WORK
Copycat
The 1983–1988 computer program Hofstadter built with Melanie Mitchell at the University of Michigan to model <em>fluid analogy-making</em> in a narrow microdomain — and, four decades later, the clearest available demonstration of what current AI architectures lack.
Copycat's task was deceptively simple: given that 'abc' changes to 'abd,' what does 'ijk' change to? Any child answers 'ijl' without effort. But solving the problem requires perceiving the relevant abstraction ('last letter,' 'successor,' 'replace') from raw material, deciding fluidly which features matter, and constructing a mapping that adjusts itself to the specific problem. Copycat's architecture deployed hundreds of small independent agents — codelets — exploring the problem space in parallel, competing and cooperating, building and tearing down representations, gradually converging on mappings that satisfied an emergent sense of coherence. The program was stochastic and deeply parallel. Run it twice on the same problem and it might produce different answers, just as two humans might parse 'iijjkk' differently depending on which features caught their attention first.
In The You On AI Field Guide
The critical feature that separated Copycat from every other AI system of its era — and from large language models of ours — was that its representations were
Keep reading with YOU ON AI
Unlock the full book, 10,000+ field-guide entries, and a 1000+ thinker library. If you have a book code, register now — it takes a minute.