Legal advice privilege and artificial legal intelligence: Can robots give privileged legal advice?

Introduction

Legal professional privilege entitles parties to legal proceedings to object to disclosing written or oral communications. The privilege has two limbs. Litigation privilege, which is capable of encompassing confidential communications between legal adviser or client and third parties, such as expert and non-expert witnesses, is not considered further in this article. The focus of this article is on legal advice privilege, which attaches to confidential communications between client and lawyer made for the purpose of giving or receiving legal advice.

The form of legal professional privilege that is now commonly known as ‘legal advice privilege’ ¹ has its origins in the 16th century, with the rationale for its existence being fully developed during the 19th century. ² The privilege attaches to ‘communications passing between a client and its lawyers, acting in their professional capacity, in connection with the provision of legal advice’. ³ Where it arises, it entitles the client to object to disclosing the communication, the right to claim or waive the privilege belonging to the client, ⁴ not to the legal adviser. In order for the privilege to arise, the communications must be confidential, though the preservation of confidentiality does not, in itself, justify the existence of the privilege. ⁵ Rather, its underlying rationale is that:

in the complex world in which we live there are a multitude of reasons why individuals, whether humble or powerful, or corporations, whether large or small, may need to seek the advice or assistance of lawyers in connection with their affairs;…the seeking and giving of this advice so that the clients may achieve an orderly arrangement of their affairs is strongly in the public interest;…[I]n order for the advice to bring about that desirable result it is essential that the full and complete facts are placed before the lawyers who are to give it; and…unless the clients can be assured that what they tell their lawyers will not be disclosed by the lawyers without their (the clients’) consent, there will be cases in which the requisite candour will be absent. ⁶

This article makes an original and distinctive contribution to discourse in this area through offering novel perspectives on and solutions to the question whether this common law privilege (developed in an age when lawyers and clients communicated either orally or via documents written using quill or dip pen and ink) can and should attach to communications between a client and a robot.

In answering this question, the following areas are investigated. First, whether the rationale underlying the existence of legal advice privilege encompasses communications between clients and robots. Secondly, whether legal advice privilege at common law may be applicable to such communications. Thirdly, whether robots may be capable of giving advice that qualifies as legal advice for the purposes of legal advice privilege. Fourthly, whether a robot is capable of being a legal adviser for the purposes of legal advice privilege. Fifthly, if legal advice privilege is not applicable to robot/client communications, whether statutory intervention extending the privilege to such communications would be desirable. Finally, if legal advice privilege cannot attach to communications between a client and an unsupervised robot providing legal advice, what are the consequences for legal advice privilege where robots are deployed by human legal advisers in circumstances in which the level of human lawyer supervision and/or understanding of the relevant algorithms is minimal? Answering this latter question also involves consideration of the adequacy of relevant professional conduct rules. Underpinning our consideration of these questions is analysis of the approach adopted by the Supreme Court in R (Prudential plc and another) v Special Commissioner of Income Tax ¹¹ when considering whether communications between accountants and their clients for the purposes of giving or obtaining legal advice were privileged. We regard consideration of the analogy between accountants and robots providing legal advice as being of significant relevance as a predictor of the likely response of the Supreme Court to an assertion that the privilege should attach to robot/client communications.

For the purposes of this article, the term ‘robot’ is used in two different senses. First, to describe software enabling a client to give instructions and receive legal advice based on those instructions without the intervention of a human lawyer. Instructions could for example be given and advice received through a question and answer decision tree type mechanism. ¹² Secondly, to describe the situation where a client gives instructions and receives legal advice based on those instructions from a human lawyer who has made use of software in formulating that advice, for example increasingly sophisticated natural language processing self-learning software. ¹³

