Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Synthesis and Properties of Optimally Value-Aligned Normative Systems
Authors: Nieves Montes, Carles Sierra
JAIR 2022 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We illustrate all of our contributions with a running example of a society of agents where taxes are collected and redistributed according to a set of parametrised norms. |
| Researcher Affiliation | Academia | Nieves Montes EMAIL Carles Sierra EMAIL Artificial Intelligence Research Institute (IIIA-CSIC) Campus UAB Carrer de Can Planas, Zona 2 08193 Bellaterra, Barcelona |
| Pseudocode | Yes | Algorithm 1: Brute-force approach to check the monotonic behaviour of an optimal normative system. |
| Open Source Code | Yes | All the code to go along with this work has been integrally developed in Python 3. It is available under an MIT license at https://github.com/nmontesg/aamas21 and as an online appendix. |
| Open Datasets | No | The paper describes how the data for the simulation is generated within the model, it does not use or provide access to a separate, pre-existing open dataset. For example: "In this society, a set of technologically-enabled agents are endowed with some initial wealth. To facilitate the exchange of resources, a common fund is set up." |
| Dataset Splits | No | The paper describes simulation sampling, not dataset splits for training/testing. For example: "To compute the expected values for the alignment, we again perform Monte-Carlo sampling with a sample of 500 paths of 10 transitions each." |
| Hardware Specification | No | The paper does not provide specific hardware details such as CPU/GPU models, memory, or specific computing infrastructure used for running the simulations. It only mentions that the code is in Python 3. |
| Software Dependencies | No | The paper states: "All the code to go along with this work has been integrally developed in Python 3." This only specifies the programming language and version, not specific libraries or solvers with their versions. |
| Experiment Setup | Yes | Table 5: Hyperparameters of the Genetic Algorithm to find the optimally value-aligned normative systems. The first six refer to hyperparameters of the GA itself, and are covered in Section 5. The latter two refer to the Monte Carlo sampling: number of state transitions per path and total amount of paths sampled to compute the alignment. Hyperparameter Value Population size 100 p (intermediate recombination) 0.25 k (elitism) 5 Maximum total iterations 500 Maximum partial iterations 50 Fitness threshold 0.9 Path length 10 Path sample size 500 |