Recommendations for the use of AI in the Judiciary (Part II)

This reality increases the importance of the regulatory debate on AI solutions applied to the Judiciary.

In the international scenario it is noticeable the existence of Artificial Intelligence (AI) systems with varied applications associated directly or indirectly to the justice systems: identification of probabilities of success or failure of proceedings before a Court; estimates of the amount of compensation to be received; support for judicial decisions; assessment of the chances of recurrence of a human person; advanced jurisprudential search engines; discovery of patterns in court decisions taken by individual judges and / or collegiate bodies; virtual assistants to inform jurisdictions or support them in legal proceedings; preparation of provisional statistics on the management of human and financial resources in the Judiciary.

This reality increases the importance regarding the regulatory debate of AI solutions applied to the Judiciary, as well as the detailed analysis of the principles organized by the European Commission for the Efficiency of Justice (CEPEJ) in the document “European Ethical Charter on use of artificial intelligence in judicial systems ”, adopted in December 2018, in order to structure the meaning of the ethical, reliable and responsible use of AI in justice systems, namely:

(1) Respect for fundamental rights: requirement that in the design and training stages be inserted in the AI ​​systems (applications that involve processing judicial decisions and data; conflict resolution; aid in decision making; and guidance to the public) rules to prohibit direct violations of the scope of protection of this principle, which encompasses legislation on fundamental rights and protection of personal data and the principles of independence of judges and the democratic rule of law.

(2) Non-discrimination: the actors involved in the development of AI systems must guarantee the ability of data processing methods not to reproduce or aggravate prohibited discrimination, especially when based directly or indirectly on sensitive data, in conjunction with the existence of corrective measures to mitigate or neutralize these risks, in addition to raising stakeholder awareness.

The data set used for the training of AI systems needs to be sufficiently representative in terms of diverse dimensions (ethnicity, gender, political opinions, religious or philosophical beliefs, sexual orientation, among others), as well as preserving information related to this phase, in order to ensure the evaluation by independent third parties and the identification of decisions made by this type of technology.

(3) Quality and safety: recommendations for – the formation of multidisciplinary teams, involving legal professionals and researchers in the areas of law and social sciences, aiming at the production of functional models; the use of certified sources for access to court decisions, which cannot be modified until processed by the AI ​​system, providing transparency as to whether there is no change in the content or meaning of the court decision being processed; and the storage of models and algorithms in a safe environment, aiming at system integrity and intangibility.

(4) Transparency, impartiality and equity: the objectives of this principle are related to the understanding by users of the results that are produced by AI (in clear and familiar language) and the possibility of auditing data processing methods. Therefore, it is recommended to provide subsets of information about the algorithm (nature of the services offered, tools developed, variables in use, training data, risk of error); the development of mechanisms that reduce bias (discriminatory attitudes towards human beings) through more diversity in data sets and approaches; and prioritizing the interests of Justice.

(5) Under the control of the user: refers to the increased autonomy of users; importance of professionals in the Judiciary (possibility of reviewing judicial decisions and data used to produce results); availability of clear and understandable information (prior processing by AI before or during the judicial process, with the right of rejection; whether or not AI-based solutions are binding; different options available; rights to legal advice and access to justice); and mandatory participation of the subjects that integrate the justice systems.

By: Wilson Sales Belchior