This article is written based upon the assumption that we will reach a time in which robots will be able to provide increasing varieties of legal advice to clients without engagement with (or with no more than nominal supervision by) a human legal adviser. Marcus (2009: 273–281) speculates on the challenges to be overcome for the computer to become lawyer; for example, whether legal reasoning and analysis, particularly at the most creative end of the spectrum, is beyond replication by computer. However, as Marcus suggests, ‘most lawyers spend most of their time doing legal analysis that is more the “fill the blanks” variety. That sort of activity might be done with some frequency by a computer’ (Marcus 2009: 275). It is accepted that as the human legal adviser/robot relationship evolves, the giving of legal advice (and the nature of legal advice) may encompass a developing spectrum of possibilities with variable forms and levels of robot/human interaction and the article seeks to take account of this. In addition to the types of software mentioned in the preceding paragraph, areas currently identified in relation to which robots are predicted to play an ever-increasing role in the provision of legal services include, ‘discovery, legal search, generation of documents, creation of briefs and memoranda, and predictive analytics’ (McGinnis & Pearce, 2014: 3065).

Does the rationale underlying the existence of legal advice privilege encompass communications with robots providing legal advice?

At one time it was believed that the rationale underlying the existence of legal advice privilege ‘was that a lawyer ought not, in honour, to be required to disclose what he had been told in confidence’. ¹⁴ If this was still regarded as the justification for legal advice privilege it could be argued that the position of human legal adviser and robots so far as legal advice privilege was concerned could be distinguished on the basis that, unlike that of a human legal adviser, the ‘honour’ of a robot cannot be impugned. Oxford Dictionaries defines ‘honour’ as ‘the quality of knowing and doing what is morally right’. ¹⁵ If the existence of legal advice privilege was still justified in terms of impugning honour, whether this is a concept that is or potentially could be applicable to a robot might have provoked an interesting debate. A similar debate, relating to ethical awareness, is encountered below when considering whether a robot is capable of being a legal adviser.

As was seen above, the existence of legal advice privilege is now justified upon the basis that ‘the seeking and giving of this advice so that the clients may achieve an orderly arrangement of their affairs is strongly in the public interest’. ¹⁶ The argument that revealing confidential communications would amount to ‘a breach of honour, and [a] great indiscretion’ on the part of a human legal adviser was defeated long ago on the basis that there is no such indiscretion when disclosure is required by law (i.e. by an order of the court). ¹⁷ Thus, it has long been clear that the privilege is a right belonging to the client, not to the legal adviser, the client being entitled to claim or to waive it. ¹⁸ As the rationale for legal advice privilege is that it is in the public interest to enable clients to arrange their affairs in an orderly way, this objective would seem to be satisfied whether legal advice was given by a lawyer, by a non-lawyer human professional with appropriate expertise (such as an accountant) or by a robot. In R (Prudential plc and another) v Special Commissioner of Income Tax, the Supreme Court considered whether documents were covered by legal advice privilege when the legal advice in them was given by accountants rather than lawyers. ¹⁹ The majority decided that at common law the privilege did not attach to legal advice given by professionals other than lawyers, any extension of privilege to non-lawyers requiring statutory intervention. However, when considering whether the rationale for legal advice privilege could encompass communications with robots, Lord Sumption’s dissenting judgment in Prudential is of particular interest. He held that legal advice given by accountants should be privileged because:

[o]nce it is appreciated (i) that legal advice privilege is the client’s privilege, (ii) that it depends on the public interest in promoting his access to legal advice on the basis of absolute confidence, and (iii) that it is not dependent on the status of the adviser, it must follow that there can be no principled reason for distinguishing between the advice of solicitors and barristers on the one hand and accountants on the other. ²⁰

The obvious difficulty in deploying Lord Sumption’s judgment in Prudential in support of the proposition that legal advice privilege could potentially attach to legal advice provided by a robot is that it was a dissenting judgment. The majority of the Supreme Court decided that the privilege did not attach to legal advice given by professionals other than lawyers. However, four of the six Supreme Court judges in Prudential did accept that, logically, the rationale underlying legal advice privilege was applicable to confidential legal advice given by professionals other than lawyers. For example, Lord Neuberger accepted that:

LAP is based on the need to ensure that a person can seek and obtain legal advice with candour and full disclosure, secure in the knowledge that the communications involved can never be used against that person. And LAP is conferred for the benefit of the client, and may only be waived by the client; it does not serve to protect the legal profession. In light of this, it is hard to see why, as a matter of pure logic, that privilege should be restricted to communications with legal advisers who happen to be qualified lawyers, as opposed to communications with other professional people with a qualification or experience which enables them to give expert legal advice in a particular field. ²³

Might legal advice privilege at common law be applicable to communications with robots providing legal advice?

