Press Release 26/2024

December 4, 2024

The AdC has published a short paper on access to AI models and competition. This is the second in a Short Paper Series expanding on the Issues Paper on Generative AI from November 2023.

The generative AI models used by consumers and businesses – specialised models – are built from other AI models – foundation models – which were adapted to specific use cases. Building specialised models requires access to foundation models.

Competition and innovation in specialised models will, therefore, depend on how open foundation models are. AI model openness is a matter of degree, with some models being more closed and others more open. The most typical form of access to models is gated access, via web interfaces and APIs, where only some components of the AI model are accessible. Ultimately, openness is a strategic decision by the foundation model developers.

Open foundation models create more choice and flexibility for downstream AI developers, and foster ecosystems of open AI development. This is crucial to promote competition and boost innovation in downstream generative AI markets, as more third parties can experiment, adapt models to specific use cases and create new products. Still, ecosystems of open AI development around a specific model may reinforce concentration, as they rely on network effects.

The fact that AI models are open or open source, does not preclude potential anticompetitive strategies. Both foundation model developers and cloud providers may have the incentives to hamper the ability of downstream developers to compete. In addition, AI model openness may be instrumental in lock-in strategies where models are initially open but closed-off in future versions, and in strategies to gain market power and then leverage it in adjacent markets.

Given the opportunities offered by generative AI, it is crucial that the sector develops in a competitive way, to the benefit of consumers. Ensuring a level playing field in access to foundation models is key to promote competition and innovation in downstream AI markets.