AI Safety Progress Loud, But Continued Vigilance Key As Risks Linger
-
AI safety has improved in recent years through increased interpretability of models, development of safety plans by major AI players, and new government regulations. However, risks remain from uncontrolled deployment and future potential for addiction/manipulation.
-
Companies like Anthropic and OpenAI have published frameworks to manage catastrophic risks from advanced AI, with strict safety levels as capabilities increase towards human-level AI.
-
Governments like the EU and US have introduced or proposed new laws and principles to ensure AI systems are trustworthy, with a focus on high-risk use cases.
-
AI safety is starting to be recognized in some sustainability frameworks, but still has limited explicit coverage. Investors have an important role to play in leading on AI safety.
-
As AI capabilities rapidly advance, frameworks and norms need to continually evolve to address new issues like addiction and viral manipulation that could emerge.