Copycat: transferable scan
Status: memory-first and source-ungrounded Recall confidence: high
Remembered model
Copycat constructs analogies in small letter-string puzzles. Rather than receiving one fixed representation and searching it, many small stochastic processes—codelets—build and challenge structures in a workspace. A Slipnet of concepts changes activation as the problem develops. Candidate groups, correspondences, and rules compete for support. A temperature-like measure links emerging coherence to randomness: weak organization permits exploration; strong organization increasingly stabilizes a solution.
Copycat's deepest transferable idea is that representation and solution are co-constructed. Which objects correspond, which differences matter, and what rule describes the change cannot always be fixed before reasoning begins.
Provisional ontology
- Workspace object/structure: a provisional representation of the current problem.
- Codelet: a small process that notices, proposes, strengthens, weakens, or tests a structure.
- Slipnet node/link: a concept and its context-sensitive associative relation to other concepts.
- Activation: temporary conceptual relevance that changes which codelets run.
- Correspondence: a proposed mapping between objects across situations.
- Slippage: a context-supported substitution of one conceptual relation for another.
- Temperature: a control signal coupling global coherence to exploratory randomness.
- Answer: the result of a mutually supporting interpretation, not merely a retrieved rule.
Transfer candidates
COPY-1— let connection discovery build several interpretations. When relating two notes or systems, propose multiple mappings, each with explicit correspondences and mismatches, before choosing a link or synthesis.COPY-2— represent analogy slippage. A transfer should state which source relation changes meaning in the target. Calling both things "memory," "attention," or "compilation" without a slippage account invites decorative analogy.COPY-3— use heterogeneous micro-critics. Small checks for scope, mechanism, counterexample, provenance, and vocabulary collision can compete and reinforce one another around a draft rather than relying on one monolithic review judgment.COPY-4— couple exploration to coherence. Early discovery should preserve diverse hypotheses; as independent constraints converge, effort should shift toward testing and consolidation. A fixed creativity setting across the whole workflow wastes either breadth or rigor.COPY-5— preserve defeated mappings long enough to diagnose ambiguity. Rival interpretations are evidence about unstable representation even when one ultimately wins.
This architecture is especially relevant to the boundary between lexical connection search and explanatory synthesis: links become useful when a coherent relation is constructed, not when two artifacts share a term.
Method worth borrowing
Copycat makes intermediate structures observable. For Commonplace, a connection or synthesis process could emit a trace of candidate mappings, supporting relations, counterevidence, confidence changes, and the final selection. Such a trace would let evaluation ask whether the system found the right answer for the right structural reason.
Non-transfer and failure modes
- Coherence is not truth. A mutually supporting interpretation can be elegant and wrong without external evidence.
- Large numbers of codelet-like calls may be expensive and difficult to reproduce.
- Global temperature can be too coarse; different subproblems may warrant different exploration levels.
- Copycat's carefully designed microworld does not demonstrate that its mechanisms scale to open-ended repositories.
Grounding questions
- Which mechanisms are properly Copycat's and which belong to the broader parallel-terraced-scan tradition?
- How are strength, activation, urgency, and temperature computed and kept distinct?
- What ablations show that co-construction or slippage is necessary?
- How does Copycat avoid premature coherent but incorrect interpretations in its task domain?