Can AI close the justice gap? Promise, evidence and limits

Artificial intelligence is often presented as a way to help the many people who cannot afford a lawyer. The Chief Justice of the United States has noted that for those who cannot afford a lawyer, AI can help. The intuition is simple: if a tool can read, summarize and explain legal material in plain language, it might give a person facing a court matter alone something they have never had before, a way to understand what is happening to them.

For plain-language explanation and organization, that promise is real. Much of the difficulty self-represented people face is not the law itself but the way it is written. Forms, notices and rules are dense, full of terms of art, and assume knowledge the reader does not have. A tool that can restate that material in everyday words, or help a person see how the pieces of their situation fit together, addresses a genuine and well-documented barrier.

The appeal is also one of scale. Lawyers are a finite resource, and free legal help is rationed. Software, by contrast, can serve many people at once at low marginal cost. That is why so much attention has turned to whether these tools can ease a shortage that decades of effort have not closed.

The research counsels caution. A Stanford study found that general-purpose AI tools hallucinate on legal questions between roughly 58 and 88 percent of the time. In other words, when asked legal questions, these widely used tools produced answers that were wrong or fabricated more often than not.

Even specialized legal-research tools, built specifically for the task and marketed to professionals, hallucinated between about 17 and 33 percent of the time. That is far better than general tools, but it is still high enough that no answer can be trusted on its face.

Errors were most pronounced at the lower-court level, which is exactly where self-represented people would most rely on them. The everyday matters that fill those courts, housing, family, small debts, are precisely where the tools are least reliable and where the user is least equipped to catch a mistake. That combination is the heart of the problem.

The lesson is not that the technology is useless, but that confidence in an answer is not the same as accuracy. A tool that sounds authoritative while being wrong a meaningful share of the time can do real harm if it is treated as a source of truth.

A defensible role for AI is triage, plain-language explanation, and helping people organize a matter, not deciding outcomes or giving advice. Helping someone figure out which kind of problem they have, what the words on a notice mean, and what documents belong together plays to the technology's strengths while keeping it away from the tasks it does worst.

The technology can widen access when it stays within those limits. The difference between a help that explains and a help that decides is not a technicality. The first leaves the person in control and able to verify; the second asks them to trust an output that the evidence says will sometimes be wrong.

Used this way, AI becomes one tool among many rather than a replacement for legal judgment. It can lower the height of the first barrier, understanding, without pretending to clear all the barriers that follow.

This is research and educational material on the use of AI to address the justice gap. It is general legal information, not legal advice, and it does not describe any individual's matter.