(Credit: The Washington Post / Contributor / Getty Images)
The White House will soon open up a 30-day public comment period on the contentious debate between so-called "open" and "closed" AI models.
These little-known terms are gaining prominence as experts debate the future of AI systems. "Open" models are made available to the public, primarily software developers, to customize and build upon. Examples include Meta's Llama 2, and Google's new Gemma models.
ChatGPT is an example of a "closed" model, in the sense that we can interact with it through an OpenAI-controlled interface and cannot access the underlying model itself, to seek out information on it or customize it. Recognizing the benefits of open-source models, OpenAI has sought to create pseudo-open products, such as its new GPTs.
The White House effort is part of President Biden's October 2023 executive order on AI, and is intended to "help us chart a policy path to promote both safety and innovation in this important technology," says Alan Davidson, Assistant Secretary of Commerce for Communications and Information and NTIA Administrator.
You can submit your thoughts on the matter to the National Telecommunications and Information Administration (NTIA) at this link, which will go live once it's published in the Federal Register.
Here are the nine questions the public can comment on. They are focused on whether large open-source models should be published online and made widely available. (Note: We edited the questions slightly for clarity, replacing the NTIA's name for open models—"dual use foundation models with widely available weights"—with "open models." Another name for open-source models is open-weight models.)
- How should NTIA define "open" or "widely available" when thinking about foundation models and model weights?
- How do the risks associated with [open models] compare to the risks associated with [closed models]?
- What are the benefits of [open models] as compared to fully closed models?
- What are the legal of business issues or effects related to open foundation models?
- Are there other relevant components of open models that, if simultaneously widely available, would change the risks or benefits presented by [open models]? If so, please list them and explain their impact.
- What are the safety-related or broader technical issues involved in managing risks and amplifying benefits of [open models]?
- What are current or potential voluntary, domestic regulatory, and international mechanisms to manage the risks and maximize the benefits of [open models]? What kind of entities should take a leadership role across which features of governance?
- In the face of continually changing technology, and given unforeseen risks and benefits, how can governments, companies, and individuals make decisions or plans today about open foundation models that will be useful in the future?
- What other issues, topics, or adjacent technological advancements should we consider when analyzing risks and benefits of [open models]?
To get the debate started, the NTIA highlights several advantages of open models. For one, they offer more transparency, as closed models have "limited or no public access to their inner workings." They also make AI more accessible to individuals and corporations that do not have millions of dollars to build and train their own systems from scratch, including small businesses and academic institutions.
Medical, pharmaceutical, and scientific research could also be "transformed" by open models, which the NTIA says support the "movement toward open science, which includes sharing research artifacts like the data and code required to reproduce results," the NTIA says.
However, open models also come with risks related to lack of oversight. Bad actors could take advantage of these widely available, powerful models to compromise security, undo civil rights progress, and even create "non-consensual intimate images" as Taylor Swift already endured. Not to mention the "mass disinformation campaigns" they could facilitate, a risk OpenAI has acknowledged as a concern during this election year.