Value Alignment or Misalignment -- What Will Keep Systems Accountable?

Thomas Arnold and Daniel Kasenberg and Matthias Scheutz

We present here a critique of a turn toward inverse reinforcement-learning as a way to guarantee ethical AI systems. We argue that a hybrid architecture, which can represent norms explicitly, is necessary for basic social coordination and moral reasoning between people and AI systems.

@inproceedings{arnoldetal17aiethics,
  title={Value Alignment or Misalignment -- What Will Keep Systems Accountable?},
  author={Thomas Arnold and Daniel Kasenberg and Matthias Scheutz},
  year={2017},
  booktitle={AAAI Workshop on AI, Ethics, and Society},
  url={https://hrilab.tufts.edu/publications/arnoldetal17aiethics.pdf}
}