Insights UK Competition and Markets Authority publishes initial review of AI foundational models (“FMs”)


FMs (such as large language models) are the result of a type of machine learning, an AI technology that employs large quantities of data. They can form the base for a wide number of applications including use in chatbots, writing code and the generation of images. An example is that used by Open AI’s ChatGPT. FMs have the potential to improve products and services and the way we search, learn, create and fulfil tasks more productively. Following engagement with several stakeholders, this paper sets out the CMA’s early views on how the development and deployment of FMs could impact competition and consumer rights.

Developing an FM requires significant computing power, access to vast quantities of training data, technical expertise and capital. The CMA believes that for the market to develop positively, there is a need for smaller developers to be able to access data and computing power on reasonable terms. Further, there may need to provide competitive constraints on the initially successful FM developers given the potential advantage these first movers may have from early involvement and economies of scale, and to ensure that FM developers with a leading position in other markets related to FMs do not use that position to exclude rival developers or rivals in those other markets. The paper refers to FM developers, such as Microsoft, Amazon and Google, who own key infrastructure for producing and distributing FMs such as data centres, servers and data repositories, as well as having a presence in a range of user-facing markets where FM technology can be integrated, such as in online shopping and search services.

As to the use of FMs, the paper states that the market is more likely to produce positive outcomes if users adopting applications in which FMs are integrated can switch between or use multiple FM providers, if FMs and the systems they use are interoperable with one another, if consumers can easily port their data between services, and if users are not subject to anti-competitive conduct by providers such as self-preferencing, tying or bundling.

Lastly, the CMA looks at potential harm to consumers. The key drivers for whether the market develops in a positive way for consumer protection include obligations to ensure FM accuracy, allocation of accountability and responsibility, providing information as to whether content is FM-generated and any associated risks and limitations, providing information to allow users to understand and critically assess FM outputs and providing sufficient safeguards against consumer harms such as fake reviews, phishing, hidden advertising and deep fakes.

To ensure that business and individual users and the wider economy benefit from the innovation AI can bring, those who develop or deploy FMs must comply with existing consumer and competition law. In addition, to ensure that consumer and competition protection remains an effective driving force as the development and deployment of FMs evolves, the CMA proposes certain guiding principles around accountability, access, diversity, choice, flexibility, fair dealing and transparency.

The CMA will now embark on a significant programme of engagement, in the UK, US and elsewhere, seeking a wide range of views from diverse groups. It will publish an update on its thinking on the guiding principles in early 2024.

To access the report, click here.