The legal profession already uses AI. The question is which tool.

In two years generative AI has become part of a lawyer's working week. Not everywhere at the same pace, but the numbers point the same way.

For a lawyer in BiH the question is no longer whether this will arrive here too. It is which tool to pick, and how to keep the decision yours.

What the surveys show

In May 2024 Wolters Kluwer surveyed 712 lawyers in law firms and corporate legal departments across the US and nine European countries. 68% of law-firm respondents and 76% of those in legal departments use generative AI at least once a week.

In the UK and Ireland the shift is faster still. In a LexisNexis survey from September 2024 of 803 lawyers and legal support staff, the share using generative AI for work rose from 11% in July 2023 to 41%. The share with no plans to adopt it fell from 61% to 15%.

What it brings

Time, first of all. In the Thomson Reuters “Future of Professionals” report from July 2024, more than 2,200 professionals in law, tax, accounting and compliance estimate that AI will save them about four hours a week within the next year, and up to twelve within five years. These are their own estimates, not a measured result.

A controlled experiment gives firmer evidence. In a 2023 Harvard and BCG study, 758 consultants did the same work with or without GPT-4. Where the model was strong, those using GPT-4 completed 12.2% more, 25.1% faster, and their work was rated more than 40% higher in quality than the control group's. The subjects were consultants, not lawyers, so this is evidence about knowledge work in general.

The other half of the same study

The same study has a result that is quoted less often. On a task outside what the model could do, consultants using AI were 19 percentage points less likely to reach the correct answer than those working without it. The authors call it the “jagged frontier”: on one side the tool helps a great deal, on the other it does harm, and the line is not visible.

In law that line is most dangerous where an answer sounds convincing. In 2024 Stanford's RegLab tested purpose-built legal research tools marketed as “hallucination-free”. Lexis+ AI and Ask Practical Law AI got about 17% of answers wrong, Westlaw AI-Assisted Research about 33%. The authors count as a hallucination both a false statement of the law and a statement cited to a source that does not support it.

How to choose a tool

None of this means you should wait. It means you choose a tool by how much of the judgement it leaves to you. A tool you can work with:

• shows the provision it relied on with each answer, so you can open and read it yourself;
• says when there is no answer, instead of offering a plausible guess;
• does not ask you to take its word for it.

In BiH this carries extra weight. Legislation is passed by the state, the entities, the cantons and the Brčko District, and each amends it at its own pace. A tool that picks the wrong level saves no time. It moves the risk onto the lawyer whose name is on the opinion.

We are building Amiro to take over the search and the reading. With every answer it opens the provision it relied on, and you make the call.