The Center for Data Innovation provided comments regarding the Competition and Markets Authority’s (CMA) initial review of foundation models, their competitive market and use cases, and principles that can guide their development and adoption.
Foundation models are AI models, such as large-language models (LLMs), pre-trained on broad data sets to perform different tasks. This technology can already produce realistic text, high-quality images, entertaining music, and error-free code. Regulatory frameworks like the UK’s “Pro-Innovation Approach to AI Regulation” will be critical in ensuring the adoption of foundation models applications in health care, education, and entertainment.
The Center would like to commend the CMA for investigating the potential impact of foundation models but also caution that the technology and market for this technology are still in their infancy and rapidly changing. To use an analogy, investigating market competition in AI foundation models today would be equivalent to the CMA investigating the search engine market in 1996 when search engines like AltaVista, Excite, Lycos, and Infoseek were popular. As such, the CMA should go slow, not act now, and adhere to the light-touch spirit of the Government’s pro-innovation approach to AI.
The following briefly addresses the three key themes of the CMA’s initial review—competition in the development of foundation models, the impact foundation models may have on competition in other markets, and consumer protection—and provides other considerations to inform the CMA’s approach to implementing the Government’s AI white paper.