The guidelines for ethical artificial intelligence will not propose any changes to the legislative framework, according to a document seen by EUROPE, which will be officially presented by the Commission on 8 April (see EUROPE 12227/6).
Instead, it identifies ethical principles for the development, deployment and use of artificial intelligence systems, namely: respect for human autonomy, damage prevention, equity and “explicability”. It also provides guidance for building trustworthy systems and provides a concrete and non-exhaustive checklist to achieve this trust. A separate document provides an updated definition of artificial intelligence, considered to consist of software (and possibly also hardware) designed by humans (...) that can either use symbolic rules or learn a digital model and can also adapt their behaviour by analysing the impact of their previous actions on the environment.
“There are already regulations in place to strengthen the reliability of AI - whether it is product safety legislation or accountability frameworks”, states the document. “To the extent that we believe that regulation may need to be revised, adapted or introduced both as a safeguard and as a facilitator, this issue will be addressed in our second initiative, a policy for investment and policy recommendations”, he continues.
Sensitive issues
These guidelines are based on a first draft published in December 2018 (see EUROPE 12162/10). They were prepared by the High Level Group on Artificial Intelligence and made use of the results of the public consultation launched earlier this year. At the time, the most controversial questions related to the degree of explanation available, the possibility of refusing certain parameters, lethal weapons or scoring systems.
On explicability, for example, the degree of explicability - transparent procedures, communicated objectives and explicable decisions - “depends strongly on the context and severity of the consequences, if the result is wrong or inaccurate”. On autonomous lethal weapons again, the document supports the European Parliament's opinion calling for an EU common position and supporting an internationally negotiated legally binding instrument to ban them (see EUROPE 12095/7). (Original version in French by Sophie Petitjean)