概率推理：實現審判智能決策的結構化進路

“審判智能決策”要求機器能夠對法律推理進行有效模擬。人工智能與三段論式法律推理具有契合性，能夠借助法律論證與深度學習等技術發現法律規則大前提，卻陷入了事實小前提的確定這一證據推理困境。中國當下的人工智能法律系統初步實現了證據分析，但未能作出準確的證據評價。基於結構化評價似然比的概率推理，有效刻畫了事實推理鏈條上證據證明力的算法模型，成為破解困境的一條可行進路。概率推理具有結構化的邏輯推理與決策框架、數字化的信念表達與事實推論、科學化的信息處理與信念結合機制等特徵，有助於實現審判智能決策的發展目標。法學界對概率推理的認識存在數字審判、合取悖論、主觀賦值等理論誤區，概率推理在司法領域的適用面臨運算複雜性、“裸統計”問題以及數據化評估證據的風險，在概率推理推行中還需要保持審慎的態度。

“Intelligent judicial decision making”requires machines to be able to effectively simulate legal reasoning. Artificial intelligence is consistent with syllogistic legal reasoning. It can discover“legal rules”as the major premise with the help of legal argument and deep learning, but it falls into the evidential reasoning dilemma of determining“ facts”as the minor premise of a case. The current legal system of artificial intelligence in China has initially realized the goal of evidence analysis, but failed to make accurate evidence evaluation. Probabilistic reasoning, based on structured evaluation of the likelihood ratio, effectively characterizes the algorithm model of the proof force of evidence in the chain of factual reasoning, and has become a feasible approach to solve the dilemma. Probabilistic reasoning has such advantages as structured logical reasoning and decision- making frame works, digital expressions of belief and fact inferences, and scientific processing about information and combination mechanisms of belief, and is therefore conducive to facilitating intelligent judicial decision making. However, probabilistic reasoning is faced with such misinterpretations in the legal circles as trial by number, conjunction paradox, and subjective assignment, as well as with oppositions and suspicions due to the complexity of operation in its application in the judicial field, the problem of“naked statistics”, and the risks of digitally assess evidence.