Balancing World Models in Decision Making
Author Information
Author(s): Ari E. Kahn, Nathaniel D. Daw
Primary Institution: Princeton University
Hypothesis
Do individuals dynamically adjust their reliance on the Successor Representation based on the predictability of their environment?
Conclusion
Participants adaptively use a mix of decision strategies, including model-based learning and the Successor Representation, which varies based on the predictability of future states.
Supporting Evidence
- Participants exhibited a significant sensitivity to reward from non-traversal trials.
- Behavior was best explained by a mixture of model-based and Successor Representation strategies.
- Participants adjusted their reliance on decision strategies based on the stability of reward probabilities.
Takeaway
People use different strategies to make decisions, and they change these strategies based on how predictable the situation is.
Methodology
Participants completed a two-stage Markov decision task where they chose between islands and boats to maximize rewards.
Limitations
The study did not distinguish between linear reinforcement learning and a mixture of separate agents.
Participant Demographics
Participants were 51 male, 44 female, and aged between 18 and 68 years, with a mean age of 36.19.
Statistical Information
P-Value
p<1eā21
Confidence Interval
95% CI = +/ā 0.101
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website