The Traditional Expert Witness: An Endangered Species in 2036?
Keynote address
Experts Direct - Expert Witness Symposium 2026, Grand Ballroom, Hilton Sydney
I am pleased to be here, although I confess that when I was invited to address a symposium devoted to expert witnesses, I experienced that slight apprehension which comes from being asked to speak to a room containing a large number of people about whose professional activities one has, from time to time, expressed views which might charitably be described as robust.
The prudent course would plainly be to spend the next 40 minutes saying reassuring things about the indispensable contribution expert witnesses make to the administration of justice, observe that artificial intelligence presents both opportunities and challenges, remind everyone that human judgment will always be important and then sit down before morning tea with the goodwill of the room substantially intact. This would have the additional advantage of allowing me to join the increasingly crowded ranks of conference speakers discussing artificial intelligence without saying anything which anyone could possibly disagree with.
I propose to do something more ambitious.
My title is The Traditional Expert Witness: An Endangered Species in 2036? The question mark is important. I have no crystal ball and, despite some of the more breathless claims presently made about artificial intelligence, nor do ChatGPT, Grok or Copilot.
I do not suggest that experts are about to disappear, and I certainly do not suggest that genuine expertise will become less important to courts. What I do want to suggest is: (a) that the form in which we presently obtain expert assistance in substantial civil litigation is not immutable; (b) that it suffers from difficulties we have recognised for decades; (c) that those difficulties contribute materially to a broader crisis in the cost and accessibility of civil justice; and (d) that the emergence of artificial intelligence gives us an opportunity to ask whether, rather than trying to improve the existing machinery, we ought to contemplate something rather more fundamental.
I am not presenting a worked-out scheme of procedural reform which I expect everyone to embrace today, but rather outlining a proposal which I consider has merit.
I have, for my sins, spent a substantial part of my judicial life attempting, with varying degrees of success, to make complex civil litigation cheaper, quicker and more comprehensible and, like most judges who persist in that activity for long enough, I have developed a healthy appreciation of the distinction between making a novel case management order and discovering six months later whether it actually did any good. I therefore approach this subject with a certain humility born of experience. But experience has also persuaded me that there are times when incremental reform ceases to be enough, and it seems to me that two closely connected problems now justify thinking rather more radically about expert evidence than we have been accustomed to doing.
The first is the cost of civil litigation. The second is whether, given what we spend upon expert evidence, it always provides anything approaching commensurate utility.
Artificial intelligence may provide the occasion for us to ask some important questions about both those problems.
The cost of justice
Earlier this year, in delivering the Sir Harry Gibbs Memorial Oration, I ventured the proposition that a civil justice system in which an ordinary middle-class person, with a good claim and something important at stake, cannot sensibly afford to vindicate a legal right is, to that extent, a system which is failing. I remain of that view.[1]
There was a time, not so very remote, when a solicitor could advise a reasonably prosperous individual or the proprietor of a modest business that he had a good commercial claim and that it was economically rational to pursue it to judgment. There is now a very substantial category of disputes in which a competent solicitor will give different advice. The prospective litigant may have suffered a serious wrong, may have a strong claim and may be able to prove it, but unless the amount at stake is very substantial indeed, or the litigant has the benefit of insurance, access to litigation funding, some representative procedure or unusually deep pockets, the economically rational advice may be that vindicating the legal right is simply not worth the candle.
We ought not to become habituated to this state of affairs. Courts are an arm of government (not providers of an expensive private dispute-resolution service for corporations and the wealthy). The practical ability to have legal rights vindicated by an independent court is part of what gives those rights their reality, and there is something distinctly unsatisfactory about maintaining a system in which a sensible citizen of modest means cannot invoke it without running the risk of financial disaster.
The problem, moreover, extends beyond the private cost borne by litigants because courts administer a finite public resource. Every unnecessary week absorbed by mega-litigation, every day spent navigating material which did not need to be generated, every elaborate interlocutory dispute which adds little to the ultimate determination of the real controversy, consumes judicial time which is then unavailable to somebody else: perhaps a small business proprietor whose livelihood depends upon the prompt determination of a dispute, perhaps parties of modest means whose resources are being steadily depleted while they wait, or, even more gravely, a person whose liberty is at stake. Efficiency in the administration of justice is therefore not some dreary managerial aspiration; it is part of the serious business of judging.
