Research library

Multi-model review Research note

Convergence is not the same as collective accuracy

A human study offers a useful warning about shared answers.

The finding

In an experiment with 144 people performing estimation tasks, exposure to others' estimates reduced diversity and increased confidence without corresponding improvements in collective accuracy. The study distinguishes convergence from learning. [1]

What it means for Pingpong

The analogy is relevant to a review chain: once an answer is visible, it can become an anchor. Pingpong explicitly permits later reviewers to depart from earlier responses. That is a reason to study the framing, not a reason to assume the influence disappears.

The limit

This is research on people, not language models. It supplies a conceptual warning rather than direct evidence about Pingpong. Our sequential reviewers share context and cannot be described as statistically independent votes.

Source

How social influence can undermine the wisdom of crowd effect

Jan Lorenz, Heiko Rauhut, Frank Schweitzer, and Dirk Helbing (2011)

PNAS 108(22), 9020-9025
DOI: 10.1073/pnas.1008636108

This is Pingpong's interpretation of external research, not a result from a Pingpong experiment or an endorsement by the authors. Editorial standard