Cody Kommers
Social & Computational Cognitive Science
My research uses methods from AI and computational modeling to understand how people think about their social world. I'm primarily interested in how recent advances in AI can help us model aspects of human behavior that had previously been difficult to model. These include questions about the perception of meaning and how we make inferences about others in complex social situations. I'm also interested in how an understanding of the mind can be applied to improve the deployment of technology for societal good.
Models of Meaning
How do people come to perceive something as meaningful?
Meaning has always been difficult to model. Models do well with numerical values, but people don't simply judge meaning on a scale from one to ten. They develop theories of what something means. Whether it's a story, an event, or a social action, this requires a process of interpretation. With new developments in large language models, we now have an entirely new means of modeling this process.
-
Work currently underway with Wil Cunningham (Toronto) and Joel Leibo (DeepMind)
-
Kommers, et al. (2015). Hierarchical reasoning with distributed vector representations.
In Proceedings of CogSci Society. PDF.
Models of Social Cognition
Applied Cognitive Science
How do we make inferences in complex social situations?
Previous approaches to understanding social cognition via multi-agent model imposed strict limits on the kinds of interactions that could be studied. With LLMs, we can look at a much broader set of complex social behavior. For example, how do people make inferences about another agent's world knowledge? We can now begin to model the ways we make sense not only of other people's situational belief states, but their underlying worldview as well.
-
Kommers, Apps, & Bird (In Prep). A novel assessment of effort and accuracy in mental state inference.
-
Kommers, Bird, & Apps. (In Prep). Prosocial motivation in an intergroup context: the role of social identity in willingness to help or harm.
-
Gershman, Zhou, & Kommers (2017). Imaginative Reinforcement Learning: Computational principles and neural mechanisms. Journal of Cognitive Neuroscience.
How can cognitive biases be used to deter adversarial actors?
As technology advances, so do the capabilities of adversarial actors such as cyber attackers. Usually these kind of adversarial attacks are deterred through technological defenses (e.g., fire walls). However, while such technology change quickly, the cognitive processes underlying adversarial behavior does not. My current applied work looks at how an understanding of cognitive biases and other psychological vulnerabilities can be used to defend against cyber attackers and other malicious digital behavior.