Expert evidence makes a formidable contribution to legal costs, and the expense does not consist simply of the fee rendered by the expert (although anyone who has encountered the charging practices of some distinguished overseas experts will appreciate that this component should not be dismissed with undue haste). Some appear to possess, in addition to their substantive field of specialised knowledge, a highly developed expertise in the temporal elasticity of the billable hour.
The real cost lies in the “superstructure” which develops around expert evidence: the search for the appropriate person; preliminary conferences; the formulation and reformulation of instructions; the preparation of assumptions; the identification and provision of documents; conferences with solicitors and counsel; the preparation of a report; responses to questions about the report; further reports; reports responding to the reports of somebody else; preparation for conclaves; the conclaves themselves; joint reports; preparation for cross-examination; and eventually the evidence.
Almost every step, considered in isolation, is entirely rational. No competent practitioner wakes in the morning with the ambition of making litigation unnecessarily expensive. Each participant is ordinarily responding sensibly to what has gone before, protecting the interests of the client and anticipating what the opponent will do next. The difficulty becomes apparent only when one stands far enough away from the individual steps to contemplate the process as a whole.
One of my favourite books is G K Chesterton’s Orthodoxy. It is ostensibly a work of Christian apologetics, although that description does not fully capture why the book remains so compelling more than a century after it was written. It is Chesterton’s attempt to explain how he reasoned his way to belief, but part of its enduring importance lies, at least for me, in his extraordinary use of paradox to expose the limitations of systems of thought which are logical within their own premises but fail to account for the strangeness and contradictions of life as it is actually lived. Chesterton’s great insight is not an attack upon reason. It is a warning about reason operating within too narrow a circle.
In the chapter entitled “The Maniac”, he makes the memorable observation: “The madman is not the man who has lost his reason. The madman is the man who has lost everything except his reason.” He then explains how the madman’s mind moves within what he calls a “perfect but narrow circle”. It is one of those characteristically Chestertonian inversions which makes one think.[2]
This phenomenon can be seen in civil procedure. Of course an applicant requires an expert, because the respondent will have one; of course the expert must prepare a comprehensive report, because the opposing expert will do so; of course there must be a report in reply, because the first exchange has revealed new matters; of course the experts must then confer, because it is necessary to identify what remains in dispute after the lengthy reports have been produced; of course there must then be a joint report identifying those disagreements; and of course counsel must prepare to cross-examine upon them.
Every step can be defended by impeccable logic by reference to the step immediately preceding it, and we can travel around Chesterton’s perfect but narrow circle without ever asking the anterior question whether there was a less expensive and more useful means of obtaining the specialised assistance the Court required.
An old problem, not a new one
The problems associated with expert evidence are old ones, and some of those best placed to judge them have expressed themselves with directness.
In 1995, Lord Woolf, in his Interim Report on Access to Justice, reproduced this passage[3]:
Expert witnesses used to be genuinely independent experts. Men of outstanding eminence in their field. Today they are, in practice, hired guns. There is a new breed of litigation hangers-on, whose main expertise is to craft reports which will conceal anything that might be to the disadvantage of their clients.
That is fairly bracing stuff (although I do not embrace it generally). Indeed, my experience is that experts generally are not consciously partisan hired guns. Most experts try to take seriously their obligations to the Court and attempt conscientiously to discharge them. The more difficult problem is structural because a process may distort the evidence which ultimately reaches the Court without anybody participating in that process consciously attempting to be partisan.
That was the aspect of the problem which I dealt with in a series of decisions in the Roundup class action, including McNickle v Huntsman Chemical Company Australia Pty Ltd (Expert Evidence) [2021] FCA 370, where I referred (at [2]) to the Victorian Law Reform Commission’s 2008 Civil Justice Review. The Commission identified three forms of what it called “adversarial bias”: deliberate partisanship, in which an expert consciously tailors evidence to assist the retaining party; unconscious partisanship, in which the expert is influenced, without necessarily appreciating it, by the forensic circumstances in which the opinion is being expressed; and selection bias, in which the litigant selects an expert precisely because that person’s honestly held views are known to support the case the litigant wishes to advance.
