The rules to adopt with AI in training

A framework is being put in place in Europe

A robot head on an empty classroom image

We often use the expression "legal vagueness" to refer to areas or subjects in which the justice system has nothing to say. However, this formulation is nonsense, because as soon as a judge is seized of a case, he or she issues a ruling that becomes part of jurisprudence and contributes to the legal framing of an issue.

So, even if the subject seems recent, there is already a framework surrounding the issue of artificial intelligence with training. However, at the beginning of 2024, the European Parliament adopted the "AI Act", which further defines the issue.

It will come into force no more than two years after its adoption. It will oblige school systems to observe modulated requirements, depending on the risks incurred by users.

Uses that pose an unacceptable risk, such as emotion recognition or social rating, will simply be banned.
High-risk uses, such as assessing learning skills, guiding users through the process and monitoring behavior, will have to comply with personal data protection law, a restrictive regime for the data used to avoid bias, and an obligation of transparency and information.
The latter must also guide so-called low-risk applications, such as conversational robots in a learning system, which must make it clear that this is a talking AI and not a human.

Running time: 11min46

Image: Alexandra_Koch from Pixabay

Learn more about this resource

See more resources from this institution