School of Computer Science THE UNIVERSITY OF BIRMINGHAM CN-CR Ghost Machine

Talk at the Bristol Visual Information Laboratory
Friday Oct 2nd, 4pm
Seminar Room, Life Sciences Building

Why are the many recent results in statistics-based machine vision misleading?

Perhaps because none of the mechanisms used can account for the roles of vision in mathematical discoveries leading to Euclid's Elements 2.5 millennia ago?

Aaron Sloman
Honorary Professor of AI and Cognitive Science
School of Computer Science
University of Birmingham

Local host: Dima Damen

As far as I know very few researchers on vision have any interest in the role of vision in mathematical discovery and reasoning (the topic that first drew me into Artificial Intelligence in 1971 as a possible route to solving old philosophical problems about the nature of mathematics, and defending Kant). Nearly four and a half decades later the problems are still unsolved. Worse, they seem to be ignored by almost all researchers in AI/Robotics, psychology, neuroscience and philosophy. A few years ago, inspired by Turing's paper on Morphogenesis published in 1952, I wondered what he might have done had he not died two years later. My tentative answer is the Meta-Morphogenesis project -- a long term multi-disciplinary project aiming to identify and explain the many transitions in biological information processing since pre-biotic molecules. Recently this led to an investigation of the evolution of many kinds of biological construction-kits (concrete, abstract and hybrid), including meta-construction-kits for producing new construction kits. I shall try to show how unsolved problems relating the amazing mathematical discoveries leading to Euclid's Elements, the many toddler theorems apparently discovered (unwittingly) by pre-verbal children, gaps in Gibson's theories of affordances, varieties of intelligence in other species, and closely related weaknesses in philosophy, psychology, neuroscience, linguistics, robotics and AI (obscured by their narrowly-focused successes) may perhaps be addressed in this project. It may turn out to require new forms of computation. In part, the talk will be an invitation to join in. The presentation will be example-based and partly audience-driven.
Note added after the talk (3 Oct 2015)
Examples discussed in the talk, and several related examples for which there was insufficient time, are presented in this (still growing) document on abilities to perceive depictions of and to think about impossible objects and scenes:
Some (Possibly) New Considerations Regarding Impossible Objects
Their significance for mathematical cognition,
and current serious limitations of AI vision systems.
(Those mechanisms are not able to explain mathematical qualia!)

Sample Background/References:

Maintained by Aaron Sloman
School of Computer Science
The University of Birmingham

This document

A partial index of discussion notes is in