This article was originally published on the OSOR platform.
During a conference in June, the French President, Emmanuel Macron, took part in a discussion where he announced an investment of €40 million in open source large language models for AI. This announcement was made alongside Arthur Mensch, one of the co-founders of the recently established and well-funded open-source AI company, Mistral AI.
Open source AI has been a central topic of debate concerning innovation and how to bring the EU up to speed in the field of artificial intelligence. With public administrations across the EU already embarking on the use of AI to digitalise their public services, questions regarding its impact and sovereign characteristics have compelled legislators to ponder how the EU can benefit from such innovation. In a 2021 report commissioned by the European Parliament, researchers examined the advantages and limitations of open source AI. When looking at cases where it has been employed in public administration, we can already discern the gap created this year with the introduction of ChatGPT and similar systems.
We have observed projects like Entrada, a Big Data platform specifically developed for constructing applications to detect botnets and other malicious systems, showing promise for the digitalisation of public administration. Other initiatives, such as CAI-X, a Danish knowledge and competence centre that facilitates connections between clinicians from hospitals and engineers and researchers at universities, have explored potential pathways forward for AI. However, as indicated by the recent surge of interest in AI projects, public administration will still need to define new approaches to harness its benefits while avoiding problems such as bias in AI systems.
During his speech, Macron declared his intention to support the emergence of European AI clusters to provide an alternative to American actors. Interestingly, he outlined a comprehensive policy that encompasses investments not only in open source AI but also in supercomputers and large language models for the French language. He also highlighted the need to reconsider intellectual property rights to allow for AI training without infringing upon the rights of the original creators and the necessity of proper training and education in Europe. Finally, he connected these investments to the potential gains his government could achieve by creating conditions conducive to the use of AI in administrations.