AI and expert evidence: reliability questions for litigators
By the ChamberOne team 4 minute read
An expert's report is only as good as the method behind it. For decades the method was the expert's training, their data and their reasoning. Now it may also include software that drafts, summarises, calculates or analyses. A litigator who instructs an expert has to know whether that happened, and how, before the other side asks the question in cross-examination.
Key points
- Ask every expert, at instruction and again before finalising, whether and how they used AI.
- Prefer measured disclosure in the report to surprise in cross-examination.
- Have experts keep records of tools, inputs, instructions and outputs.
- Apply the same scrutiny to the other side's expert and to your own instructions.
Why it matters to the court
Courts admit expert opinion because the expert has specialised knowledge that the judge lacks, and they weigh it by looking at how the opinion was reached. If part of that reasoning was produced by a tool the expert did not fully understand, the weight of the opinion may fall, however sound the conclusion.
The rules on expert evidence differ by country and court, including duties on experts to the court and requirements about disclosing the basis of an opinion. Check the guidance from your regulator or bar, and the rules and practice directions of the court concerned, on expert evidence and on any provisions about AI.
Uses an expert might make of AI
The question is not whether the expert used AI at all, but how.
- Drafting. Producing or polishing text of the report.
- Summarising. Reading large volumes of documents or literature.
- Analysis. Running data through a model, such as pattern detection or forecasting.
- Research. Finding literature or comparators.
- Calculation. Using software, including AI-assisted software, to produce figures.
- Transcription and translation. Turning recordings or foreign-language documents into text.
Questions to ask your own expert
Ask at the time of instruction, and again before the report is finalised.
- 1
Have you used or do you intend to use AI tools in preparing this report? Which, and for what?
- 2
Did the tool produce any text, figure, citation or conclusion that appears in the report?
- 3
How did you check each output? What did you correct?
- 4
What information did you enter, and was any of it confidential to our client or the case?
- 5
Could you explain how the tool reaches its results, to the extent that you rely on them?
- 6
Are the inputs, instructions and outputs retained, so that they can be shown if required?
- 7
Is the opinion your own, reached by your own reasoning, and could you defend each step without the tool?
Disclosure of methods
Where an expert relied on AI for something substantive, the safer course is to say so in the report, with a plain description of what was done and how it was verified. Leaving it out, and having it emerge in cross-examination, damages credibility far more than a measured disclosure would have.
Ask the expert to keep a record of their method: the tool and version, the instructions given, the data entered and the outputs received. Agree early how it will be stored. If the other side or the court asks for it, you want to be able to supply it, subject to any privilege or confidentiality issues on which you should take advice.
Tell experts about confidentiality. Material from your client must not be entered into an unapproved public tool.
- Say so in the report, with a plain description of what was done and how it was verified
- Keep a record of tool and version, instructions, data entered and outputs
- Agree early how the record will be stored
- Tell experts about confidentiality
- Leave it out and let it emerge in cross-examination
- Enter client material into an unapproved public tool
Assessing the other side's expert
The same questions apply in reverse. Consider asking the other side, through the procedure your court provides, whether their expert used AI and how. Look for signs in the report: generic wording, citations you cannot find, inconsistencies in figures, unexplained changes in terminology.
Verify every authority and data source cited. Where a model has been used for analysis, ask for the inputs and for an explanation of how the result was reached. Consider whether your own expert should review the method.
Preparing for cross-examination
- Rehearse the AI questions with your expert as you would any other part of their methodology.
- Make sure they can explain, in plain words, what the tool did and what they did.
- Make sure they can say what they would have concluded without it.
- Check they can identify limitations of the tool and where it may have erred.
- Make sure what they say in court matches what the report and their records say.
An expert who says confidently that they 'just used it to tidy the text', when records show it generated an analysis, is in difficulty.
Your own use of AI in preparing the case
Be careful with instructing letters and drafts. If you use AI to summarise documents for your expert, make sure the summary is accurate and complete, and that the expert has access to the underlying documents. An expert who relies on your summary inherits its errors.
Keep records of what you sent. Do not use AI to draft or rewrite an expert's opinion for them. An opinion must be the expert's own, and drafting by lawyers is a sensitive area in itself.
A hypothetical example
Suppose a construction dispute turns on delay. The claimant's expert produces a report with a detailed analysis of programme records. On a routine question, she mentions that she used a tool to cluster thousands of site diary entries before reading a sample by hand.
Because the question was asked early, the report is amended to describe the step, the sample she read and how she checked the clusters. In cross-examination, she explains the method and the checks, and the other side has little to attack. Had the point surfaced for the first time at the hearing, the same method would have looked like something hidden.
The lesson for the instructing litigator is that the answer was not the problem. The surprise would have been.
- The report is amended to describe the step, the sample read and the checks
- In cross-examination she explains the method and the checks
- The other side has little to attack
- The same method would have looked like something hidden
- The surprise, not the answer, is the problem
Where expert evidence and AI tools overlap
Not every use is a concern. Spell-checking, formatting and transcription raise different questions from analysis that shapes the opinion. Agree with the expert at the outset which uses are acceptable and which need your prior approval, and write it into the instructions. That way the line is drawn before the report is written, and not argued over afterwards.
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 in Witness Bundle and Proof of Evidence Preparation: What to Automate and What Not To
A practical guide for litigators on which parts of witness bundle and proof of evidence preparation AI can handle, and which parts it must not touch.
6 minute readHallucinated citations: what court sanctions teach about checking AI output
The patterns behind fake authorities reaching courts, and the supervision and checking controls that would have caught them.
5 minute readE-discovery and technology-assisted review explained for non-specialists
How predictive coding and generative review differ, why validation matters and how to document the process.
5 minute read