ChamberOne

AI and access to justice: an honest assessment

By the ChamberOne team 5 minute read

Most people who have a legal problem never speak to a lawyer. Cost is part of the reason, and so are distance, fear and not knowing that the problem is a legal one. It is tempting to say that AI will fix this. It might help. It might also produce a new class of people who rely on confident, wrong answers and have nobody to complain to. This article tries to separate the realistic gains from the hopeful ones.

Key points

  • Cost is only one barrier to justice, and AI helps with some barriers far more than others.
  • The most reliable gains are plain-language explanation and faster routine drafting reviewed by a lawyer.
  • AI information is not advice, and public users have no regulator or insurer behind a chatbot's mistakes.
  • Test tools on local law and local languages, where performance is likely to be weaker.
  • Keep public-facing tools narrow, reviewed and backed by a clear route to a person.

What the access problem actually is

Access to justice is often reduced to price, but price is one of several barriers. People do not recognise a problem as legal until it is late. They do not know which kind of lawyer to see. They cannot take time off work to attend court. They distrust institutions. Some live far from any adviser, and some speak a language in which legal help is thin.

A tool can address some of these and not others. Treating all of them as one problem leads to overclaiming, so it helps to take them one at a time.

Suppose a legal aid clinic in a regional town sees the same three problems every week: tenancy notices, small debts and simple family forms. The barrier for most of its clients is not that they cannot afford a letter. It is that they do not know what a notice means or what to do in the fourteen days after it arrives. A tool that explains the notice in plain language and lists the questions to bring to the clinic addresses the real barrier. A tool that drafts a legal response and sends it unreviewed would not.

Where AI can realistically help

The strongest case is the plain-language layer: explaining what a letter means, what a form is asking, what the usual steps in a process are, and what to bring to a first appointment. A person who arrives prepared makes better use of a short meeting with a lawyer or advice worker.

  • Translating legal wording into everyday language, and sometimes into another language.
  • Helping people organise their facts and documents before they seek advice.
  • Pointing people to the right type of service or the right court or office.
  • Helping legal aid providers and small firms draft routine letters and forms more quickly, so each adviser can help more people.
  • Reducing the admin that sits between a client and the lawyer's actual work.

Where it cannot replace advice

Advice is not information plus a conclusion. It involves asking the right follow-up questions, noticing what the person has not said, weighing the options against their circumstances and priorities, and being accountable for the result. A tool answers the question it is given, in the terms it is given, and can sound equally sure whether it is right or wrong.

Some situations are especially unsuited to automated help: anything involving liberty, children, immigration status, a home, or a deadline that is about to run. In these the cost of a wrong answer falls on the person least able to bear it, and the person asking is often the one least able to spot a mistake.

A useful test for any proposed use is to ask what happens if the answer is confidently wrong, and who would find out. If the answer is that nobody would find out until a deadline has passed, the use needs a human in the loop.

Where AI can help
  • Plain-language explanation of letters and forms.
  • Organising facts and documents before advice.
  • Pointing people to the right service or court.
  • Faster routine drafting for providers.
Where advice is still needed
  • Asking follow-up questions and weighing options.
  • Liberty, children, immigration, a home or a deadline about to run.
  • Any answer that could be confidently wrong without anyone noticing.
Where AI helps and where a person must advise

Who bears the risk when the answer is wrong

When a lawyer gives bad advice there is a regulator, a complaints process and usually insurance. When a general-purpose chatbot gives bad information to a member of the public, there is generally none of those. The user carries the loss.

That asymmetry matters for anyone building or offering legal help powered by AI. If a legal aid organisation or a firm puts a tool in front of the public, it should be clear about what the tool is for, who has reviewed its content, what happens when a person's case is outside its scope, and how a human can be reached. Check the guidance from your regulator or bar on offering services to the public through automated tools.

Quality varies by place and language

Tools perform best where there is a lot of written law and commentary in the language the tool knows well. Coverage of smaller jurisdictions, local procedure, and languages with less published material is likely to be weaker. A tool that gives a fluent answer about one country's law may give a fluent and wrong answer about another's.

Firms and advice providers in smaller markets should test any tool against questions to which they already know the answers, and test it again whenever the tool or the law changes.

What a firm or provider can do now

The safest gains sit behind the scenes, where a lawyer reviews everything. Faster drafting of routine documents, quicker intake summaries, and better organised files let a team serve more people at the same quality. A modest, supervised use of tools often does more for access than a public-facing chatbot would.

If you do offer self-service help to the public, keep it narrow. A tool that explains how an application process works, and where to find the form, is easier to make reliable than one that tells people what their rights are.

  • Define the scope in writing and show it to the user.
  • Have a lawyer review the content and the answers, and re-test regularly.
  • Give a clear route to a person, and say how long it takes.
  • Keep records, so mistakes can be traced and corrected.
  • Tell users that the output is information, not advice, and that it may be wrong.

The risk of a two-tier system

One uncomfortable possibility is that people with means get a lawyer using tools, and people without get only the tool. If automated help becomes the default for those who cannot pay, errors will concentrate on exactly those people.

That is a reason for caution rather than for abandoning the idea. It suggests that funders, regulators and the profession should ask what level of human oversight is appropriate where a service is aimed at people who cannot easily tell whether it is right.

This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.

Read next

AI and access to justice: an honest assessment | ChamberOne