On June 28, 2024, the European Commission (EC) held a workshop on competition in virtual worlds and generative artificial intelligence (AI). The event followed the EC’s January 2024 call for contributions on the level of competition within these two sectors.
The workshop included keynote speeches from the EC Commissioner for Competition Margrethe Vestager, the Director-General of the EC Directorate-General for Competition (DG COMP) Olivier Guersent, President of the French Competition Authority Benoît Cœuré, as well as panel discussions involving industry representatives, academics and other competition authority officials. This article summarizes the key points made.
- The EC will continue to monitor market concentration, anticompetitive behaviour, and relationships between large tech companies and start-ups active in these sectors, and explore whether these arrangements could have a negative effect on competition. The EC is also monitoring distribution channels.
- Discussants noted the importance of access to key inputs for the development of both virtual world and AI technologies, and discussed the number of ways to potentially limit barriers to entry including public investment.
- The applicable competition tools are fit for purpose but should remain adaptable and innovative in order to address the emerging issues in virtual worlds and generative AI.
Margrethe Vestager: Competition Authorities’ Focus on Market Concentration, Anticompetitive Behaviour and Partnerships
EC Executive Vice-President in Charge of Competition Policy Margrethe Vestager emphasised that competition authorities need to be on guard over market concentration, anticompetitive behaviour and new types of partnerships between tech companies, but should continue to apply traditional merger and antitrust rules. Ms. Vestager noted that the EU Digital Markets Act (DMA) can regulate AI even though AI is not listed as a core platform service in itself; AI is covered where it is embedded in designated core platform services such as search engines, operating systems and social networking services.
Ms. Vestager commented that the commercialisation of AI and its powerful tools will be led by a few companies that already have a lot of market power, and these firms could leverage powerful network effects to control emerging markets.
According to Ms. Vestager, a major risk is large tech players leveraging their market power across different markets within their ecosystems This could lead to practices such as tying and bundling by dominant firms, blocking AI competitors from accessing essential resources, and preventing customers from switching.
In March, the EC sent formal information requests under its antitrust rules to several large tech players. The EC has now reviewed the replies and is sending follow-up requests for information, including on agreements between large tech companies and start-ups, to understand whether certain exclusivity clauses could have a negative effect on competition, Ms. Vestager said.
Another risk is that large tech companies could make it difficult for smaller foundation model developers to reach end users, Ms. Vestager explained. The EC is closely monitoring distribution channels to make sure that businesses and consumers still have a wide range of choice among foundation models. The EC has sent requests for information to better understand the effects of arrangements to pre-install small foundation model on certain devices.
Ms. Vestager added that the EC also has a number of other preliminary antitrust investigations ongoing into various practices in AI-related markets.
In terms of merger control, Ms. Vestager echoed the views of some of the panellists that partnerships between large companies and small AI developers are generally pro-competitive because they give access to the necessary components which allow AI systems to be developed. But such agreements should be monitored to ensure fair play because they can sometimes create entrenched market positions. The EC recently reviewed one of these partnerships from a merger control angle, but concluded that it was not reviewable under the EU merger control rules. Ms. Vestager added that the EC will continue to monitor relationships between all the key players in this fast-moving sector, including so-called “acqui-hires,” where one company acquires another mainly for its talent.
Olivier Guersent: Contributions on Competition in Virtual Worlds and Generative AI
EC Directorate-General for DG COMP Oliver Guersent summarised some of the feedback received in response to the EC’s call for contributions on competition in virtual worlds and generative AI. The EC received just over 50 submissions on virtual worlds and approximately 120 on generative AI.
The contributions noted that access to key inputs could become potential barriers to entry or expansion in the relevant markets. Key inputs for the development and deployment of generative AI systems were considered to be data; chips, computer infrastructure and cloud capacity; and technical expertise. Crucial inputs for virtual world technology were considered to be high-speed networks, cloud computing, data, AI and intellectual property rights. A number of contributions also called for caution in ensuring that applicable regulation is deployed at the right time — neither too early nor too late.
Benoît Cœuré: Role of Competition Enforcement in Fostering Competitive Generative AI Market
The President of the French Competition Authority Benoît Cœuré outlined the findings set out in the authority’s opinion on the competitive functioning of the generative AI sector, which was published on June 28, 2024.
