“The AI Act Shouldn’t Stifle the Market Dynamics“

Opinion

Opinion Piece by ZEW President Achim Wambach

Regulation of the AI market must be faster and more market-oriented in order to prevent the formation of new oligopolies and ensure fair competition.

The major tech giants are poised to dominate the AI market. Europe shouldn’t rely solely on regulation, but must instead improve market conditions for its own corporations. An interview from the Börsenzeitung with ZEW President Professor Achim Wambach, conducted by Stephan Lorz, head of the Economic Policy Department.

BZ: Professor Wambach, the antitrust authorities have allowed the US tech giants to grow into oligopolies and ultimately reacted too late to concentration of power in the markets. Do you share the concern that this might now also happen in the AI market?

In the Monopolies Commission’s 2015 report, we actually assumed that the digital market had not yet solidified and was developing extremely dynamically. New stakeholders and products were constantly emerging. We were therefore of the opinion that this had to be allowed to take effect. After all, we didn’t know what would ultimately emerge and considered the antitrust authorities’ ability to control abuse – extended to platform markets – to be sufficient to put an end to any violations of competition rules.

BZ: But that didn’t work out as planned, did it?

That’s right. The market went on to concentrate on a few dominant players such as Alphabet including YouTube, Microsoft, Amazon, Facebook/Meta including Instagram, and Twitter, which happened faster than expected. The momentum declined rapidly. We realised that abuse control was not enough because it takes far too long to change anything in a timely manner, with lawsuits lasting six to eight years.

BZ: Clearly a misjudgement at the time?

In hindsight, of course it’s easy to say: yes. But I want to make it clear once again: It is necessary to let a market develop first. At the time, there were 10, 15 or more small new digital companies trying their hand at new business ideas. And the – now large – corporations that had bought up some of them didn’t yet know which ones would be successful. In such a phase, you can’t point the finger at a single company and brand it as anti-competitive.

BZ: So, what have we learned from this?

Many companies, large and small, are also emerging in the field of artificial intelligence with ever new business ideas. And even now we have to ask ourselves: In what situations does competition law suffice? Where in the market do we already need to brandish a regulatory bludgeon, knowing full well that this would slow down the momentum? But our experiences of the first digital wave will certainly have an impact – and our range of instruments has expanded.

BZ: Is the AI market still in the experimental stage, or has it long since moved on? I’m thinking of the cooperation between OpenAI and the French company Mistral AI, both with Microsoft.

These “cooperations” are indeed a problem. You could even see them as a precursor to a merger. At the very least, influence is being exerted. But even here, it’s not clear yet what will fly in business terms. And what are the platform effects?

BZ: Any ideas yet?

When it comes to AI, the question of data seems to me to be the decisive factor: Whoever has more data will also be able to create better AI. This is where we need to start. The antitrust authorities must ensure that data is freely available to all competitors. And in the case of specialised applications, companies must ensure that their data does not flow to the big players.

BZ: How do we get started?

The often heard “simple” solution of breaking up the company is not an option in my opinion. In the German rail sector, we would never think about removing the market power of Deutsche Bahn. There is an indivisible network, and it has to be regulated so that competitors can find traction and become successful. As a result, there is competition on the railways. So breaking up Facebook or Google would not help either. There would then be a northern and southern Facebook, for example, and at some point one would win out again over the other. Data-driven network effects also have an impact on AI platforms. We must therefore prevent them – just like with search engines or sales portals – from benefiting from broad knowledge, while favouring their own products and services and putting competitors at a disadvantage.

BZ: Is the antitrust toolkit sufficient for this?

During the first wave of digitalisation, we recognised that the previous abuse control was not sufficient. This is why the new section 19a of competition law was established in Germany, which provides more powers for supervision. The Digital Markets Act (DMA) has also created new instruments at the European level, such as the ban on self-preferencing.

BZ: But what is the point if slews of lawsuits delay the decisions of the supervisory authority while the companies concerned create new industry precedents?

Perhaps 19a will speed things up because the legal process has been shortened. The plaintiffs now have to go directly to the Federal Court of Justice. We’re still talking about years and not weeks, but the penalty, for example with the DMA, is quite high, which is why the vast majority of companies are more likely to comply. That alone will lead to changes in behaviour.

