GenAI: EU Commission EVP Vestager at the French Autorité de la Concurrence

Many thanks for your kind invitation to address this very important topic.

Everything about Artificial Intelligence is exciting and, to be honest, a little bit scary. So many policy questions are still waiting for an answer, and today’s Webinar is an important contribution to finding these answers.

To start with, we need to consider the macroeconomic dimension, which Daron Acemoglu’s paper analyses. So let me warm you up to his presentation by discussing a few of the highlights. I think he offers a very important note of caution on the risks versus the benefits of AI. He suggests the productivity gains from deploying AI may actually turn out to be more modest than expected; while at the same time he warns against the risks of a widening gap between capital and labour income.

Daron’s paper also offers a number of suggestions for policy makers, such as progressive taxation or education and training programmes for workers to adapt to the new environment. Let me add another instrument that I happen to know very well: Competition policy. If AI and related markets remain contestable, competition will stimulate greater productivity and innovation for AI and the related sectors. Healthy markets will help bring down prices for consumers, and that in turn can help spread the benefits of AI more widely.

At the same time, we know how quickly these markets are growing and changing, and that tipping points can be reached quickly. For a regulator, being able to act quickly means being informed early about what is going on.

That is why the Commission has already become active in the competition aspects of AI. Last year, we started investigating the sector using antitrust and merger tools. And we reached out to the market through a call for contributions to better understand the AI sector. Each of these initiatives has improved our understanding of this highly dynamic area.

On 19 September we published a competition policy brief summarising the results of a call for contributions; a workshop we held in June; and our market investigations. It’s an interesting read – we can share the link later.

The first point to note is that generative AI is not happening in a vacuum. The field is already dominated by large digital players who might want to leverage their dominance in parallel markets, or use foreclosure strategies to gain unfair advantages. Too much ‘laissez faire’ could create the conditions for market tipping.

There are three types of risks we need to look out for:

• First: upstream markets. Bottlenecks and early consolidation of market power can give large digital players control of key inputs. Large digital players could also enjoy preferential access to AI’s essential components, and block competitors from accessing them;
• Second: downstream markets. Where big tech controls the distribution and commercialisation of AI applications. This can take many forms, like exclusivity clauses, self-preferencing, refusal to supply, tying or bundling, non-competes or lock-in strategies;
• Third: killer or reverse killer acquisitions to stop competition before it begins. I would also put in this category predatory hiring of talent (i.e. foreclosing competitors by hiring their AI key staff) and abusive non-compete clauses.

These risks have led us to initiate a number of investigations into partnerships in the sector, namely those where big companies set up partnerships with small AI developers. To be clear: These partnerships are important. They give AI developers access to the necessary components and allow AI systems to be developed.

So in principle, I consider these deals to be pro-competitive. But they can sometimes entrench market positions, especially through exclusivity rights. So we need to keep an eye on them to ensure fair play.

From a merger perspective, AI partnerships raise two questions.

First: can a partnership can be scrutinised as a concentration? Under the EU Merger Regulation, that is only the case if one of the partners obtains a controlling influence over the other.

• This question was very central to our review of Microsoft’s partnership with OpenAI. The partnership was not formally notified, but DG COMP closely scrutinised the situation of control over OpenAI even before the firing and re-hiring of the OpenAI CEO, and before Microsoft got an observer seat on the OpenAI board.
• After a detailed consideration, we decided over the summer that Microsoft does not control OpenAI.
• But let me stress that our rules allow us to keep a close eye on this partnership. We can and will continue to monitor any changes in the relationship.
• At the same time, OpenAI benefits greatly from the significant investments it receives from Microsoft. As long as it stays independent and may choose to enter into other or additional partnerships, these investments do not constitute a structural change in the market that would merit merger review.

The second question is whether AI partnerships that actually involve an acquisition of control are reviewable, either at the European or national level.

• This question is now all the more important following the Court’s judgment in the Illumina / GRAIL case. As you know, that judgment invalidated our revised approach to Article 22 referrals.
• This revised approach was adopted in order to close a jurisdictional gap: today, highly valued acquisitions of innovative and competitively significant start-up companies active in the tech and digital sectors typically do not meet the turnover thresholds of the EU Merger Regulation.
• The Court’s judgement provides an important legal clarification, but it does not change the fact that we need to examine these mergers. We are therefore already reflecting on how we can best ensure that sub-threshold deals with a significant competitive impact in Europe can be reviewed in the future.

In the short to medium term, it remains possible for Member States with jurisdiction under their national rules, as well as for Member States without merger control rules (like Luxembourg), to refer a merger to the Commission.

A number of Member States have adopted so-called call-in provisions which allow their national authorities to assert jurisdiction over a case, insofar as the conditions set out under their national law are fulfilled.

And then there are Member States that use either deal-size thresholds or market share thresholds, which allow them to review mergers where a target has little or no revenues.

For AI, as for other areas, this means the Commission will have to work closely with the national competition authorities of these Member States, to make sure there is still adequate scrutiny of transactions involving companies whose turnover does not reflect their competitive potential.

• For example, in the case of the acquisition of certain assets of Inflection by Microsoft, seven Member States submitted to the Commission a request to review the case.
• While all Member States decided to withdraw their requests following the Illumina/GRAIL judgment, this sequence still shows that partnerships and/or agreements akin to those entered into by Microsoft and Inflection may constitute a structural change in the market and amount to a reviewable transaction under the EU Merger Regulation.

For competition enforcers, it’s clear what must be done. We must continue to monitor transactions in the AI sector and assess our options on how to address new challenges. Further change will come, and we have to be able to react to that quickly.

Merger control is not the only tool where we have to do that.

• Where the partnership constitutes in fact a merger, the Court’s Towercast judgment allows Member States to carry out an ex-post review under antitrust rules.
• Of course, ex-ante reviews under EU or national merger control rules are preferrable, because they provide greater legal certainty, they are timelier, and they are more effective.
• Furthermore, and in particular in case of partnerships that do not constitute a merger, companies have to play by all competition rules, including antitrust rules and the DMA.

This is of course only one piece in the puzzle. Other regulatory and legal issues remain open – from the risks of fraudulent ‘deepfakes’, to racial and gender discrimination, to copyright enforcement and everything in between.

This is yet another area where Competition Enforcers need to put their ‘advocacy’ hats on: We must make sure that the full range of AI-related regulations are consistent with the principles of fair competition and a level playing field.

We have still only seen a fraction of what AI will do to our economies and our societies – it’s an exciting thing to watch! For my part, I know I will be keeping “a close eye on AI”.

Thank you.

Source – EU Commission