Local

The next big challenge in AI could be teaching computers how to think like humans

UC Davis professor Charan Ranganath collects brain-imaging data with two MRI machines at UC Davis’ Sacramento campus and Center for Neuroscience as part of a project to create humanlike artificial intelligence.
UC Davis professor Charan Ranganath collects brain-imaging data with two MRI machines at UC Davis’ Sacramento campus and Center for Neuroscience as part of a project to create humanlike artificial intelligence. wko@sacbee.com

Charan Ranganath has dedicated his professional life to studying how the human brain works. Now the UC Davis psychology professor is pushing that research into new frontiers by teaching computers how to slow down and mimic that same complicated but imperfect organ.

Ranganath and his team are putting to use a $7.5 million grant from the Department of Defense to develop a computational model that emulates how memory functions in the real world. The DOD is hoping to use that model to help predict and prevent terrorist activities as well as for other national security purposes. Ranganath is taking on that task by leading a team of five other scientists from Princeton, Harvard, Washington University in St. Louis and New York University.

“Human memory constantly adapts to new situations, and we often fill in the blanks of our memory with other information,” said Ranganath, 46. “I wanted to create a model just like the human brain that is intelligent but not necessarily accurate.”

Most well-known artificial intelligence models such as Apple’s Siri and Amazon’s Alexa can solve problems through statistical methods and computation formulas but are centered around problems such as reasoning, planning and predicting, researchers said. Sam Gershman, psychology professor at Harvard, said this team’s model will take a step beyond those capabilities.

“The key strength to the human brain is flexibility, and that is what gave humans an edge over machines,” Gershman said. “We want to build a model that emulates how event cognition works in the brain.”

Ranganath will start the project this summer by spending a month collecting brain imaging data with two MRI machines at UC Davis’ Sacramento campus and the school’s Center for Neuroscience. More than 20 students will show about 180 subjects visual presentations and get their real-time reactions. The team will also track the subjects’ eye movements and the electrical activity in their cerebral cortexes to analyze brain activity. Four other institutions will also collect patient data, Ranganath said.

Analyzed brain activity will then be developed into a computational model that will develop its own understanding and make predictions unlike current artificial intelligence models that respond within the boundaries of their stored data, Ranganath said. Gershman said the new model will feature imperfect memory like that of a human brain.

The DOD is hoping to use the model to analyze surveillance video footage and catch signs of terrorist activity instead of having humans constantly monitor screens, Ranganath said. The department will look for more ways to employ the model in national security projects with improvements over time.

Ranganath has been doing his part by leading dozens of UC Davis students at the school’s Dynamic Memory Lab. He often goes across the street to an MRI lab, where he and his students spend hours trying to find one slight difference that may occur in one-three-hundredths of a second.

“We may make seriously wrong assumptions along the way, but that’s good,” Ranganath said. “At the end of the day, we are going to have a result, and scientific research is all about testing our assumptions one after another.”

The study could also prompt enhancements in understanding memory disorders, including Alzheimer’s disease and dementia, and develop new research methods for further developments in the field, Ranganath said.

Sanmay Das, an artificial intelligence researcher at Washington University in St. Louis, said the UC Davis research has exciting potential for artificial intelligence.

“This is a very hard problem to solve, and understanding how humans use their brains can give us insights on how to solve many complex issues,” Das said.

Walter Ko: 916-321-1436, @juntaeko

  Comments