AI and expert evidence, assistance is not authorship

By Stephen Hawes MFPWS MCIOB MCIArb AssocRICS MAPM MCABE C.Build E

Stephen Hawes and Co. Chartered Building Consultancy

Artificial intelligence is no longer something expert witnesses can sensibly leave in the background or treat as a professional taboo. It is already close enough to the preparation of reports, chronologies, document reviews and technical analysis that the profession now has to deal with it properly. The Ciarb Guideline on the Use of AI in Arbitration and the Judicial Office AI guidance both point to the same practical position, AI may assist professional work, but the person using it remains responsible for what is produced, relied upon and put forward.

That should not surprise any experienced surveyor as we have always had to take responsibility for the work that goes out and has our name on it. The fact that a tool is quicker, more polished or more technically impressive does not change that. If an expert signs a report, the opinion has to be their opinion, thus they must understand it, be able to explain it, and be able to defend it when the questions become uncomfortable.

There is, however, another side to this which should not be overlooked. The risk is not only that an expert may use AI badly. There is now also the emerging risk that a professional may be criticised for not using AI at all where a careful and controlled use of it could have helped the case, report or matter they are involved in. Recent industry discussion around AI liability has raised the striking possibility that professionals could in due course face negligence arguments for failing to use AI tools appropriately where those tools could have aided the work they have been instructed to undertake.

That is an important point for we surveyors, as if AI can help organise a substantial bundle, test a chronology, identify repeated references, check consistency, or reduce the risk that a material document has been missed, then simply refusing to engage with it may not always look prudent or in the best interest of the party instructing them. It may put the surveyor, the appointing party and the process itself at a disadvantage, however, that does not mean the surveyor should hand the work load over to AI completely, it means the surveyor has to use their own judgment.

In that sense, the position the industry finds itself in is not entirely a new one, as for many years now in construction, property and technical disputes, many experts and expert practices have always used junior surveyors, assistants, technicians, document reviewers, administrative staff and wider professional team members to help prepare expert evidence. That support has included collating photographs, preparing chronologies, formatting appendices, reviewing documents, extracting measurements, checking schedules and assisting with early report drafting.

There is certainly nothing automatically wrong with that, as in a large complicated matters, it may be the only sensible way to deal with the volume of material put forward. The difficulty comes when the support used starts to replace the expert’s own engagement with the evidence and his or hers own true professional opinion. Most experienced practitioners will have seen, or at least suspected, reports where too much of the work appears to have been prepared by the team behind the instructed expert rather than by the expert themselves. This is not a new but an old problem that has bene evident in the industry for many years now. AI simply gives it a new shape and a faster engine.

The named expert in any instruction must still understand the material provided, test the reasoning, correct the draft where necessary, consider the awkward points as well as the helpful ones, and sign only where the opinion expressed is genuinely their own independent expert opinion. If they cannot do that, the report is exposed, however well it reads.

AI should be treated in the same practical way. It can help organise documents, summarise communications, prepare a draft chronology, identify repeated references, check consistency in terminology, or improve the structure of non opinion material. Used sensibly, particularly where the papers are extensive, that can be useful and helpful. It may even help the expert make better use of their own time and experience by reducing the clerical donkey work.

But it becomes dangerous when the tool is allowed to decide what matters, weigh competing explanations, draft the causation reasoning, or frame the final opinion. That is no longer assistance, that is the expert function being performed by something or someone else and then enters the territory of not being the experts own true independent opinion.

It is highly possible that the real question will often be whether there has been too much AI, or not enough AI. Too much reliance risks weakening the independence and credibility of the expert opinion. Where too little use may also become difficult to justify where careful use could have helped the expert review the material, improve the organisation of the evidence, or interrogate the documents bundle more efficiently, thus undertaking their duty to the best of their capabilities.  The correct balance at this moment in time is to use AI enough to support and enhance professional judgment, as a supplement if you like, and like with any supplement, it is not to be completely relied upon, but, still not to the extent that the reasoning, independence and final opinion disappear and stop being truly the expert’s own work.

Expert evidence is not ordinary advice dressed up in formal language. It is opinion evidence intended to assist a court, tribunal, arbitrator, adjudicator or other decision maker. In the civil courts, CPR Part 35 restricts expert evidence to what is reasonably required to resolve the proceedings, requires the expert to assist the court on matters within their own expertise, and Practice Direction 35 requires expert evidence to be the independent product of the expert, uninfluenced by the pressures of litigation, or any commercial pressures.

