Oliver Selfridge (1926-2008) was a British-born American computer scientist often called the “father of machine perception.” He helped organize the 1956 Dartmouth workshop that launched AI as a field, and worked across MIT’s Lincoln Laboratory, Project MAC, Bolt Beranek and Newman, and GTE Laboratories. His most influential idea is the Pandemonium model, presented in the 1959 paper “Pandemonium: A Paradigm for Learning,” whose original text is preserved by AAAI’s AITopics archive.
Pandemonium proposed a system of simple parallel agents Selfridge called “demons,” each watching for a particular feature and “shouting” with a strength proportional to how well it matched. Higher-level demons listened to lower ones and combined their cries into a decision, and the system could improve itself by adjusting which features mattered. The paper opens by claiming a model that “can adaptively improve itself to handle certain pattern recognition problems” that cannot be fully specified in advance.
This layered, feature-detecting, self-improving design prefigured much of what came later: the hierarchical feature detectors of convolutional neural networks and the general notion that perception can be learned rather than hand-coded. Selfridge spent more than fifty years in the field, but Pandemonium remains his clearest mark on the lineage that runs to modern deep learning.