SA grant supports scientist's work toward helping computers grasp the human mind

For centuries philosophers and others have grappled with causal relationships, usually from the position of a determinist: If A causes B, and B causes C, then if A happens, C will happen.

Steven Sloman, associate professor of cognitive and linguistic sciences (left), sees a new school of philosophers emerging in the 21^st century — from, of all places, the ranks of computer scientists and statisticians. "They look at causal relationships differently," Sloman says. "They think about them ‘probabilistically.’

"Cause may make something more probable, but it doesn’t guarantee it," argues Sloman, who recently received a NASA grant to pursue his interest in causal reasoning. "If I offer you a piece of cheesecake so that you will do something for me, there’s no guarantee it will happen," he says. "Perhaps offering the cheesecake will make it more probable, but you might refuse the cheesecake. Or you might take the cheesecake and still not do what I want. There’s an element of uncertainty."

That element of uncertainty is one of the toughest problems computer scientists and statisticians face. In recent years they have been developing ways of dealing with it in causal inferences by turning to causal models built out of probabilistic ideas, Sloman notes.

Sloman approaches human behavior and learning in much the same way. In his work for NASA, Sloman is testing a framework for causal modeling as a language for describing human reasoning. His research is expected to help the space agency better understand the relationship between human being and machine. It could have implications for designers of the next generation of software as well.

"It would be useful to be able to tell the machine how people think," Sloman says. "With causal reasoning, we’re not just predicting what will happen, but also what might have happened. This may change the way people develop software. Eventually we may be able to make the computer more compatible with the way we think."

Human-computer interaction is, in fact, a growing field, with many institutions devoting entire departments to it and several national journals focusing exclusively on the subject. Sloman looks forward to NASA-sponsored conferences where he can share his results with others and possibly initiate collaborations. Although his grant runs year to year, he expects it to last three years and to total more than $400,000. The project falls under the auspices of NASA’s Intelligent Systems Program for Human-Centered Computing.

A postdoctoral research associate, David Lagnado, is working full time on the project, studying how people learn causal relationships. Undergraduate Daniel Mochon also participates in the NASA project.

"Having an adequate model of how people reason in complex domains is critical for the development of effective human-machine systems," says Sloman. "With optimal information, we can promote a more effective dialogue between people and the machines they are using."