It is the latter two which seem to me to be more significant. A genuine “hired gun” is comparatively easy to spot. The subtler difficulty arises when everybody is behaving properly. Suppose there are ten eminent economists holding a range of respectable opinions upon a controversial economic question. A competent solicitor acting for an applicant will naturally attempt to retain the economist whose honestly held view best advances the applicant’s case; an equally competent solicitor acting for the respondent will, with equal propriety, do precisely the same thing from the opposite direction. Neither economist need depart by a millimetre from an honestly held professional opinion, and neither solicitor need do anything inconsistent with the highest standards of professional conduct, but by the time the process reaches the Court we have managed, through a succession of entirely rational and proper decisions, to manufacture a polarised expert controversy.
We can, of course, remind experts of their duty to the Court, require declarations of independence and promulgate increasingly emphatic codes of conduct, but if the problem arises in part from the structure through which expert evidence is generated, one cannot expect a declaration at the beginning of a report to dissolve the problems produced by that structure. Experimental and survey research in forensic psychology provides some support for the existence of retaining-side and self-selection effects, although its disciplinary setting and methodological limits counsel against indiscriminate generalisation to every field of expert evidence.
Australian courts have certainly not been idle in confronting these difficulties. Over several decades we have developed what is probably one of the more sophisticated collections of procedural responses to expert evidence in the common law world, including court-appointed experts, referees, conclaves, independently facilitated conferences, joint reports and concurrent evidence, together with increasingly active judicial control of the questions upon which expert evidence is to be received.
I have used these techniques, sometimes in combination, and have often seen them work very well. Concurrent evidence is perhaps the least ambitious of these tools but has proved to be very commonly used and can be useful because it changes the dynamics of expert testimony: instead of two performances separated by days or weeks and mediated entirely through counsel, the judge can have the experts together, exploring the same issue, questioning one another and exposing with considerable speed whether a supposed disagreement is fundamental, semantic or (by far most commonly) dependent upon different factual assumptions.
But the success of these techniques itself brings into relief an underlying problem. If the great utility of the conclave is that, after two expensive reports have been prepared, it reveals that the experts agree about 80% of the subject and that much of the remaining disagreement arises because they were given different assumptions, one might reasonably ask: why did we wait until after the reports were written to discover this?
Roundup and the limits of procedural repair
My experience in Roundup is worth dwelling upon because it illustrates both the possibilities and limitations of procedural reform.
The case concerned an allegation that Roundup products were carcinogenic and that exposure increased the risk of developing non-Hodgkin lymphoma. The scientific issues were obviously going to be complicated (indeed I described them as only marginally less complicated than the theory of relativity).
When the matter first came before me for case management, I proposed that, instead of immediately proceeding by the conventional process under which each side assembled its own scientific forces and eventually presented the Court with competing accounts, the Court appoint a series of referees, at staggered stages, to inquire into and report upon scientific questions.
The parties opposed the proposal with vigour. This did not entirely surprise me. Both sets of experienced and highly competent solicitors opposed the proposed course. Like most people, they liked control (although they were sufficiently polite to say they were enthusiastic about innovation - provided it occurred somewhere else).
In fairness, there was a genuine difficulty peculiar to that case because part of the applicant’s case was that Monsanto had itself affected the integrity of the scientific literature by initiating and funding research which disputed the alleged carcinogenicity of Roundup and by seeking to undermine contrary research. If the integrity of the scientific corpus itself was part of the controversy, there was an obvious difficulty in asking a referee simply to synthesise that corpus, and I ultimately concluded that the allegation, to the extent it mattered, should be determined in open court by a judge.
But what followed was not a meek return to common practice. I attempted to construct a process in which the experts would meet at an early stage in the presence of an independent barrister, and would themselves assist in identifying the questions requiring expert consideration and the areas upon which agreement could already be reached. Individual reports would follow, then further independently facilitated conclaves, producing joint reports which were to be as short as practicable, expressed in plain English and directed specifically to identifying what was agreed, what was not agreed and the reasons for disagreement. There were also restrictions upon communications between the parties and their solicitors because, if we really mean what we say when we describe an expert as owing a paramount duty to the Court, it seemed sensible to create procedural conditions in which that independence could operate in substance.
An assessor was later appointed to assist the Court with scientific and technical matters. His function was carefully confined: his advice was not binding, any new theory or opinion had to be disclosed and discussed with counsel, and every finding remained the Court’s.
