Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

AI on Trial: Legal Models Hallucinate in 1 out of 6 (or More) Benchmarking Queries

May 31, 2024 - hai.stanford.edu
The article discusses the use of artificial intelligence (AI) tools in the legal profession and their reliability. It highlights that while many lawyers plan on using AI for tasks such as reviewing documents and drafting contracts, these tools have a tendency to "hallucinate" or produce false information. The article cites a study by Stanford RegLab and HAI researchers that tested the reliability of two AI legal research tools, LexisNexis's Lexis+ AI and Thomson Reuters's Westlaw AI-Assisted Research and Ask Practical Law AI. The study found that while these tools reduced errors compared to general-purpose AI models, they still produced incorrect information a significant amount of the time.

The article also discusses the challenges of using retrieval-augmented generation (RAG) in AI legal tools, including the difficulty of legal retrieval and the risk of retrieving inapplicable authority. It emphasizes the need for transparency and rigorous benchmarking of legal AI tools, given their current opacity and the ethical and professional responsibility requirements for lawyers. The article concludes by calling for public benchmarking and rigorous evaluations of AI tools in the legal profession.

Key takeaways:

  • AI tools are increasingly being used in the legal profession, but they have a documented tendency to 'hallucinate' or produce false information. This is a significant concern as it can lead to incorrect legal judgments and conclusions.
  • Despite claims from providers, AI-driven legal research tools like LexisNexis's Lexis+ AI and Thomson Reuters's Westlaw AI-Assisted Research still produce incorrect information a significant amount of the time.
  • Retrieval-augmented generation (RAG) is seen as a potential solution to the hallucination problem, but the study found that even RAG systems are not hallucination-free due to challenges unique to the legal domain.
  • The study highlights the need for transparency and rigorous benchmarking of legal AI tools, as the current lack of transparency makes it difficult for lawyers to comply with ethical and professional responsibility requirements, and for practitioners to responsibly adopt these tools.
View Full Article

Comments (0)

Be the first to comment!