Farmers can recognize a better explanation even when they cannot produce it alone. That is the finding of Locals know more than it seems, a 2026 Royal Society paper co-authored by Émile Servan-Schreiber. On 2 June, he recommended the theme issue containing the study. UM6P’s announcement described 18 contributions across animal, human and artificial collective intelligence; he had highlighted his farming-community and language-model papers on 27 April.
A field that studies how groups decide
In a Royal Society interview published on 22 May, the guest editors traced the project to a workshop at UM6P in Rabat. Sarah Alami explained that research in behavioural ecology and evolutionary anthropology had developed alongside, rather than in close conversation with, the newer collective-intelligence field. Bringing those traditions together meant examining what groups achieve, as well as what individuals can do.
Cathal O’Madagain recalled a discussion at a 2024 Rabat conference in which participants converged on group decision-making as the field’s central subject. That connects practical questions about organizing teams with research on cooperation, cultural transmission and computer systems. The farming paper makes that connection concrete.
What isolated answers can miss
Published in April 2026, Locals know more than it seems is by Talib El Aissati, Ghizlane Goubraim and five colleagues, including Servan-Schreiber. It reports farming-community samples of 203 in Morocco, 198 in Mali and 120 in Ghana. Participants explained why familiar agricultural techniques work, then evaluated a shortened, anonymous set of explanations supplied by their peers. Scientists independently rated the explanations.
The anonymity mattered: participants could judge an answer without knowing whether it came from an influential neighbour. After reviewing alternatives, their preferred explanations scored better against the scientists’ evaluations; less experienced participants showed greater improvement. The authors interpret this as a division of knowledge across the community: people need not produce the strongest explanation themselves to recognize and select it.
This was an explanation task, not a forecast of crop yields or a test of election probabilities. Its finding concerns how existing knowledge is elicited. A survey that stops at the first individual response can therefore miss knowledge accessible through a second round of comparison.
The connection to forecasting
For forecasting, the useful question is how a collection process allows people to contribute information they hold and recognize information they lack. This study tests one such process. The issue’s separate experiment on human crowds and language-model forecasts addresses a different task and has its own performance measures. Keeping the two apart preserves the value of both: collective intelligence is a research question about group design, with results that must be checked in the setting where they were measured.
