nucleation

LIT-733: In Epistemic Networks, Is Less Really More?

2 min read

Properties5
StatusActive — read 2026-10-05 (NOTE-559): read in full from the copy on Cailin O'Connor's website. Worth reading as the robustness check on the Zollman effect. Over a wider parameter space, sparse networks beat dense ones only where inquiry is hard: small groups, small batches of data, and nearly indistinguishable alternatives. Elsewhere sparsity only slows learning. The authors keep Zollman's transient-diversity point as robust and read these models as "how-potentially" stories.
Tagsepistemology, network-science, philosophy-of-science
CorrectsLIT-724 — The Communication Structure of Epistemic Communities, LIT-735 — The Epistemic Benefit of Transient Diversity
Cited as “paper” byNOTE-559 — In Epistemic Networks, Is Less Really More?
Cited as “source” byTHEORY-143 — When a group learns only about the options its members pursue and good evidence is scarce, sparser communication makes it more likely to settle on the better option, at a large cost in speed; when evidence is plentiful, sparsity only slows learning, THEORY-148 — A group learning by trying needs diversity in what its members pursue that lasts long enough to survive misleading early evidence and then gives way to consensus; limited communication and stubborn priors each supply it, and together they make it permanent, so the group never settles

Graph View

Sarita Rosenstock, Justin Bruner & Cailin O’Connor (2017), Philosophy of Science 84(2):234–252 — DOI-10.1086/690717

Key takeaways

  • The “Zollman effect” is the advantage of a cycle over a complete network in converging on the better action. Zollman (LIT-724, n. 7) said the network ordering holds “for any setting of the parameters”. This paper finds otherwise (§3).
  • Success gap. With 10 agents and 1,000 trials a round, the effect falls below 2% at pB = 0.51 and is zero from pB = 0.525. Zollman used 0.501 (Figure 2).
  • Data per round. Fewer trials per round increase the effect: small batches let misleading streaks occur (Figure 3).
  • Group size. The effect fades as networks grow. At 100 agents the complete network converged correctly 99.12% of the time against the cycle’s 100%, in 20 rounds against 1,977 (Figures 4–5).
  • Exploratory agents. In Kummerfeld & Zollman’s ε-greedy models both the effect and its reverse (dense networks better when agents explore) disappear as the better action becomes easier to identify (§4).
  • Diagnosis (§5.1). Network structure matters only when inquiry is “difficult”: similar alternatives, small groups, small batches. There, “decreasing connectivity only slows learning without providing any benefit” once agents have enough good data.
  • What survives (§5.2). Zollman’s observation that transient diversity is necessary (LIT-735) “holds for all the models discussed”. The models are “how-potentially” stories, and the authors lower their confidence that such network effects regularly occur in real communities.
  • A better remedy. Where data are scarce, stubborn or exploratory agents, or standards for how much data a judgment needs, prevent premature settling “apart from ignoring good data”.

Standing in the record

Filed on 2026-10-05 at the owner’s request, as the robustness check on the literature an essay’s §8.2 cites. It corrects the scope of Zollman’s claim, not its mechanism, and the record’s account THEORY-143 is stated with the scope it leaves. Longino’s SEP entry (LIT-725) reports this paper’s finding.

Backlinks