Safety Cases: How to Justify the Safety of Advanced AI Systems
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2024
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866913270251651072 |
|---|---|
| author | Clymer, Joshua Gabrieli, Nick Krueger, David Larsen, Thomas |
| author_facet | Clymer, Joshua Gabrieli, Nick Krueger, David Larsen, Thomas |
| contents | As AI systems become more advanced, companies and regulators will make difficult decisions about whether it is safe to train and deploy them. To prepare for these decisions, we investigate how developers could make a 'safety case,' which is a structured rationale that AI systems are unlikely to cause a catastrophe. We propose a framework for organizing a safety case and discuss four categories of arguments to justify safety: total inability to cause a catastrophe, sufficiently strong control measures, trustworthiness despite capability to cause harm, and -- if AI systems become much more powerful -- deference to credible AI advisors. We evaluate concrete examples of arguments in each category and outline how arguments could be combined to justify that AI systems are safe to deploy. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2403_10462 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Safety Cases: How to Justify the Safety of Advanced AI Systems Clymer, Joshua Gabrieli, Nick Krueger, David Larsen, Thomas Computers and Society Artificial Intelligence As AI systems become more advanced, companies and regulators will make difficult decisions about whether it is safe to train and deploy them. To prepare for these decisions, we investigate how developers could make a 'safety case,' which is a structured rationale that AI systems are unlikely to cause a catastrophe. We propose a framework for organizing a safety case and discuss four categories of arguments to justify safety: total inability to cause a catastrophe, sufficiently strong control measures, trustworthiness despite capability to cause harm, and -- if AI systems become much more powerful -- deference to credible AI advisors. We evaluate concrete examples of arguments in each category and outline how arguments could be combined to justify that AI systems are safe to deploy. |
| title | Safety Cases: How to Justify the Safety of Advanced AI Systems |
| topic | Computers and Society Artificial Intelligence |
| url | https://arxiv.org/abs/2403.10462 |