An interdisciplinary research team at NTHU has recently unveiled Taiwan’s first artificial intelligence (AI) system using natural language processing technology to predict child custody judgements in divorce cases. Once the particulars of the case are entered into the system, it predicts with 90 percent accuracy which parent will be awarded custody of the children.
The research team is led by Prof. Wang Daw-wei of the Dept. of Physics and the National Center for Theoretical Sciences (NCTS) and Associate Prof. Lin Yun-hsien of the Institute of Law for Science and Technology (ILST). On September 20 they presented their system at a workshop on AI and society, which was held at NTHU and organized by the ILST. On the same day they launched a test version of their system on the Internet (https://custodyprediction.herokuapp.com/), and are currently gathering feedback from those who have tried it out.
Wang said that the team used more than 2,000 judgments made between 2015 and 2017 as models for the system to refer to when making its predictions. Lin said that their system is designed to help couples in the midst of an acrimonious divorce to understand the likely outcome of a court case involving child custody, thereby making it easier for them to reach an agreement without litigation, and also reducing the large backlog of child custody cases clogging the judicial system.
Wang said that similar AI systems are already being used overseas to assist judges in such areas as determining the likelihood of recidivism and for handling credit card disputes, and that the AI system developed at NTHU can be extended for use in other types of legal cases. He explained that in divorce cases, child custody is determined by civil law in accordance with the principle of the best interests of the child, but doing so requires taking into account a wide range of factors of varying importance, depending on the case, including the wishes of the husband, wife, and children; and the parents’ economic status, work situation, relationship with their children, support system, and place of residence.
Wang said that the team developed a natural language processing technology similar to the one used for speech recognition by Apple’s Siri virtual assistant, enabling users without specialized legal knowledge to enter a case by either selecting from several options on a series of webpages, or by verbally describing the situation, whereupon the system provides the most likely judgement should the case go to court.
Team member Li Yalun said that the system uses a special labeling method and data amplification technology for which they have applied for a patent, as well as an efficiency-enhancing network of algorithms which mimics the human nervous system, and is widely recognized as the most suitable algorithm for dealing with natural language.
The NTHU research team which has developed an AI system predicting the outcome of child custody judgements in divorce cases.
Team members conferring with co-leader Prof. Wang Daw-wei of the Dept. of Physics.
Team members conferring with co-leader Associate Prof. Lin Yun-hsien (center) of the ILST.
Lin said that the research team is planning to develop an expanded version of the system based on input provided by legal scholars, lawyers, judicial personnel, and social workers.
Some of the other research projects presented at the ILST workshop were on the application of AI in such areas as art, legal ethics, self-driving cars, and international human rights.