NTIA seeks feedback on risks, rewards of ‘open-weight’ AI models

An agency within the Commerce Department is seeking public feedback on the openness of artificial intelligence tools, as the Biden administration looks to accelerate the use of these tools in and out of government.

The National Telecommunications and Information Administration (NTIA) launched a request for comment Wednesday, looking for more feedback on the benefits and risks of “open-weight” AI systems.

Comments are due within 30 days of publication in the Federal Register.

AI model weights are a distillation of what an AI model has been trained to understand, and how that model continues to behave.

Ben Buchanan, White House Special Advisor for Artificial Intelligence, told reporters that open-weight models “have a lot of potential for innovation,” but also raise some important questions around whether the safeguards in them will endure or can be easily removed.

“We recognize there’s a lot of technical complexity in the AI landscape. It’s changing very quickly. And probably on no issue is that more true than on this question, as the AI executive order puts it, of models with weights that are widely available,” Buchanan said in a call Wednesday.

Machine learning algorithms train a model, using large amounts of data, to recognize patterns and generate appropriate responses.

As models learn, the values of their weights change over time to reflect new knowledge.

“Open-weight” AI models allow developers to build upon and adapt previous work, without training their own AI models from scratch. Administration officials said this makes AI tools more available to small companies, researchers, nonprofits and the public.

“If a person has access to a model’s weights, that person does not need to train the model from scratch,” the Commerce Department wrote in a press release Wednesday. “Additionally, that person can more easily fine-tune the model or adapt it towards different goals, unlocking new innovations but also potentially removing safeguards.”

Alan Davidson, NTIA administrator and assistant secretary of commerce for communications and information, said NTIA is seeking public feedback, because early conversations about AI openness “have engendered fear about making the most advanced models widely available, without adequate restrictions or safeguards against misuse.”

“There is a risk of the ability of malicious actors to use open models to engage in harm, due to, for example, their affirmative misuse of those models or failures of effective oversight that had been built into models or a lack of clear safe guidance, safeguards or accountability mechanisms. So there have been real concerns about open foundation models, as it were.”

Among its questions, NTIA is specifically seeking input on how making AI model weights more or less open will impact innovation, competition, safety, security, trustworthiness and national security concerns.

The agency is also asking what role the federal government should have in supporting or restricting the availability of AI model weights.

Commerce Secretary Gina Raimondo said in statement that the Biden administration, through the request for comments, is taking a closer at “ensuring the safety, security, and trust in these systems.”

“AI is an accelerator – it has the potential to make people’s existing capabilities better, faster, and stronger. In the right hands, it carries incredible opportunity, but in the wrong hands, it can pose a threat to public safety. That’s why, under President Biden’s leadership and direction, we’re acting quickly and precisely to keep pace with this rapidly evolving technology to ensure safety, while protecting innovation,” Raimondo said.

President Joe Biden’s AI executive order from October 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.

Davidson said the agency is focused on making sure that important technologies are deployed in the service of the public, and that “nowhere is that more important than in the explosive growth of artificial intelligence systems.”

“Responsible AI innovation we know is going to bring enormous benefits to people. It’s going to transform every corner of our economy. But we also know that we will only realize the promise of AI if we also address the serious risks that it raises today,” Davidson said.

He added that greater access to AI model weights might also increase competition within this emerging tech industry, and ensure “that the most important AI systems are not controlled exclusively by a small set of companies.”

“Policymakers have recognized that we need expert inquiry on these very hard technical and policy issues if we are going to reap the full benefits of the AI revolution,” Davidson said.

NTIA held its first listening session on the topic at the Center for Democracy and Technology last December.

“It’s clear to the experts that this is not a binary issue. There are gradients of openness. And that will be something we will be exploring as we move forward and try to find solutions again, that promote both innovation and safety in AI tools,” Davidson said.

Copyright © 2024 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.