In-depth reporting and analytical commentary on artificial intelligence regulation. No legal advice.

Hungarian competition authority GVH sees huge opportunities in AI, raises concerns over lack of support of “minor” languages

Context: The Hungarian Competition Authority (GVH) launched an AI market study last winter, with the original target of releasing a study in the summer (July 7, 2024 ai fray article).

What’s new: While potential concerns over competition in the AI sector were emphasized at the start of the project, the study released earlier this week (press release) reflects a balanced perspective on both the tremendous opportunities created by AI and the market dynamics that watchdogs must keep an eye on. Large parts of it read more like an economic policy paper, full of constructive recommendations, than an abstract indictment of market actors. With the Hungarian language being spoken by relatively few people, many of whom are not fluent in a second language, the GVH also explains the potential implications of AI systems being optimized for the world’s most widely-spoken languages. But it notes that two Hungarian LLMs are in the works.

Direct impact & wider ramifications: Unlike some other reports by competition authorities, not only but also in the AI context, this one does not hint at imminent enforcement action, though the agency will keep an eye on market dynamics. It talks about the need to invest and educate in order to seize the AI opportunity. That also includes investments in high-performance computing.

The GVH report explains the basics of the AI technology stack, different levels of AI and various key applications. It also sums up what other competition regulators, such as the UK Competition & Markets Authority (CMA), have said in their reports.

It’s not that they aren’t concerned about potential market concentration. The GVH is definitely going to be watching developments closely. But the study is clearly not designed to build a case for near-term or massive intervention.

The GVH makes recommendations that are all about economic and industrial policy, and not about proposing new legislative initiatives for the purpose of giving competition watchdogs greater powers:

  • Promoting the use of AI-based solutions and services in the domestic business sector, with a particular focus on SMEs.
  • Provision of up-to-date and transparent information to consumers.
  • The revision of the national AI strategy of Hungary, with particular attention to the development and support of specialised training courses; the rational integration and coordination of AI related research directions and the government’s AI strategy; the continuous development of an adequate IT infrastructure, including the already available but limited supercomputing capacities; and the creation of an ecosystem for AI development (data economy, education for technology applications, benchmark evaluation system, etc.).
  • Targeted support for the development of Hungarian language models.

Put differently, they say “invest and educate” rather than “enforce and regulate,” though there is no reason to doubt they wouldn’t be enforced to do the latter to address market failures if necessary.

Based on the holistic and forward-looking nature of the report, it would probably make sense for the Hungarian government to get the GVH more closely involved in the development and implementation of its national AI strategy, simply because of the knowledge of the subject and expertise in the field of market mechanics that it could contribute.

The part about language models for “minor” (i.e., less widely spoken) languages is an interesting one that was not found in the reports some of the GVH’s counterparts have released. Those other competition regulators don’t have to worry about it because their countries’ languages are spoken by far more people than Hungarian (13-15 million people). And it’s not just the number of (native) speakers, but also about how difficult a language is to learn and to speak and write correctly (the report notes that machine translations into the Hungarian language contain many grammatical errors).

Another key factor is whether the local population is fluent in other languages. To ai fray‘s founder’s knowledge, many Hungarians living near the Austrian border speak German. The GVH report says, however, that most Hungarians do not speak a second language. That is a major difference from, say, the Netherlands or Scandinavian countries, where most movies and TV series are not synchronized but merely subtitled. In those countries, English is virtually everyone’s second language, to the extent that Scandinavian secretaries are, on average, more articulate in English than, for instance, German (let alone southern European) executives.

The GVH is profoundly concerned that many Hungarians could be left behind if they can’t interact with AI systems in their own language, or only with shortcomings such as grammatically incorrect answers. The report notes that global AI providers don’t make any adaptations of their products for the small Hungarian market, and they feed their LLMs with content that is available in the world’s major languages. But at least there are two LLMs under development in Hungary.

ai fray agrees with the GVH that this is an important issue, and the European Union, which has two dozen official languages (plus many of its member states have additional ones, such as regionally recognized languages in Spain), should also support efforts to have AI systems trained in more languages than just the top 5 or top 10 in the world.

The GVH report properly distinguishes between economic and cultural implications. Both are valid concerns.

Actually, AI technology is an opportunity to preserve such cultural heritage and to level the playing field between major and minor languages in certain respects, but some markets may not be much of a priority for major global players.