The majority in Prudential held that a decision to extend the ambit of legal advice privilege to encompass legal advice given by non-legal professionals was a matter for Parliament. The issue was regarded as one of policy which was best left to Parliament and Parliament had already chosen to legislate in the area of legal advice privilege, for example, by extending privilege to other professions. ²⁴ In addition, there was a risk of uncertainty regarding which professions would be encompassed by legal advice privilege had the appeal been allowed. ²⁵ So, legal advice privilege can only be applicable to communications between a client and a robot provided that this does not amount to an extension to the common law privilege which, following Prudential, would require statutory intervention. In other words, the circumstances must be such that a robot could properly be regarded as a legal adviser at common law. This seems to be so even though Lord Sumption (using words that on their face would appear to be as applicable to developments in legal technology as to accountants) suggested that courts should be wary of leaving matters to Parliament where decisions can be made at common law which reflect Parliament’s intentions in the light of modern developments. ²⁶

In order to determine whether communications with a robot may fall within legal advice privilege (i.e. whether the common law privilege is applicable to such communications) it is necessary to consider two matters. First, whether robots are capable of giving legal advice for the purposes of the common law privilege. Secondly, even if they are so capable, whether it is feasible that robots might be admitted to the legal professions, membership of which, the decision of the House of Lords in Prudential made clear, is a necessary requirement for privilege to attach.

Can a robot give advice that qualifies as legal advice for the purposes of legal advice privilege?

Communications between robot and client would only be privileged if they comprised the giving or receiving of legal advice. So, what constitutes ‘legal advice’ in this context and is a robot capable of giving it? The nature of legal advice for the purposes of legal advice privilege was considered by the House of Lords in Three Rivers District Council and others v Governor and Company of the Bank of England (No 6) (Three Rivers). ²⁷ The case essentially required the court to determine what sort of communications between client and lawyer are protected by legal advice privilege and whether this might include what was described as ‘presentational advice’. In Three Rivers, the parameters of legal advice privilege are quite widely drawn. Referring to judgments from Balabel v Air India, ²⁸ their Lordships considered that legal advice is not limited to ‘telling the client the law; it must include advice as to what should prudently and sensibly be done in the relevant legal context’. ²⁹ As there must be a relevant legal context, legal advice privilege does not apply to all solicitor/client communications whatever their nature. For example, where a lawyer advises a client about business or financial matters, a relevant legal context may be lacking with the result that legal adviser/client communications will not be privileged. ³⁰ There will be situations where it is difficult to determine whether or not the advice has a relevant legal context. In Three Rivers, Lord Scott suggested that the question to ask was ‘whether the advice relates to the rights, liabilities, obligations or remedies of the client either under private law or under public law’. ³¹ If it did not so relate, then the communications would not be privileged, but if it did, then a secondary question was whether the communication had taken place in circumstances which came within the policy justification for privilege. ³² Baroness Hale referred to the relevant legal context as being ‘…one in which it is reasonable for a client to consult the special professional knowledge and skills of a lawyer, so that the lawyer will be able to give the client sound advice as to what he should do [or] not do, and how to do it…’. ³³ Lord Carswell expressed the view that ‘all communications between a solicitor and his client relating to a transaction in which the solicitor has been instructed for the purpose of obtaining legal advice will be privileged…provided that they are directly related to the performance by the solicitor of his professional duty as a legal adviser of his client.’ ³⁴

