Following our series of CIMCYC Sessions, we are pleased to announce our next AI, Mind and Brain session on Thursday, 22nd May (fully online this time). As in other sessions, we will have an unguided discussion, so do come with ideas, questions and doubts for discussion (5 minutes of thinking time). Find below the full reference of the paper, the link to a podcast summary of the paper and a couple of questions to inspire discussion (via Notebook LM and ChatGPT).
AI, Mind and Brain: A large-scale examination of inductive biases shaping high-level visual representation in brains and machines (concrete example with implications for the family of DNNs)
Paper: Conwell, C., Prince, J. S., Kay, K. N., Alvarez, G. A., & Konkle, T. (2024). A large-scale examination of inductive biases shaping high-level visual representation in brains and machines. Nature communications, 15(1), 9383. https://doi.org/10.1038/s41467-024-53147-y
Podcast summary: https://drive.google.com/file/d/18qvv3PlWaxYPteKRvMpNl1fYr3q_8GYh/view?usp=sharing
Kickstarting questions
- How 'brain-like' are today's AI models really?
The paper shows that very different models (CNNs, Transformers, contrastive learning, vision-language models) can all achieve similar levels of brain predictivity. What does this suggest about the specificity—or lack thereof—of our current brain-model comparison methods? Are we truly modeling the brain, or just matching general visual similarity?
- Why does the type of training data matter more than the model architecture or objective?
The authors found that the visual diet—the images a model is trained on—had a larger impact on brain predictivity than whether the model used language, self-supervision, or even whether it was a CNN or transformer. How might this relate to human visual development? What parallels can we draw with the role of visual experience in shaping brain representations?
No prior experience in AI is needed –just curiosity and a willingness to discuss ideas. Save the date!
*Please contact elisaherguido@ugr.es prior to the meeting in case you need confirmation of your attendance.
Date: Thursday, 22nd May 2025
Location: online session (https://meet.google.com/sgz-quvr-rma)
Time: 10.00 am
Upcoming CIMCYC Sessions
-
Wednesday, 18th June (AI, Mind and Brain session): Inferring DNN-Brain Alignment using Representational Similarity Analyses can be Problematic (potential problems with RSA for this)