The EDPS takes a tricky stance as regards a number of the options envisaged within the proposal. For example, the authority as soon as once more emphasised that classifying a number of makes use of of AI as “excessive threat” isn’t sufficient in circumstances the place such makes use of pose unacceptable dangers to basic rights (see para. 7 of the opinion). This features a.o.:
- any use of AI to hold out any kind of “social scoring”;
- any use of AI for automated recognition of human options in publicly accessible areas, corresponding to of faces, gait, fingerprints, DNA, voice, keystrokes and different biometric or behavioural alerts;
- using AI to deduce feelings of a pure individual aside from sure well-specified use-cases, specifically for well being or analysis functions;
- any use of AI techniques categorising people from biometrics into clusters in line with ethnicity, gender, political or sexual orientation, or different grounds for discrimination prohibited below Article 21 of the EU Constitution of Elementary Rights.
In response to the EDPS, such makes use of ought to be prohibited as they’re intrusive and have an effect on human dignity.
The EDPS additionally notes that the AI Act proposal exempts operators of high-risk AI techniques already available on the market or in use earlier than the AI Act’s applicability, besides in circumstances when these techniques are topic to vital adjustments of their design or objective or in case of “substantial modifications” (para. 12 of the opinion, see additionally Article 83(2) of the AI Act proposal). Nevertheless, the EDPS finds this resolution unclear, resulting in authorized uncertainty and a few high-risk AI techniques by no means falling throughout the scope of the AI Act. The EDPS recommends eradicating this exemption and making use of the AI Act to current high-risk AI techniques on the date of its applicability.
What’s extra, the EDPS means that the notion of AI “suppliers” ought to be additional clarified, and possibly (explicitly?) embrace AI operators who retrain pre-trained AI techniques. Though coaching is a basic a part of AI growth, the present proposal doesn’t clearly state whether or not actions corresponding to retraining or steady coaching ought to be thought of as a part of AI system ‘growth’. Because of this, it’s unsure whether or not operators participating in such actions may very well be assigned the standing of “suppliers” of AI techniques (para. 15-19 of the opinion).
Lastly, the authority shared particular suggestions on easy methods to make clear the proposal’s provisions on EDPS roles and duties as a notified physique, market surveillance authority and competent authority for the supervision of the event, provision or use of AI techniques by EU establishments, our bodies, places of work and companies (para. 29 et seq.).
* Updated info on the legislative course of yow will discover right here.
