Work

Displacement of Discriminating Questions by Expected Response Costs: Why Some Questions Are Easier to Ask

Year: 2026
Type: article
Abstract: This conceptual paper introduces Displacement of Discriminating Questions by Expected Response Costs, a mechanism through which the anticipated consequences of an answer can change which question a person pursues. Someone may recognize the question that would best inform a decision and turn to another because asking the first could delay a purchase, strain a relationship, or create a duty to act. The substitute can yield useful information while leaving the decision’s central uncertainty unresolved. The first case concerns the Metropolitan Museum of Art’s 1972 purchase of the Euphronios krater. Trustee Arthur Ochs Sulzberger asked whether the vase had been tested. Director Thomas Hoving promised a test of the clay and later recalled keeping his doubts about the ownership papers to himself. Dating the clay could establish when it was fired. Establishing who had owned the vase required examining the documents. One person raised the first question and another withheld the second. The possibility that a provenance inquiry would have put the purchase at risk is the paper’s inference. The second case concerns an investment in Azerbaijan examined in United States v. Kozeny. On a recorded call, investor Frederic Bourke asked what someone should do if told that an official had been paid. His question makes the consequences of knowing explicit. Whether those consequences led him to pursue a different question remains open. A thought experiment supplies the comparison missing from both historical records. The same committee member considers the same purchase and the same evidence, while the expected delay from checking the ownership papers changes. Part of Recursive Inquiry and Encounter in the On Human Modeling series, the paper places question selection alongside research on deliberate ignorance and considers when a less demanding inquiry may be proportionate. One possible application is AI evaluation. A reviewer who recognizes that test questions may have appeared in a model’s training data might request another breakdown of the scores if checking for overlap would delay release. Making that check routine would allow evaluators to examine whether the anticipated delay changes what reviewers ask.
Language: English
Open Access status: green
OpenAlex ID: W7236725009
FWCI:
Cites:
Cited by:
Related to: 10