A/78/310 and organizational measures. A right to explanation should provide for additional transparency. 31 34. In the report mentioned earlier, 32 the Special Rapporteur had noted that, in cases in which data subjects are subjected to automated decision-making or profiling, they should be able to understand the way in which the information concerning them will be processed (whether artificial intelligence is involved, for example) and with meaningful information about the logic involved and the significance and the envisaged consequences. 35. On that point, the Spanish Data Protection Agency has pointed out that “[t]he word ‘meaningful’ […] must be understood as information which, once provided to the data subject[s], makes them aware of the type of processing that their data is undergoing and provides certainty and trust as to the associated results”. 33 36. The Agency has also stated that: Complying with this obligation by offering technical references on the implementation of the algorithm may be obscure, confusing or lead to information fatigue. Sufficient information should be provided to enable the subjects to understand the behaviour of the processing. Although it will depend on the type of [artificial intelligence] component used, an example of the types of information that may be relevant to the data subject would be: • Details about the data used for decision-making beyond just the category, especially information regarding the duration of use of the data (how old t he data are). • Relative importance or weight given to each of the data in the decision -making. • Quality of the training data and the type of models used. • Profiling activities conducted and their implications. • Error or precision values according to the specific metrics used to measure the validity of the inference. • Existence or non-existence of qualified human supervision. • References to audits, especially audits on possible deviations of inference results, as well as the certification or certifications of the [artificial intelligence] system. For adaptive or evolutionary systems, the last audit conducted. • If the [artificial intelligence] system includes information referring to identifiable third parties, the prohibition of processing such information without legitimization and the consequences of doing so. 34 37. The European Data Protection Supervisor has issued an opinion suggesting that if the Commission were to put forward a new artificial intelligence-specific regulatory framework, a certain number of reasonable safeguards should apply to all artificial intelligence applications, regardless of the level of risk, such as having technical and __________________ 31 32 33 34 12/20 European Data Protection Board and the European Data Protection Supervisor, Joint Opinion 5/2021 on the proposal for a Regulation of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), 18 June 2021, p. 17. Available at https://edpb.europa.eu/system/files/2021-06/edpb-edps_joint_opinion_ai_ regulation_en.pdf. A/77/196, para. 55. Spanish Data Protection Agency, Adecuación al RGPD de tratamientos que incorporan Inteligencia Artificial. Una introducción, February 2020. p. 24. Available at: https://www.aepd.es/sites/default/files/2020-02/adecuacion-rgpd-ia.pdf. Ibid. 23-15851