摘要： 法律发现并非单纯的规则适用活动，而是一种包含先在理解、直觉与价值评价的心理活动，算法驱动的法律发现是否以及如何可能，取决于法律发现作为心理活动能否被算法承载并复刻。拟人化复刻要求算法与人的思维路径和判断标准保持结构的同一性，仅结果层面的统计近似并不足以构成判断依据。在此基础上，法律发现的实现条件被分解为心理活动的可承载性与语言图示的可转译性两个层面：前者的依据在于发现的心理活动可被逐项辨识，其中可被结构化的部分能够交由算法承载；后者通过法律概念、规范与事实关系、权衡强度以及符合性四类要素的结构化转译得以实现。由此构造出以多方案生成、实践理性筛选和过程留痕为核心特征的算法驱动法律发现模型，并形成由场景、类型、原则与理性自持构成的限度体系。

关键词: 法律发现｜司法人工智能｜法学方法论｜拟人化复刻｜实践理性

Abstract: Legal discovery is not merely an activity of rule application, but a psychological activity involving prior understanding, intuition, and value evaluation. Whether and how algorithm-driven legal discovery is possible depends on whether legal discovery, as a psychological activity, can be carried and replicated by algorithms. Anthropomorphic replication requires that algorithms maintain structural identity with human thinking paths and judgment standards; mere statistical approximation at the result level is not sufficient to constitute a basis for judgment. On this basis, the conditions for realizing legal discovery are decomposed into two levels. This constructs an algorithm-driven legal discovery model with multi-scheme generation, practical rational screening, and process tracing as its core features, and forms a system of limits composed of scenarios, types, principles, and rational self-restraint.

Key words: legal discovery｜ judicial artificial intelligence｜ legal methodology｜ anthropomorphic replication｜ practical reason