Considering the meaning of legal advice suggested in Three Rivers, the question that arises is whether the product of an automated process can come within these parameters and constitute legal advice for the purposes of legal advice privilege. Currently, automated legal services are delivered in a variety of ways. At one end of the spectrum is the no-cost, decision tree type automated offer, that uses a fixed menu of options and directs the user to a relevant form or process they can then use to try and resolve their legal problem. ³⁵ At the more sophisticated end of the spectrum, there are the increasingly complex systems that can process huge amounts of data to respond to questions phrased in natural language. ³⁶ In between, there are hybrid offers, combining some purely online document creation services with access (at a cost) to a network of lawyers giving advice online. ³⁷

Does the way in which automated technological processes work preclude them from giving what is regarded as legal advice for the purposes of privilege? For example, the IBM Watson type model relies on brute force processing to analyse data using word association and then calculate the probability of an answer being accurate (The Law Society of New South Wales, 2017: 42). At the lower end of the spectrum in terms of complexity is the decision tree style process. Considering the judgments in Three Rivers, the key component of legal advice is the existence of a relevant legal context. Essentially, the communications must relate to private or public law rights, liabilities, obligations or remedies and must have been made in circumstances in which it was reasonable to seek the ‘special professional knowledge and skills of a lawyer’, the communications being directly related to the performance of the lawyer’s ‘professional duty as a legal adviser’. If the output of the automated process enabled the client to regulate their affairs in accordance with the law, then the underlying policy justification for legal advice privilege would be met. This could be the case with both a decision tree limited options process and a more sophisticated natural language model, provided that the circumstances were such that it would be reasonable to consult a lawyer’s special knowledge and skills and the communications between the client and the robot were such as would fall within the ambit of performance of the duties of a legal adviser.

As is true of a human legal adviser, the fact that a robot does not explicitly provide legal advice to the client would not mean that the interaction between it and the client in a relevant legal context would not implicitly amount to the provision of legal advice for the purposes of legal advice privilege. For example, in a conveyancing process, presenting the client with a contract to sign amounts to implicit legal advice that the contract is correctly drafted and complies with instructions. ³⁸ The position would appear to be the same where a robot, upon the basis of information provided by the client, drafts a document for the client to sign.

Whether the automated software can perform a solicitor’s ‘professional duty’ or possesses the professional skills of a lawyer is unclear. Professional duty in this context may mean duty to the client in accordance with relevant professional body codes of conduct or simply encompass the work that a solicitor normally carries out in a professional capacity. What is meant by the professional skills of a lawyer could simply encompass the skills required to perform the work that a solicitor normally carries out rather than relate to, for example, the specific skills required in order to qualify as a solicitor. Software could be programmed to comply with codes of conduct and, arguably, skills relevant to a particular area of practice are already demonstrated by legal advice software in order to adequately perform and advise on that area.

The parameters of legal advice for the purposes of legal advice privilege are not framed by human reasoning, analysis and application. They are framed by context and underlying rationale. It should therefore be possible for automated legal advice software to give what constitutes legal advice for the purposes of legal advice privilege. The problem is that whilst the context in which advice is sought may be a relevant legal one, and the advice itself may qualify as legal advice, if the advice is sought from a robot it, like advice sought from an accountant, will not be privileged unless the robot is a lawyer.

Can a robot be a legal adviser for the purposes of legal advice privilege?

Is it conceivable that their Lordships in Prudential might have been persuaded that there were circumstances in which a robot could properly have been classified as a legal adviser at common law and (unlike an accountant) fall within the ambit of the common law privilege without the need for statutory intervention? Lord Neuberger suggested that legal advice privilege ‘only applies to communications in connection with advice given by members of the legal profession, which, in modern English and Welsh terms, includes members of the Bar, The Law Society, and the Chartered Institute of Legal Executives…(and, by extension, foreign lawyers)’. ³⁹

