This is a repository associated with AI-powered boosting via decision aids: a project on automatically discovering interpretable descriptions of decision-making strategies and conveying those to people. This project is a part of a larger endeavor on improving human decision-making, and even larger endeavor of rationality enhancement and life improvement science.
Navigate to RL2DT to learn more about AI-Interpret, an algorithm that automatically creates descriptions of decision strategies formalized in terms of (metalevel) reinforcement learning policies (Skirzyński et al., 2021).
Navigate to DNF2LTL to learn more about DNF2LTL, a method for transforming the static output of AI-Interpret into a procedural description (Becker et al., 2022).
Navigate to the experiment folders in order to see the data gathered in experiments on the quality of AI-Interpret (Skirzyński, et al., 2021), the experiments showing people's preference towards procedural descriptions of decision strategies (Becker et al., 2022), as well as experiments showing the performance of AI-powered boosting via decision aids on naturalistic tasks (Becker et al., 2022).
Becker, F., Skirzyński, J,, van Opheusden, B., & Lieder. F. (2022). Boosting human decision-making with AI-generated decision aids. Computational Brain & Behavior (2022). Available at https://link.springer.com/article/10.1007/s42113-022-00149-y
Skirzyński, J., Becker, F., & Lieder, F. (2021). Automatic discovery of interpretable planning strategies. Machine Learning, 110(9), 2641-2683. Available at https://link.springer.com/article/10.1007/s10994-021-05963-2