Learning and Obeying Conflicting Norms in Stochastic Domains

Daniel Kasenberg

Artificial agents will need to be able to reason about and obey human moral and social norms. Additionally, agents must be able to learn norms, both from instruction (eg, by natural language interaction with humans) and by observing the behavior of humans or other agents. This is necessary because (1) humans have many moral and social norms, perhaps too many to be able to pre-program; and (2) moral and social norms may vary between cultures, and across time.

@inproceedings{kasenberg18aiesdc,
  title={Learning and Obeying Conflicting Norms in Stochastic Domains},
  author={Daniel Kasenberg},
  year={2018},
  booktitle={Proceedings of the Student Program, 1st AAAI/ACM Conference on Artificial Intelligence, Ethics, and Society},
  url={https://hrilab.tufts.edu/publications/kasenberg18aiesdc.pdf}
  doi={10.1145/3278721.3278791}
}