For a robot to be classified as a legal adviser would require professional bodies to open their membership to robots of the relevant type. As The Law Society has recognised, the debate about replacing human lawyers with artificial intelligence gives rise to ‘questions about what the core values of the legal profession are and what they should or could be in the future’ (The Law Society of England and Wales, 2018: 13). But is it likely that a robot would ever be able to emulate the essential attributes of a practising lawyer such that it could potentially satisfy all of the prerequisites to qualification as a solicitor, barrister or legal executive? For example, qualification as a solicitor currently involves meeting the ‘day one outcomes’ through completion of prescribed academic and vocational training, including a period of work based learning. ⁴⁰ Even if it were possible for a robot to evidence completion of the required elements of assessment and training, it seems unlikely that a robot could demonstrate all the day one outcomes. Although a robot may arguably be able to demonstrate knowledge, analytical and practical skills such as drafting, the day one outcomes include, for example, the ability to recognise personal and professional strengths and weaknesses, to develop strategies to enhance professional performance and to work effectively as a team member. In addition, the ability to behave professionally and with integrity and to identify issues of culture, disability and diversity are required. It seems unlikely that a robot could effectively participate in the elements of training designed to meet these outcomes or indeed demonstrate these attributes. It may, within the foreseeable future, be impossible to design a robot that could meet all of the relevant prescribed characteristics to qualify as a lawyer. Remus and Levy make the point that the complexity of some legal tasks, such as human interaction where skills of emotional intelligence are involved, is such that they are unlikely to be reduced successfully to a set of coded instructions:

Unscripted human interaction falls into this category because it often depends on formulating responses to unanticipated questions and statements. This, in turn, requires recognizing the broader context in which words are being used-not only the surrounding words…but the identity and motivation of the speaker and the purpose of the communication. (Remus and Levy, 2017:512)

In considering whether a robot could potentially be a member of one of the legal professions, another issue that would require resolution under the current state of the law is whether a robot is capable of being a ‘person’. A similar problem was encountered when women first applied to be solicitors, even though women, unlike robots, have clearly always been persons within the normal usage of that term and undoubtedly possess the same professional attributes as their masculine counterparts. In 1919, the Court of Appeal held that a woman could not be admitted as a solicitor. Section 2 of the Solicitors Act 1843 referred to a ‘person [being] admitted and enrolled and otherwise duly qualified as an Attorney or Solicitor, pursuant to the Directions and Regulations of this Act.’ Section 48 provided that ‘every word importing the Masculine Gender only shall extend and be applied to a Female as well as Male…unless…there by something in the Subject or Context, repugnant to such Construction’. The Court of Appeal held that, ‘the Act of 1843 confers no fresh and independent right, because it does not destroy a pre-existing disability’. ⁴³ The position was rectified by The Sex Disqualification (Removal) Act 1919, which provides that

A person shall not be disqualified by sex or marriage from the exercise of any public function, or from being appointed to or holding any civil or judicial office or post, or from entering or assuming or carrying on any civil profession or vocation, or for admission to any incorporated society (whether incorporated by Royal Charter or otherwise),…

The Law Society recently posed the question, ‘Far enough into the future, will AI/robots be sufficiently advanced to deserve “personhood”?’ and suggested that ‘the Common Law approach allows judges to evolve the law and, for some, it is an overreach to call for new laws when existing ones can be applied in, or transitioned to, new contexts’ (The Law Society of England and Wales, 2018: 14). In view of the reluctance of the Supreme Court to extend the ambit of legal advice privilege, it is suggested that this is an area where legislation would be required if robots were to be treated as persons.

Is it desirable to extend legal advice privilege by statute to encompass legal advice provided by a robot?