I confess that I was pleased to see that this use of an assessor and our deep dive into the depths of the relevant scientific fields attracted some attention in the United States. Writing for the American Council on Science and Health, a writer described the proceeding in an article entitled “The RoundUp Judge Who Got It”, a title which I received with the measured detachment expected of a judicial officer, before forwarding it to all my mates.
Her opening description was perhaps more generous than I would have ventured myself: “Imagine a judge reviewing all the key scientific evidence regarding RoundUp’s alleged carcinogenicity - not an IARC or EPA monograph - but the studies themselves. Let’s say he also had a truly impartial expert, an ‘assessor,’ to teach him the science behind the research. Well, this actually happened.” The article’s interest lay in the method: an Australian court had attempted to examine the studies themselves, with independent scientific assistance, while leaving the adjudicative responsibility where it belonged (Barbara Pfeffer Billauer, “The RoundUp Judge Who Got It”, American Council on Science and Health, 29 August 2024).
I mention this history not because I regard this case as some neglected masterpiece of civil procedure. The costs were still very large and there are things I would now do differently. But these orders were an attempt to deal pragmatically with a difficult case, and parts of the proposed regime themselves required reconsideration. I mention it because it illustrates how far judges have already gone in attempting to repair the conventional model. When one has deployed almost every available procedural means of improving expert evidence, and still encounters cases in which the process is disproportionately expensive or ultimately illuminates relatively little, one begins to wonder whether the next step should be another refinement of the same process or whether we should ask a more basic question.
That question is simple: “What do we actually need an expert witness to do?”
Assumptions, assumptions
The answer appears obvious until one attempts to disaggregate the activities which have traditionally been bundled together under the description “expert evidence”. A retained expert possesses specialised knowledge acquired through study, training or experience; identifies and assimilates relevant literature or data; decides what material is material; applies specialised knowledge to assumed facts; undertakes calculations or modelling where necessary; chooses between methodologies; forms conclusions; explains those conclusions in a report; and ultimately defends them under cross-examination. Historically it was entirely sensible to bundle these activities together because there was no realistic alternative. If the Court required access to a large and complicated body of specialised knowledge, it needed a human being who had spent years acquiring and organising it.
But what is clear is that this practical constraint is rapidly disappearing.
I used event studies in securities class actions as an example in the Gibbs Oration because they illustrate the point rather well. An event study has an impressive appearance. There are regressions, coefficients, confidence intervals and measures of statistical significance, sometimes expressed to several decimal places, a degree of precision which commands particular respect from those of us (like me) whose mathematical education reached its modest intellectual apogee some decades ago. Two distinguished economists can produce reports containing an impressive quantity of mathematics and reach materially different conclusions.
But when one strips away the surface difficulty, the difference almost always depends upon choices and assumptions: the estimation window; the benchmark index; the factor structure; the event window; the treatment of confounding information; or, critically, what information the market understood a particular announcement to convey. Mathematical precision can disguise the extent to which the model rests upon contestable premises.[4]
Any judge with substantial experience of expert evidence will recognise the broader phenomenon. One can spend days examining sophisticated differences between two experts only to discover that the real dispute lies in the factual assumptions upon which their respective analyses have been constructed. Once the relevant factual premise is resolved, a considerable part of the apparent expert controversy simply falls away.
This is where I think artificial intelligence becomes important, because it can assist in reversing the conventional sequence. Why should the first step necessarily be for each party to commission an opinion? Why should we not first identify the assumptions upon which the answer depends? Some assumptions will be agreed; some may be capable of determination by the Court at an early stage; and others may remain contested, in which event there is nothing conceptually difficult about asking for an analysis upon alternatives. If A is ultimately found, what follows? If B is found, what follows? The Court can then determine the factual controversy and apply the corresponding specialised analysis.
Indeed, one does not need artificial intelligence to perceive the merit of this approach. Singapore has already moved considerably in this direction.
