allen-institute-for-ai
allen-institute-for-ai - organisation named in the AI Incident Database harm index.
What changed
A publicly accessible research model that was trained via Reddit threads showed racially biased advice on moral dilemmas, allegedly demonstrating limitations of language-based models trained on moral judgments.
Timeline
- 12 Dec 2021Research Prototype AI, Delphi, Reportedly Gave Racially Biased Answers on EthicsIncident82%
PSF mapping
No PSF domains are mapped to this entity in the current public record.