These “open-weight” models allow developers to build upon and adapt previous work, broadening AI tools’ availability to small companies, researchers, nonprofits, and individuals. This may accelerate the diffusion of AI’s benefits and the pace of AI safety research, but it may also increase the scale and likelihood of harms from advanced models.
President Biden’s Executive Order on Artificial Intelligence directs NTIA to review the risks and benefits of large AI models with widely available weights and develop policy recommendations to maximize those benefits while mitigating the risks.
The Executive Order directs NTIA to discuss benefits, risks, and policy choices associated with dual-use foundation models, which are powerful models that can be fine-tuned and used for multiple purposes, with widely available model weights. The Request for Comment asks for public feedback about how making model weights and other model components widely available creates benefits or risks to the broader economy, communities, individuals, and to national security.
Model weights reflect distillations of knowledge within AI models and govern how those models behave. Using large amounts of data, machine learning algorithms train a model to recognize patterns and learn appropriate responses. As the model learns, the values of its weights adjust over time to reflect its new knowledge. Ultimately, the training process aims to arrive at a set of weights optimized to produce behavior that fits the developer’s goals. If a person has access to a model’s weights, that person does not need to train the model from scratch. Additionally, that person can more easily fine-tune the model or adapt it towards different goals, unlocking new innovations but also potentially removing safeguards.
The Request for Comment seeks input on a number of issues, including:
- The varying levels of openness of AI models;
- The benefits and risks of making model weights widely available compared to the benefits and risks associated with closed models;
- Innovation, competition, safety, security, trustworthiness, equity, and national security concerns with making AI model weights more or less open; and
- The role of the U.S. government in guiding, supporting, or restricting the availability of AI model weights.
Comments are due within 30 days of publication of the Request for Comment in the Federal Register. The responses will help inform a report to the President with NTIA’s findings and policy recommendations.