Monday, May 19, 2025, 2:30 – 4:00 pm EDT, Banyan/Citrus

Organizers: Akihito Maruya, State University of New York; Amy Bucklaew, University of Rochester; and Brady Roberts, University of Chicago (VSS Student-Postdoc Advisory Committee); Shin’ya Nishida (VSS Board of Directors)
Moderator: Akihito Maruya, State University of New York
Speakers: Frank Tong, American University; Michael F. Bonner, Johns Hopkins University; Kohitij Kar, York University

In recent years, AI has made remarkable progress, becoming increasingly accessible and implementable even for individuals without extensive expertise in computer science. Visual AI, a subset of artificial intelligence, empowers machines to interpret and understand the visual world. Recent advances have demonstrated AI’s value in modeling the visual cortex, predicting neural responses, simulating eye-tracking behavior, and analyzing psychophysical data. As AI technology becomes widely adopted, it is critical to understand the principles of its successes as well as its challenges.

This review begins by exploring how AI has empowered visual neuroscientists to unravel aspects of visual processing that were previously beyond reach. We will also examine how the similarity between AI and human vision can be quantified. While AI models can mimic human visual processing to some extent, they often produce percepts that deviate significantly from human perception, such as susceptibility to hallucinations or inversion effects. Understanding these differences raises an intriguing question: how can visual scientists help guide AI to align more closely with human visual perception? We will delve into the key differences between AI and human vision, uncover the reasons for these disparities—such as biases in training data and fundamental computational differences—and explore strategies to make AI systems emulate human visual processing more effectively.

Finally, while AI is a rapidly evolving technology with the potential to revolutionize research and innovation, it also brings substantial ethical challenges. For instance, when tools like ChatGPT generate code, the output is often built upon the contributions of others, yet those contributions may not be adequately recognized. This underscores the importance of addressing issues like training data bias, privacy concerns, and the steep learning curve required to grasp foundational AI principles. In this review, we will highlight these challenges and provide insights into fostering a deeper understanding of AI ethics, emphasizing the responsibility of integrating AI into scientific workflows thoughtfully and equitably.

Frank Tong

Vanderbilt University

Frank Tong is a Centennial Professor of Psychology and Professor of Ophthalmology and Visual Sciences at Vanderbilt University. He completed his PhD studies at Harvard University (1995-1999) working with Ken Nakayama and Nancy Kanwisher. His early research capitalized on functional MRI to investigate the neural bases of face processing and visual awareness, followed by the development of novel techniques to decode feature-selective responses from the human visual cortex to characterize their role in attentional selection and visual working memory. In recent years, he has been captivated by noteworthy similarities and striking divergences between the human visual system and current deep neural network models. His research has been recognized by YIA awards from the Vision Sciences Society, Cognitive Neuroscience Society, and the Troland Award from the National Academy of Sciences. Frank has previously served on the VSS Board of Directors and currently serves on the NIH Neuroscience of Basic Visual Processes study section panel.

Michael F. Bonner

Johns Hopkins University

Mick Bonner is an Assistant Professor of Cognitive Science at Johns Hopkins University, where he leads the Cognitive Neuroscience & Deep Learning Group. His work uses computational methods, including deep neural networks and advanced statistical techniques, in combination with neuroimaging and behavioral studies to understand the visual system of the human brain. The goal of this work is to identify the statistical principles that govern the representations of visual cortex and to build theoretically grounded models of how these representations are computed from sensory inputs. Before joining the Cognitive Science Department at Johns Hopkins, Mick completed a PhD in Neuroscience and a postdoctoral fellowship at the University of Pennsylvania.