Upon the assumption that, for the foreseeable future, it will not be possible for robots to qualify as members of one of the legal professions (thus preventing legal professional privilege at common law from encompassing legal advice provided by robots), the next question is whether statute should extend legal professional privilege to such advice. It is suggested that there are two reasons why such legislation might not be desirable. First, in a rapidly evolving environment of automated legal and other professional services, it would be extremely difficult to draft a provision which provided an adequate degree of certainty concerning which form of automated service benefitted from the protection of legal advice privilege. This equates with a point made by Lord Neuberger in Prudential when declining to extend the ambit of the privilege at common law to accountants. In his dissenting judgment, Lord Sumption distinguished between persons ‘whose profession ordinarily includes the giving of legal advice’ and ‘other advisory professions whose practitioners although not lawyers require some knowledge of law’. ⁴⁴ Lord Neuberger regarded this distinction as ‘carry[ing] with it an unacceptable risk of uncertainty and loss of clarity in a sensitive area of law’. ⁴⁵ He believed that requiring the courts to draw this distinction would require them ‘to delve into the qualifications or standing, and maybe into the rules and disciplinary procedures, of a particular group of people to decide whether the group constitutes a profession for the purpose of LAP’. ⁴⁶ Similarly, he pondered whether the issue of whether a profession ordinarily included the giving of legal advice ‘[s]hould.be judged by reference to the profession generally, a particular branch of the profession or the practice of the particular member of the profession…’ and ‘suspect[ed] that much of the advice given by most members of those professions could not infrequently be characterised as “legal” in nature by some people but not by others’. ⁴⁷ The difficulties identified by Lord Neuberger would equally be encountered by: Parliament in attempting to draft a provision extending legal professional privilege to robots; providers of automated services in attempting to determine whether the relevant statutory provision applied to some of the services that they provided; and potentially the courts when required to determine whether communications with an automated service fell within the ambit of the statutory provision. It is suggested that in order to provide certainty, statute would be required to legislate in terms of specific forms of automated advice (as it has done in the past when extending the ambit of legal professional privilege to specific professions such as patent attorneys, trade mark attorneys and licensed conveyancers) ⁴⁸ but that an attempt to legislate by adopting a more general formulation, such as that which Lord Sumption’s approach to human advisers might suggest, would result in significant uncertainty. In the ever-changing environment referred to above, even that would be problematic, as technological change potentially outpaces legislative definition.

The second reason why legislation to confer legal advice privilege upon legal advice provided by robots might not be desirable concerns the relationship between the rationale for legal advice privilege and the ability of consumers to instruct specialist, cheap or free forms of legal advice. Some clients may not know of or care about the existence of legal advice privilege or may be prepared to forego it in order to obtain the service they want at the price they prefer. As Lord Scott observed in Three Rivers, ⁴⁹ in ‘many cases clients would have no inhibitions in providing their lawyers with all the facts and information the lawyers might need whether or not there were the absolute assurance of non-disclosure’. ⁵⁰ Some consumers already choose to consult other professionals for legal advice, such as accountants providing fiscal legal advice, even though their communications are not protected by privilege. Moreover, as was indicated above, recent reform proposals by the SRA, approved by the Legal Services Board (LSB), ⁵¹ will authorise solicitors to work in unregulated firms even though the SRA appears to accept that whether or not legal advice privilege will arise in relation to communications with such a person is uncertain (Solicitors Regulation Authority, 2017: 62–66). The SRA does not intend ‘to provide a commercial advantage to any type of firm’ and believes that ‘regulated firms employing solicitors will continue to provide a strong “brand”; the difference is the ability to provide the full range of legal services (including reserved activities), the availability of legal professional privilege (LPP), and a range of consumer protections that are unrivalled by any other profession, either in the UK or internationally’ (Solicitors Regulation Authority, 2017: 59). Thus, the SRA and, by approving its proposals, the LSB, both seem to accept that consumers are entitled to choose to instruct legal service providers even though it is uncertain whether legal advice privilege will attach to communications with those providers. Legislating to extend legal advice privilege to encompass communications with robots providing legal advice would go against the trend of supporting increased consumer choice in the provision of legal services and would not reflect the reality that markets commonly include a variety of product at different pricing levels that possess different attributes. For this reason, legislative change to this effect does not appear to be desirable and the drafting issues referred to above would seem to suggest that it is not practicable.

