Modeling of moral decisions with deep learning

Area of Science:

• Artificial Intelligence
• Computational Ethics
• Machine Learning

Background:

• Autonomous vehicle ethical dilemmas, exemplified by the Moral Machine Experiment, pose significant challenges.
• Hierarchical Bayesian (HB) models have been previously employed to address moral decision-making.
• Advancements in machine learning offer new approaches to modeling complex human behaviors.

Purpose of the Study:

• To apply a deep learning method for modeling human ethics in the context of artificial intelligence dilemmas.
• To compare the performance of deep learning models against the established hierarchical Bayesian approach.
• To evaluate the effectiveness of these models in predicting moral decisions of simulated populations.

Main Methods:

• Utilized a deep neural network approach to learn ethical patterns from observed data.
• Employed simulated populations of Moral Machine participants for testing predictive capabilities.
• Compared deep learning model predictions with those generated by the hierarchical Bayesian model.

Main Results:

• Deep neural networks demonstrated effectiveness in learning population-level morality through observational data.
• The deep learning approach outperformed the hierarchical Bayesian model, particularly in scenarios with model mismatches.
• Both methods were tested for their ability to predict moral choices in simulated populations.

Conclusions:

• Deep learning models show significant promise for understanding and predicting group morality in AI ethical contexts.
• Deep neural networks offer a powerful alternative to traditional statistical methods for modeling complex ethical decision-making.
• Further research can leverage these findings to develop more ethically aligned AI systems.