Source-linked AI summary
Neither hype nor gloom do DNNs justice
Felix A. Wichmann, Simon Kornblith, Robert Geirhos
TL;DR
The commentary argues that neither exaggerated enthusiasm nor pessimism adequately assesses DNNs as models of vision science. It calls for models that provide both prediction and explanation, while recognizing that current limitations may change as DNNs evolve.
Problem
DNNs are assessed amid exaggerated claims, pessimistic critiques, and a tendency to prioritize either prediction or explanation rather than both.
Method
The commentary evaluates DNNs in vision science by contrasting current limitations with recent progress and by considering prediction, explanation, and image-computability as model desiderata.
Results
The commentary concludes that behavioural differences between DNNs and humans are snapshots in time, with several former limitations already reduced by newer models.
Takeaways & Limitations
Progress requires moving beyond both hype and gloom while carefully exploring similarities and differences between human vision and rapidly evolving DNNs.
Takeaways & Limitations
Current DNNs fail on many psychological tasks, and it remains unknown whether predominantly discriminative DNNs will suffice as computational models of human vision.
Abstract
from arXiv · showhide
Neither the hype exemplified in some exaggerated claims about deep neural networks (DNNs), nor the gloom expressed by Bowers et al. do DNNs as models in vision science justice: DNNs rapidly evolve, and today's limitations are often tomorrow's successes. In addition, providing explanations as well as prediction and image-computability are model desiderata; one should not be favoured at the expense of the other.