AI and High-Stakes Professional Services

Overview

Using legal and professional tasks as research settings, this area studies reasoning, source verification and reliable AI-assisted methods. It examines how models retrieve and use authoritative sources, handle uncertainty and support human review. Hong Kong’s bilingual common-law environment provides a practical setting for this work.

Methods

Stepwise reasoning verification and correction: check intermediate reasoning in legal judgment tasks, connecting language models with structured verification through LegalReasoner.

Agent trajectory analysis and hallucination profiling: distinguish legal-content errors from tool-use and procedural failures across multi-step trajectories through LexAgentHallu.

Authoritative retrieval and bilingual reasoning: study source retrieval, citation and uncertainty in Hong Kong’s bilingual common-law setting through LexiHK.

Human–AI collaboration and review: establish workflows that allow professionals to verify and correct outputs, and use agents to simulate complex dispute resolution.

Research Questions

How can authoritative sources be distinguished from plausible generated text?

Where do reasoning and tool-use errors arise?

How can professionals verify and correct model outputs?

How can legal reasoning and citations remain reliable in a bilingual common-law setting?

Resources & Outputs

LegalReasoner, LexAgentHallu and LexiHK research practice; “How well can large language model agents simulate complex legal dispute resolution?” (Artificial Intelligence and Law, 2026). Planned programme: LexiHK Research and Evaluation Platform.