Observation Interference in Partially Observable Assistance Games
Authors: Scott Emmons, Caspar Oesterheld, Vincent Conitzer, Stuart Russell
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, in Section 7, we use an experimental model to investigate tradeoffs the assistant faces when deciding whether or not to interfere with observations. In line with our theory, we find that observation interference allows the AI assistant to communicate private information, but it comes at the cost of destroying useful information. Measuring this tradeoff, we find that having more private information leads to a stronger incentive to interfere with observations. We run a Monte Carlo simulation with 30,000 trials to calculate the expected payoff in each setting. |
| Researcher Affiliation | Academia | 1Center for Human-Compatible AI, University of California, Berkeley 2Foundations of Cooperative AI Lab, Carnegie Mellon University. Correspondence to: Scott Emmons <EMAIL>. |
| Pseudocode | No | The paper describes methods and proofs using mathematical notation and conceptual descriptions but does not include any clearly labeled 'Pseudocode' or 'Algorithm' blocks or figures with structured, code-like steps. |
| Open Source Code | No | Section H 'Code Assets' mentions: Our experiments use the Python software libraries Matplotlib (Hunter, 2007), Num Py (Harris et al., 2020), pandas (pandas development team, 2020; Wes Mc Kinney, 2010), and seaborn (Waskom, 2021). This refers to third-party libraries used, not the authors' own implementation code for their methodology. There is no explicit statement or link indicating that the authors' own source code is provided. |
| Open Datasets | No | We study a game where selecting the best action requires combining private observations known only to H and private observations known only to A. The game has d products. Each product i has two attributes, Hi and Ri, drawn i.i.d. from Unif(0, 1). This indicates that data for the experiments is simulated or generated, not sourced from a publicly available dataset. |
| Dataset Splits | No | The paper describes a simulated environment where product attributes are 'drawn i.i.d. from Unif(0, 1)' and experiments are run using 'a Monte Carlo simulation with 30,000 trials'. Since the data is generated via simulation rather than being a fixed dataset, there are no traditional training, validation, or test splits mentioned or applicable for reproduction. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware used to run the experiments, such as GPU or CPU models, memory specifications, or cloud computing platforms with their configurations. It only mentions software libraries used for the experiments. |
| Software Dependencies | No | Section H 'Code Assets' states: 'Our experiments use the Python software libraries Matplotlib (Hunter, 2007), Num Py (Harris et al., 2020), pandas (pandas development team, 2020; Wes Mc Kinney, 2010), and seaborn (Waskom, 2021).' While these libraries are named, specific version numbers for these libraries or the Python interpreter used in the experiments are not provided in the text, which is required for a reproducible description. |
| Experiment Setup | Yes | In Section 7.1 'Experiment Details', the paper specifies key parameters and methods: 'We consider a game with d = 5 products. We vary R s number of interferences k {0, 1, 2, 3, 4}. We run a Monte Carlo simulation with 30,000 trials to calculate the expected payoff in each setting.' It also defines the human's policy: 'Definition 7.1. H s Boltzmann selection policy chooses products by a Boltzmann distribution over ˆHi, the observed product values: πH(ai) exp(β ˆHi). The parameter β controls H s rationality.' and specifies values for it: 'We fix A to have 2 private observations. We do a logarithmic sweep over H s rationality coefficient β {0.01, 0.03, 0.1, 0.3, 1, 3, 10, 30, 100}.' |