Last fall, President Biden signed the wide-ranging Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence (“the Order”), which we first blogged about here. The Order tasked federal agencies with developing industry-specific AI guidance and best practices.
Six months after President Biden signed the Order, the Department of Commerce, in a press release, made significant announcements about efforts to implement the Order. In particular, the Department’s National Institute of Standards and Technology (NIST) has released four draft publications focused on managing risk and developing standards for AI. All four NIST draft publications sought public feedback before releasing final versions later this year. NIST also announced the GenAI program which will issue a series of challenge problems focused on detecting AI-generated content. Following guidance released in February, the U.S. Patent and Trademark Office (USPTO) has published a request for public comment (RFC) seeking comments on the patentability of AI-assisted inventions.
The AI RMF Generative AI Profile (NIST AI 600-1)
- The NIST AI 600-1 specifies 13 unique risks generative AI poses. Those risks include more accessible access to information about creating dangerous weapons, a lower barrier to entry for hackers and cybercriminals, and near effortless production of hateful and abusive content. The paper goes on to offer 400 actions that developers can take to mitigate those risks. The guidance is intended as a companion resource to NIST’s AI Risk Management Framework (AI RMF).
Secure Software Development Practices for Generative AI and Dual-Use Foundation Models (NIST Special Publication (SP) 800-218A).
- This paper identifies the risks posed by training generative AI systems, including poisoning, bias, homogeneity, and tampering. In addition to identifying the dangers of training AI systems, the paper also offers strategies to address those risks.
Reducing Risks Posed by Synthetic Content (NIST AI 100-4)
- This publication discusses methods for detecting, authenticating, and labeling AI-generated, synthetic content. The paper focuses on the risks of unknown synthetic data and technical approaches to solve it based on the use case and context.
A Plan for Global Engagement on AI Standards (NIST AI 100-5)
- This paper suggests universal AI standards, in addition to coordination and information sharing across borders.
The NIST GenAI program
- This new program will release public challenges with the main goal of increasing information integrity and guiding the safe and responsible use of digital content. The first pilot challenge is to identify and label AI-generated, synthetic content (text, image, video, and audio) from content made by humans.
The USPTO also sought public comment to determine whether an invention made with the help of AI is patentable.
AI-assisted innovations raise novel questions about what qualifies as prior art and the level of skill of a person who has ordinary skill in the art. The USPTO hopes to use these comments to develop further guidance, inform the USPTO in the courtroom, and provide technical advice to Congress.
The explosion of generative AI has brought with it many new challenges and opportunities. The Department of Commerce, implementing Biden’s executive Order, released many reports and opportunities for public comment, along with challenge problems to harness AI’s challenges without stifling the opportunities.
