‘Designing Minds’ Workshop Examines Future of Artificial Intelligence

Workshop brought together researchers from Northwestern, Microsoft, IBM, and Google

In 2011 IBM’s artificially intelligent computer system, Watson, soundly defeated its human competition on the television quiz show Jeopardy!. It knew its Olympic history, Beatles songs, literary characters, and rare diseases and finished a two-day winning streak several thousand dollars ahead of its opponents.

“Everybody was rocked when they watched that,” said Ken Forbus, Walter P. Murphy Professor of Electrical Engineering and Computer Science. “Someday we all might have our own personal Watson—a cognitive system that works with you.”

Ken ForbusThe past, present, and future of artificial intelligence, including Watson, was the theme of the McCormick School of Engineering’s daylong workshop “Designing Minds: New Forms of Cognitive Systems for Work, Learning, and Play.” Organized by Forbus, the workshop examined the prospects for, and impacts of, creating software systems that operate more like human minds. The event took place Wednesday, September 24 in the James L. Allen Center on the Evanston campus.

“Technology affects everything,” said McCormick Dean Julio M. Ottino, who moderated one of the panels. “What we see here today is going to change everything. There are ideas in this room that someday will look like magic to people outside of the field.”

Moderated by Northwestern linguistics professor Gregory Ward, the first panel focused on cognitive science. Dedre Genter, the Alice Gabrielle Twight Professor of Psychology in the Weinberg College of Arts and Sciences and Steve Franconeri, associate professor of psychology in Weinberg, discussed how the mind works and the difficulties of modeling its complexity with computer systems. 

John Laird, professor of computer science at the University of Michigan, presented a timeline of the evolution of artificial intelligence systems that included Soar, a cognitive architecture that he co-created. “We are in the golden age of cognitive architecture,” Laird said. “But there is still a lot of research to be done.”

Dean Ottino moderated the second panel, which focused on software assistants. Eric Horvitz from Microsoft Research and James Fan from IBM’s Watson team presented. Horvitz showed video demonstrations of artificial intelligent systems that he employs in his office building, including a direction-giving robot in the lobby, an interactive voice in the elevator, and, outside of his office, a human-like “personal assistant,” which recognizes faces and chats easily with Horvitz and visitors.

“We want to build systems that people actually need,” Horvitz said. “We want to build models that represent what the user is thinking. When you do this correctly, it’s magical.”

Fan presented ways that IBM is working to “take Watson beyond Jeopardy!” and expand the system beyond trivia. Now researchers are trying to adapt Watson for the medical domain, so healthcare providers can give it a list of symptoms and find possible diagnoses. Fan said a major challenge lies in the complexity of medical language, which requires Watson to undergo continuous learning.

“The Jeopardy! challenge revealed that we are entering a new era of computing,” Fan said. “Our goal is to produce a system that can take in rich scenarios and interact with humans.”

The second half of the workshop opened with a panel examining cognitive science in entertainment. Moderated by Barbara O’Keefe, dean of the School of Communication, the panel featured McCormick electrical engineering and computer science associate professors Bryan Pardo and Ian Horswill, who discussed how intelligent software systems are changing the way people create and interact with art, video games, and music.

Pardo showcased a series of audio tools he developed in Northwestern’s Interactive Audio Lab that function using the hallmarks of human communication. One tool, Synthassist, programs audio synthesizers using similar vocal imitations. Another program, Reverbalize, manipulates reverberation using natural language. 

“These are intelligent, interactive systems that work with users through natural language, giving examples, and evaluation — letting you know that you like this or don’t like this and by how much,” Pardo said. “And with these interactive paradigms, we are rethinking how these media production tools can be interacted with and used.” 

Horswill added that in video games, artificial intelligence is becoming an expressive medium in itself, with artists becoming AI designers by moving beyond previous design goals to focus on crafting psychologies. “We’ve gone from describing moving characters from drawings on the page to abstract geometric representations,” Horswill said. “Why not create moving characters by defining them through their psychologies? Why not have AI be the expressive medium through which we create these characters?” 

The next panel focused on the role of cognitive systems in education. Penelope Peterson, dean of the School of Education and Social Policy, moderated the on the role of cognitive systems in education. The panelists, Peter Norvig, director of research at Google, Chris Riesbeck, associate professor of electrical engineering and computer science at McCormick, and Matt Easterday, assistant professor at the School of Education and Social Policy, shared their own views of how cognitive systems have changed learning.

Discussion centered around the issue that education, historically, has been slow to embrace technology. But the panelists agreed that while conventional wisdom assumes that human tutors are more effective than software systems acting in the same role, intelligent tutoring systems have shown similar or even improved rates of effectiveness. 

According to Norvig, the promise of intelligent tutoring systems can be seen in the rise and success of massive open online courses (MOOCs). He explained that a MOOC’s unstructured approach encourages mastery learning by allowing the student to work at their own pace, which he believes yields a better result than a more rigid, traditionally taught course.  

“A coloring book is only beautiful when you color it in a unique way. It’s not a sketchpad. It’s not a blank page,” Norvig said. “There is a structure the author has put in, and a combination of the author and user together come up with the result. That’s what we want to be able to figure out — what should be guided and what should be free play.”