Parametrically Retargetable Decision-Makers Tend To Seek Power

Authors: Alex Turner, Prasad Tadepalli

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical We develop a simple, broad criterion of functional retargetability (definition 3.5) which is a sufficient condition for power-seeking tendencies. Theorem 3.6 (Multiply retargetable functions have orbit-level tendencies). If f is (Θ, A n B)-retargetable, then f(B | θ) n most: Θ f(A | θ). Proof outline (full proof in Appendix B).
Researcher Affiliation Academia Alexander Matt Turner, Prasad Tadepalli Oregon State University {turneale@, tadepall@eecs.}oregonstate.edu
Pseudocode No The paper describes various decision-making procedures conceptually and formally, but it does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code No The paper does not contain any statements about making source code available or provide links to a code repository.
Open Datasets No The paper analyzes decision-making in environments such as the 'Pac-Man video game' and 'Montezuma s Revenge (MR).' While these are well-known game environments, the paper does not provide concrete access information like specific links, DOIs, repository names, or formal citations with authors and year for a dataset used in their analysis.
Dataset Splits No The paper is theoretical in nature and does not conduct experiments involving dataset splits, thus no validation split information is provided.
Hardware Specification No The paper does not specify any hardware details or computing resources used for its analysis or any potential simulations.
Software Dependencies No The paper does not list specific software dependencies or their version numbers that would be required to replicate any aspects of the work.
Experiment Setup No The paper is theoretical and does not describe an experimental setup with hyperparameter values, training configurations, or other system-level settings.