Bruce Burns (1995), “Fluid Concepts and Creative Analogies: A Review”, AI Magazine 16(3), Fall 1995, pp. 81–83 — DOI-10.1609/aimag.v16i3.1155
Key takeaways
- What the book is. A collection of articles, “many already published”, by Hofstadter and members of the Fluid Analogies Research Group (Chalmers, Defays, French, McGraw, Mitchell), held together by Hofstadter’s prefaces. For “a quick tour”, chapters 2 (Jumbo), 4 (high-level perception) and 5 (Copycat).
- The computational idea: the parallel terraced scan. Codelets are chosen probabilistically from a coderack; some spot possible structures, others build or destroy them, and each run changes the odds for the rest. As coherent structure forms, “the temperature is lowered”, so it becomes harder to tear down: “a coherent structure emerges without the need for any top-level control”.
- The psychological idea: cognition is high-level perception (“cognition is recognition”). Representation and processing interact, so leaving representation to a later module misses the point; the example is BACON, which is handed the data Kepler had to find.
- Why little impact on AI, criticism 1: microdomains. Every program works on letter strings, table settings, word puzzles or letter fonts. Burns grants that real-world programs also restrict their domains, but microdomains “can contain hidden assumptions”, and it is unclear how far the programs succeed by leaving representation to low-level perception.
- Criticism 2: argument by analogy and intuition. The symbol-grounding objection is pressed against SME but “quickly dismissed” for Copycat’s SIGMA symbol; Burns turns the same argument back. The evaluation preface says AI research is “90 percent” intuition; Burns replies that the other 10 percent should not be ignored, and that this stance “risks irrelevance”.
- Verdict. Recommended, “because it is eloquently written and provocative”, and because Burns thinks the basic ideas “might be on the right track”.
Standing in the record
Filed on 2026-10-02 at the owner’s request. NOTE-375 is the close reading, and it placed the review: Active, as a reading of a review.
How it bears on LIT-428. LIT-428, Hofstadter and FARG’s book, is
Deferred because no lawful full text could be read, and it already quotes
this review as “the one account of the content I read”. This entry is that
review in its own right. The reading checked LIT-428’s account of the review
against the text, and they agree: the bibliographic details (Basic Books,
New York, 1995, 518 pp., ISBN 0-465-05154-5), the two underlying ideas, the
codelet-and-temperature description and the quotation “without the need for
any top-level control”, “cognition is recognition”, the two criticisms and
the recommendation. LIT-428 notes that Burns calls the evaluation preface
“Preface 5” where the catalogue’s contents note puts it before chapter 9.
The review adds a check on that: it calls the Eliza-effect preface “Preface
4”, which agrees with the catalogue if the prefaces are numbered by the
chapter they precede (that preface precedes chapter 4). On that numbering
the evaluation preface would be Preface 9, so the discrepancy is in the
review’s “Preface 5”, or the book numbers its prefaces another way. The book
is still needed to settle it.
The review is quoted here as the reviewer’s account, not as the book’s. It does not make LIT-428 read.
Tags. cognition for analogy and high-level perception, complex-systems
for the emergent, controller-free architecture it describes (as on LIT-428),
and philosophy-of-science for its argument about how AI research should be
evaluated and what intuition can carry.
Boundary. It carries no instruction for machine-learning practice. It compares the terraced scan with neural networks and genetic algorithms, and suggests learning codelets with genetic algorithms, as a possibility. Its point about evaluation (the Eliza effect, separating a program’s success for interesting and uninteresting reasons) is about evaluating AI models in general; an anthology reader on evaluation might want it, as LIT-428 says of the book’s evaluation preface. That is a judgement for the anthology.