A different starting point
Under Singapore’s Rules of Court 2021, the starting point is not an assumed entitlement to expert evidence. Expert evidence may not be used without the Court’s approval, and before approving it the Court must consider whether the evidence will contribute materially to determining an issue involving specialised knowledge and whether the issue can instead be resolved through an agreed statement of facts or submissions based upon mutually agreed materials. So far as possible, the parties are to agree upon one common expert and upon the issues and common set of agreed or assumed facts to be referred to that expert; where agreement cannot be reached, the Court determines those matters.[5]
Justice Kannan Ramesh, a Judge of the Appellate Division of the High Court of Singapore, has explained that the use of a common set of agreed or assumed facts is directed to the recurring problem of parties providing experts with separate sets of instructions and facts which tend towards their respective forensic positions. The Singapore approach is attractive because instead of beginning with the proposition that each side has an expert and asking how the resulting evidence should be managed, the process begins by asking whether expert evidence is necessary, what precise issue requires it and upon what factual foundation the expert should proceed.
The expert witness of 2036
Let me therefore engage in some speculation and imagine a substantial commercial proceeding in 2036.
A complicated economic or accounting issue arises. The Court first requires the parties to identify precisely what question of specialised knowledge requires resolution. Expertise may be necessary at that stage, perhaps indispensably so, but the expert’s function is not to become a witness. The expert instead operates as a technical adviser, assisting the lawyers and perhaps ultimately the Court to formulate the real question, identify the relevant data, expose the assumptions upon which an answer will depend and distinguish matters of genuinely contestable professional judgment from matters capable of objective analysis.
I hesitate to invent another legal label, because the law has already produced enough labels. It may be enough to describe the person as an “independent technical adviser”, whose principal value at that stage lies in helping to ensure that the technical question is the right one.
The factual assumptions are then agreed, determined where that can sensibly occur at an early stage, or stated in the alternative. The relevant datasets are identified and, where necessary, challenged. The recognised methodologies are exposed. A secure, transparent and appropriately validated artificial intelligence system then undertakes the task of interrogating the relevant literature, analysing the data, performing calculations, applying specified methodologies and identifying the consequences which follow from alternative assumptions.
At this point (and this is critical), the human expert does not disappear.
Instead of spending months constructing an analysis from scratch, the expert may be asked to interrogate the analysis which has been produced. What relevant literature has been overlooked? Is the dataset appropriate? Is the methodology orthodox? Does a recognised alternative methodology produce a materially different result? Has the system treated correlation as causation? Has it failed to appreciate a limitation which an experienced practitioner in the discipline would immediately recognise? Is there an apparent consensus in the published literature which disguises a serious and respectable minority view?
There is then a procedural question. Once the artificial intelligence system has, upon the disclosed assumptions and by reference to the identified body of specialised knowledge, produced its answer, a party wishing to contest that answer might ordinarily be required to seek leave to do so by application, identifying with precision the respect in which the process or result is said to be unsound. The party would remain free to call expert evidence if leave were granted, but would do so at a real risk as to costs. If, after adjudication, the judge accepted that the answer had been produced by the proper application of a transparent and appropriately validated artificial intelligence model, the unsuccessful challenge would justify an adverse costs order and, in a suitable case, an order on an indemnity basis. Such a rule would preserve the right to challenge machine-assisted analysis while discouraging the reflex production of costly expert evidence merely because a party disliked the answer.
At the end of the process, the expert would perform a different role from the one with which we are familiar: explaining why an apparently convincing answer is somehow wrong. This requires sophisticated human judgment.
The analogy with an experienced silk is reasonably close. The value of an experienced appellate advocate lies in knowing which authority matters, which apparent distinction is spurious, which superficially attractive submission will collapse when pushed one step further and, above all, which point among 20 is the point upon which the case actually turns.
The highest form of expertise is not the possession of information. It is judgment about information.
The expert at either end
This leads me to what I think is a more plausible prediction than the simple proposition that artificial intelligence will replace expert witnesses. I suspect the expert of 2036 may become more important at either end of the litigation process while becoming less prominent in its expensive middle.
At the beginning, experts may become involved much earlier, assisting in the proper formulation of pleadings, exposing hidden technical assumptions, identifying what information is actually required, distinguishing factual controversies from scientific or technical controversies and helping formulate questions capable of sensible resolution.
At the other end of the process, the expert may increasingly become an auditor, validator, critic and explainer of machine-assisted analysis.
