Aligning AI With Shared Human Values
arXiv:2008.02275
Abstract
We show how to assess a language model's knowledge of basic concepts of morality. We introduce the ETHICS dataset, a new benchmark that spans concepts in justice, well-being, duties, virtues, and commonsense morality. Models predict widespread moral judgments about diverse text scenarios. This requires connecting physical and social world knowledge to value judgements, a capability that may enable us to steer chatbot outputs or eventually regularize open-ended reinforcement learning agents. With the ETHICS dataset, we find that current language models have a promising but incomplete ability to predict basic human ethical judgements. Our work shows that progress can be made on machine ethics today, and it provides a steppingstone toward AI that is aligned with human values.
arXiv commentICLR 2021; the ETHICS dataset is available at https://github.com/hendrycks/ethics/
References in corpus (15)
- Gaussian Error Linear Units (GELUs)
- HuggingFace's Transformers: State-of-the-art Natural Language Processing
- Language Models are Few-Shot Learners
- Longformer: The Long-Document Transformer
- Equality of Opportunity in Supervised Learning
- A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks
- Deep reinforcement learning from human preferences
- The Measure and Mismeasure of Fairness
- Reformer: The Efficient Transformer
- Fairness Through Computationally-Bounded Awareness
- Online Learning with an Unknown Fairness Metric
- Fairness Through Awareness
- Probably Approximately Metric-Fair Learning
- Human Instruction-Following with Deep Reinforcement Learning via Transfer-Learning from Text
- Interactive Fiction Games: A Colossal Adventure
Cited by in corpus (11)
- Large AI Models in Health Informatics: Applications, Challenges, and the Future
- Natural Adversarial Examples
- On Transferability of Prompt Tuning for Natural Language Processing
- Value Kaleidoscope: Engaging AI with Pluralistic Human Values, Rights, and Duties
- ValueNet: A New Dataset for Human Value Driven Dialogue System
- Natural Language Processing with Commonsense Knowledge: A Survey
- Macro Ethics Principles for Responsible AI Systems: Taxonomy and Future Directions
- What Would Jiminy Cricket Do? Towards Agents That Behave Morally
- TAPE: Assessing Few-shot Russian Language Understanding
- Mini Minds: Exploring Bebeshka and Zlata Baby Models
- The Goofus & Gallant Story Corpus for Practical Value Alignment