November 21, 2023
Germany, France and Italy have reached an agreement on a strategy to regulate artificial intelligence. The agreement comes on the heels of infighting among key European Union member states that has held up legislation and could potentially accelerate the broader EU negotiations. The three governments support binding voluntary commitments for large and small AI providers and endorse “mandatory self-regulation through codes of conduct” for foundation models while opposing “un-tested norms.” The paper underscores that “the AI Act regulates the application of AI and not the technology as such” and says the “inherent risks” are in the application, not the technology.
“The European Commission, the European Parliament and the EU Council are currently negotiating how the bloc should position itself in this new field,” CNBC reports, noting that “the Parliament presented an ‘AI Act’ in June, with the aim of averting safety risks from AI applications and avoiding discriminatory effects, but without slowing down the innovative power of this new technology in Europe.”
Financial Times reports that the “UK will refrain from regulating AI ‘in the short term.’”
During recent discussions, the Parliament proposed to make the code of conduct binding only “for major AI providers, which are primarily from the U.S.,” CNBC says, adding that “three EU governments have warned against this apparent competitive advantage for smaller European providers,” which “could lead to less trust in the security of these smaller providers and therefore fewer customers,” concluding “the rules of conduct and transparency should be binding for everyone.”
The paper mandates that developers of foundation models must “define model cards, which are used to provide information about a machine learning model,” writes Reuters. “The model cards shall include the relevant information to understand the functioning of the model, its capabilities and its limits and will be based on best practices within the developers community.”
An AI governance body would help create guidelines and check the application of model cards, according to the paper. Initially, there would be no sanctions for model card violations, but if patterns of conduct are identified over a period of time, a sanctions system could be effected, Reuters says.
In October, the U.S. State Department “redlined” the June version of EU’s AI Act warning that “the legislation put forward by the 27-nation bloc might inadvertently favor large AI companies at the expense of smaller startups, as well as curb investment in AI technology due to its onerous compliance costs,” according to PYMNTS.