California State Senator Josh Becker has just proposed the California AI Transparency Act (CAITA), and given that I am advising him on this bill (ala my work with his team on last year’s California Delete Act that was successfully signed into law), I wanted to provide a few quick thoughts on it in this blog post. I have also included the text of Senator Becker’s press release on the topic and will provide additional analysis and commentary in subsequent blog posts in the coming weeks.
In terms of context, many of the current AI legislative proposals focus on AI bias that can lead to discrimination or how AI can be exploited for nefarious purposes such as deep fake pornography. This bill focuses on AI transparency in light of the significant advancements in generative artificial intelligence (GenAI) — AI that generates new content, such as images, text, or other forms of data, rather than simply analyzing or processing existing information. Namely, it is becoming increasingly difficult to know if content is human-generated or machine-generated (i.e., “synthetic”). As U.S. Senator Brian Schatz (D-Hawai‘i) noted when commenting on the downside of GenAI: “People deserve to know whether or not the videos, photos, and content they see and read online is real or not.”
Not surprisingly, this concept was also encapsulated as part of Europe’s AI Act in Article 52 which states in part:
“Providers shall ensure that AI systems intended to interact with natural persons are designed and developed in such a way that natural persons are informed that they are interacting with an AI system, unless this is obvious from the circumstances and the context of use.”
Senator Becker’s proposed California AI Transparency Act (CAITA) would require large GenAI providers to do some of the following:
(a) provide a content verification mechanism on their website (and/or a menu inside their mobile app) to allow Californians to be able to upload a text file, video, URL, image, etc., and ask the GenAI provider the following simple question: “Did you generate some or all of this?” The response would be Yes or No, and if Yes, it would then approximate what percent was generated by the GenAI vendor (e.g., “ChatGPT generated approximately 85% of this text”);
(b) include a clear and conspicuous disclosure identifying AI-generated content and AI chatbots that they create; and
(c ) make any of their developers and third-party licensees take reasonable steps to prevent systematic content publication without disclosures.
With the enactment of this proposed law, we would have a world where consumers could quickly and easily determine if photos or videos are real or a deepfake, teachers could check if homework assignments were generated by GenAI or original material, reporters could better sniff out disinformation, consumers can determine when interacting with a business in a technical support online chat if they are communicating with a human or an AI chatbot, etc. The combination of having clear and conspicuous labeling of AI-generated content and the backstop ability to still “trust but verify” by going directly to the creators of GenAI content to check if they created content will sidestep multiple issues in detecting synthetic content, including the well-documented “false positives” that third-party applications typically have when detecting GenAI-created content or people purposely “photoshopping” out watermarks or deleting meta-data.
CAITA is partly modeled after U.S. Senators Schatz and Kennedy’s proposed AI Labeling Act (with CAITA adding the content verification mechanism and other enhancements). Senator Kennedy said this of his bill: “Our bill would set an AI-based standard to protect U.S. consumers by telling them whether what they’re reading, seeing or hearing is the product of AI, and that’s clarity that people desperately need.” The same sentiment would certainly apply to Senator Becker’s CAITA bill.
The bottom line is this bill gives consumers another potential weapon to detect human versus synthetic. If passed, this would be the United States’ first and most comprehensive AI transparency law. Below is the full text of Senator Becker’s press release on CAITA. More details to come.
State Senator Josh Becker Proposes Groundbreaking Legislation Empowering Consumers to Identify AI-Generated Products
The California Artificial Intelligence Transparency Act will be a first-of-its-kind bill that would protect consumers by giving them the ability to determine if materials have been generated by AI
SACRAMENTO, CA — In a move to champion transparency and empower consumers, California State Senator Josh Becker (D-Menlo Park) will introduce in the 2024 legislative session the California Artificial Intelligence Transparency Act, a groundbreaking bill aimed at providing individuals with the ability to determine whether a product has been generated by artificial intelligence (AI).
Senator Josh Becker is a leading advocate for technological innovation and consumer rights and the author of the nation’s premier consumer data privacy law, the Delete Act.
“Artificial intelligence has become an integral part of our daily lives, influencing the products we use. It is crucial that consumers have the right to know if a product has been generated by AI. In my discussions with experts, it became increasingly clear that the ability to distribute high-quality content made by generative AI creates concerns about its potential misuse,” said Senator Becker. “AI-generated images, audio and video could be used for spreading political misinformation and creating deepfakes. My legislation will aim to advance provenance, transparency, accountability, and empower individuals to make choices aligned with their values.”
“As AI becomes better at producing realistic outputs and AI-generated media becomes more and more common online, we must create tools for consumers to understand whether content they are interacting with is real or AI-generated,” said Tom Kemp, a privacy rights expert and author of Containing Big Tech: How to Protect Our Civil Rights, Economy, and Democracy.
Becker’s legislation will seek to address the growing influence of AI in product creation and ensure that consumers have the information they need to make informed choices. As AI continues to play a pivotal role in various industries, the bill emphasizes the importance of disclosure and consumer awareness.
The bill will require large generative AI providers to watermark the images, videos, and audio created by their models and provide a platform for consumers to ask these companies whether content was created by them. The introduction of this legislation marks a significant step toward establishing clear guidelines for AI-generated products, setting a precedent for other states and jurisdictions to follow.
Senator Becker is committed to ensuring that consumers are equipped with the knowledge they need in an increasingly AI-driven world. He has been meeting with academics and experts in the field of generative artificial intelligence to better understand the emerging technology and necessary guardrails. In conclusion, Senator Becker is partnering with experts in an effort to establish transparency in this quickly-evolving space.
Last summer, the Biden-Harris Administration secured voluntary commitments from leading AI companies to manage the risks posed by AI. This included the commitment to develop robust technical mechanisms to ensure users know when content is AI-generated, such as a watermarking system.
“These commitments are encouraging and demonstrate the technical ability to provide this much needed transparency, but it was only the best actors that committed to this best practice,” said Senator Becker. “California is ready to once again take the lead in the space of consumer protection in innovation and technology.”
###
Senator Becker represents the 13th Senate District covering portions of Santa Clara and San Mateo counties and includes the cities of Atherton, Belmont, Brisbane, Burlingame, East Palo Alto, Foster City, Half Moon Bay, Hillsborough, Los Altos, Los Altos Hills, Menlo Park, Millbrae, Mountain View, Pacifica, Palo Alto, Redwood City, San Bruno, San Carlos, San Mateo, South San Francisco