The French authority concluded that the sector was characterised by high barriers to entry, and that the position of certain operators in other markets linked to generative AI could give rise to a range of competitive advantages, such as preferential access to inputs and advantages linked to vertical and conglomerate integration.
Mr. Cœuré identified a number of competitive risks, such as those associated with:
- Abuse by chip providers, such as price-fixing and discriminatory behaviour.
- Lock-in by major cloud service providers.
- Data access.
- Access to a skilled workforce.
- Open-access models that may lead to users being locked-in in some cases.
- The presence of companies in several markets that could result in self-preferencing and tying.
- Collusion.
In relation to the risks associated with minority investments and partnerships by large digital platforms, Mr. Cœuré noted that such arrangements were to be expected as the sector developed, but that some may raise risks such as weakened competition or lock-ins. Such partnerships should be reviewed under the applicable merger control rules or, if not, other antitrust rules may be applicable.
The opinion issued by the French authority puts forward a number of non-legislative recommendations, including to make the existing regulatory framework more effective, to increase transparency on investments by large tech platforms and to increase access to computing power through the development of public supercomputers.
Panel on Competition Dynamics in Virtual Worlds
Speakers were Fabiana di Porto (associate professor of innovation law and regulation, Sapienza University of Rome and adjunct professor fintech law, University LUISS Guido Carli, Rome); Joan O’Hara (senior vice president of public policy, XR Association); Sébastien Massart (director of strategy, Dassault Systèmes); Mariana de la Roche (member of the board, INATBA and director, Validvent). The panel was moderated by Cani Fernández Vicién (president of the Spanish Competition Authority).
Competitive Dynamics
Virtual worlds will impact the way people and businesses operate. The panellists noted that virtual worlds have the potential to bring efficiencies to many markets, by lowering entry barriers and encouraging innovation, for example. The use of virtual twins (i.e., virtual representations that reflect the behaviour of a physical product or service) is one way in which the technology could be used to accelerate innovation by digitally testing new production methods.
In general, virtual worlds are still in the early stages of development and new entrants should be encouraged and enabled. Some emerging challenges to competition include: the dominance of incumbent firms, which may deter new entrants; the significant investment required to successfully enter the market and compete; and interoperability between different platforms for users.
Panellists pointed to potential exclusionary conduct in the sector, such as tying and self-preferencing, as well as potential exploitative abuses, such as a platform exploiting data gained from the platform’s buyers and sellers.
Interoperability and Standardisation
As the virtual world space continues to evolve, issues of interoperability are becoming increasingly prominent, and a large proportion of the session was devoted to discussing the possible merging of different virtual worlds and systems. Virtual worlds are currently siloed but, as the sector continues to mature and develop, either full or partial interoperability should be made available in order to grant users the ability to operate cross-platform. This discussion recognised that interoperability may be defined or achieved in different ways and that there is not yet a broad consensus on how much interoperability is required.
The need for interoperability should be balanced against other considerations such as privacy and security. One panellist observed that AI and blockchain technology have the potential to make virtual worlds more secure and user-centric.
Interoperability and standardisation can be imposed either by regulation and/or achieved upon initiative of market players, based for example on user demand and preferences. The EU already provides many examples of a regulatory approach to a limited extent, both in the DMA and the Data Act.
Panel on Competition Dynamics in Generative AI
Speakers were Stefan Wagenpfeil (professor of software engineering and IT management, Private University of Applied Science-PFH Gottingen), Dominique Costesec (senior competition counsel, Google), Blanche Savary de Beauregard (general counsel, Mistral AI) and Thomas Senderovitz (senior VP data science, Novo Nordisk). The panel was moderated by Carlota Reyners Fonatana (director of information technology, communication and media unit at DG COMP, seconded to the cabinet of the EC president as acting digital advisor).
Competitive Dynamics
The current landscape for generative AI is competitive and vibrant, characterised by the sharing of models and research.
There was broad consensus among the panellists that the key inputs for the development and deployment of generative AI systems are computing power, high quality data, investment and talent. Some panellists considered that access to sustainable energy was quickly becoming an additional input given the vast amounts energy required to build and train AI models. It was observed that startups required access to all of these inputs simultaneously in order to successfully access the market.
Only a few players have access to these key inputs, which can represent barriers to entry. Competition authorities should ensure that such companies do not use their position to distort competition by limiting or denying access to key inputs to competitors. The integration of all businesses, including SMEs, into the evolving landscape is needed to ensure innovation and choice.
