The UK’s Competition and Markets Authority set out specific principles to help guide AI regulations and companies that develop the technology.
The CMA, which acts as the primary antitrust regulator for the UK, focused its attention on foundation models: AI systems like OpenAI’s GPT-4, Meta’s Llama 2, and other large language models that form the basis for many generative AI use cases.
Companies making foundation models should follow seven principles. That includes making sure developers and businesses that use these models are accountable for the output that consumers are given, ensuring broad access to chips and processors and the training data needed to develop these AI systems, and offering a diversity of business models by including open and closed models. The CMA also said companies should provide a choice for businesses to decide how to use the model, offer flexibility or interoperability to switch to other models or use multiple models at the same time, avoid anti-competitive actions like bundling or self-preferencing, and offer transparency into the risks and limitations of generative AI content.
The CMA developed the principles following an initial review before it launches a series of dialogues with consumer and civil society groups, foundation model developers like Google, Meta, OpenAI, Microsoft, Nvidia, and Anthropic, foundation model users, and academics.
The agency said providing principles for the development and deployment of foundation models is necessary to protect competition and prevent low-performing AI systems from proliferating.
“The impact of foundation models could allow a wider range of firms to compete successfully, perhaps challenging current incumbents,” the CMA said in its review. “Vibrant competition and innovation could benefit the economy as a whole through increased productivity and economic growth.”
It added that if competition is weak, “a handful of firms gain or entrench positions of market power and fail to offer the best products and services or charge high prices.”
While the CMA admitted AI regulation raises broader questions on copyright and data privacy and protection, it chose to focus on competition and consumer protection to help guide the current development of the technology.
Governments around the world have been looking into different ways of regulating generative AI. The European Union, in its proposed AI Act, also focused on foundation models and requiring companies to comply with transparency rules. China’s recent AI rules mandate AI companies to register with the government and promise not to offer anti-competitive algorithms.
Meanwhile, the US is still figuring out how to approach AI regulation, though some policymakers hope to have its rules out by the end of this year.