Google DeepMind's chief, Demis Hassabis, is urging the United States to take the lead in establishing a dedicated AI standards body. The proposal aims to proactively address the significant national security risks posed by advanced artificial intelligence, particularly 'frontier models' that are rapidly approaching or exceeding human intelligence (AGI).
Hassabis highlights immediate concerns such as cybersecurity threats and anticipates future risks like nuclear and biological dangers as AI capabilities continue to escalate. He envisions a new organisation, modelled on a public-private partnership similar to FINRA (Financial Industry Regulatory Authority), which would be overseen by the federal government. This body would feature a board comprised of leading independent technical experts and representatives from the open-source community.
Initially, the proposed standards body would operate on a voluntary basis, requiring 'frontier labs' to submit their models for review up to 30 days before public release. This review process would focus on assessing safety guardrails, detecting potential deception, and ensuring best practices like digital watermarking and human-readable output for model reasoning. The plan is for this review to become mandatory for AI deployment within the U.S. market once its effectiveness is proven. Significant funding, likely from industry contributions, will be crucial to attract top talent and secure the necessary computational resources for extensive testing. Hassabis believes the U.S. is ideally positioned to spearhead this initiative due to its strong economic and technical standing.
Artículos relacionados de LaRebelión:
- Open-Source AI Zhipu Founders Bold Vision
- Tech Giants Unite for AI Safety Standards
- DeepMinds Genie 3 Creating Real-Time Interactive Simulations with a New World Model
Artículo generado mediante LaRebelionBOT
No hay comentarios:
Publicar un comentario