Private sector companies are asking the federal government to prioritize diverse investments across technology areas that inform the development of U.S. artificial intelligence systems, along with an effort to take the lead on standards development and to pursue industry-government partnerships.
In comments sent to the Office of Science and Technology Policy and Networking and Information Technology Research and Development National Coordination Office, industry leaders in AI innovation — such as Google, IBM, Anthropic and Amazon — told policymakers what they would like to see in the forthcoming 2025 National AI R&D Strategic Plan.
The four companies touched on several common themes, most prominent of which was the need for sustained federal investment in the myriad science and technology efforts related to AI systems. For Google, this means continued investment in algorithm and chip design, scaling advanced manufacturing infrastructure and workforce development.
“The new Strategic Plan should launch a comprehensive initiative to educate America’s science students and scientific workforce on new AI technologies, supporting the uptake of AI as the next scientific instrument,” Google wrote in its comments.
Google also advocated for the U.S. to take a strong role in cultivating the standards by which emergent AI systems will operate and be evaluated. It said that engagement with industry partners and federal investment could help ensure U.S.-led AI standards are adopted worldwide.
“These are vital for consistent assessment, preventing race-to-the-boom dynamics and promoting adoption of U.S. technology,” Google’s letter read.
Amazon agreed with Google’s stances, emphasizing the importance of each layer of the AI technology stack. The company also noted that its extensive collaborations with academia to innovate in AI, quantum computing, robotics, cloud computing software, semiconductors and more each need to see continued federal support.
“The federal government plays a critical role in AI research and development, and can take steps to ensure that scientific advancement continues to advance at the pace necessary to maintain U.S. leadership on AI,” Amazon’s letter read.
Amazon also echoed Google’s push for strong U.S. government engagement in standards development with the help of industry partners.
“This public-private cooperation promotes standards that are both practical and balanced, incorporating interested stakeholder perspectives while supporting U.S. technological leadership,” the letter said. “Federal agencies should ensure that international standards address their key operational priorities, which may include security, testing, metrics, and physical safety, while remaining flexible enough to accommodate rapid technological advancement and promote innovation.
IBM concurred with Google and Amazon’s stance of continuing federal support for AI-related research and development, specifically recommending the government fund research into smaller, open source, task-specific models and their cost-benefit tradeoffs against larger generalized models to inform usage in federal missions.
IBM also highlighted the need for the federal government to prioritize more niche AI model architectures as opposed to more general transformer models, or the machine learning foundation models underpinning leading AI platforms like ChatGPT.
“In order to ensure American dominance of AI well into the coming decades, the federal government should prioritize research beyond transformer-based architectures, which, while powerful, present challenges in areas such as explainability, energy efficiency, and hallucination risk,” IBM’s comment said. “Alternative approaches — such as models inspired by neuroscience that incorporate memory hierarchies and dynamic control — offer promising solutions.”
IBM echoed the need for government involvement in defining industry standards and proliferating them abroad. The company also notably recommended excluding third party audits from those standards, which the company asserts can create burdensome regulatory bottlenecks in a given AI model’s early-stage development. It criticized legislative efforts that seek to incorporate this approach to AI risk mitigation, recommending that the government focus on testing in adversarial and worst-case scenarios in post-training AI model environments.
“While some legislative efforts have proposed to address these risks through third-party auditing and validation requirements … we continue to urge policymakers to address risks presented by specific applications of AI,” IBM wrote to OSTP and the National Science Foundation. “AI regulation must be contextualized — what works in one domain may not transfer to another.”
For Anthropic, the most important thing the government can do is continue federal funding for research, particularly for work in mechanistic interpretability — how AI models work through reverse engineering — and leveraging AI as a tool for scientific discovery, echoing Google’s posture on establishing AI as the major scientific instrument.
Anthropic also specified that critical AI research needs to be democratized through partnerships between the public and private sector, as well as academia, and asked the NSF to continue working with academia and nonprofits.
“It’s important that this work is not confined to the private sector and that the government and academia both take a role in developing and advancing the field of mechanistic interpretability,” Anthropic’s comments read. “These investments would help ensure that AI development proceeds with transparency and accountability, and build the foundations for responsible governance of increasingly powerful systems.”
Ensuring the federal government maintains strong support of critical research projects comes following the widespread cancellation of multiple NSF-led grants by the Trump administration. Some of the grants focused on AI, including projects meant to cultivate and expand access to AI education.