Upon the assumption that legislation to extend legal advice privilege to robots providing legal advice is not desirable, it could be argued that a proportionate response to the provision of legal services by robots would be to impose a requirement for a clear health warning concerning the absence of legal advice privilege. This would enable the consumer to make an informed choice whether to obtain traditional legal services or whether, at reduced cost, to use a robot even though privilege would not be applicable. In England and Wales, not all areas of legal practice are reserved to lawyers. A commercial undertaking may currently offer legal services in unreserved areas in a number of ways: through an unregulated human adviser or entirely via automated software. For example, accountants regularly give tax advice which, if given by a lawyer, would be privileged. In either case, as a lawyer is not involved in giving advice then privilege will not arise. However, in the former case there is still contact between non-legal adviser and client and the opportunity for a client to ask or the adviser to explain that legal advice privilege is not available. (This does not guarantee, of course, that such a question will be asked and/or that such guidance will be provided.) Where an entirely automated process is used to give advice, legal advice privilege will not be available and a consumer using this type of service may well have no idea that this is the case.

Whether a client makes use of a human non-legal adviser or an automated system, it may be that, as has already been suggested, for the expertise that the non-legal human adviser (e.g. an accountant) can provide or the convenience and benefit that a low-cost online advice service can offer, the consumer is happy to trade off having the protection of legal advice privilege. Sheppard points out that both LegalZoom and Rocket Lawyer include disclaimers on their websites advising customers that communications are not protected by attorney-client privilege. He suggests that ‘At this price point, there appears to be consumer demand for the product. LegalZoom claims to have had over three million customers. Rocket Lawyer, another online legal documents creator, claims to have created over three million documents for customers…’ (Sheppard, 2015: 1840). This means that clients using such services, for example to assist when forming a business or drafting an employment agreement, would not be able to rely on legal advice privilege. Provided that they have both read the disclaimers and have understood what it is that they are giving up, they have made an informed choice. It may be that in the context of communications with a robot for such purposes there will often be little disadvantage to the client in not being able to claim the privilege, given the limited nature of the communications that are likely to take place in the context of the current level of sophistication of such robots.

A counter-argument to the imposition of a mandatory warning requirement where robots give unprivileged legal advice is that no such requirement is currently imposed upon human non-legal advisers, such as accountants, when they give legal advice. Indeed, in relation to its proposals concerning solicitors in unregulated firms, the SRA has suggested that:

[i]t is down to the individual solicitor to make it clear to their clients what level of protection that client has and where such protections would be appropriate and/or relevant. In most circumstances this will not be an issue, but there may be occasions when a solicitor working in a non LSA regulated firm should advise their client on the benefits of privilege. This may include advising them of the option to seek advice from a solicitor in a regulated firm in order to make sure that this attracts privilege. (Solicitors Regulation Authority, 2017: 66)

There is currently no mandatory warning where legal advice is given by a human non-legal adviser. It seems that such a warning will not be required when, in future, such advice is given by solicitors in unregulated firms. It would seem anomalous to require such a warning where legal advice was given by a robot. One argument might be that such a mandatory warning requirement should be imposed both on robots and on humans. Yet accountants, for example, would undoubtedly regard it as wholly unfair were they required to add to the illogical commercial disadvantage of legal advice privilege not attaching to communications with their clients, made for the purposes of giving or obtaining legal advice, a requirement that they were required to spell this out. Indeed, in areas where the non-existence of privilege might be little or no disadvantage to the client, could the imposition of such a health warning potentially chase clients away from specialist, efficient, low-cost or free services, either towards more expensive services that provide them with no tangible advantage or even discourage them from using any service, which would act contrary to the rationale underlying the existence of the privilege? If the existence of a specialist, cheap or free automated system results in clients utilising the system to arrange their affairs in accordance with the law, then the rationale underlying the existence of the privilege is achieved by another means. Moreover, an attempt to legislate to impose mandatory health warnings would give rise to the same problems of legislative drafting etc that were identified above, i.e. the issue of how to define the forms of automated legal services (or indeed human service providers) to which such a warning requirement would apply. Consequently, the imposition of such a mandatory health warning requirement does not appear to be desirable or practicable.

