In part 2 of this Exclusive Interview, Lisa Latts explains the process behind the artificial intelligence Watson in a conversation with Diabetes in Control Publisher Steve Freed.
Lisa Latts MD, MSPH, MBA, FACP is Deputy Chief Health Officer for IBM Watson Health.
Transcript of this video segment:
Freed: Is Watson really an artificial intelligence? I mean, who actually does the thinking?
Latts: Yes. So, Watson is really an artificial intelligence. And this is not the robot that you see in Terminator. This is basically, Watson will think — will be able to draw conclusions. So, the thing that Watson can do is it can understand vast amounts of data, so it can read both structured data and unstructured data. So, structured data are things like we used for that claims analysis that I mentioned where it’s reading in claims and we can do things with it because we can do analytics on it. But Watson can also understand unstructured data, so that’s data that’s trapped in the electronic health record or in medical literature, or in places where — if you’re human, you can sit down and read it, but it takes so much time. Watson can read large amounts in seconds and be able to ingest it, and then draw conclusions. And so, the thing about artificial intelligence is that while it can read it and ingest it, it’s like a baby. It doesn’t know what it means, so it’ll draw conclusions but those conclusions may be wrong. So, humans have to then teach it over time just like you would teach a child, “That’s incorrect,” or, “That’s correct.” And then, as it learns more and more, it gets smarter and starts to draw conclusions that are more and more correct over time. And so, we used artificial intelligence in our partnership with Medtronic to develop Sugar.IQ which is in exclusive use with their Guardian Connect Continuous Glucose Monitor.
Freed: So, I have an interesting question and I don’t know if Watson can answer.
Freed: But if you go back to 1950, we had one medication for type 2 diabetes, sulfonylureas. And then it took us 50 years to get a second medication called metformin. So, it took us all that time to get two. As of today we have 6 million possible combinations. So, for the family practitioner to sit down with a patient and come up with the right possible combination is basically impossible.
Freed: That’s where something I would imagine like Watson can actually calculate, put their insurance in, and actually calculate what’s the most effective treatment at a reasonable cost. How far are you guys away from that?
Latts: So, we don’t have that kind of tool for diabetes yet but we do have something very similar for oncology where we have a tool that we developed in partnership with Memorial Sloan Kettering in New York, and what the oncology told us is it takes in all that patient information, as you mentioned, although we don’t have insurance in there currently. And it reads through the individual’s characteristics, looks at the individual cancer, and then comes up with recommendations for both chemotherapy, endocrine therapy, and radiation therapy based on the expert training that it received from MSK. So, we have that sort of thing for oncology. We do not have it for diabetes yet.