Symposium on Explanation
in Neuroscience and Artificial Intelligence
10/02/2021 9am - 5pm EST
Introductory Lecture by Jessica Thompson, Mila
Panel session moderated by Max Puelma Touzel, Mila
Register here: https://www.crowdcast.io/e/senai2021/register
Schedule
9:00 Welcome
9:05-9:45 Jessica Thompson - Introduction to scientific explanation of cognitive capacities [slides]
9:45-10:30 Ida Momennejad - Navigation Turing Test: Moving Beyond Benchmark Chasing & Toward Human-like AI
10:30-11:15 Andrew Saxe - Caricaturing the essential: Minimal and toy models as a route for understanding deep networks and the brain
11:15-12:00 Cameron Buckner - Adversarial examples and the deeper riddle of induction
Break
1:30-2:15 Lotem Elber-Dorozko - Can identifying correlations teach us what the brain computes?
2:15-3:00 John Krakauer - The Cognitive-motor interface
3:00-3:45 Rosa Cao - Making sense of mechanism: How neural network models can explain brain function
3:45-5:00 Panel Discussion moderated by Max Puelma Touzel
Motivation
Cognitive science, broadly defined as the study of cognitive capacities, including methodologies from psychology, neuroscience and computer science, is currently grappling with several philosophical questions concerning the primacy and integration of evidence from these various fields. Can cognitive behaviour be reduced to neural firing patterns? Are deep learning models of the brain explanatory? What does it mean for something to be ‘represented’ in a population of neurons? How do we know the right level (or levels) at which a phenomenon should be explained?
This event brings together practitioners in transdisciplinary neuroscience and philosophy to discuss the convergence of methodology, technology, and goals at the intersection of neuroscience, AI, and psychology. Our goal is to have the spectrum of positions clarified and identify and articulate the points of disagreement and the areas for collaboration. We encourage participants to think broadly about what set of scientific approaches should be explored and how best to integrate them.
Themes:
Reductionism and emergence
Representation and computation
Explanation and understanding
Suggested Reading
Ida Momennejad - Learning Structures: Predictive Representations, Replay, and Generalization
Andrew Saxe - If deep learning is the answer, what is the question? (arxiv version)
John Krakauer - Neuroscience needs behaviour: Correcting a reductionist's bias
Lotem Elber-Dorozko - Integrating computation into the mechanistic hierarchy in the cognitive and neural sciences,
Lotem Elber-Dorozko - Manipulation is key: on why non-mechanistic explanations in the cognitive sciences also describe relations of manipulation and control
Cameron Buckner - Deep Learning: A Philosophical Introduction
Cameron Buckner - Understanding adversarial examples requires a theory of artifacts for deep learning
Rosa Cao - Making sense of mechanism: How neural network models can explain brain function
Organization
This event is organized by Jessica Thompson, Max Puelma Touzel, and Zeke Williams
This event is supported by UNIQUE