Initiatives to limit barriers to entry were also discussed. These include the creation of public supercomputers (which can be used free of charge in exchange for a contribution to open science, enabling operators to access computing power for training or fine-tuning generative AI models) and the emergence of technological innovations that reduce the need for computing power and data, such as synthetic data (generated by AI) that can partially replace real data and the use of smaller, more cost-effective language models. Whether these initiatives will truly encourage new entry remains to be seen.
The panellists agreed that authorities should recognise the necessity of partnerships for the development of the sector because they provide AI start-ups with access to critical inputs. Such partnerships should be seen as procompetitive, and over-enforcement in this area may deter large tech firms from entering into further partnerships. Partnerships in the AI space may bring other benefits; for example, they could lead to the creation of industry standards on aspects such as data exchange.
Open-Source Models
Open-source models can be reused or fine-tuned by other operators, which can reduce barriers to entry in AI markets. However, the term “open-source” in the AI sector covers a wide range of situations and not all open-source AI models are actually open. The majority of open-source AI models only make available the model’s weights, which can have a beneficial impact on competition for fine-tuning and deployment but does not significantly contribute to reducing entry barriers for players wishing to train foundation models. Fully open-source AI models (which make available the model’s weights, system source code, architecture, training data and training processes etc.) have now become the exception.
There was a broad consensus that open-source models would continue to co-exist with closed-source models given that each serves different needs. What ultimately matters, it was noted, is that there is sufficient diversity.
Compliance with Copyright
The rise of generative AI has triggered a debate about the appropriate protections for copyrighted data. This is because training data used to build these AI models often include copyrighted content such as online media. Panellists noted that this space was regulated to a limited extent; for example, EU copyright law has two exceptions that allow for text and data mining.
Panel on the Role of Competition Enforcement in Generative AI
Speakers were Ana Sofia Rodrigues (commissioner and former chief economist, Portuguese Competition Authority), Alex Stratakis (partner, Van Bael & Bellis), Sebastiano Toffaletti (secretary general, European Digital SME Alliance), András Tóth (vice president, Hungarian Competition Authority). The panel was moderated by Inge Bernaerts (director for strategy and policy at DG COMP).
International Cooperation
Competition authorities are undergoing a rapid learning process to quickly understand the emerging AI ecosystem across various sectors of the economy. This is a global challenge and so authorities worldwide are working together and reinforcing each other.
The panel opened by noting that regulating AI to ensure that consumers receive its benefits is not the sole job of competition authorities but requires a cross-sectoral approach. The competition tools should be used in a way that is consistent with other industrial and regulatory policies. At the same time, regulatory tools such as the EU’s AI Act should be implemented in a pro-competitive way. In this context, it was pointed out that the AI Act itself could act as a barrier to entry for smaller companies with more limited resources in light of the high compliance costs required.
Risks for Competition
The panel discussed the high input costs required to enter AI markets, in particular access to high-quality data. The risks relating to data access have been known for some time but these risks have only recently started to materialise; for example, we have seen various recent actions brought in relation to copyright. One panellist pointed out that the introduction of licence agreements with content owners is a recent trend, as a result of which developers are granted different levels of access to proprietary data.
The panellists picked up on a point discussed earlier in the workshop relating to ways to limit barriers to entry in these markets. Suggestions included the use of open-source public large language models and regulating generative AI as if it were a public utility to ensure access to key inputs on non-discriminatory terms.
The panellists also discussed the need to review key transactions in the AI space under the merger control rules. One of the panellists queried whether mergers in the AI space should be assessed against a “balance of harms” standard instead of “balance of probabilities” given the dynamic assessment involved.
Finally, the vice president of the Hungarian Competition Authority noted that the ongoing AI market study in Hungary, which is expected to be finalised by October 2024, is looking at the need to improve the accuracy of non-English language generative AI tools which have been trained on lower quantities of data than English language equivalents. He added that there is a shortage of AI experts in the smaller EU member states.
Professional support lawyer Elizabeth Malik contributed to this article.
This memorandum is provided by Skadden, Arps, Slate, Meagher & Flom LLP and its affiliates for educational and informational purposes only and is not intended and should not be construed as legal advice. This memorandum is considered advertising under applicable state laws.