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

Area of Science:

• Computational Social Science
• Machine Learning
• Collective Behavior

Background:

• The
• wisdom of crowds
• suggests larger groups make more accurate decisions.
• However, this assumption may not hold under certain conditions, such as correlated individual information.

Purpose of the Study:

• To investigate collective decision-making dynamics using a machine learning framework.
• To compare a synthetic population model with an ensemble machine learning model.
• To identify conditions where collective accuracy declines with increasing group size.

Main Methods:

• Utilized an ensemble machine learning framework.
• Replicated conditions of correlated information within groups.
• Employed machine learning models like decision trees and support vector machines.

Main Results:

• Demonstrated that collective accuracy can decrease as group size increases when information is highly correlated.
• Identified specific circumstances where the
• wisdom of crowds
• principle is limited.
• Showcased limitations of collective models within machine learning.

Conclusions:

• Larger groups do not always yield better decisions, particularly with correlated information.
• Machine learning ensemble models can reveal nuances in collective decision-making.
• Findings offer insights for decision-making in data-scarce environments.