Safeguarding the safeguards: How best to promote AI alignment in the public interest
| Autor: | Guest, Oliver, Aird, Michael, hÉigeartaigh, Seán Ó |
|---|---|
| Rok vydání: | 2023 |
| Předmět: | |
| Druh dokumentu: | Working Paper |
| Popis: | AI alignment work is important from both a commercial and a safety lens. With this paper, we aim to help actors who support alignment efforts to make these efforts as effective as possible, and to avoid potential adverse effects. We begin by suggesting that institutions that are trying to act in the public interest (such as governments) should aim to support specifically alignment work that reduces accident or misuse risks. We then describe four problems which might cause alignment efforts to be counterproductive, increasing large-scale AI risks. We suggest mitigations for each problem. Finally, we make a broader recommendation that institutions trying to act in the public interest should think systematically about how to make their alignment efforts as effective, and as likely to be beneficial, as possible. Comment: Update Dec-15: Added a missing acknowledgement and fixed minor formatting errors |
| Databáze: | arXiv |
| Externí odkaz: |
|