National Security and Open Weight Models
Risks and Benefits of Open Foundation Models
Center for Strategic and International Studies
Remarks of Alan Davidson
Assistant Secretary of Commerce for Communications and Information
National Telecommunications and Information Administration
Washington, D.C.
March 21, 2024
As prepared for delivery
Thank you, Greg for that introduction and for moderating the upcoming panel. And thank you to CSIS – and particularly Jim Lewis, who I got to know and admire when he was at the Commerce Department, Georgia Adamson and the rest of the team here – for all you have done to make this event possible.
At NTIA, tech policy is what we do. Our focus is making sure important technologies – from broadband to spectrum to emerging innovations like AI – are developed in the service of people and progress.
Nowhere is that more important than in the explosive growth of artificial intelligence systems.
Responsible AI innovation can bring enormous benefits to people. It is going to transform every corner of our economy, from advances in medicine to precision agriculture.
But we will only realize the promise of AI if we also address the serious risks it raises today.
The Biden-Harris Administration is moving with urgency to engage on these issues.
President Biden’s AI Executive Order, the most significant government action to date on AI, brings the full capabilities of the U.S. government to bear in promoting innovation and trust in AI. The Commerce Department is playing a leading role in the Administration’s AI work.
And we at NTIA, are doing our part.
We are here today because the EO gave us an important homework assignment: Weighing the benefits and risks of dual use foundation models with widely available model weights. These “open-weight models” – advanced AI models with key parameters made openly available – raise serious policy questions.
We have heard already about the potential marginal benefits offered by these open-weight models: Benefits for increased competition and innovation, improved research capacity and perhaps greater security.
We have also heard that this type of openness can create marginal new risks, including hindering efforts to control risks from misuse of these powerful new systems. It could also make it more difficult to hold to account those who would use these systems for harm.
To better understand the landscape of these difficult policy issues, we are seeking broad input. Just last month we put out a public request for comment.
One thing we have already learned is the importance of focusing on the marginal or differential risks and benefits of open weights. For example, we need to measure the risks of open-weight models relative to the risks that already exist today from widely-available information, or from closed models.
We have also been encouraged to hear that this is not a binary choice of “open” vs. “closed.” Rather there is a broader “gradient of openness” that we need to consider and that may offer broader options for policy.
I hope today’s conversation will dive into some of these questions. We are particularly interested in hearing about the international implications of these powerful systems, and the national security considerations raised by widely available model weights.
As we note in the Request for Comment, we want input on:
- What do you see as the marginal risks and benefits associated with widely available model weights? How should we think about those risks and benefits from an international perspective?
- How might foreign adversaries use open-weight models to exacerbate security risks? How do those marginal risks compare with closed models?
- In what ways could open-weight models support U.S. national security interests?
As you can see, we are digging into this topic – and we have a tight deadline for doing so. We hope you all will participate in our comment proceeding – due by March 27.
To close, let me once again say we are grateful to CSIS for hosting today, and to all of you for your continued partnership and engagement.
This is an important moment, and important work. NTIA is happy to have this assignment. We look forward to delivering policy recommendations to the President that protect America’s security while promoting innovation and competition in new AI systems.
Thank you.