Those principles remain important in arbitration, even where the procedural framework is different. An arbitrator may not be applying CPR Part 35 as a code, but they still need independent technical assistance, this a report that reads well but cannot be explained by the person who signed it is of little real value.

Take a building defect dispute involving damp and mould around window reveals, the possible causes may include omitted insulation, thermal bridging, incomplete airtightness, rain penetration, defective workmanship, inadequate ventilation, heating patterns, high internal humidity, product substitution, later alteration, or a combination of primary and secondary factors.

AI may assist in arranging disclosed communications and documents, grouping photographs by date, summarising inspection notes or preparing a chronology. That may well save time and may reduce the risk of missing a material sequence or repeated reference. But it should not be asked to decide whether the mould was caused by workmanship, design, occupation, ventilation or water ingress, because that is not administrative support. That is the central expert question and should be reasoned, considered and reflect the experts own professional opinion.

The expert has to do the hard part. Was the reveal opened up. Was insulation present. Did the cavity closer match the specification. Were moisture readings taken. Was thermal imaging carried out. Were ventilation rates measured where relevant. Were photographs taken before remedial works. Had the condition been altered before inspection. Which facts are known. Which facts are assumed. Which matters remain uncertain, can an absolute certain opinion be given, or a necessarily qualified one.

That analysis cannot be delegated to AI and then adopted because the wording sounds plausible. Building pathology often turns on small physical details, sequence, exposure, workmanship, material substitution, maintenance and use, thus a confident paragraph may still be wrong if it has not been checked against the actual evidence.

The same applies to drafting. There is no absolute problem with AI assisting with draft versions of non opinion material, provided the expert checks it properly (more than once) and is happy with it before issuing. The risk is that fluent wording can find its way into the report before the expert has fully tested the reasoning or considered contrary material that is on record. That is not an AI problem alone, It is an old fashioned drafting problem made faster and easier to miss and get wrong.

A well-considered expert report should demonstrate the basis on which the opinion is considered, reasoned and then given. It should distinguish between fact, assumption, inference and opinion. It should identify where information is missing, where the expert cannot reach a firm conclusion, and where alternative explanations or counter arguments remain possible. A report that hides uncertainty may look stronger to the client, however, it is usually weaker when tested.

Transparency also matters where it affects the evidence relied upon. If another person carried out a material inspection, measurement, test or experiment used in the report, that should be properly identified. Routine assistance with organisation, drafting support or document handling does not automatically change the nature of the expert opinion, whether the assistance comes from a team member or an AI tool. The important point is always whether the expert has checked the work, understood the evidence and reasoning, and can explain and defend the opinion as their own if put in a position where they must do so.

Confidentiality requires the same level of care. Expert instructions, draft opinions, photographs, privileged communications, personal data and commercially sensitive documents should not be placed into an uncontrolled AI system simply because it is convenient for the surveyor. An expert using AI should know what tool is being used, what information is being entered, how confidentiality is protected, and whether that use is permitted by the instructions and the procedural framework.

The answer is not to prohibit AI from expert work, as this would be unrealistic and, in some cases, may itself become difficult to defend. The answer is to control it with the same discipline that should apply to any assistance used in preparing an expert report, whether that assistance comes from a junior colleague, a technical assistant, administrative support or an AI tool.

Any expert using AI should have sensible internal controls within their practice. Those controls should address what material may be used, how confidential information is protected, how any output is checked against the source material, and how the named expert satisfies themselves that the final report is accurate, properly reasoned and their own true independent opinion. The point is not to create a separate performance around AI. The point is to make sure the expert remains in control of the work being signed and that they have been instructed to undertake.

The Civil Justice Council Guidance for the instruction of experts in civil claims remains useful because it reinforces the need for experts and those instructing them to understand best practice when complying with Part 35 and court orders. The same professional discipline should inform expert work in arbitration and other dispute forums.

The signature at the end of an expert report is not a formality. It is the point at which the expert confirms that the opinions expressed are their true and complete professional opinions. That confirmation cannot be delegated to an assistant, a colleague or any AI system.

AI will become part of expert witness practice. Used carefully, it may reduce clerical burden, improve organisation and help the expert manage large volumes of material. Used carelessly, it may produce a polished report that fails under questioning because the expert cannot explain the reasoning as their own.

The question is no longer as simple as whether assistance was used, or whether AI was avoided. The better question is whether the expert used the tools available with proper judgment, remained in control, understood the material, exercised independent professional opinion, and signed a report that was truly their own work.