How can we build AI systems that can learn any set of individual human values both quickly and safely, avoiding causing harm or violating societal standards for acceptable behavior during the learning process? We explore the effects of representational alignment between humans and AI agents on learning human values.

In this work, we propose anefficient,gradient-based algorithm tometa-learn PThyperparameters.

An important aspect of the abovedefinition is the notion of neighboring or adjacent datasets.

The results of our survey are shown in Table 10.