Abstract : In our digital and connected societies, the development of social networks,online shopping, and reputation systems raises the questionof how individuals use social information, and how it affects theirdecisions. We report experiments performed in France and Japan,in which subjects could update their estimates after having receivedinformation from other subjects. We measure and model the impactof this social information at individual and collective scales. We observeand justify that when individuals have little prior knowledgeabout a quantity, the distribution of the logarithm of their estimatesis close to a Cauchy distribution. We find that social influence helpsthe group improve its properly defined collective accuracy. We quantifythe improvement of the group estimation when additional controlledand reliable information is provided, unbeknownst to the subjects.We show that subjects' sensitivity to social influence permitsto define five robust personality traits and increases with the differencebetween personal and group estimates. We then use our datato build and calibrate a model of collective estimation, to analyze theimpact on the group performance of the quantity and quality of informationreceived by individuals. The model quantitatively reproducesthe distributions of estimates and the improvement of collective performanceand accuracy observed in our experiments. Finally, ourmodel predicts that providing a moderate amount of incorrect informationto individuals can counterbalance the human cognitive biasto systematically underestimate quantities, and thereby improve collectiveperformance.

Reference : How social information can improve estimation accuracy in human groups. Jayles, B., Kim, H-R., Escobedo, R., Cezera, S., Blanchet, A., Kameda, T., Sire, C. , and Theraulaz, G., Proceedings of the National Academy of Sciences USA (published 6 November 2017)

DOI : www.pnas.org/cgi/doi/10.1073/pnas.1703695114

CNRS press release (in French)