UCF Scientist Programming Human Stereotypes, Biases into Computers

For better or worse, people tend to employ biases and stereotypes to make quick decisions about how to interact with others.  A University of Central Florida researcher is working to enable computers to better predict and simulate such behavior.

Gita Sukthankar, an assistant professor of Computer Science, has received a prestigious Faculty Early Development CAREER award from the National Science Foundation to study ways to inject realistic group dynamics into online games and simulations. The NSF Career award, which is awarded to some of the nation’s top young faculty members, totals $435,000 for five years.

“We are trying to push the envelope in the gaming worlds and make characters more human-like,” Sukthankar said.

The work will expand the state of the art in computerized human activity recognition and advance psychological theory on group dynamics. While the applications will initially be welcomed by the computer gaming community, they will also lead to more life-like simulation programs for medical training and the military.

For Sukthankar, who received a bachelor’s degree in psychology before earning her master’s degree and Ph.D. in robotics, the melding of those disciplines is a natural.

“We want computerized characters to behave in the most realistic way possible,” she said.  While that capability will make computer simulation and even gaming programs much more lifelike, getting there will require years of research in complicated statistical-machine learning.

Sukthankar and graduate students in UCF’s Intelligent Agents lab will build on earlier research into group dynamics by tracking how social interactions between humans impact group performance. They will then translate that data to computational models.

A potential jaywalker serves as an example of the challenge. Behavioral studies have shown that an individual’s decision whether or not to jaywalk is largely influenced by the actions of the people nearby. Sukthankar’s group wants to transfer elements of that decision-making process to computers so that they can accurately predict a person’s actions based on who else is present. For instance, a police officer or authority figure might deter a person from jaywalking. But if a group of well-dressed businesspeople begins walking, it’s more likely others will follow.

Using their statistical models, Sukthankar and her students will impose such human characteristics as bias and stereotyping into simulated characters.