As artificial intelligence technology rapidly advances, a group of Princeton researchers is asking what human minds can teach us about developing smarter AI and what AI can in turn teach us about how human minds work.
Princeton University has launched Natural and Artificial Minds, a research initiative with goals of answering questions about how both humans and machines think and learn. The initiative aims to support interaction between researchers in the cognitive sciences and those working on AI to accelerate discoveries on both fronts.
“Natural and Artificial Minds re-envisions theory-driven cognitive research in the age of artificial intelligence,” said Tania Lombrozo, professor of psychology and associate of the Department of Philosophy and the University Center for Human Values and co-director of the research initiative.
Answers to fundamental questions about natural minds and artificial minds won’t come from psychology and cognitive science alone, nor from artificial intelligence and engineering alone, said Lombrozo. “Bringing together the people, methods and ideas across fields will allow us to spark innovation concerning these broad, interdisciplinary questions,” she said.
Sarah-Jane Leslie, Class of 1943 professor of philosophy and the Center for Statistics and Machine Learning, co-directs the initiative with Lombrozo. Leslie is leading a project that aims to develop and test AI models of human cognitive function for payoffs in two directions. “One is providing clues to how we could improve AI systems going forward,” she said. The other is to understand how natural minds function with such efficiency. “This is really only possible through the collaboration of experts in cognitive science and experts in artificial intelligence.”
Sarah-Jane Leslie, co-director of Natural & Artificial Minds, speaks at the initiative’s lightning talks on Sept. 27. Photo credit: Sameer Khan
Natural and Artificial Minds is one of three research initiatives that Princeton recently established to comprise the Princeton Laboratory for Artificial Intelligence. The other two are AI for Accelerating Invention and Princeton Language and Intelligence.
Re-envisioning cognitive research
The two-day launch for the latest AI Lab research initiative kicked off on Sept. 26 with a keynote lecture from pioneering neural network researcher and Stanford University psychology professor James McClelland. On, Sept. 27, eight Princeton faculty members presented flash talks demonstrating how their research pushes forward the interaction of cognitive science and artificial intelligence. The talks came from an interdisciplinary array of researchers, including those in engineering, psychology, neuroscience, computer science and philosophy. Among them were:
Tom Griffiths, Henry R. Luce professor of Information Technology, Consciousness, and Culture of Psychology and Computer Science and the director of the AI Lab, is using an area of statistics called Bayesian models to explore learning in human cognition and AI. Griffiths said this approach can help explain human inductive bias – the way humans process and draw general conclusions with only small amounts of data.
Nathaniel Daw, Huo Professor in Computational and Theoretical Neuroscience, is exploring whether modern artificial intelligence can help answer the question of how the human brain solves difficult tasks. These tasks include decision making, particularly choices in situations of uncertainty – think games or gambling. Daw is also interested in the process of deliberation. AI, he said, offers flexible models that allow for data-driven understanding of how natural minds approach decisions.
Tania Lombrozo is studying the phenomenon of “learning by thinking,” a type of learning in which humans gain new insights in the absence of novel observations. Specifically, Lombrozo has been studying learning from explaining, in which people gain new knowledge by explaining something either to themselves or to another person. It’s not just humans who engage in learning by thinking, she said. Evidence shows AI systems can learn this way too. Lombrozo said that understanding how machines learn as they work could offer insights into both natural and artificial minds.
Naomi Ehrich Leonard Edwin S. Wilsey Professor of Mechanical and Aerospace Engineering director of the Council on Science and Technology is interested in decision making in natural minds, including animals that behave with social cohesion. Decision making in the real world must be fast and flexible – fast both in divergence from one decision and in convergence to a differing decision. Leonard is working on modeling this decision making as a nonlinear dynamical process with mathematical models that could be applied in robotics and other areas.
Tom Griffiths, Director, Princeton Lab for Artificial Intelligence, speaks at the initiative’s lightning talks on Sept. 27. Photo credit: Sameer Khan
“I was really pleased by the excellent attendance from departments across campus and to see the level of engagement among students,” said Lombrozo of the launch.
“Historically, the fields of artificial intelligence and cognitive science have been intertwined, and each has benefitted from insights from the other,” said Leslie. “We look forward to fostering interdisciplinary connections at Princeton that will turbo charge the cross-pollination of ideas in the age of AI.”