The Cognition Project Seeking Universal Laws: Roger Shepard
Apr 1, 2026
Roger Shepard, pioneering cognitive psychologist who developed multidimensional scaling and studied mental rotation. He talks about reconstructing psychological spaces, the birth of multidimensional scaling, the mental rotation discovery, debates on similarity representations, and his search for universal laws of generalization.
AI Snips
Chapters
Transcript
Episode notes
Discovering Non-Euclidean Mental Spaces
- Roger Shepard switched from math and philosophy to psychology after being intrigued by Fred Atney's finding that perceived similarity spaces were non-Euclidean.
- He pursued geometry-based psychological measures, motivated by the L1 (city block) vs L2 (Euclidean) metric difference found in pitch and other stimuli.
How Multidimensional Scaling Reveals Mental Maps
- Shepard developed multidimensional scaling to recover psychological spaces from similarity or confusion data by arranging points so ranked dissimilarities match distances.
- He used gradient descent on computers at Bell Labs to get rigid, metric solutions even from only rank-order (nonmetric) data.
Phone Dialing Experiment Cut Errors in Half
- At Bell Labs Shepard tested dialing sequences and found reversing unfamiliar digits halved errors and time, but system engineers rejected the change for customer disruption reasons.
- This applied experiment used lots of available subjects (mail carriers) and showed human factors gains often clash with large-system inertia.