AI Model Registries: A Foundational Tool for AI Governance

post by Elliot Mckernon (elliot), Deric Cheng (deric-cheng), Gwyn Glasser (gwyn-g) · 2024-10-07T19:27:43.466Z · LW · GW · 1 comments

This is a link post for https://www.convergenceanalysis.org/research/ai-model-registries-a-foundational-tool-for-ai-governance

Contents

  Executive Summary of Our Proposal
    The Case for a Model Registry
    Proposed Design of a Model Registry
    Proposed Implementation of a Model Registry
    Structure of the full report
  Conclusion
None
1 comment

This post consists of the Executive Summary and Conclusion of our proposal for model registries as a foundational tool for AI governance, authored by Elliot McKernon, Gwyn Glasser, Deric Cheng, and Gillian Hadfield. The full report is available through the link above.

Executive Summary of Our Proposal

In this report, we propose that national governments should implement AI model registries as a foundational tool for AI governance. By model registry, we mean a centralized database of frontier AI models that includes standard commercial and specific safety-relevant information about these models and their deployers. Developers would be required to report any qualifying models and their information to the registry before public deployment. 

In Parts I, II, and III, we explore and make recommendations on the purpose of such a registry, what information it should store, and how to practically implement and administer it, respectively. 

In this executive summary of our proposal we provide a concise, high-level summary of each of our conclusions, without argumentation, analysis, or evidence. To understand why we make each of these specific recommendations, we encourage readers to read the full section on each topic. 

The Case for a Model Registry

AI model registries can serve as a foundational lever to increase regulatory visibility, support legal action, and manage societal risks.  In other industries, registries successfully serve this same purpose for products and services associated with notable economic impacts or risks to society, as we detail in Registries are a basic, common governance tool.  However, while some nations are taking early steps to develop model registries, as we detail in What AI model registries currently exist?, the current standards for frontier AI registration are not yet substantial enough to bring AI oversight into parity with other industries.

We identify four high-level objectives that motivate the adoption of frontier model registries:

• A registry will facilitate the monitoring of frontier AI technology, providing governments with increased regulatory visibility into the capabilities and risks of leading AI models.
• A registry will provide a key mechanism for regulatory enforcement of AI models, enabling governments to accurately pinpoint models subject to regulation.
• A registry will enable the development of new regulation and serve as a foundational governance hub, allowing governments to classify models and create regulation based on specific capabilities or characteristics. 
• A registry will foster public sector field-building by promoting the use of common standards, providing structured information on AI for policymakers, and encouraging the development of the technical skills and knowledge required to manage AI systems.

Crucially, a registry can achieve these four important goals efficiently and without hobbling innovation. We elaborate on these benefits in What value does a model registry provide to governments?. 

Proposed Design of a Model Registry

Based on our research detailed in Part II, we propose that an effective AI model registry should adhere to the following design principles to achieve the goals listed above:

• A model registry should be minimal, and aim to only require the information needed to fulfill the described purposes. 
• A model registry should not include licensing requirements or mandatory standards. It should primarily consist of reporting existing information about an AI model, and require minimal additional overhead for developers.
• A model registry should be interoperable and conform to international standards that minimize the regulatory burden on registry administrators and AI developers.
• The bar for inclusion into a model registry should be low enough to capture the next generation of highly capable frontier models, but above the current generation of models (those deployed before the publication of this report).
• Models should be required to be registered prior to deployment.
• The registry should support categorizing models into families, and allow developers to maintain the model information for only the most capable models in each key measurable dimension to minimize overhead.
• Developers should be required to revisit their registry entries twice a year, either confirming that the information remains accurate or updating it to reflect any changes.
• An effective model registry should contain information including: 
  • Basic information on the developing organization
  • Open-source status of the model
  • Model size in parameters
  • Compute used during training, retraining, and post-training
  • Training data: amount, type, and provenance
  • A high-level description of model architecture
  • General information about the hardware used for development
  • A description of the security standards protecting key components of the AI model
  • The mechanism and results of any model evaluations or benchmarks conducted by the developer
  • A description of the functions of the model
  • A summary of post-deployment monitoring techniques used.

Proposed Implementation of a Model Registry

Based on our research detailed in Part III, we propose that an effective AI model registry should meet the following implementation principles: 

• A model registry should be enforced by implementing a system to fine AI developers a percentage of annual turnover for non-compliance.
• A model registry should require third-party users of frontier AI models to verify that those models have been registered.
• A model registry should be overseen directly by governments with minimal outsourcing to third-parties. 
• A model registry should be implemented at the national level, but remain interoperable with international standards.
• A model registry should be pragmatically confidential and secure.

Structure of the full report

In Part I, we explore why AI models require greater governance and introduce model registries as a potential governance tool. We explore the benefits a registry could provide to governments and society and the risks that should be mitigated in designing and implementing a model registry. 

In Part II, we research and make recommendations on how to design an effective registry: which models should qualify for inclusion on the registry, and what information developers should submit to the registry about their models.

In Part III, we research and make recommendations on how to practically implement an effective registry: how it should be administered, whether its information should be public or private, and how to ensure developers share accurate information. 

For each topic, we share our research, weigh benefits and risks, and conclude by making specific recommendations.

Conclusion

AI has advanced dramatically in the last decade, and its impact on our everyday lives, our economy, and our society is likely to continue growing. This rapid development has outpaced governmental capacity to establish basic insight and design effective regulation for AI, in line with insight and regulation in other industries. 

Experts disagree about the future of AI. However, few if any expect AI to be less prominent in a decade than it is today, and its prominence today already warrants basic governmental oversight to ensure public safety and economic stability. Registries are a standard governmental tool to establish such oversight and to inform future policy-making. 

Our full report recognizes the need for lightweight and efficient governmental oversight, and so our proposal minimizes the burden on both developers and governments by recommending injunctive action in the market as the primary mechanism to ensure compliance. We recognize the value of innovation and the need for care when dealing with commercially sensitive information. We recognize the need for confidentiality and careful protection of hazardous information. We recognize the difficulty developers face in evaluating the capabilities and risks of their models. 

Crucially, though, we recognize that AI development will have a huge impact on society in the coming decades. Governments need to establish basic insight, and our proposal grants that insight without undue burden or risk. 

We urge policymakers and AI developers to collaborate in implementing national model registries, as they offer a critical first step towards responsible AI governance that balances innovation with public safety.

1 comments

Comments sorted by top scores.