Technology Blog
Data and Models: A Quote Book from the Tech Summit on AI
The FTC’s Tech Summit on AI[1] convened three panels that each highlighted different layers of the AI tech stack: hardware and infrastructure, data and models, and front-end user applications. This second Quote Book is focused on data and models. This post outlines the intended purpose of the quote book, a summary of the panel, and relevant topics and actions raised by the FTC.
Purpose of the quote book
A key component of the FTC’s AI work is to listen to people on the ground with knowledge of the effects of innovation in real time, including the engineers building next-generation cloud-computing platforms, the data scientists training AI models, the venture capitalists who are funding new innovative startups or the startups building companies to improve consumers’ lives. As policymakers debate the benefits or risks of new technologies, these voices can sometimes be lost in otherwise dense technical, policy, or legal discussions. The FTC’s Tech Summit is one component of our effort to listen and engage with a variety of perspectives.
This “Quote Book” aims to reflect and compile quotes from the participants, aggregated into common themes around AI model development. The “Quote Book” is a resource to quickly distill various perspectives on topics, from ways to enable competition and innovation to potential consumer concerns like data privacy, labor issues, deceptive messaging, and more.
Overview of the panel on data and models
“This current AI race is based on certain assumptions about both scale and speed as a proxy for progress. And it's a view that's based on narrow benchmarks, it's one that never really properly contends with the longer term environmental or labor impacts, or the impacts on our information environment.” - Amba Kak, Executive Director, AI Now Institute
The panel discussion underscored that the methods of data collection and model development have implications for both competition and consumer protection. The panelists discussed how dominant firms have access to large amounts of public or private data through existing product lines or through changing terms of service. On the consumer protection front, they discussed how large volumes of data are being used to train AI models, which raises a number of key questions for regulatory agencies and other policymakers to evaluate, such as: What are the legitimate business purposes for collecting, using, or retaining data? Are there types of data that should not be collected, used, or retained? Do consumers know how their data is being handled, and can they do anything about it? And what happens if companies misrepresent, or don't fully disclose their privacy and confidentiality practices?
On competition, panelists expressed that incumbent tech firms have access to large amounts of data through the existing product lines, and that there are challenges associated with competing in AI development related to access to data, and access to resources and investment needed to compete. This raises questions about whether the AI models will be developed and deployed in a way that fosters competition and introduces new competitive pressures on incumbents, or whether those challenges associated with access to data and other resources might steer AI development in a direction that protects or enhances market power.
“Venture capital, and the tech giants have a very large role in picking what startups are going to win and lose, coming out of this AI boom.” - Stephanie Palazzolo, Journalist, The Information, covering artificial intelligence
The FTC will need to be vigilant in evaluating these issues as the agency pursues its joint competition and consumer protection mandate.
The topics raised during the panel are not new for the FTC. The agency has an existing track record of addressing consumer facing harms due to AI-generated technologies, as well as competition concerns.
Methods of data collection and use: The FTC has a long history of enforcing the law with regards to the collection and use of consumer data. Most recently, the FTC’s proposed orders against Cerebral,[2] Avast,[3] X-Mode,[4] and InMarket[5] seek to ensure these companies comply with the law. Regarding AI models, the agency released a blog[6] that highlights any company who adopts more permissive data practices through surreptitious, retroactive amendments to its terms of service could be unfair or deceptive. In another blog[7] the agency asserted that “model-as-a-service companies that fail to abide by their privacy commitments to their users and customers, may be liable under the laws enforced by the FTC.”
Impact of generative AI model development on the Creative Economy: Additionally, in October 2023, the FTC hosted a virtual round table on the Creative Economy and Generative AI.[8] During the event, working creative professionals representing artists, writers, actors, musicians and people in other creative fields noted that while there are benefits to AI, such as potentially aiding their own work, they also expressed concerns: data collection without consent, nondisclosure, competing for work with AI, style mimicry, and fake endorsements. Additionally, the FTC submitted a comment to the U.S. Copyright Office, in which it identified several issues raised by the development and deployment of AI that implicate competition and consumer protection policy.[9] As stated in a related publication, “targeted enforcement under the FTC’s existing authority in AI-related markets can help to foster fair competition and protect people in creative industries and beyond from unfair or deceptive practices.”[10]
Competition at the model layer: The FTC has also taken action to deepen its understanding of competition among some AI model developers, including recently issuing 6(b) orders to five companies requiring them to provide information regarding recent investments and partnerships involving generative AI companies and major cloud service providers.[11] The agency’s 6(b) inquiry scrutinizes corporate partnerships and investments with AI providers in order to build a better internal understanding of these relationships and their impact on the competitive landscape.
“Just as we’ve seen behavioral advertising fuel the endless collection of user data, model training is emerging as another feature that could further incentivize surveillance,” FTC Chair Khan recently said in her Tech Summit remarks.[12] “The FTC’s work has made clear that these business incentives cannot justify violations of the law. The drive to refine your algorithm cannot come at the expense of people’s privacy or security, and privileged access to customers’ data cannot be used to undermine competition.”
[1]/news-events/events/2024/01/ftc-tech-summit
[8]/news-events/events/2023/10/creative-economy-generative-ai
[11]/news-events/news/press-releases/2024/01/ftc-launches-inquiry-generative-ai-investments-partnerships
[12]/system/files/ftc_gov/pdf/2024.01.25-chair-khan-remarks-at-ot-tech-summit.pdf