Generative AI can “imitate the reasoning of the lawyer by establishing statistical models from his database“, in a logic of analysis of previous specific cases but she does not give meaning to her answers and she is not able to assess the relevance of her own answer. According to the rapporteurs, its use leads to training needs in order to better understand the responses.
Additionally, generative AI tools provide inconsistent answers. It appears essential that any generative AI solution mentions, after each result, the risk of error what the tool can commit and the need to systematically verify said result.
The use of generative AI tools also involves data confidentiality issueespecially in the area of justice.
Overall, AI must be supervised to meet the requirements of General Data Protection Regulation (GDPR)as well as European regulation on artificial intelligence (RIA), which entered into force on 1is August 2024. This classifies, for example, the algorithms used by the administration of justice among software.high risk“.
-The rapporteurs make 20 proposals, including in particular:
- to systematically inform the user of the risk of error;
- to refine the Légifrance search engine to allow users of the public legal information service to formulate their questions in natural language;
- or even to develop a certification or a public label for legal publishers who undertake to respect certain good practices.
Belgium
Related News :