On June 29, 2023, the Federal Trade Commission (“FTC”) posted a blog to its website expressing concerns about the recent rise of generative artificial intelligence (“generative AI”). To get ahead of this rapidly developing technology, the FTC identified “the essential building blocks” of generative AI and highlighted some business practices the agency would consider “unfair methods of competition.” The FTC also underscored technological aspects unique to generative AI that could raise competition concerns.

What is Generative AI?

Traditional AI has existed in the marketplace for years and largely assisted users in analyzing or manipulating existing data.  Generative AI, on the other hand, represents a significant advance with its ability to generate entirely new text, images, audio, and video. The FTC notes that this content is frequently “indistinguishable from content crafted directly by humans.”

What are the “essential building blocks” of generative AI?

The FTC identified three “essential building blocks” that companies need to develop generative AI. Without fair access to the necessary inputs, the FTC warns that competition and the ability for new players to enter the market will suffer.

• Data. Generative AI models require access to vast amounts of data, particularly in the early phases where models build up a robust competency in a specific domain (for example, text or images). Market incumbents may possess an inherent advantage because of access to data collected over many years. The FTC notes that while “simply having large amounts of data is not unlawful,” creating undue barriers to access that data may be considered unfair competition.

• Talent. Creating generative AI requires unique skillsets, including in-depth understanding of machine learning, natural language processing, and computer vision. The FTC cautions that powerful companies should not attempt to “lock-in” workers and stifle competition through the use of non-competes. Permitting workers to move freely enables a competitive and innovative marketplace.

• Computational Resources. Computational Resources (or simply, “compute”) represents the specialized computers and graphical processing units (“GPUs”) necessary to train and develop generative AI. The market for these specialized chips is already highly concentrated, and demand far outstrips supply. Computational resources are therefore  expensive to acquire and retain, and many new entrants turn to cloud providers instead. Because few firms offer significant cloud services, this trend also presents anti-competitive risks.

What might be considered “unfair methods of competition?”

Because the market for generative AI is rapidly evolving, the FTC put market participants on notice about potential unfair methods of competition. These include both traditional anti-competitive business strategies, as well as technological challenges unique to generative AI that could “supercharge” anti-competitive harms.

• “Bundling” and “Tying.” By no means unique to generative AI, bundling and tying occur when a large incumbent packages new products together with existing products. The FTC counsels that this can “potentially distort[] competition.”

• Ecosystem Services. When incumbents offer new services as part of an “ecosystem” of services, they might “funnel” their users towards their own products and away from competitors. Market incumbents that offer both compute and generative AI products may also stifle competition by locking competitors out accessing compute—an essential input into creating innovative generative AI products.

• Mergers & Acquisition. Incumbent firms can quickly consolidate power by acquiring suppliers of the necessary inputs for generative AI or simply purchasing threatening competitors.

• Network Effects. Incumbents often benefit from the “network effect” due to users’ familiarity with their products. However, generative AI can “supercharge” this effect because the technology, by its nature, improves by rapid iteration and frequent interaction. A “positive feedback loop” can quickly entrench incumbents through consumer familiarity and superior product capability, stifling competition.

• Open-Source. Access to low-cost open-source AI models has allowed for the proliferation of generative AI tools. The FTC warns, however, that lurking behind these open-source resources may be anti-competitive threats. Specifically, models that begin as open-source but where access is later limited (“open first, closed later” strategies) could draw in business and data up front and lock up customers after the model closes.

What’s next?

Like many other regulatory agencies, the FTC has signaled that it is closely watching the development of generative AI. As with all emerging technologies, generative AI comes with enormous potential to create value. The FTC has indicated that preservation of a “vibrant market” for generative AI is a top priority, and that it will utilize the “full range of tools to identify and address unfair methods of competition.”