Denver — Today, US Senator Michael Bennett of Colorado called on leaders of major technology and generative artificial intelligence (AI) companies, including Meta, Alphabet, Microsoft, Twitter, TikTok, and OpenAI, to use AI-generated content. Measures to limit the proliferation of AI-generated content that is specifically labeled and intended to mislead users.
“Online misinformation and disinformation is nothing new. But the sophistication and scale of these tools are rapidly evolving, surpassing existing safeguards.” Bennett wrote in a letter. “It used to be that creating a plausible deepfake required a great deal of technical skill. Today, generative AI systems have democratized that capability, with floodgates open to anyone who wants to use or abuse the technology. It’s open.”
In his letter, Bennett points to several examples of AI-generated content causing alarm and market turmoil, as well as the emergence of AI-generated content in political social media posts. Bennett pointed out that OpenAI CEO Sam Altman’s testimony before the Senate Judiciary Committee cited AI’s ability to spread disinformation as a major concern, saying that It points out the danger of letting information go unchecked.
“Americans should know that images and videos are products of generative AI models, and platforms and developers have a responsibility to label such content appropriately.” wrote Bennett. “Falsified images can derail the stock market, suppress voter turnout, and undermine the American public’s confidence in the authenticity of election materials.”
Bennett acknowledges the initial steps tech companies are taking to identify and label AI-generated content, but stresses that these existing steps are voluntary and easily circumvented. doing. He provided a framework for labeling AI-generated content and concluded by asking companies to provide identification and watermarking policies and standards, as well as a commitment to remove AI-generated content for the purpose of misleading users. rice field.
“Continued inaction jeopardizes our democracy. Generative AI can support new creative endeavors and produce amazing content, but these benefits are the It doesn’t come at the cost of destroying reality.” Bennett concluded:
Bennett has repeatedly advocated for digital regulation, youth online safety, and greater protection for emerging technologies.Bennett last week gave a speech on the Senate floor To advocate the creation of a new federal agency that can regulate digital platforms and AI.Bennett reappeared last month Digital Platform Commission Actfirst bill in parliament create Dedicated A federal agency tasked with overseeing large technology companies, protecting consumers, promoting competition, and protecting the public interest.
In June Mr. Bennett said, Global Technology Leadership Act Establish a Global Competition Analysis Bureau that can assess how the United States is doing relative to other countries in key emerging technologies such as AI, inform U.S. policy, and strengthen U.S. competitiveness. He recently introduced oversee emerging technology laws and the Evaluate AI Law Ensure government use of AI is compliant with fundamental rights and civil liberties. In May, Bennett joined his colleagues, introduction The REAL Political Ads Act mandates disclaimers for political advertising in federal campaigns that use AI-generated content.
The text of the letter is available here the following.
Dear Mr. Zuckerberg, Mr. Musk, Mr. Altman, Mr. Chu, Mr. Pichai, Mr. Nadella, Mr. Mostak, Mr. Holz and Dr. Amodei.
I am writing this with concern about your company’s current identification and disclosure policy for content generated by artificial intelligence (AI). Americans need to be aware that images and videos are products of generative AI models, and platforms and developers have a responsibility to label such content appropriately. This is especially true for political communications. Falsified images can disrupt the stock market, suppress voter turnout, and undermine the American public’s confidence in the authenticity of campaign materials. The continued creation and distribution of AI-generated content without a clear and readily understandable identifier poses an unacceptable risk to public debate and electoral integrity.
Online misinformation and disinformation is nothing new. However, the sophistication and scale of these tools are rapidly evolving, surpassing existing safeguards. In the past, creating a plausible deepfake required considerable technical skill. Today, generative AI systems have democratized their capabilities, opening the floodgates to anyone who wants to use or abuse the technology.
We have already seen evidence of generative AI being used to create and share fake images. Some of these are relatively mild, like Pope Francis in a large white down jacket. Others are more anxious. In May, an AI-generated image of an alleged explosion at the Pentagon went viral, triggering a decline in major stock indices. Fake news accounts redistributed these images alongside real news outlets. RTRussian state media organization.
The proliferation of AI-generated content poses special problems for political communications. In a recent testimony before the Senate Judiciary Committee, OpenAI CEO Sam Altman cited the ability of AI models to provide “one-on-one interactive disinformation” as a top area of concern. We are at the beginning of this era.
In June, the official Rapid Response Twitter account of 2024 Republican presidential candidate Ron DeSantis, Florida Governor Ron DeSantis, shared an image that experts said was believed to be AI-generated. Both official and unaffiliated accounts that support former President Trump are posting AI-generated content targeting political opponents.
In May, I worked with colleagues to introduce the REAL Political Advertising Act, which mandates a disclaimer for federal election campaign political ads that use AI-generated content. However, as the political media increasingly moves from regulated television, print and radio advertising to freely available social media, broader disclosure requirements will be required.
AI system developers and platforms must work together to combat the proliferation of unlabeled AI content. Developers should strive to watermark videos and images when they are created, and platforms should work on attaching and disclosing labels when distributing them. Addressing this unique threat requires a multifaceted approach.
Companies are starting to take steps to better identify AI-generated content to users. For example, nonprofits like the Partnership on AI have published recommended guidelines. Microsoft has promised to watermark AI-generated content, and Google will begin attaching written disclosures to Google Images. OpenAI’s DALL-E 2 adds a watermark to the images it generates, and Stable Diffusion embeds the watermark in the content by default. Midjourney, Shutterstock, and Google are working to embed metadata indicators into AI-generated content.
However, these policies are still easily circumvented or, surprisingly, rely on voluntary compliance. Google’s process for labeling AI-generated images from third-party systems relies on self-disclosure. Stable Diffusion’s open source structure allows users to avoid watermarking code. Watermarks on the DALL-E 2 are discreet and easy to remove. Also, some platforms such as Meta, Twitter, and TikTok have existing policies regarding AI-generated images and videos, but such content will continue to appear in users’ feeds.
Platforms need to update their policies towards a world where everyone has access to generative AI tools. AI-generated videos and images must be clearly and conspicuously labeled, and if users do not comply, they must label AI-generated content themselves. Platforms must consider certain rules regarding official political accounts and must release periodic reports detailing efforts to identify, label, or remove AI-generated content.
Similarly, generative AI system developers should scrutinize whether their models could be used for manipulation or misinformation, conduct public risk assessments, and take action to identify and mitigate these vulnerabilities. You have to make a plan. Users cannot be expected to dive into the metadata of every image in their feed, nor can the platform encourage users to infer the credibility of content shared by political candidates, parties, and their supporters. should not be forced.
Continued inaction endangers our democracy. Generative AI can support new creative endeavors and create amazing content, but these benefits don’t come at the cost of destroying our shared reality.
To that end, we ask you to answer the following questions by July 31, 2023.
For generative AI developers:
-
What technical standards, capabilities, or requirements do you currently employ for watermarking or identifying content created using your system?
-
When were these standards, features and requirements developed?
-
When were these standards, features, or requirements last updated?
-
What audit processes does your organization have in place to assess the effectiveness of these standards, features, or requirements?
-
What tracking systems do you currently have in place to monitor the distribution of content created using your system?
-
What tests and assessments will you use to estimate potential capabilities related to misinformation, disinformation, persuasion, and manipulation before deploying your model?
-
What process do you use to estimate risks associated with misinformation, disinformation, persuasion, and manipulation? Under what circumstances might concerns about these risks delay access to generative AI systems? or restrict?
For social media platforms and search engines:
Thank you for your attention to this important issue. We look forward to your reply.
Sincerely,
