Melanie Mitchell, Douglas R. Hofstadter (1990), Physica D: Nonlinear Phenomena 42(1–3):322–334 — DOI-10.1016/0167-2789(90)90086-5
Standing in the record
Filed on 2026-10-01 at the owner’s request for Douglas Hofstadter’s work in this record: the Fluid Analogies Research Group’s programme that analogy-making is the core of cognition, of which Copycat is the founding program. It is the journal paper the group itself cites for Copycat’s mechanics.
Unavailable because no lawful full text could be reached (see the history
note), so no NOTE is filed. What the record can say without it comes from
Chalmers, French & Hofstadter (LIT-414), read in full, which describes
Copycat and sends readers here “for further details”. On that account:
- Copycat solves letter-string analogies (“if abc changes to abd, what does xyz change to?”) by building its own representations of the strings, not by being handed them.
- Many small agents (codelets) build and break bonds, groups and correspondences in parallel and nondeterministically. A network of concepts (the Slipnet) is activated by what they find and biases which codelets run next. A computational temperature, falling as coherent structure accumulates, sets how random their choices are.
- The xyz problem is the showcase. The successor of z does not exist in Copycat’s non-circular alphabet, so the program hits a snag, raises its temperature, breaks structure, and can re-perceive xyz as a predecessor group to answer wyz. LIT-414 is careful that it gives xyd more often than wyz, and attributes “further details” of this run to this paper.
The title’s claim, that understanding emerges, is what a reading would test. It places the paper in the record’s emergence cluster: McKenzie’s survey (LIT-141) counts self-organisation and downward causation among the marks of emergence, and Copycat’s Slipnet is explicitly a top-down influence on bottom-up agents. Whether “emergence” here means more than “no single codelet does it” is the question.
It also sits beside Minsky’s account of a mind as many small, mindless agents (LIT-429), filed in the same session: Copycat is a working program in that spirit, though this record has not read whether the paper cites Minsky. Hofstadter’s book-length account of the group’s programme, Fluid Concepts and Creative Analogies (1995), is being filed beside this work.
What a reading should check: the run statistics it reports for its problems, the exact definition of temperature, whether it compares Copycat’s answer distributions with people’s (its companion paper says the distributions are “qualitatively similar”, citing Mitchell’s 1990 dissertation), and what it means by “understanding”.