Source: NYTimes, May 2013
… the more recent work he was doing at I.B.M., called WatsonPaths, was the direction he thought artificial intelligence research needed to go to make further advances.
Much of artificial intelligence today, he said, focuses on mining vast amounts of data to make predictions. Those predictions are based on statistical probabilities and patterns — a certain symptom is highly correlated with a certain disease, for example.
“But in a purely data-driven approach, I can explain my decisions,” Dr. Ferrucci said. “People are so enamored with the data-driven approach that they believe correlation is sufficient.”
The Big Data formula, he noted, has proved to be “incredibly powerful” for tasks like natural-language processing — a central technology behind Google search, for instance.
WatsonPaths, by contrast, builds step-by-step graphs, or paths, that trace possible causes rather than mere statistical correlations. In the case of medicine at the Cleveland Clinic project, for example, the paths go from an observation of symptoms to a conclusion about the diagnosis of a disease and treatment.
That approach is a hybrid of the Big Data tools, which sift through troves of medical literature, and logic tools to identify likely chains of inference — what humans see as logical explanations for the “why” of things. The approach is also a step in the direction of classic artificial intelligence, which relied on knowledge rules and relationships, to create so-called expert systems. The blend combines elements of what Dr. Ferrucci termed “my 30-year journey in A.I.”
(see a sample medical WatsonPaths picture HERE)