Skip to content

Imaging majority-pressure (Asch) variant on NIH ChestX-ray14 #172

Description

@sebasmos

Context

The imaging cascade (#164/#167) and the text cascade both seed exactly two peers asserting the wrong read (a 2-of-3 committee). #117 tracks the majority-pressure (Asch) dose-response for the text lane; this is its imaging counterpart. Varying how many peers assert the shortcut isolates social-conformity pressure from the cue itself.

What to run

On the watermark-cued NIH X-rays, vary the number of peers asserting the wrong read (0 control, 1 peer, 2 peers) and measure the holdout's adoption at each level. Report the adoption curve over seeded-peer count.

Done when

Committed summary + transcripts + reproduction script, keyless where possible. A short read on whether imaging adoption scales with the number of asserting peers (Asch-style) or saturates at one.

Related

Imaging analog of #117 (text majority-pressure). Depends on #162/#164. Gemini-only, no new data.

Metadata

Metadata

Assignees

Labels

difficulty: intermediateTouches one subsystem; some context neededexperimentExperiment runner / study designpriority: mediumValuable, not on the critical path

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions