This paper aims to present the path dependence for the standardization of artificial intelligence. To this end, it starts from the growing innovation in the Brazilian judiciary, especially with the use of artificial intelligence (AI), highlighting that in 2021, 41 projects using AI were mapped (National Council of Justice, 2021), although its unrestricted use could jeopardize human rights and the democratic rule of law (Liu et al., 2019). Thus, the study is motivated by understanding the standardization of the use of AI in the judiciary, considering recent advances and the normative contours established. In this way, the research is based on institutional theory as a backdrop, more specifically on path dependence as theoretical foundations for understanding innovation in the judiciary. The methods used include a documentary analysis focused on legislation related to innovation and AI in the National Council of Justice (CNJ) and the Labor Courts as part of a historical analysis, which is also portrayed. Path dependency is discussed as an explanatory mechanism for the institutionalization of innovation, considering the history and traditionalist culture of the judicial environment that influences its capacity for innovation. The historical and documentary findings are presented and related to the stages of institutionalization and the phases of path dependence, according to Sydow (2020). The challenges for the research focus on the broad temporal spectrum addressed and the extensive documentary analysis. The expected results of the study include the identification of patterns and trends in the adoption of AI in the Brazilian judiciary, as well as an understanding of the forces that drive or limit this adoption. In short, the normalization of innovation (with a focus on AI) in the judiciary takes place while preserving social and legal stability, supported naturally by the path dependence highlighted in the judiciary.

The Brazilian judicial system faces the challenge of delays in processing millions of legal cases. A new system, developed as a result of a research project in Natural Language Processing (NLP) applied to the case files of the Federal District Judicial Section (SJDF), is providing greater efficiency in process analysis. The ALEI-1G (Intelligent Legal Analysis - 1st Instance) consists of Artificial Intelligence (AI) modules that optimize the work of the SJDF. This article describes how the AI models integrated into ALEI-1G were trained and how they are assisting the SJDF in identifying new Objects of Action (OA), selecting processes related to the same OA in the archive, and classifying processes for batch decisions. The GOA (Objects of Actions Manager) module uses an unsupervised AI model that allows process selection based on similarity levels (from 0 to 1). This approach allows for flexible definition of similar processes by legal experts. Meanwhile, the supervised AI model COA (Objects of Actions Classifier), responsible for classifying processes for minuting, achieves an Average F1-Score of 0.94. These performances of the AI models integrated into ALEI-1G demonstrate precision comparable to tasks previously performed manually. Furthermore, time is a crucial variable for the court's procedural flow: while manual reading of a process for grouping or classification takes minutes, ALEI-1G performs these tasks in a matter of seconds.

The SABIA project encompasses the research and development of machine learning methods to tackle the challenge of similarity among appeals filed with the Brazilian Superior Labor Court. The complexity of labor cases adds a layer of difficulty, as they often involve a wide array of claims. Therefore, cases can be similar in certain aspects while being distinct in others. This paper introduces an innovative approach using machine learning, data science techniques, and heuristics to identify similar appeals, whether for voting purposes or to aid in establishing precedents. Additionally, we share insights into the solution's workflow, which will allow for the evolution of outcomes through user feedback collection and model retraining.

The main subject of the proposed article and presentation should be a a general look at binding provisions in Switzerland regarding publication and anonymization of courts` decisions.

In the first part of the presentation is to consider if Swiss courts (on the federal and cantonal level) are obliged to publish their decisions – if so, then what kind of decisions? It is also to disguise how far reach these obligations and if courts are allowed to publish more than regulated in the general provisions.

In the second part of the presentations is to present and analyze main provisions excluding general transparency regarding courts` decisions in Switzerland. These “interests of confidentiality” are divided in following categories: confidentiality interests of natural persons (eg. protection of privacy, data protection, prohibition of discrimination), confidentiality interest of legal persons (eg. trade and manufacturing secrets) and confidentiality interests of the state (including confidentiality interests of the judiciary system itself).

The third part of the presentation is dedicated the tensions between IT possibilities of anonymization (especially data mining and AI solutions) and legal regulations referring to data protection and other confidentiality interests. For instance: some countries are considering penalization of de-anonymization using IT solutions – is this idea really useful?