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. |