“Europe is entangled in a regulatory network that lacks clarity”. Thus Markus Reinisch, Vice President Public Policy for Europe of Meta, speaking to Adnkronos on the sidelines of the Comolake conference about the reasons that prevent the company from launching in the EU its new multimodal AI model Llama 3.2.
“Perhaps the current European system of rules works in a sector that evolves much more slowly, but is not suitable for the technological one. This approach risks blocking the growth of European companies and makes life very difficult for companies that aim to bring potential growth drivers in Europe”.
It is not clear either what kind of data can be used to train AI models in the EU, nor to what extent it can be used, explains Reinisch. And then there is the issue of timing: the conversations between Meta and the EU institutions began in the summer and the process will continue until at least the first quarter of 2025″. A similar process in the United Kingdom, however, took “only a few weeks”, the manager remarks, highlighting the consequences in terms of competitiveness.
As Europeans “open source is the best thing that could have happened to us”. Reinisch continues. The executive explains why open AI models can open up new opportunities for European companies.
“The inconvenient truth is that the development of foundation models is incredibly expensive,” explains the executive, speaking of the large and versatile AI models underlying modern chatbots. Meta currently has an investment rate of around 40 billion dollars per year. “We are not even the largest investors, the value of the investments of the top four operators is around 100 billion in 2024.”
This is not a good sign for Europe, because companies and institutions would have difficulty even raising similar sums of money. Not to mention the cost and amount of energy needed to train foundation models, Reinisch continues.
Given that the development of foundation models is a huge barrier for the EU, the executive reasons, European companies should exploit open source systems to avoid the development phase and do “what they do best, which is building verticals”.
The example he gives are the Ray-Ban glasses developed in collaboration with Meta, equipped with sensors and internal AI, built with “European skills and construction quality, on a technological platform developed in the USA”. Today they are the “most successful series of Ai devices” on the market, Reinisch remarks.