When the crowd gets it wrong – the limits of collective wisdom in machine learning

Abstract

This study examines collective decision-making dynamics using a machine learning framework, drawing parallels between a previously established synthetic population model and a newly introduced ensemble machine learning counterpart. Grounded in the “wisdom of crowds” principle, the research explores scenarios where the accuracy of group decisions may unexpectedly decrease as group size increases, particularly when individuals share highly correlated information. By replicating these conditions with machine learning ensembles, such as decision trees and support vector machines, the study identifies circumstances where collective accuracy declines, challenging the assumption that larger groups inherently make better decisions. The findings reveal the limitations of collective models in machine learning and provide valuable insights for data-scarce environments.