ALLIANCE BETWEEN THE US AND THE UK TO IMPROVE AI SAFETY TESTING AND COOPERATION (GS Paper 3- Science and Technology)
Why in news?
- The UK and the US have formally committed to working together to create advanced artificial intelligence (AI) model tests.
- This agreement intends to improve AI safety and security and is a result of promises made at the Bletchley Park AI Safety Summit.
Information Exchange:
- Mutual exchange of crucial information on AI capabilities and risks is paramount, alongside joint efforts in foundational technical AI safety and security research.
- This partnership also entails harmonizing strategies for the secure implementation of AI systems.
Immediate Operationalization:
- The collaboration will be promptly put into action, facilitating seamless coordination between the two nations.
- Given the rapid proliferation of AI systems, this collaboration is indispensable, given the array of opportunities and substantial risks they pose to societal structures.
Synchronized Scientific Approaches:
- The US and UK will closely collaborate to synchronize their scientific methodologies, expediting the development of robust evaluation frameworks for AI models, systems, and agents.
Unified Approach:
- Both countries aim to establish a unified approach to AI safety testing and leverage their capabilities to effectively manage associated risks.
- This involves conducting joint testing exercises on publicly accessible models and exploring personnel exchanges between their respective AI Safety Institutes.
US Consultation on Open-Source AI Models:
Impact of Executive Order:
- Following the executive order on AI system deployment by the Biden administration, the National Telecommunications and Information Administration (NTIA) in the US has initiated consultations on risks, benefits, and potential policies concerning widely available foundation models.
Industry Insights:
- Meta and Open AI, leading AI developers, have offered perspectives on open-source AI models.
- Meta underscores the significance of open-source innovation, while OpenAI emphasizes the advantages of releasing models through APIs and commercial products to enable ongoing study and risk mitigation.
