As the use of artificial intelligence (AI) models continues to grow, the Biden administration has recognized the need to monitor and regulate these models to prevent potential disasters and ensure public safety. The National Telecommunications and Information Administration (NTIA) recently released a report emphasizing the importance of understanding the risks posed by open-weight models and developing strategies to mitigate those risks.
Open-weight models, as defined by the NTIA, are foundation models with publicly released weights or parameters that users can download. These models are different from open-source models, which can be replicated under an open license. The NTIA acknowledges that both open and closed models have risks that need management, but open models may present unique opportunities and challenges in terms of risk reduction.
To address the risks associated with open-weight models, the NTIA suggests three main areas of focus. First, collecting evidence on the capabilities of these models to monitor specific risks is crucial. This will help in understanding the potential dangers and developing appropriate regulations. Second, evaluating and comparing indicators of risk will enable policymakers to make informed decisions. Lastly, adopting policies that specifically target the identified risks will help ensure the safe and responsible use of AI models.
The Biden administration’s approach to regulating open-weight models may be similar to the European Union’s AI Act. The EU has taken a use-case-based approach to regulate AI models, focusing on the level of risk associated with specific applications rather than the models themselves. For instance, the EU has imposed hefty fines for companies using AI for facial recognition. By considering the potential dangers of public AI models, the U.S. may follow in the footsteps of the EU and adopt similar regulations.
Kevin Bankston, a senior advisor on AI governance with the Center for Democracy and Technology, commended the NTIA for taking the time to carefully consider how to police AI models. Bankston believes that there is currently insufficient evidence of novel risks from open foundation models to warrant strict restrictions on their distribution. This suggests that developers of AI models can still release their model weights, albeit under increased scrutiny.
However, it is important to note that the AI Executive Order issued by the Biden administration is not yet regulated. While some lawmakers and states have proposed potential policies, the NTIA is still in the fact-finding phase. Developers of AI models may not face immediate restrictions, but they should be prepared for potential changes in regulations as the government continues to gather information and adapt based on their findings.
Assaf Melochna, the founder of AI company Aquant, believes that the NTIA’s observations will not have a significant impact on model developers. He suggests that developers can still release their model weights at their discretion, but they should expect increased scrutiny. Melochna emphasizes the need for federal agencies to remain flexible and adapt to the rapidly evolving AI sector.
In conclusion, the Biden administration’s focus on monitoring and regulating open-weight AI models demonstrates its commitment to ensuring the safe and responsible use of AI technology. By learning from the EU’s AI Act and carefully considering the potential risks, the U.S. government aims to strike a balance between innovation and public safety. Developers of AI models should stay informed about the evolving regulatory landscape and be prepared to adapt their practices accordingly.
