• Recognition of chatbots as search engines opens door to designation
• Portrayal of AI as search or virtual assistants may spark litigation
• Experts prefer ‘cautious’ approach to regulate AI

The European Commission’s (EC) recent decision to recognise AI chatbots as providers of online search sparks new questions over their potential designation under the Digital Markets Act (DMA), according to an analysis by this news service.

If chatbots are considered search engines for the purposes of benefitting from a DMA provision that forces Google to hand out search data to rivals, does that mean they are also viewed as search engines when it comes to determining whether they should be treated as regulated gatekeepers?

Several digital and competition experts say that chatbots such as ChatGPT could qualify for DMA designation as search engines, and hence face data-sharing and ranking obligations if they meet the law’s thresholds. But the same experts also argue that a more cautious approach would be preferable.

When the DMA debuted in November 2022, it did so as a law designed to rein in the market power of digital gatekeepers in ambits such as online advertising, social media, operating systems and online search.

The boom of AI services occurred soon after, and as this technology has been transforming the digital ecosystem at breakneck speed since then, the Commission has progressively incorporated AI considerations into its DMA probes.

The Commission is “hugely interested” in understanding how AI services function, and is requesting lots of information with that purpose, said a source familiar with the matter.

This becomes clear when looking at some of the latest DMA investigations into Google. The Commission has mandated that the company must make Android interoperable with rival AI assistants, and has also specified how Google should comply with an obligation to share search data with other search engines.

When it comes to the data sharing decision last month, the Commission specified that Google’s parent, Alphabet, shall share the data with any third party that meets the “online search engine” (OSE) definition in the DMA.

“Alphabet shall not exclude third-party undertakings, such as undertakings providing AI chatbots with OSE functionalities, to the extent that these undertakings provide an OSE … even if the OSE is provided as part of a broader service,” the decision reads.

In the DMA context, the definition of an online search engine is used to establish which platforms should be caught by the law. If a platform meets this definition, as well as the DMA’s thresholds for number of users, and turnover or market value, it must abide by its rules.

So, could AI chatbots be treated as search engines under the DMA?

According to Alba Ribera, assistant professor in law and technology at VU Amsterdam, they “absolutely” can. “If the Commission believes that is the approach that it must take, it can designate an AI provider as an online search engine,” she said.

AI chatbots can meet the definitions for several categories of DMA core platform services, an EC official said.

This news service has previously reported that they could also be viewed as “virtual assistants”.

However, the EC official continued, they could also be considered as something that is not in the current list of core platform services – a more encompassing definition that would not only cover search, but a broader set of the functionalities that an AI chatbot can deliver.

The legal debate around AI designation is still ongoing, the source familiar with the matter said, but added that if the EC designates AI services as virtual assistants or search engines, “everybody will litigate this”.

The definitions of search engines and virtual assistants were not drafted with AI in mind, this source added.

The delineation game

If the Commission ends up creating a new, tailor-made category to designate AI services, the providers of these services could use the regulation’s legal ambiguity in their favour.

The boundaries between different core platform services in the DMA are not always crystal clear. For example, when the law started to be implemented there was a debate on whether TikTok was a social network or a video-sharing platform.

This is important because certain obligations of the DMA apply to some core platform services but not to others. The Commission must therefore discern how to delineate these services in an adequate way.

If they are designated, AI firms could provide “alternative delineations” of their services, said Ribera at VU Amsterdam. The EC would need to at least respond to those arguments, which could trigger debates before court, she added.

According to Ribera, AI companies could play a “game” in which they say, for example, that they should be designated as online search engines because that is more favourable to them in terms of the obligations they would have to face.

The road ahead

For the scholar, there is enough leeway within the DMA to accommodate AI enforcement. However, it is also about having the right obligations in place that are suited for this industry, she added.

The source familiar with the matter argued that to properly capture AI services, the DMA should be amended. For that to happen, the Commission would need to carry out a market investigation that could take around 18 months, and it would also need to get the support of the European Parliament and Council for legislative changes.

During this process, the EC could introduce new obligations to the law that would be more suited to regulate the AI ecosystem.

If what the Commission wants is to capture AI in the DMA as soon as possible, due to political pressure or other reasons, legislative changes is not the best solution because it would take time, Alba Ribera said.

She added, however, that the “cautious” approach is the most advisable route to take right now. “Especially because we have seen the leaps that AI is taking in the market. We can regulate AI today, but tomorrow we might have a different iteration and a different manifestation of AI that is even more harmful than the initial iteration,” she said.