OpenAI wants US to lead global AI standards

OpenAI is urging the United States to spearhead a global initiative to establish technical standards for advanced artificial intelligence (AI), particularly focusing on highly autonomous systems that can improve themselves through recursive self-improvement (RSI). In a blog post detailing its aspirations for future AI development, OpenAI stated that automated AI research could expedite the pace of AI innovation considerably.

However, the company cautioned against fully autonomous RSI unless humans can maintain a level of control over the process. RSI pertains to AI systems gradually assuming more responsibilities in developing subsequent generations of AI, potentially accelerating AI progress significantly. OpenAI believes the advantages of automated AI research include reduced costs for advanced intelligence, accelerated scientific discoveries, and enhanced AI safety research.

The company also noted that an autonomous AI researcher could potentially evolve into an autonomous AI safety researcher. While OpenAI sees potential benefits, it acknowledged the increasing risks associated with AI systems developing their own capabilities without proper safeguards, such as potential loss of human control and reduced alignment between AI outcomes and human values.

OpenAI emphasized the need for common AI safety standards and believes the United States, with its robust AI industry and global influence, is ideally positioned to spearhead this effort. The company advocates for the development of shared safety benchmarks, risk assessment methodologies, and guidelines for human oversight of automated AI research.

OpenAI proposes leveraging existing networks of national AI safety institutes across countries like India, the UK, Japan, South Korea, Singapore, France, Germany, Canada, Australia, and Kenya to collaboratively establish these global standards. The proposed standards would focus on AI capability benchmarks, risk evaluation, human oversight of automated AI research, and incident reporting.

OpenAI stressed that these standards would not be prescriptive or mandatory approval systems, allowing individual governments to determine how to integrate them into their respective legal frameworks. The company also called for inclusive participation from diverse stakeholders, including open-model and closed-model developers, independent researchers, and academia, ensuring that the standards do not favor specific companies or nations.

Additionally, OpenAI highlighted the importance of secure channels of communication between governments and critical infrastructure operators to share information on AI-related national security concerns, vulnerabilities, and emerging threats. OpenAI framed AI development as a matter of pacing rather than setting a fixed speed, emphasizing the necessity of ensuring alignment research and safety measures remain ahead of AI capabilities.

The company believes that leading in this endeavor could position the United States as a central player in shaping a coherent global AI framework, as opposed to witnessing a fragmented and conflict-ridden system emerge.