Deep Ganguli
Coverage of Deep Ganguli in the Nexus archive.
- Anthropic's moral compass architect suggested AI overcorrection could address historical injustices
Anthropic's philosopher Amanda Askell proposed that AI models could intentionally overcorrect stereotypes to address historical injustices, as outlined in a 2023 paper. The experiment showed varying discrimination rates against Black students based on training conditions, highlighting ethical challenges in AI alignment. Anthropic faces scrutiny over its ethical AI claims amid controversies like withholding its Mythos model and clashes with the Department of War.
- Anthropic's moral compass architect suggested AI overcorrection could address historical injustices
Anthropic's AI philosophy lead, Amanda Askell, proposed that AI models could intentionally overcorrect for historical injustices by adjusting outputs based on human input. A 2023 paper highlighted experiments showing models trained with human corrections reduced racial bias in student evaluations. Anthropic faces ethical scrutiny amid debates over AI's societal impact and recent controversies over restricting its technology.