Stance on The Proposal for a Regulation Laying Down Harmonised Rules on Artificial Intelligence – Artificial Intelligence Act

Mesarcik, M., Solarova, S., Podrouzek, J., Bielikova, M.

This document is the KInIT Stance on the proposal of the European Parliament and of the Council Laying down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act) and Amending Certain Union Legislative Acts.

Executive summary:

The definition of AI systems shall be more precise to cover the use of predefined techniques and
not only the development. Furthermore, the influence of generated outputs shall be considered
and material criteria for techniques added as a binding prerequisite.

The scope of prohibited practice shall be clarified as the current scope may be extensive and
additionally contains loopholes for exploitation. More evidence shall be presented as per specific
prohibited practices by the legislator.

Several crucial areas of high-risk AI systems are absent in Annex III namely AI systems used in
the context of environmental protection, climate change and transport. Furthermore, attention
economy (including social networks) and transportation shall be specific areas. From the procedural
point of view, due to technological advancement, the EU shall opt for a more dynamic approach
in terms of updating Annex III.

AIA shall make direct reference to the HLEG AI work on trustworthy AI. Specifically, acknowledgement
of ALTAI in recitals and obligation to conduct ethical assessment shall be part of the binding text
of the regulation.

AIA shall contain specific rules for legacy AI systems that are not currently covered by the proposal
considering the feasibility of compliance. Furthermore, the notion of significant change shall be
provided.

We are of the opinion that transparency obligations shall be covered in a more comprehensive
manner. Considering the state-of-the-art, more nuanced approach including delegated acts
considering technological development shall be preferred.

Deepfakes require a more specific approach in the regulation and shall be considered as a high-risk
AI system considering real-world scenarios severely violating fundamental rights and freedoms.
Text-based deepfakes shall be also included in the regulation.

Remote biometry shall be addressed more specifically with precise rules instead of a general ban
on the use of remote biometry. Many concerns are already mitigated by existing laws. Further
concerns shall be regulated considering fundamental rights at stake.

Stronger EU oversight is needed in case of monitoring of compliance with the AI regulation. KInIT
supports the creation of a more active EU supra authority for this purpose.

Public authorities shall not be excluded from imposing administrative penalties. The exception in
question shall be interpreted narrowly.

Cite: Mesarcik, M., Solarova, S., Podrouzek, J., Bielikova, M. Stance on The Proposal for a Regulation Laying Down Harmonised Rules on Artificial Intelligence – Artificial Intelligence Act. Kempelen Institute of Intelligent Technologies. September 2021. DOI: 10.5281/zenodo.12567879