BZ: Is it not possible to break the market power of individual companies currently pushing into the AI market by using technical requirements? For example, with interconnectivity, such as between telephone companies, or an obligation to carry over data.

This is exactly what the DMA is trying to achieve: interconnectivity and data portability. This could take away the market dominance of individual corporations. In the telephone sector, I can also choose between Telekom and – let’s say – Vodafone; and yet the users can all reach each other. That would weaken the network effects. I would like to see a little more pressure from the market surveillance here.

BZ: But why is that not ultimately expected of Facebook, Instagram, and similar corporations?

There are already some requirements concerning interconnectivity and data portability for messaging services. But the DMA has only recently taken effect. The only thing missing is experience.

BZ: Right now we’re talking about the market dominance of US corporations. Would we be talking about it the same way if European corporations had been similarly successful? German Googles, Facebooks, or OpenAIs, for example.

I think so. But we shouldn’t limit the discussion only to regulation. We have to see for ourselves how we can create more dynamism in this country. And to achieve that, we will have to address the still-delayed broadband expansion as well as our weak start-up culture. Not to mention the low degree of digitalisation in administration along with the high level of bureaucracy. Regulation is always good for markets where a certain lethargy exists. If they are dynamic, as in the digital and AI markets, we in Germany are still lagging behind.

BZ: But success begets more success: The tech giants saturated with oligopoly billions are currently exploiting the AI market much faster than new start-ups are. How can we prevent these second-round effects?

Well, we do have an asset that we can use to put these tech conglomerates under pressure, and that’s our European Single Market. Entry into this market is very attractive, which means we can persuade the tech giants to play by our rules. Europe’s position in this matter is much more powerful than Australia’s, for example, who is of course also fighting against the market power of Facebook & Co.

BZ: How great is the risk of new oligopolies emerging in the AI market?

There is certainly a reasonable concern. We’re seeing huge investments from the tech giants that can afford it. They’ve already developed a sizeable lead. Not least because they’re able to dip into their massive global data reserves that they’ve developed for decades. They also have the necessary computing capacities. However, they still don’t have the requisite data expertise for individual industry applications. Domestic companies should build upon this in order to develop their own AI systems. The more general AI applications such as ChatGPT aren’t capable of that.

BZ: How much of a brake is German data protection law in this context?

It is true that with 16 different data protection authorities, we do not make it easy for ourselves. In fact, data protection already allows sufficient leeway, but I would like to see the authorities work more towards solutions and less towards prohibitions.

BZ: Despite the presence of major digital players such as Google and OpenAI/Microsoft, how can the AI market be made fair and open to new players?

We already have decent regulations with the Digital Markets Act (DMA), the Digital Services Act (DSA), the General Data Protection Regulation (GDPR), and with the AI Act. These must first be allowed to take effect. From a German and European perspective, data access is already the crucial point to create a fair market. I’m thinking of government and health data. Competitors must be able to access this data. Conversely, they must also be able to access the data collected, for example, by Google through its web search.

BZ: The recently passed AI Act has been hailed as laying the foundation for the new AI market by providing legal clarity. Do you agree with this praise?

It all depends on how it is implemented. With the GDPR, we in Europe were initially quite happy from a competition perspective, because we assumed that there would finally be uniform data protection rules. But no such luck! Each country has implemented the GDPR differently, with Germany being particularly strict. And because the authority in Ireland is smaller and decisions are postponed there, it is still worthwhile for companies to set up there. Let’s hope that this will be different with the AI Act.

BZ: What is your specific concern?

My concern is that the provisions could pose an entry barrier, especially for smaller companies. The AI Act must not stifle market dynamics – as has happened in the financial sector, where fintechs have been slowed down by regulation.

BZ: In light of the GDPR experience: Isn’t it time to abolish federalism in the digital sector and regulate everything uniformly at federal level?

Legally, this is not so easy. From a practical point of view, there is certainly much to be said for giving responsibility to a single entity. Especially since the European digital single market is still not complete. Just because you open an online branch in Germany doesn’t mean you can do the same in France. Tax law is also different. Here, we are not making any progress at all. And this sets us apart – despite our size – from the US domestic market. That's why it’s easier for companies there to grow to sizes that raise competition issues here.

This interview originally appeared in the Börsenzeitung (in German)