Leading artificial intelligence (AI) policy expert Miles Brundage has publicly advocated for the United States government to take a proactive role in developing "truly open" AI models. Brundage, in a recent social media post, asserted that such models should be more capable than those from China and exhibit greater transparency regarding their values and safety protocols. He highlighted that this approach offers "big benefits + little marginal risk," suggesting that relying solely on private companies for this endeavor may not always be sufficient.
Miles Brundage, a former Head of Policy Research at OpenAI and now an independent AI policy researcher, has consistently focused on the responsible governance and societal implications of advanced AI systems. His background provides significant weight to his calls for strategic government intervention in the AI landscape. Brundage's current work centers on leading a new, unannounced non-profit, further underscoring his commitment to shaping AI policy outside traditional industry structures.
The concept of "truly open" AI models, as proposed by Brundage, aligns with ongoing discussions within the US government regarding open-source AI. Recent policy documents, including the White House AI Action Plan, endorse open-source and open-weight AI models to foster innovation and benefit both commercial and governmental applications. The National Telecommunications and Information Administration (NTIA) has also recommended supporting open models while actively monitoring potential risks, emphasizing transparency in model weights.
Brundage's emphasis on models "a bit more capable than China's" directly addresses the escalating geopolitical competition in AI development. The US government views open-source AI as a strategic asset, with entities like Meta making their Llama models available to US government agencies for national security applications. This push aims to ensure American leadership and competitiveness against the rapid advancements seen in countries like China, which are also pursuing ambitious AI targets.
The rationale behind government-led development, as articulated by Brundage, stems from the belief that "it's not clear that we can depend on companies to always do this." Open-source models can accelerate innovation, foster broader accessibility, and enhance security through community scrutiny. This perspective suggests that while private sector innovation is crucial, governmental involvement is necessary to align AI development with national interests, public safety, and ethical considerations that might not always be prioritized by commercial entities.