The AdC has released a paper addressing the competition risks associated with access to generative AI models, marking the second installment in its Short Paper Series following an Issues Paper from November 2023.

Generative AI models, utilized by both consumers and businesses, are derived from foundation models that are tailored for specific applications. The development of these specialized models hinges on the accessibility of foundation models, which can vary in their openness. Access is often gated through web interfaces and APIs, limiting the components available to users. The degree of openness is a strategic choice made by the developers of foundation models.

Open foundation models enhance choice and flexibility for downstream AI developers, fostering an ecosystem conducive to open AI development. This openness is vital for promoting competition and innovation in the generative AI market, as it allows third parties to experiment and adapt models for various applications. However, the reliance on network effects within these ecosystems can also lead to increased market concentration.

Despite the potential benefits of open or open-source AI models, there are risks of anticompetitive behavior. Developers of foundation models and cloud service providers may have incentives to restrict competition among downstream developers. Furthermore, the initial openness of AI models could be a precursor to lock-in strategies, where models become closed in future iterations, allowing developers to gain market power and exploit it in related markets.

To harness the opportunities presented by generative AI, it is essential for the sector to evolve competitively, ensuring that consumers benefit from a fair playing field in accessing foundation models, which is crucial for fostering competition and innovation in downstream AI markets.