US government wants more “open” AI

The US Commerce Department agency, National Telecommunications and Information Administration (NTIA), has released a report that endorses the development and use of open-weight AI models (which allow developers to build and fine-tune the existing model but restricts full access, so they can’t see its inner workings) but has also called for “new capabilities” to evaluate and monitor the associated risks that come with ‘open’ models. 

The report says the government shouldn’t restrict open-weight models, because they allow independent developers, researchers, and small companies to build new, innovative technology, that they couldn’t otherwise afford to build, as the big tech giants can. This will leave room for new, healthy competition in the AI space, and boost innovation in the US, helping it take the lead in the AI race. 

But while they recognize the competition and innovation benefits of using open models, they also see that the ‘open’ nature of these models poses a higher risk than the closed ones, and have urged the government to introduce an ongoing program that assesses the safety of AI models, impose restrictions if needed, and develop risk-specific indicators that signal when policies need to be changed.