The European Parliament (EP) has recently[i] voted to adopt its negotiating position in a plenary session on the Artificial Intelligence (AI) Act.
Essentially following a risk based approach, the discussions over rules span around ensuring that the developments and use of AI applications and systems in Europe would in theory comply with EU rights and values including “human oversight, safety, privacy, transparency, non-discrimination and social and environmental well-being”.
In a nutshell, next to a revised definition of an AI system in line with the OECD version, the proposed to-do list, targeting providers and deployers among other actors, contains the following:
- ban on emotion-recognition AI;
- ban on “real-time” and “post” remote biometric identification and predictive policing in public spaces;
- ban on biometric categorisation systems using sensitive characteristics;
- ban on social scoring;
- ban on untargeted scraping of facial images, from the internet or CCTV footage, for facial recognition purposes;
- new set of restrictions for general purpose AI and foundation models;
- new set of restrictions on recommendation algorithms on social media;
- assignment of recommender systems to the “high risk” category, whereby placing higher scrutiny on recommender systems on social media platforms as to how they work. Consequentially, tech companies could be held more liable for the impact of user generated content.
Notably, the ban on “post” remote biometrics identification would be subject to the exception of law enforcement upon prior judicial authorisation in the context of serious crimes.
Furthermore, those generative AI systems based on foundation models, such as ChatGPT, would have to comply with transparency requirements and put in place effective safeguarding mechanisms against illegal content. In the case of use of copyrighted data for training models, detailed summaries would need to be made publicly available. Registration in the EU database will also be obligatory for foundation models.
Importantly, alongside defining responsibilities across AI value chain of various actors involved, the EP proposes the development of non-binding standard contractual clauses to regulate rights and obligations in line with each actor’s level of control in a given value chain.
Taking into account that the AI Act is set to be also applicable to providers and users of AI systems located outside of the EU – provided that the output produced is intended to be used in the EU, these developments are pivotal for the Swiss market.