Main moments
On Tuesday, Foreign Minister Espen told Barth Eide, in his semi-annual report to Parliament on important EU and EEA matters, that the EU's Artificial Intelligence Regulation (AI Act) “looks set to become a good regulatory framework”. The truth is that lobbying pressure from German and French companies threatens an essential part of the regulation, namely the regulation of the most powerful KI systems.
So-called foundation models are models such as GPT-4 (which is part of ChatGPT), which are trained on huge amounts of text and images. With good language skills, they can quickly adapt to new tasks. Therefore, they are used by many small companies to improve their products. Proposals for the AI Act included regulation of such models. It would mean that big tech companies, such as OpenAI/Microsoft and Google Deepmind, would have to ensure that the models they offer meet security and documentation requirements.
After intense lobbying pressure from KI companies such as Mistral and Aleph Alpha, France and Germany have reversed. They will now exempt the basic models from regulation.
Undermines the whole point
Giving exemption to the largest models undermines the whole point of the regulation, namely to regulate Risky KI Models without to slow innovation and development in areas where there is less risk.
It increases the risks associated with the most powerful models because base models are difficult to insure against bad behavior and can be used for evil purposes. Securing these models must be the responsibility of the manufacturers and not the users.
It also favors the few most powerful companies at the expense innovation and development among smaller players. It deprives the big companies of responsibility and puts the burden on the many small companies that won't have the expertise or ability to adequately secure their own systems.
Unless Germany and France reverse, the regulation could become an irresponsible regulatory framework that favors powerful companies at the expense of the safety of most people. If the regulations are to be in line with our interests, the Norwegian authorities and other countries must act.
More from Langsikt

A sea of data
Norway is in the process of building a digital twin of the ocean, but we use it primarily for reporting, not for innovation.

School should teach kids how to navigate the future
We will not be able to live without a mobile phone, nor will we do without artificial intelligence. Yet school protects children from the tools they must learn to master.

The Ten AI Commandments Link Principles to Politics
The committee's report proposes 44 concrete measures for Norway to build competence, capacity and supervision in the KIC economy.

The machine can find shortcuts in the tank
A machine has improved one of computer science's most fundamental ideas -- Dijkstra's shortest path algorithm. It marks a turning point, where the boundary between human insight and machine computation is blurring.
