ChamberOne

Seven AI myths versus reality for lawyers

By the ChamberOne team 4 minute read

Lawyers who talk about AI tend to sit at one of two poles. One group believes it will fix everything, and the other believes it is a fad. Both groups hold beliefs that do not survive an hour at a desk with the tools. Here are seven of the most common, each set against what happens in practice. The aim is not to settle the argument for you, but to give you better questions to ask the next time somebody makes a bold claim at a partners' meeting.

Key points

  • AI makes mistakes that are hard to spot, so check against the source every time.
  • A tool that is sometimes wrong can still help if checking is quicker than doing.
  • It does parts of tasks, and responsibility, judgement and client trust stay with the lawyer.
  • Test claims on your own work: specific task, specific tool, total time, data handling.

Myth one: AI never makes mistakes

Reality: it makes them often, and they are hard to spot because the writing is fluent. Tools can misstate a rule, misread a clause and invent a case with a convincing citation. The mistakes do not look like mistakes. That is why checking against the source is not optional.

A useful test is to ask a tool a question you know the answer to, in an area you know well, and read its reply closely.

Myth two: AI always makes mistakes, so it is useless

Reality: a tool that is sometimes wrong can still save time, if checking is quicker than doing. A first draft of a routine letter, an index of a bundle or a list of dates from a contract can be checked faster than produced. The test is the total time, including checking, against the baseline. Some tasks pass and some do not.

Myth one: never makes mistakes
  • Mistakes are common and hard to spot because the writing is fluent
  • Tools can invent a case with a convincing citation
  • Checking against the source is not optional
Myth two: always wrong, so useless
  • A sometimes-wrong tool can still save time
  • Drafts, indexes and date lists can be checked faster than produced
  • Test the total time, including checking
Two opposite myths about mistakes

Myth three: it will do my job

Reality: it does parts of tasks. It does not take responsibility, hold a client's hand, read a judge, decide what to concede or stand behind advice. Some tasks within a lawyer's job will shrink, and the mix of work may change, particularly for those who start their careers. Nobody can say exactly how.

The safer bet is to learn what tools do well and to move toward work where judgement and accountability are what clients pay for.

Myth four: my data is safe because the vendor says so

Reality: safety depends on terms and architecture, and a sentence in a sales deck is not a contract. Ask where text goes, whether it is stored, who can read it and whether it trains a model. Read the written terms. Consumer tools and business tools often differ greatly on these points.

Myth five: it is only for big firms

Reality: small firms often have the most to gain from tools that take on routine reading and drafting, because they have fewer people to spread it over. They also have fewer resources to manage risk, so they need simpler rules. A one-page policy, a short list of approved tools and a habit of checking can be enough to start.

Myth six: clients will not accept it

Reality: clients vary. Some ask for it and some ask you not to use it. Most care that the work is right, private and fairly priced. A firm that explains plainly how it uses tools and what it checks will usually be heard. A firm that hides its use risks losing trust if it comes out.

Myth seven: a policy will keep us safe

Reality: a policy sets expectations. It does not prevent a rushed trainee from pasting the wrong document. Safety comes from the policy together with training, approved tools, technical controls and a culture in which people report mistakes. Check what people actually do, not just what is written.

A policy that is never tested is a hope. Run a short exercise: give a few staff a realistic task and see what they do.

Before you decide: waiting is also a choice

Waiting is a choice, and it is not a neutral one. Staff already use tools on their own phones, clients already ask questions, and opposing counsel already send drafts that a tool helped produce. A firm without a position leaves each person to make their own, with whatever data rules they happen to guess at.

You do not have to adopt everything. You do have to decide what is allowed, what is not and who checks. Making that decision for one task, such as summarising long documents, is a modest first step, and you can stop if it does not work.

Whichever pole you sit at, try a tool on a closed matter and keep the notes. A small, honest experiment will tell you more than another round of opinions.

How to test a claim at your own desk

  • Pick one real, closed task and write down how long it took and who did it.
  • Run it with the tool, timing everything, including the checking.
  • Ask a colleague to compare the two results without knowing which is which.
  • Note every error the tool made and how you found it.
  • Decide whether to use it for that task, and write down why.

What the myths have in common

Each myth treats AI as one thing with one behaviour. In practice, tools differ, tasks differ and the quality of checking differs. The useful questions are specific. Which task? Which tool? What does it cost in checking time? What happens to the data? Who is responsible for the result?

A sceptic who tries a tool on three real tasks and an enthusiast who checks three outputs line by line will both learn more than they will from a week of arguing. Put the question to the test and decide with evidence from your own work.

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

Seven AI myths versus reality for lawyers | ChamberOne