Focus on the CNCDH's recommendations to promote an ambitious legal framework for AI

Focus on the CNCDH’s recommendations to promote an ambitious legal framework for AI

4 juillet 2022

On April 21, 2021, the European Commission presented its draft regulation of AI. More recently, last December, Slovenia, which held the EU Council Presidency at the time, submitted a compromise text concerning this draft law aimed at prohibiting social scoring, framing the use of biometric identification and high-risk AI systems. It is in this context that the National Consultative Commission on Human Rights (CNCDH) adopted last April an “Opinion on the impact of artificial intelligence on fundamental rights” and published 19 recommendations.

The CNCDH is the French national institution for the promotion and protection of human rights. Created in 1947 on the initiative of René Cassin, Nobel Peace Prize winner, the CNCDH is accredited to the United Nations and is independent. It advises the public authorities and monitors France’s international commitments in the field of human rights and international humanitarian law. Its mission is also to educate and raise awareness about human rights.

To advise public authorities and inform their decisions, it publishes studies, reports and opinions.

The opinion on the impact of artificial intelligence on fundamental rights

While the draft regulation of AI is in the process of being adopted and the Council of the EU continues its work, the CNCDH published this opinion adopted unanimously on April 7. The CNCDH is in full agreement with the proposals made by Slovenia last December and submits a total of 19 recommendations, summarized below:

Recommendation 1: The CNCDH recommends that institutional communication use more neutral and objective terminology than “artificial intelligence”, such as “algorithmic decision support systems” (ADSS).
Recommendation 2: The Commission recommends strengthening, within the proposed EU regulation on AI, the provisions to ensure the establishment of a binding legal framework, guaranteeing the effective respect of fundamental rights. Furthermore, the CNCDH recommends the adoption, within the framework of the Council of Europe, of a “Convention 108+ on AI”.
Recommendation 3: The CNCDH recommends that a human rights-based approach be taken into account in the reforms underway, as long as they are intended to guarantee the respect of fundamental rights.
Recommendation 4: The CNCDH recommends the prohibition of the use of choice interfaces when their purpose or effect is to manipulate users to their detriment by exploiting their vulnerabilities.
Recommendation 5: The CNCDH recommends prohibiting any type of social scoring implemented by government agencies or by any company, public or private.
Recommendation 6: The CNCDH recommends prohibiting the remote biometric identification of persons in the public space and in places accessible to the public, while allowing its use by way of exception, provided that it is strictly necessary, appropriate and proportionate for the prevention of a serious and imminent threat to the life or security of persons and to that of works, installations and establishments of vital importance.
Recommendation 7: The CNCDH recommends that further thought be given to identifying the benefits and limits of using AI in legal proceedings.
Recommendation 8: The CNCDH recommends prohibiting the use of emotion recognition technologies, while admitting their use by exception when they aim to reinforce the autonomy of individuals, or more broadly the effectiveness of their fundamental rights.
Recommendation 9: The CNCDH recommends that the user of an AI system assess the impact of the use of this system on fundamental rights and, in case of identified risks, proceed to their evaluation taking into account the probability and the seriousness of the latter.
Recommendation n°10: Depending on the risks generated by an AI system on fundamental rights in a particular context of use, the CNCDH recommends to ensure, prior to the decision to use it, a consultation of the stakeholders, according to adapted modalities, by including, for example, the staff representatives and, more widely, the persons targeted by the AI system.
Recommendation n°11: The CNCDH recommends to set up a supervision of the AI system, according to a procedure that may vary according to the risks of infringement of fundamental rights as identified by the impact study, in order to maintain a continuous vigilance on the part of the user with regard to the effects of the system, especially its discriminatory effects.
Recommendation 12: The CNCDH recommends that public investment be encouraged in the design of training and information tools accessible to the greatest number of people.
Recommendation n°13: The CNCDH recommends the organization of national consultations based on the model of the National Consultative Committee on Bioethics.
Recommendation n°14: The CNCDH recommends that the French National Education system reinforce the training of students on the technical, political and societal issues of artificial intelligence and to propose, to this end, educational materials for teachers.
Recommendation 15: The CNCDH recommends
– guarantee human intervention for the control of individual decisions resulting from an AI system according to modalities corresponding to the level of risk of the latter;
– to ensure the effectiveness of the intervention through appropriate training and information of the intervener on the characteristics of the system, without imposing any particular constraint when he deviates from the recommendation issued by the AI system;
– ensure that users of the public service systematically maintain alternative access to a human agent.
Recommendation n°16: The CNCDH recommends recognizing the right of AI system users to set their own criteria, in particular in order to determine the selection and presentation of the content received, and more generally in the case of human-machine interactions.
Recommendation n°17: The CNCDH recommends to systematically inform people when they are exposed or brought to interact with an AI system and, when they are subject to a decision, that this decision is based, if necessary, partly or totally on an algorithmic processing.
Recommendation 18: The CNCDH recommends that the person concerned should be guaranteed the right to have any individual decision based wholly or even partly on algorithmic processing reviewed by a human being, as soon as it has significant consequences for that person.
Recommendation n°19: The CNCDH recommends that administrations communicate in an intelligible form information on the functioning of the algorithm, as well as on the possible part played by human intervention in the decision-making process. It also recommends that consideration be given to extending this obligation to private organizations.