How should a normal person decide what's true?: Network epistemology

The simulation that broke both stories

In 2010, three philosophers of science ran simulations of how beliefs move through investigator communities. Agents experimented, shared results, updated. The finding upended intuition: communities with less connectivity sometimes converged on truth faster. Full connectivity let persuasive early results cascade before counter-evidence accumulated. Partial connectivity preserved independent investigation. The optimal epistemic structure was not the most open or closed. It was the most strategically partitioned.

That breaks both stories. The institutional trust camp says trust the credentialed consensus. The simulations show credentialed consensus can lock in errors precisely because the network is too connected. When everyone reads the same journals, a wrong result propagates through the entire tree before anyone checks the roots. This is the replication crisis — half to two-thirds of published findings in psychology could not be reproduced.

The nurse is already networked

The personal verification camp says trace the chain yourself. The simulations show isolated agents converge on truth more slowly and sometimes never converge. Human knowledge is irreducibly social. The nurse deciding about her tap water is not choosing between trust and verification. She is integrating signals — her neighbor at the plant, a Facebook group, the newspaper that no longer exists. She produces a judgment that feels individual. It is distributed computation on a social network she did not design.

The question is how to design the network so it produces accurate beliefs. The failures — fact-checking that convinced the already-convinced, media literacy programs whose graduates identified bias only in opposing sources — targeted nodes. The problem lives in the edges.

Where we concede ground: Our framework describes but does not prescribe. The nurse needs a yes or no by Tuesday.

What would change our mind: Trained individuals in bad networks outperforming untrained individuals in good ones.

Read the full synthesis: How should a normal person decide what’s true?