m not a medical doctor. My background is in computer science and computational biology.

Related: Meet the researcher investigating the future of AI in medicine

Is it unusual for someone who’s not a medical doctor to be the chief of something at a hospital?
Well, in this case the role is very new. But I do work with a lot of medical doctors, of course. I meet with them to discuss their burning questions about AI and assess how machine learning could help address challenges they’re facing. Those sessions go both ways—they tell me what they care most about, and we tell them what’s possible. Doctors are fantastic people. It’s amazing that they manage to do any research given their busy schedules.

You must have fairly packed days as well, with the shiny new role. What does the chief AI scientist at UHN do?
I have three main focuses. The first is research—developing AI models that can parse health-related data, publishing papers and working with other hospitals. The second is education. There’s a lot of hype around AI. I want to demystify tools that can assist with note-taking and administrative tasks, and I want to help physicians know what’s true and false about more long-term possibilities, such as AI that can assist with surgical procedures. The third pillar is what we call adoption. There are lots of research papers on AI out there. While there are a few AI tools being used in UHN hospitals, we’re not seeing widespread adoption in Canada. As we know, our health care system is in crisis. If AI can help improve the system, it’s important that we explore that.

Some hospitals in the US, like Mayo Clinic, are already using AI to help diagnose heart disease and strokes. Why do you think Canadian hospitals have been so hesitant?
Well, the US government has invested a lot into basic research on AI in hospitals. Their investment is the highest in the world, which is why they’re the leader in this field. In Canada, we’ve fallen behind in terms of our investments in new science. In addition to that, AI can digest huge amounts of data, but it’s bulky and complicated. There can be a lack of trust from physicians because they don’t fully understand how it works. Then there are ongoing legal issues, like ensuring that our infrastructure for storing patient data continues to abide by Canada’s privacy laws, and of course ethical issues around the possibility of biases in the data we use to train AI.

Right, we know that AI tends to take on human biases. For example, an algorithm used to predict crime in the US disproportionally labelled Black defendants as “high risk.”
We’ve seen that with algorithms used for matching donor organs with potential patients. They tend to match more men than women because the data that was used to train them was mostly collected from men. These are life-or-death scenarios, so it’s extremely important that we develop unbiased tools.

How do you troubleshoot for that sort of thing?
First, you try to collect data in an unbiased way—making sure you have sufficient representation of different kinds of patients. Toronto hospitals have an advantage with that because it’s such a diverse city. Then, we need to be testing these tools in more than one setting. At UHN, we try something out in one of our hospitals and then bring it into other hospitals or even reach out to colleagues in the US or Europe. That way, we can verify that it works well for different groups of people.

Related: Scenes from the AI revolution in university education

How do you imagine AI will change regular people’s experiences in the health care system? I’m guessing fully robot-operated surgeries are still a ways out.
Yes, but there are some real examples of how the patient experience is already changing. One is this app we co-developed called Medley. It’s a regular mobile app—you can get it at the App Store—and it helps nurses at the Peter Munk Cardiac Centre remotely monitor patients with heart conditions. A patient inputs certain measurements at home, and the app uses AI to flag any abnormalities or changes. Then a nurse can call the patient to check in. Another example is a recent product we developed at UHN, which is a language model, sort of like ChatGPT. We connect it to a microphone, and it can listen to a conversation between a doctor and a patient. When they’re done talking, the model can immediately summarize that conversation into clinical notes. That way, instead of spending the whole appointment glued to their computer, typing, doctors can actually look their patients in the eyes.

Could these kinds of changes mean that patients won’t be stuck waiting in emergency rooms for 12 hours?
That’s the goal, yes.

In both of your earlier examples, physicians are still in the equation. Is it safe to say you’re not working on an algorithm that could replace them entirely?
Exactly. We’re not making decisions for doctors; we’re helping them make more-informed decisions while spending more time with their patients. AI is the perfect tool for precision medicine. For example, there is this concept called the “digital twin.” We have all this data, which represents millions of past patients. So, when a new patient comes in, we can find their digital twin from the database—the person who is most similar to them based on, say, genomic profiling and medical history. We can learn from their twin’s experience. If they tried certain drugs that failed, maybe we should avoid those drugs. If they benefited from a specific combination of treatments, maybe we should try similar treatments for this new patient.

Okay, fast forward ten years. How do you hope to see hospitals using AI?
One thing I’m excited about is the potential for medically equipped AI training. We have a shortage of medical staff due to a lack of decent training programs. AI could be a part of the solution there. Language models could be used to simulate doctor-patient dialogue, which trainees could critique. Virtual reality could create 3-D models of tumours, which surgical residents could interact with. Also, we can develop AI to synthesize vast amounts of literature and help doctors make the most informed decisions. But I don’t think current AI models are accurate enough for that yet.

In some industries, like the movie-making biz, there has been a lot of pushback against letting AI take on formerly human responsibilities. Should health care workers be worried for their jobs?
I know that’s a fear we hear a lot about, but I haven’t seen it in health care. The president of the American Medical Association said, “It is clear to me that AI will never replace physicians—but physicians who use AI will replace those who don’t.” I think that’s very accurate, because AI can greatly enhance a doctor’s ability to manage their patient care and daily workflow.

You don’t think there will be any Luddite hold-outs?
There’s a famous 2021 survey from the Peter Munk Cardiac Centre, which found that over half of physicians are suffering from burnout. Cumbersome tasks like note-taking are some of the driving factors behind that, and those are things we know AI could handle. Anything we can do to lower that percentage would only benefit our health care system.

This interview has been edited for length and clarity.