You probably remember Watson, the IBM supercomputer, from its stint on Jeopardy where it soundly whooped its human competition. The megamind is more than parlor tricks and trivia, however, and IBM has reached out to many industries and companies to see how this uber-smart "learning" computer that can interact and answer questions might be able to help shape the world going forward. One of those is the Cleveland Clinic, which is where Watson began as a medical student late last fall. We talked with Dr. John-Eric Jelovsek, who is co-leading the Clinic project, to see how far Watson has progressed to becoming our supreme medical ruler of the world.
Vince Grzegorek: So, from what I've read elsewhere, there's a language problem with Watson in trying to teach it medical terminology. What are some examples and how do you fix it?
Dr. Jelovsek: What we're dealing with now with Watson is being to translate subtle human nuances of human language that we use in medial lingo. We gotta make the system understand that. So, really, what's at the heart of it, at this early development with Watson, is just a pure human language issue. For example, when someone says, "I've been burning up all night," we translate that as they have a fever. But it takes a human to figure that out. Right now, the system doesn't understand that. You can see logically what that means, but you have to understand in the context of using a computer system in a healthcare environment, certain words and phrases we translate differently than if we were on a game show. We forget about that fact, that we interpret language in context of what we're doing, so when you're talk about a computer trying to understand human language, context must be painted, really. So that's what we're doing, or one of the major areas that we are doing. The language and recommendations have to be interpreted in the context of medicine. You have to teach that, it doesn't magically happen. You have to go and give it example after example where Watson thought this but it really means this and this is why.
How does that work, exactly?
We're giving the system lots of examples as we go through medical expertise, and the next time, to some extent, it leverages the adaptive learning the system has.
How long is this going to take—to make Watson a complete medical student? Is it accelerated because Watson is, ya know, a supercomputer and not someone who fell into medicine for the wrong reasons?
Four or five years, actually. You'd be surprised. It's going to take about as long as normal.
Part of what Watson's goal in all this is has to do with the sheer amount of information out there for doctors. Updates and new studies are always coming out, and it's not longer the guy with the black bag and a stethoscope. And, I'm assuming, doctors are feeling overwhelmed by the data and unprepared to tackle treatments with new advances. And Watson can process and link information far quicker and better?
Oh, absolutely. Any physician will tell you it's not feasible any longer to stay up to date with the scientific literature. It's becoming unfeasible to keep up with the most basic scientific literature. There's progress in so many areas and journals and non-healthcare journals that we never had access to before. We can't keep up with that, so any system that brings those together, or at least can rapidly scan them and bring them to light when it's most important, that's a huge asset. It's just enormous, and then you add in societal pressures of seeing more patients, lower reimbursement for the care, working long hours, and you add that all together, it becomes impossible, even if you had access to an hour of summary that was relevant to you, it's nothing.
So, is that a component of why Watson can be transformative—the ability to link research and case histories and diagnoses and symptoms and come up with an answer that a doctor may never have thought of or would have taken much longer to arrive at?
There's lots of different goals. Even if you get a system as accurate as a human—an average healthcare provider—you've essentially built a computerized physician, and that is a huge feat. If you can't do any better or make to that level, to some extent I'm not going to rely on that. But if you're as accurate, you start asking, "How can we use this to amplify the current physicians?"
So, we're not looking at House, the robot doctor?
Well, we want to do better than that, and we believe we will. Avoiding the biases and mistakes that humans make, for instance, or not being able to keep up with the latest developments in medicine. Obviously, humans are susceptible to fatigue; computers are not. So, there are a lot of benefits.
But that's why it's going to take four or five years?
I guess some of it is based on different challenges. So, it's not surprising to see what's happening. It's really kind of cool. And it's not that long, when you think about it. Just to learn the sheer vocabulary that's used in medicine, the pneumonics, and all of this is a human language – how long would it take you to learn a completely new language? We want that language to be interpreted correctly and give accurate answers, which is longer than it would take humans to learn. When someone gives you data, sometimes it can take a while to come up with the answers. We just don't have the data capacity. The computer has access to the data, and once it learns the language nuances, it'll make decisions faster.
And so it could be House?
Right now, a lot of that is unknown. We don't know what it will do.
But you have high hopes to not just be an average healthcare provider.
Once it learns the medicine, then you can see how it can use existing data to take care of patients. That's why it's exciting. It takes it through medical school. Alright, now let's see if you can help take care of a patient rather than just knowing what you're supposed to do, which is essentially what we're doing with it now.
What does this mean for the future of computers assisting in healthcare?
So, that's a wide open field right now. Once it understands things, we can't help but to envision the role we see it playing. In our discussions, we see it in any number of areas, from a pure information support system where it can provide information quicker and in a more easily digested fashion, and all done accurately. That in and of itself is a huge leap forward. That's in some sense electronic medical records machine that has information searching built into it. The other way to see it is can it be used as a support person? We think of it as an additional team member. Watson has the ability to understand human language. It can also give that information back. Now you're saying this is like having a smart team member in the room. If you could partner in that, provide information and not only seek information, if it could provide an opinion or counseling, that would be tremendous. Any time technology can do that, it also becomes an educational tool. If it can tell us the right answer fairly accurately, we can learn how it came up with that answer.
I, for one, welcome our new computer overlord.