JohnMark Taylor

I'm a visual neuroscientist in the Kriegeskorte Lab currently working as an AI Safety Fellow at Anthropic. In my work, I use neuroimaging and deep learning methods to study how intelligent visual systems, whether biological or artificial, can flexibly encode and combine different features of their visual experience to subserve adaptive behavior. Additionally, I maintain TorchLens, an open-source Python package for inspecting, visualizing and intervening on any operation in any neural network; version 3.0 extends it beyond PyTorch and makes it usable by AI agents. I am also using it to build the Model Menagerie, an atlas of every neural network architecture ever described, from McCulloch and Pitts in 1943 to today's frontier models.

I did my PhD in the Harvard Vision Lab, working with Yaoda Xu. Before that, I studied face processing as a lab manager for Gregory McCarthy, and self-control in capuchin monkeys as an undergrad with Laurie Santos.

I'm also co-editing, with Anne Sereno, The Binding Problem in the 21st Century: Perspectives across Mind, Brain, and Machine (Cambridge University Press, 2027), with an amazing lineup of contributing authors.

Contact: johnmarkedwardtaylor@gmail.com