Visual Computing for Industry, Biomedicine, and Art 2020
Moral Decisions
Modeling of Moral Decisions with Deep Learning
Current or former lab member. Select a name to see all of their projects.
Abstract
MIT’s Moral Machine Experiment posed the autonomous-vehicle dilemma to a large public, and its results were modeled with a hierarchical Bayesian approach. This paper applies deep learning to the same problem, learning human ethics from observed decisions, and compares it with the Bayesian model. Both are tested on predicting the decisions of simulated populations of participants.
Neural networks proved effective at learning group morality through observation, and they outperformed the Bayesian model when its assumptions did not match the population.
What’s in the repository
- The generated moral principle vectors used in the tests, as MATLAB files
- Training and test sets for an abstract scenario and for the autonomous-vehicle scenario
Citation
@article{wiedeman2020modeling,
title = {Modeling of Moral Decisions with Deep Learning},
author = {Wiedeman, Christopher and Wang, Ge and Kruger, Uwe},
journal = {Visual Computing for Industry, Biomedicine, and Art},
volume = {3},
pages = {27},
year = {2020},
doi = {10.1186/s42492-020-00063-9}
}