MIT CSAIL AI Action Plan Recommendations – 2025

Note: IPRI was glad to work with CSAIL director Daniela Rus and CSAIL researchers (PIs) to develop these comments.

Executive Summary

MIT CSAIL has been pioneering the future of computing and AI for over six decades. This is CSAIL’s lab-wide response to the US government’s Request for Information on the Development of an Artificial Intelligence (AI) Action Plan.

The United States is well-positioned to lead the world in AI innovation but must act to build on the organic advantages we have in basic science, a thriving technology industry, and a society infused with stable laws and democratic values. We offer the following recommendations based on our analysis of the underlying computer science challenges, market conditions and the public policy context in which AI systems operate. Our submission focuses on five key areas:

1. 1. Support for basic research in AI is essential, as we are still in the early stages of AI technology development and application. Investing in fundamental AI research will drive the next wave of breakthroughs, ensuring that the U.S. remains at the forefront of innovation, scientific discovery, and real-world AI integration. For all the excitement around this generation of Large Language Models, the rate of growth in performance and efficiency is slowing down. In order to continue to make foundational advances in AI architectures we need ongoing 6.1 research support from agencies like the NSF and DoD. Any reduction in these funds, particularly in the face of proposed tax cuts, would severely undermine the U.S.’s ability to maintain its competitive edge and secure its future in AI.
   2. Set a goal of achieving Artificial Super Intelligence (ASI) to reach beyond the limitations of current Large Language Models (LLMs) by integrating real-world awareness, common-sense reasoning, agentic computation, and physical intelligence. Investing in these advancements will ensure that AI systems evolve from statistical pattern recognition to true autonomous reasoning and physical-world interaction, unlocking transformative capabilities across industries and scientific disciplines. Achieving this vision will also require AI hardware innovation, with a co-design design approach where AI algorithms and hardware evolve together for optimizing performance. As AI adoption expands, innovations in low power chips and edge computing will be crucial for real-time applications.
   3. Increase investment in AI applications for scientific discovery to spur greater understanding of how to apply powerful AI tools to a wide range of society’s needs. Advancing AI-driven research will enable the development of high-accuracy, high-efficiency models that can accelerate breakthroughs in physics, biology, chemistry, medicine, and engineering. The research community has a vital role here to pioneer new uses, proving their value and derisking them for market application.
   4. Maintain a stable regulatory framework that ensures reliability, security, and legal compliance, leveraging existing sectoral regulations unless they prove insufficient for AI-specific risks. Many high-value, high-sensitivity applications of AI such as healthcare, transportation, and financial services already have stable regulatory requirements that apply to services both offered with or without AI. We must ensure that there are adequate resources to enforce these laws to maintain public confidence in Ai technologies. A key research priority is to build technical tools – privacy-enhancing technologies, resilience and safety metrics, and explainability techniques that make it easier for service providers to use AI while adhering to legal requirements and give the public confidence that these innovative new services are reliable.
   5. Invest in robust job transition and retraining programs to ensure that workers displaced by AI can continue contributing meaningfully to the economy. Expanding access to AI-driven workforce development initiatives—including reskilling in emerging technology fields, automation management, and digital economy jobs—will help mitigate disruptions and create new opportunities. Additionally, strengthening mathematics and computer science education at the high school level is critical to preparing the next generation for an AI-driven future, ensuring a highly skilled workforce that can innovate, adapt, and compete on a global scale. While AI job displacement will be less rapid than doomsayers predict, it will nevertheless represent a substantial shift to the tasks being done by U.S. workers and will lead to job losses for some. Integrating AI into work has the potential to help workers if they can leverage new tools to make themselves more productive, and to meaningfully hurt their earning ability if they don’t. Research should be pursued to identify which skills and professions will be most advantageous for workers as AI automation proceeds.

The full submission is available as a PDF at:
https://internetpolicy.mit.edu/reports/2025/MIT-CSAIL-AI-USG-RFI-2025-Final-Web.pdf