Last week the UK's CMA published its Update Paper on AI foundation models (FMs), which follows its Initial Report published in September 2023 and coincides with CMA CEO, Sarah Cardell’s, speech on AI FMs in Washington DC last week. Earlier today, the CMA published its fuller Technical Report, which details the analysis behind last week’s publications.
The update signals heightened concerns over AI and a desire to take effective early action to address risks to competition and consumer protection. This represents a significant change in tone from the cautious and considered approach to any intervention in AI markets of its initial report on foundation models of last year, which emphasised a desire to “guide not force” FMs towards more positive outcomes. Significantly, it has introduced a new acronym to describe the most significant firms – “GAMMAN” (Google, Apple, Microsoft, Meta, Amazon and Nvidia) – in which Nvidia joins other “Big Tech” businesses as front and centre of the CMA’s focus.
The CMA's key messages
The Update Paper outlines the CMA’s findings on the ongoing development of FM-related markets. It recognises the significant pace at which FMs have developed since September 2023, noting that over 120 FMs have been released in that period (in addition to updates to existing models), and that these are increasingly complex and capable (including – critically – smaller models). The CMA also notes increased uptake of FMs and related services among consumers and businesses.
Objectively, such significant new entry from a diversity of firms would appear on its face to demonstrate competition working very well. However, compared to September’s report, the Update Paper takes a decidedly sterner tone: as Sarah Cardell summarised: “when we started this work, we were curious. Now… we have real concerns".
Notwithstanding the findings of strong existing competition, the CMA expresses concern that “winner takes all” dynamics could arise. Its concerns focus on the inputs to FMs (in particular compute, data and expertise) and the interconnections between FMs and the platforms on which they are released and their connections with other digital tools. While the CMA recognises that partnerships can be pro-competitive and that they “may be an essential ingredient for the success of independent developers”, it cautions each must be “assessed on its individual facts and merits”.
The CMA states clearly its intention to use all tools at its disposal to prevent the risks it identifies eventuating, not only through its work in the ongoing AI review, but with its merger and markets tools, as well as with its soon-to-be-received digital regulation and consumer law powers under the Digital Markets, Competition and Consumers Bill.
The CMA also updates its guiding AI principles, which are intended to prevent anticompetitive outcomes or outcomes that harm consumers in FM-related markets. The updated draft includes a new emphasis on the facilitation of choice through interoperability.
GAMMAN: what’s in an acronym?
Among the most interesting aspects of the report is the inclusion of a new member in the CMA’s “Big Tech Club”, formerly known as GAFA, then GAMMA. Nvidia, the leading supplier of accelerator chips is now officially a member of the newly coined “GAMMAN” group. The CMA recognises (as it did in its initial report) the availability of AI accelerator chips as a critical input to the FM value chain and the CMA is planning a paper on AI accelerator chips, to be published later this year.
While the coining of a new acronym may seem trivial, such shorthand for the collection of the most significant firms have a long and chequered history in UK competition enforcement. Once a term catches on, it can be hard to displace and can loom large in enforcement (e.g. the “big four” grocers and the “big six” energy suppliers). The CMA’s inclusion of Nvidia in the GAMMAN “club” recognises the critical importance of chips to the development of FMs and is doubtless a signal that Nvidia will come under significantly more scrutiny from the CMA in the years ahead.
What’s next?
The CMA has expanded its workplan throughout 2024 to further consider the impact of FMs on markets, including with plans to:
- publish a paper to examine the role in the FM value chain of AI accelerator chips;
- continue joint research in the Digital Regulation Cooperation Forum (DRCF) on consumers’ understanding and use of FM-related services, as well as participate in the DRCF AI and Digital Hub pilot – which will provide businesses with tailored support with the AI regulatory framework – due for launch in Spring 2024;
- publish a joint statement with the Information Commissioner’s Office on the interaction between competition, consumer protection and data protection in FMs; and
- publish a further update report on AI FMs in Autumn 2024, which will coincide with the expected commencement of the UK’s new digital markets regime.
Competition and the broader AI regulation debate
The CMA’s work is well in line with both domestic and global trends: in the UK, just yesterday the Digital Regulation Cooperation Forum published a new paper on Fairness in AI, while at the global level, EU and US antitrust officials convened last week in the EU-US Fourth Technology Competition Policy Dialogue.
Growing concern and enhanced scrutiny of AI is also a broader theme in the UK: just a few days after the CMA’s Update Paper, it was reported that the UK Government is reconsidering its “light touch” approach to AI regulation and is likely to bring forward regulation that could impose limits on developments of FMs to address AI safety concerns.
The CMA’s report says its principles are intended to complement the Government’s approach, but it does acknowledge potential trade-offs between different policy objectives. For example, while the CMA is clear that open-source models are important for competition, these could clearly raise other risks around AI safety. Similarly, the CMA cautions imposing regulatory obligations on firms could make it harder for smaller firms to comply and therefore compete.
How these tensions will be resolved is far from clear but what is certain is that regulatory attitudes are evolving at an unprecedented pace – outstripped only by the pace of change AI FMs themselves.