OpenAI proposes that 'cutting-edge AI needs regulation' and has released its own policy proposals for regulating high-performance AI.
OpenAI has released a regulatory proposal for 'frontier AI,' which is cutting-edge, high-performance AI, titled 'Democratic Governance of Frontier AI: A blueprint for a federal framework.' This policy proposal outlines how the U.S. government should evaluate, ensure the safety of, and responsibly manage frontier AI.
A blueprint for democratic governance of frontier AI | OpenAI
As AI technology advances, we've already become accustomed to using it to draft emails, find bugs in programs, and summarize long documents. On the other hand, there are concerns that AI could be used to refine cyberattack procedures or handle information that could lead to biological exploitation. As AI's capabilities improve, it's inevitable that not only will its value as a convenient tool increase, but the risks it poses to society as a whole will also grow.
OpenAI explains that the decision to release AI should not be left solely to companies, but rather a federal framework based on democratic procedures is needed. They argue that a growing number of different rules in each state will not only complicate how companies respond, but will also fail to adequately address the AI risks that cross national and state borders, thus necessitating a federal-level approach.
The proposed regulations announced this time focus on three main points: 'establishing a nationwide system requiring safety assessments and transparency reports for frontier AI,' 'developing a professional evaluation system centered around the Center for AI Standards and Innovation (CAISI),' and 'a strategy to enhance defense capabilities in areas such as cybersecurity and public health across the entire government.'
Specifically, these measures include assessing significant risks, independent audits, reporting major incidents, protecting undisclosed model weights, and protecting whistleblowers. Model weights are internal data that make up the capabilities of the AI, and their leakage could lead to the misuse of undisclosed models, so protection is necessary.
OpenAI also states that while the most powerful AIs should undergo pre-release evaluation, government agencies should not act as 'gatekeepers' granting or denying their release. They envision a system where evaluation results and security measures are made public, with the ultimate responsibility resting with the development companies.
OpenAI also pointed out that 'AI capabilities may advance at a faster pace than existing systems and rules can be established.' They called for early institutional development, emphasizing the need to have mechanisms in place to assess safety and clarify accountability before frontier AI has a significant impact on the economy, scientific research, and national security.
Related Posts:
in AI, Posted by log1d_ts