When Law Meets AI...
Professor Sieh-Chuen Huang, a dedicated scholar in civil law practices within the Department of Law, has extended her focus to the application of information technology in the legal field. Her journey took her from using statistics to analyze a substantial number of court decisions to connecting statistics with information science. To gain a deeper understanding of the role of machine learning and AI in legal decisions, Professor Huang enrolled in programming language courses offered by the NTU Computer and Information Networking Center. Professor Huang explained that by employing machine learning technology to analyze a large volume of court decisions, it becomes possible to identify the most influential factors in the court's judgments on similar cases. These efforts can lead to the development of a model capable of predicting potential judgment outcomes based on the provided parameters.
Reliability and Trustworthiness of AI
Are the judgments made by AI accurate? This may be a common question among the public. In response, Professor Huang stressed the significance of data collection in training AI models. Legal provisions evolve over time to adapt to societal changes. Therefore, including judgments from too long ago in the training dataset may yield results that are not applicable to contemporary times. For example, one of her main research interests is the Family and Succession Law (under the Civil Code), which has undergone several amendments in response to societal changes. Consequently, a judgment made 50 years ago might not be suitable for training an AI model. Researchers must carefully filter available judgments, including only the most recent ones from the last amendment in the training dataset.
What advantages does AI have in legal trials, then? To answer this question, Professor Huang pointed out that AI, unlike humans, does not experience fatigue, making it capable of rendering judgments on thousands of cases consecutively without rest. AI judgments are also unaffected by subjective emotions or physiological conditions. Therefore, if an ideal model can be developed, its overall performance is likely to surpass that of human judges. However, Professor Huang noted that the purpose of a trial extends beyond the correctness of the judgment. Beneath the surface of a judgment lies the ability to "define the rights and obligations of the parties involved and resolve disputes." Previous court verdicts and decisions carry the weight of authority from experienced professionals, instilling confidence and maintaining social order. AI does not possess the same level of authority for the public at this time, and there is still a considerable distance to cover before AI can fully automate the process of judgment.
Even if an automatic AI trial model is successfully built, concerns remain about the potential misuse of AI as a tool for surveillance or oppression by ill-intended parties. Given these unresolved issues, Professor Huang believes that a more optimal approach for now is a "human-machine collaboration" model. AI predictions based on historical data can serve as a reference for judges during trial proceedings. If the AI prediction contradicts the judge's intuition, the judge can then scrutinize the details of the current case and determine how it differs from previous cases. Subsequently, the judge can make their final judgment with confidence.