Deep Learning for Prediction of Judicial Outcome – LAWBOT

LAWBOT is first, an applied research project in law, on the use of automated natural language processing techniques. The LAWBOT project aims to create an artificial case-law intelligence capable of predicting the judicial outcome for a given case, by imitating the decisions previously rendered by the courts on similar cases. LAWBOT is based on an artificial neural network for the deep learning of textual characteristics predictive of the judicial outcome. The project highlights five results.
- Firstly, the provision of 24,000 decisions annotated by lawyers on 120 classes of claim.
- Second, the creation of an automatic annotator which, from the annotated decisions, will automatically generate a large-scale classification of the legal decisions made public on a daily basis.
- Third, based on the large volume of classified decisions, predict the class of claim, the outcome and the sum allocated to the claimant using artificial intelligence (AI) models.
- Fourth, generate from the IA models legal reasons, in other words, summaries of court decisions highlighting the link between the facts which led to the dispute and its outcome.
- Fifth, measure the ethical, psychological and economic impacts of using a predictive justice AI.
LAWBOT aims to produce fundamental experimental knowledge on the very nature of law, and its epistemology. Indeed, the formulation of predictive models is possible, if and only if, two fundamental hypotheses on law and jurisprudence are verified. First hypothesis, the polysemy of language does not constitute an insurmountable obstacle to the modeling of jurisdictional decisions in a quantifiable form of computable data. Second hypothesis, there are sufficient statistical correlations to formulate a prediction on the outcome of the decision from explicit factors present in the text, without the need to resort to hidden variables likely to induce legal uncertainty, such as factors non-quantifiable human beings (personality of the parties, performance of lawyers, prejudices of judges). It is assumed that the magistrates are rational and consistent, and tend to judge in the same way, the cases considered - from their point of view - as being similar. From these two hypotheses, the quantification of past case law should make it possible to predict the decision of a given judge in a given case, based on the existence of a correlation link between a known dependent variable (the result judicial), and unknown independent variables (the explicit statement of the dispute represented as a combination of quantifiable factors). It is a classic experimental approach, identifying a statistical correlation link between a known result and controlled variable factors.