Generalization and Search in Risky Environments.

Area of Science:

• Cognitive Psychology
• Decision Science
• Computational Neuroscience

Background:

• Understanding decision-making under risk is crucial for various fields.
• Previous research has focused on exploration-exploitation trade-offs, but less on avoiding negative outcomes.
• The need to navigate environments with potentially harmful options requires adaptive search strategies.

Purpose of the Study:

• To investigate how individuals search for rewards in environments with high-stakes risks.
• To model human search behavior using computational approaches.
• To examine the adaptability of human decision-making to varying levels of environmental risk.

Main Methods:

• Participants engaged in modified multi-armed bandit tasks with spatially correlated rewards and avoidance criteria.
• Behavioral data was analyzed and compared against a Gaussian process function learning model with safe optimization.
• Leave-one-block-out cross-validation was employed to assess model fit and behavioral adaptation.

Main Results:

• Participant search behavior closely matched the predictions of the Gaussian process function learning algorithm.
• Individuals demonstrated adaptive sampling strategies, adjusting their search based on the perceived riskiness of the environment.
• The core function learning mechanism remained consistent despite behavioral adjustments.

Conclusions:

• Human decision-making in risky environments involves sophisticated adaptive search strategies.
• Gaussian process function learning provides a viable computational framework for understanding these behaviors.
• Individuals can effectively modify their approach to mitigate risks while pursuing rewards, showcasing cognitive flexibility.