If that occurs, what may gradually contract is the enormous middle section of the present process in which two teams separately pay distinguished people to digest substantially the same literature, interrogate substantially the same data, produce lengthy reports, respond to the other lengthy reports, meet eventually to determine which parts of those reports they did not really disagree about, prepare another report recording the remaining disagreements and then spend days being cross-examined about them.
I do not suggest that this description applies to every case. But anyone who maintains that it does not apply to many cases (particularly large commercial cases) has had a markedly different experience of civil litigation from mine.
So the future may involve fewer expert opinions but more expert assistance.
The obvious difficulties
There are, of course, formidable objections to what I have been suggesting, and any serious consideration of this subject has to confront them. There is the so-called “black-box” problem. If nobody can explain why a machine has produced a conclusion, it is difficult to see how that conclusion could safely play a material part in judicial fact-finding. A person whose rights are to be affected by analytical material must have a meaningful opportunity to understand and challenge it, and a judge cannot abdicate the responsibility of assessing its reliability merely because the underlying technology is complicated.
But that seems to me an argument about the characteristics which a system must possess before it is fit for use, rather than an argument that no such system could ever be fit for use. There would plainly need to be reliable provenance, transparency as to source material, disclosure of material assumptions, reproducibility where appropriate, a capacity to identify methodological choices and an adequate audit trail, together with safeguards for privileged, confidential and commercially sensitive information and, importantly, an intelligible allocation of human responsibility for the decision to use the output.[6]
But there is also a logically anterior problem which Roundup illustrates particularly well: who determines the integrity of the corpus upon which the machine acts? If the published scientific literature has itself been affected by bias, commercial influence, defective research or selective publication, a machine capable of digesting it with superhuman efficiency may merely produce error at unprecedented speed. Garbage in, garbage out is not a problem solved by artificial intelligence.
Human black boxes
There is, however, a danger that we impose upon machines standards of transparency and neutrality which we have never seriously demanded of human beings.
We speak gravely about algorithmic bias as though bias were invented in Silicon Valley, although the Victorian Law Reform Commission was discussing adversarial bias in 2008. We worry, properly, that an artificial intelligence system may depend upon assumptions which have not been exposed, although expert reports depend upon assumptions; and we worry that an AI system may state an erroneous proposition with remarkable confidence (although I hesitate to suggest to a room containing lawyers and expert witnesses that confident error is a phenomenon with which the administration of justice has had no previous acquaintance).
Indeed, one of the potential virtues of a properly designed machine-assisted process is that it may expose choices which are presently concealed beneath the authority of a polished expert report. A model which requires the assumptions, data, source material and methodological choices to be specified may, in some circumstances, be more transparent than an opinion resting partly upon an expert’s accumulated experience and professional intuition, particularly when the expert understandably finds it difficult to reconstruct precisely how years of experience have informed a particular evaluative conclusion.
In my view, the choice is not between imperfect artificial intelligence and a Platonic ideal of perfectly objective human expertise. It is between an emerging technology with serious and identifiable deficiencies and an existing human process with manifold deficiencies of its own, some of which we have been documenting for generations.
Why now?
The forms of procedure with which we grew up are not sacred objects handed down on tablets. Their virtue lies in whether they enable a court to identify the real controversy, afford procedural fairness and determine that controversy according to law at a cost bearing some rational relationship to what is at stake.
I have reached the age at which one becomes particularly suspicious of propositions beginning with the words, “But we have always done it this way”.
There is a passage in Giuseppe Tomasi di Lampedusa’s The Leopard which has become famous precisely because it captures one of the paradoxes of institutional life. Tancredi tells the Prince of Salina, in Archibald Colquhoun’s translation, “If we want things to stay as they are, things will have to change.”
There is something in that paradox for the administration of justice. If what we wish to preserve is a system in which independent judges can determine controversies fairly but without excessive cost, then we may occasionally have to be quite radical about changing the machinery by which those controversies are presented to them.
Conclusion
I intended at the beginning to be provocative, and I hope I have succeeded.
The provocation has a serious purpose. Civil litigation has become far too expensive, and one of its most elaborate features is a method of obtaining specialised assistance. At some point, intellectual honesty requires us to contemplate the possibility that the answer is not simply another refinement of the existing process.
I hope (and I am quietly confident) your role will look rather different in 2036 from how it looks in 2026.
Thank you.
