Abstract:

Empowered by self-attention mechanisms within pre-trained language models, generative artificial intelligence demonstrates powerful linguistic capabilities. Nevertheless, concerning judicial application, its textual-meaning determination function cannot fully align with judicial adjudication practice. Legal semantic theory mistakenly frames semantic disputes as the core of legal conflicts and ignores that legal disputes essentially turn on axiological-theoretical controversies. This obscures the core dimension of judicial adjudication, namely exploring value implications embedded within legal norms, and generates theoretical misunderstandings regarding artificial-intelligence judicial deployment. In hard cases, gaps emerge between rule semantics and substantive normative values. When formal linguistic meanings of legal rules diverge from substantive value intentions behind underlying legal norms, core judicial work shifts toward interpreting substantive normative meanings. Therefore, generative artificial intelligence lacks capacity for in-depth substantive-level understanding of legal values and cannot independently render judicial decisions. Accordingly, future legal artificial-intelligence tools should position themselves as instrumental aids within judicial practice and mainly apply to simple-case handling scenarios.

Key words: generative artificial intelligence, judicial application, semantic theory, hard cases, legal artificial intelligence, judicial philosophy