Many agreeing answers may originate from a single information source and therefore contain less evidence than they appear to.
In an information cascade, other people’s visible choices can outweigh private information. A related risk in AI environments is that several documents appear independent while sharing the same source.
Many agreeing answers may originate from a single information source and therefore contain less evidence than they appear to.
In cascade models, people observe earlier actions and receive their own private signals. Once the public history is strong enough, a rational next actor may follow the previous choices and ignore their own signal. Later agreement then becomes visible without adding a corresponding amount of new information.
Bikhchandani, Hirshleifer and Welch (1992) derive cascades within a specified model of sequential choices. Laboratory behaviour resembling a cascade does not by itself establish that participants use the assumed Bayesian reasoning; Huck and Oechssler (2000) show why task design and decision rules matter. The organisational application to sources with shared ancestry is related, but is not necessarily a formal cascade.
Observed actions compress earlier private information. When derivative artefacts share an origin, visible consensus grows faster than independent evidential weight.
AI transfer to investigate: AI makes derivative evidence extremely inexpensive: summaries, agents and documents can reproduce the same source chain. This amplifies an old problem of social learning and dependence.
Proposed test, not yet conducted: Ask users to assess a claim using several genuinely independent sources versus several derivative artefacts from a single source. Make ancestry explicit in one condition. Measure confidence and source weighting.
Consensus is not a property of numbers alone. We need to know how that consensus acquired its informational basis.
Shared-source dependence is a related risk. A formal cascade requires later choices to cease following private information under the specified sequential model.
Bikhchandani, Hirshleifer & Welch (1992), full article
Bikhchandani and colleagues describe the sequential model and conditions for fragility. Huck and Oechssler show that matching choices do not by themselves establish the assumed reasoning. Their 1999 working paper is used here; the final 2000 wording has not been fully compared. Shared source dependence is a related but separately testable problem.
Help improve this case
We review your contribution before updating the text or sources.