Breakthrough in Legal Tech: Active Generative AI Poised as Milestone for Legal Sector
Artificial intelligence has made great progress recently and its implementation in the legal area to improve process management and optimize times has advanced. According to Ana Carla Bliacheriene, Law professor in the Public Policy Management courses at the School of Arts, Sciences and Humanities (EACH) at the University of São Paulo, the use of artificial intelligence in the area is not new, as in 2017 the issue was regulated by the Judiciary.
After the launch of generative AI models (GPTs) in 2022, a series of new possibilities in the use of these technologies were opened — with pre-generative models, there was a need for structured data, focused only on excess processes and greater flow of information. management, without the ability to read and understand to generate results. “Generative AI has the ability to read procedural pieces, not just text, but images, in addition to the ability to learn and create. Within the process, it is extremely useful so that it can carry out a preliminary scan and previously present a report to that server, with all the documents that are in that process”, he explains.
Active generative AI
Unlike commercial generative AI models — such as ChatGPT, for example —, in which the user must ask the AI to receive answers, researchers at USP developed the active generative AI model, which is already being tested. In it, information is previously presented to the user, regardless of their command, since the technology was trained to perform repetitive and usual processes in the procedural flow.
With its use in the Judiciary, there are fears about the use of this technology, which are issues of bias and error. Regarding the first, the professor states that, to mitigate and control this issue, there must be training and analysis of the database used, so as not to generate mathematical models originating from discriminatory data.
On the issue of error, she understands that there is a difference between general-purpose generative technologies and active generative AI, which would be for professional use and trained with court databases, with data from a specific area. Furthermore, its use needs to be intensive and massive. “It is in this sense that we created this model at the University of São Paulo, in which it is not only trained on specific databases but also has transparency, traceability, information security and the possibility of mass use, which is what the Judiciary need”, he concludes.