Source
Kirsh (1995), LIT-1076 (Artificial Intelligence 73, 31–68), read in NOTE-871: §§2–6, footnotes 8, 9 and 11, Table 1. Kirsh (1996), LIT-1057 (Adaptive Behavior 4, 415–452), read in NOTE-872: sections 4 and 5, footnotes 3 and 6.
What was actually shown
A classification, from observation. The 1995 paper sorts uses of space into “spatial arrangements that simplify choice; spatial arrangements that simplify perception; and spatial dynamics that simplify internal computation.” Its evidence is “videos of cooking, assembly and packing, everyday observations in supermarkets, workshops and playrooms, and experimental studies of subjects playing Tetris”; much is anecdotal. Its agents are experts arranging their own workspace in the medium and short term; long-term pre-built environments are set aside.
Informational, not physical. An informational arrangement “achieves nothing that could not equally well be achieved by asking an oracle”, and is slower than one (footnote 8). The 1996 paper makes the same split in its formalism: some actions change a task graph’s topology or metric, others “leave the state space formally intact but that reduce the number and cost of mental operations needed for task success”.
Savings stated as search complexity. A strong cue at each choice point cuts search “from b^n to n”; knowing an assembly order turns a doubly exponential problem into an exponential one; hierarchical grouping reduces visual search “from O(N) to O(log N)”. These are worst-case counts for illustration.
Scope and set-up. A heuristic cue “normally applies to many choice points”. Of the salad maker’s rows, “the set-up costs could be amortized over the whole task. By contrast, in leaving a plug in the sink, or in placing the next item to be bagged close by, the cost of set-up must be paid off by the savings in the next few actions.” Footnote 9 points to Tarjan’s amortized analysis of self-adjusting data structures as “a more formal model”. The 1996 paper’s one tie of a saving to later uses: a sorted card hand’s saving “will reappear every time a change in intentions is made”.
Investment, risk and upkeep (1996). Doing work now for later use “depends on whether the expected benefits to be collected some time in the future outweigh the current costs multiplied by some interest factor to compensate for risk.” Footnote 3 lists the risks: whether the result is used, whether the changes “will still be present later”, and whether the agent survives. Arrangements decay: “Entropy teaches us that objects tend to scatter”, and the state “must be maintained or enforced”. Labour cost varies over time, so work is best done when it is cheap, and more of it when the next chance is uncertain.
The only numbers. 1996 pilot studies without sample sizes: counting coins, errors above 50% untouched, about 35% with pointing, “nearly 20 percent” with free movement; finding the longest of 20 sticks without touching them took “more than 55 percent more time”.
What this does not say
- How much any arrangement saves. “The actual reduction in computation achieved by the various methods I shall discuss does not, in general, lend itself to meaningful quantitative estimation.” The salad cue’s cost-benefit is named as an empirical matter and not carried out.
- That arranging is memoization. Tarjan’s amortization spreads the upkeep of a structure over operations; Kirsh’s analogy is to continual rearrangement. A cluster stores an index that makes search cheap, not an answer.
- That the 1996 economics is a model. The interest factor is verbal, with no rate, formula or reuse frequency; the task graph prices only physical transitions, and of congeniality “To date, there is no general theory to report”. “Amortize” occurs once there, a gloss on chimpanzees leaving nut-cracking stones in place.
- That stale results are a cost. The named failures are disturbance, clutter and decay of the arrangement; a stored result that has become wrong for the query does not appear.
- That it holds when arranger and user differ, or under deadline. Neither paper has such a case.
Connections
- QUESTION-095: this is the “maintain structure that makes search cheap” strategy. Kirsh 1995 draws the link to algorithmic amortization, not to amortized inference, which postdates it. The time-varying cost of labour is the cockpit’s reason for setting speed bugs at low workload (LIT-1066).
- THEORY-360 is the moment-of-use case; THEORY-339 is the stored-result case with the store in the head.
- CLAIM-303’s break-even has the same form as the scope rule here, with an upkeep term; Kirsh supplies neither its arithmetic nor a test.
- CLAIM-296 and CASE-085: “If I can’t find the garlic, I am not likely to think of rubbing it on the salad bowl—I miss some available actions.”
Note of 2026-10-30: what holds it back
Kept Proposed by the twenty-fourth status review. The title is not a report of Kirsh. It asserts, in the record’s voice, that agents arrange their surroundings this way and that set-up is recovered over the choice points an arrangement serves. Only the size of the savings is hedged. Neither paper measures a saving or tests the scope rule (NOTE-871, NOTE-872). The condition asks for that measured study, which is what the title asserts, so it stays (ADR-065 §6).