Including the cost of irreducible uncertainty in the Policy Compression framework
- Author
- Alvaro Garrido Perez (UGent) , Pieter Simoens (UGent) , Amrapali Pednekar (UGent) and Yara Khaluf (UGent)
- Organization
- Abstract
- AI decision-support systems can benefit from anticipating biases in human decision-making. Many such biases may arise from human cognitive limitations. The Policy Compression framework models decision-making as a trade-off between reward maximization and the cognitive cost of encoding state-dependent action policies, formalized as the mutual information between states and actions (policy complexity). We argue that this account is incomplete because it treats conditional entropy–the irreducible uncertainty about which action should be selected given a state–as costless, even though empirical evidence suggests that it modulates reaction times. We therefore extend the framework by defining cognitive cost as the sum of policy complexity and a weighted conditional-entropy term, governed by a new parameter, η. The resulting optimal policy retains the standard exponential form but becomes sharper as increases, allowing policy precision to vary more independently of reward sensitivity. This modification implies that the standard Policy Compression framework may underestimate the cognitive cost of action selection, and it has the potential to better account for biases in human decision-making. At the same time, it introduces additional complexity for fitting the model to human data, which future work will need to address.
Downloads
-
publisher version.pdf
- full text (Published version)
- |
- open access
- |
- |
- 322.21 KB
Citation
Please use this url to cite or link to this publication: http://hdl.handle.net/1854/LU-01KX0A76HFQXEWM4KJAB9KX6SB
- MLA
- Garrido Perez, Alvaro, et al. “Including the Cost of Irreducible Uncertainty in the Policy Compression Framework.” Proceedings of the 5th International Conference on Hybrid Human-Artificial Intelligence, edited by Maryam Alimardani et al., vol. 423, IOS Press, 2026, pp. 132–44, doi:10.3233/faia260499.
- APA
- Garrido Perez, A., Simoens, P., Pednekar, A., & Khaluf, Y. (2026). Including the cost of irreducible uncertainty in the Policy Compression framework. In M. Alimardani, T. Lenaerts, A. Meyer-Vitali, A. Nowé, J. Vennekens, & S. Wang (Eds.), Proceedings of the 5th International Conference on Hybrid Human-Artificial Intelligence (Vol. 423, pp. 132–144). https://doi.org/10.3233/faia260499
- Chicago author-date
- Garrido Perez, Alvaro, Pieter Simoens, Amrapali Pednekar, and Yara Khaluf. 2026. “Including the Cost of Irreducible Uncertainty in the Policy Compression Framework.” In Proceedings of the 5th International Conference on Hybrid Human-Artificial Intelligence, edited by Maryam Alimardani, Tom Lenaerts, André Meyer-Vitali, Ann Nowé, Joost Vennekens, and Shenghui Wang, 423:132–44. IOS Press. https://doi.org/10.3233/faia260499.
- Chicago author-date (all authors)
- Garrido Perez, Alvaro, Pieter Simoens, Amrapali Pednekar, and Yara Khaluf. 2026. “Including the Cost of Irreducible Uncertainty in the Policy Compression Framework.” In Proceedings of the 5th International Conference on Hybrid Human-Artificial Intelligence, ed by. Maryam Alimardani, Tom Lenaerts, André Meyer-Vitali, Ann Nowé, Joost Vennekens, and Shenghui Wang, 423:132–144. IOS Press. doi:10.3233/faia260499.
- Vancouver
- 1.Garrido Perez A, Simoens P, Pednekar A, Khaluf Y. Including the cost of irreducible uncertainty in the Policy Compression framework. In: Alimardani M, Lenaerts T, Meyer-Vitali A, Nowé A, Vennekens J, Wang S, editors. Proceedings of the 5th International Conference on Hybrid Human-Artificial Intelligence. IOS Press; 2026. p. 132–44.
- IEEE
- [1]A. Garrido Perez, P. Simoens, A. Pednekar, and Y. Khaluf, “Including the cost of irreducible uncertainty in the Policy Compression framework,” in Proceedings of the 5th International Conference on Hybrid Human-Artificial Intelligence, Brussels, Belgium, 2026, vol. 423, pp. 132–144.
@inproceedings{01KX0A76HFQXEWM4KJAB9KX6SB,
abstract = {{AI decision-support systems can benefit from anticipating biases in human decision-making. Many such biases may arise from human cognitive limitations. The Policy Compression framework models decision-making as a trade-off between reward maximization and the cognitive cost of encoding state-dependent action policies, formalized as the mutual information between states and actions (policy complexity). We argue that this account is incomplete because it treats conditional entropy–the irreducible uncertainty about which action should be selected given a state–as costless, even though empirical evidence suggests that it modulates reaction times. We therefore extend the framework by defining cognitive cost as the sum of policy complexity and a weighted conditional-entropy term, governed by a new parameter, η. The resulting optimal policy retains the standard exponential form but becomes sharper as increases, allowing policy precision to vary more independently of reward sensitivity. This modification implies that the standard Policy Compression framework may underestimate the cognitive cost of action selection, and it has the potential to better account for biases in human decision-making. At the same time, it introduces additional complexity for fitting the model to human data, which future work will need to address.}},
author = {{Garrido Perez, Alvaro and Simoens, Pieter and Pednekar, Amrapali and Khaluf, Yara}},
booktitle = {{Proceedings of the 5th International Conference on Hybrid Human-Artificial Intelligence}},
editor = {{Alimardani, Maryam and Lenaerts, Tom and Meyer-Vitali, André and Nowé, Ann and Vennekens, Joost and Wang, Shenghui}},
isbn = {{9781643686707}},
issn = {{0922-6389}},
language = {{eng}},
location = {{Brussels, Belgium}},
pages = {{132--144}},
publisher = {{IOS Press}},
title = {{Including the cost of irreducible uncertainty in the Policy Compression framework}},
url = {{http://doi.org/10.3233/faia260499}},
volume = {{423}},
year = {{2026}},
}
- Altmetric
- View in Altmetric