valalgn 0.0.4

ValAlgn: Value Alignment

Description

Welcome to the documentation of the Value-Alignment (valalgn) package. This package provides tools and functionalities for computing the alignment of a normative system with respect to a value (or a set of values). The package provides two modules:

1. valalgn.sampling computes the value alignment of a set of norms with respect to a value using simulation and approximate methods. It also includes functionalities for value-driven optimization of norms and evaluation metrics [1].

2. valalgn.asl implements the Action Situation Language (ASL) and interprets a set of norms written in this language and builds their operational semantics as an extensive-form game. This game can then be analyzed using standard game theoretical tools and solution concepts to derive the alignment of the norms that generated it [2].

References

[1]

Montes, N., & Sierra, C. (2022). Synthesis and properties of optimally value-aligned normative systems. Journal of Artificial Intelligence Research, 74, 1739–1774. https://doi.org/10.1613/jair.1

[2]

Montes, N., Osman, N., & Sierra, C. (2022a). A computational model of Ostrom’s Institutional Analysis and Development framework. Artificial Intelligence, 311, 103756. https://doi.org/10.1016/j.artint.2022.103756

Credits

Nieves Montes (IIIA-CSIC)