Moral Machine or Tyranny of the Majority?

Authors: Michael Feffer, Hoda Heidari, Zachary C. Lipton

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical In this paper, we analyze the mechanism of averaging preference vectors across a population, focusing on stability, strategic effects, and implications for fairness. To clarify the fairness properties of the mechanism, we concentrate our analysis on an stylized setting in which within-group preferences are homogeneous. We emphasize this assumption is only meant for simplifying the analysis and revealing fundamental limitations of the simple preference vector averaging mechanism. The problems we identify here do not simply disappear in more complicated settings with within-group variation in preferences. Moreover, to isolate the role of the aggregation mechanism, we assume participants can directly report their preference vectors. With these assumptions in place, our analysis makes the following key observations: (i) even when preferences are reported truthfully, the fraction of cases where the minority prevails is sub-proportionate (i.e., less than α); (ii) the degree of subproportionality grows more severe when the divergence between the two groups preference vectors is large; (iii) as with most averaging-based approaches, this mechanism is not strategy-proof; (iv) whenever a pure strategy equilibrium exists, the majority group prevails on 100% of cases; and (v) last but not least, while other stable and incentivecompatible mechanisms do exist (e.g., the randomized dictatorship model), they come with other fundamental shortcomings.
Researcher Affiliation Academia Michael Feffer, Hoda Heidari*, Zachary C. Lipton* Carnegie Mellon University {mfeffer,hheidari,zlipton}@andrew.cmu.edu
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks. It focuses on mathematical proofs and theoretical analysis.
Open Source Code No The paper does not provide concrete access to source code (no specific repository link, explicit code release statement, or mention of code in supplementary materials) for the methodology described. The paper is theoretical in nature.
Open Datasets No The paper refers to the 'Moral Machine dataset' (Awad et al. 2018) as background for the problem, but its own analysis is based on a 'stylized setting' and theoretical model, not on an empirical dataset. Therefore, it does not provide access information for a dataset used in its own research.
Dataset Splits No The paper is theoretical and does not involve empirical experiments with datasets that would require training, validation, and test splits.
Hardware Specification No The paper is theoretical and does not describe empirical experiments that would require specific hardware specifications.
Software Dependencies No The paper is theoretical and does not mention any specific software dependencies with version numbers for its analysis or models.
Experiment Setup No The paper is theoretical and does not describe an empirical experimental setup with hyperparameters or system-level training settings.