Model-Based Wisdom of the Crowd for Sequential Decision-Making Tasks.

Area of Science:

• Cognitive Science
• Behavioral Economics
• Collective Intelligence

Background:

• The 'wisdom of the crowd' phenomenon suggests group decisions can outperform individual ones.
• Sequential decision-making tasks, like the Balloon Analogue Risk Task (BART), optimal stopping, and bandit problems, present unique challenges for collective intelligence.
• Traditional behavior-based approaches, relying on majority opinion, may falter when crowd diversity diminishes over time.

Purpose of the Study:

• To evaluate the effectiveness of behavior-based versus model-based approaches in sequential decision-making tasks.
• To identify the limitations of behavior-based collective intelligence in dynamic environments.
• To demonstrate the predictive power and generalizability of a cognitive model-based approach.

Main Methods:

• Assessed crowd behavior using majority decisions in BART, optimal stopping, and bandit problems.
• Developed and inferred individual cognitive models from observed behavior across tasks.
• Utilized inferred models to predict individual and collective responses in novel task scenarios.

Main Results:

• Behavior-based approaches performed poorly in BART and bandit tasks, showing a tendency towards extreme crowd decisions.
• Model-based approaches demonstrated robust performance across all three sequential decision tasks.
• The model-based approach successfully generalized predictions to new problems lacking behavioral data.

Conclusions:

• Collective intelligence in sequential tasks is undermined by the loss of opinion diversity in behavior-based methods.
• Cognitive model-based approaches offer a more reliable and generalizable framework for understanding and predicting collective decision-making.
• This research provides a foundation for applying advanced modeling techniques to real-world collective intelligence problems.