What are the consequences for legal advice privilege where robots are deployed by human legal advisers in circumstances in which the level of human lawyer supervision is minimal?

Upon the assumption that robots will not be admitted to membership of one of the legal professions in the foreseeable future, does the use of a robot by a law firm to deliver legal services have the potential to defeat a claim of privilege in circumstances in which the level of human supervision of the robot and/or understanding of how the robot performs its tasks is minimal? Clearly, lawyers regularly use databases such as Lexis or Westlaw as research tools and it seems unlikely that this, equating with the use of the traditional paper-based law library, has potential to negate the existence of legal advice privilege. Equally, paralegals and trainee solicitors, acting under the supervision of qualified solicitors, are frequently involved in the provision of legal services, and it has never been suggested that their involvement provides a threat to the existence of legal professional privilege. ⁵² Conversely, legal advice privilege does not attach to communications where a human other than a legal adviser (e.g. an accountant) does not merely act as a conduit for communications between legal adviser and client but is required to bring material into existence. ⁵³

What would the position be if a robot, deployed by a law firm and supervised by a human lawyer, was so sophisticated that it was capable of receiving the client’s instructions in their original form and of providing full legal advice based on those instructions without the intervention of a human legal adviser? If such a robot was merely used as a tool by a human legal adviser, who considered its recommendations and then made a decision relying on them, on any other relevant information and on the human legal adviser’s own expertise, the position would seem to equate with the use by a lawyer of a sophisticated version of traditional legal databases. There would seem to be no reason why the use of the robot in this way would prevent legal advice privilege from arising. What, however, if legal advice produced by such a robot was merely rubber stamped by a human legal adviser through whose hand the instructions and the advice had passed? It may be that the legal adviser had little or no understanding of how the robot had reached its conclusions (and perhaps, if inexperienced, had little or no understanding of the instructions or the advice). This would appear to give rise to ethical/professional conduct issues and, we would assert, should also prevent legal advice privilege from attaching to communications with the robot.

In its report on The Future of Law and Innovation in the Profession, a Commission of Enquiry established by The Law Society of New South Wales posed the following questions:

[w]here a lawyer provides a legal service that has been supported by technology…can [the duty to deliver legal services competently] be discharged if the lawyer does not have, at the very least, a basic understanding of how that technology works?…[T]o what extent should [the] lawyer be required to understand the workings of the algorithms and the integrity of the data used to produce the legal work?…[T]o what extent should [the] lawyer be required to understand the technologies used…to ensure data security? (The Law Society of New South Wales, 2017: 41)

Is there a point at which advice provided by a robot under the nominal supervision of a human lawyer should not be regarded as being privileged at common law because, to paraphrase Lord Neuberger, it could not properly be said to be ‘given by a member of the legal profession’? ⁵⁴ Upon the assumption that the robot has not been admitted to one of the legal professions, we would assert that legal advice privilege should not attach in circumstances in which there is no effective supervision of its work by a member of the legal profession. The use of an identical robot by an accountancy firm would not attract privilege even if the accountant understood the workings of the robot and contributed their own skill and knowledge to the ultimate advice given. For privilege to attach in circumstances where there is no effective supervision by a lawyer, merely because the robot is deployed by a law firm, would seem to provide law firms with a commercial advantage which the decision of the Supreme Court in Prudential does not appear to justify.

Potential objections to the existence of legal advice privilege based upon the lack of supervision of robots by lawyers could be countered by modification of the professional conduct code for solicitors so as to explicitly cover required minimum levels of supervision where technology is heavily used in a law firm to give almost entirely automated advice. A more difficult issue is how such rules should deal with the issue of lack of understanding of what the technology being supervised is actually doing and how it is doing it. As was suggested above, a point may presumably be reached at which such lack of understanding renders supervision by a lawyer purely nominal. This lack of understanding may also affect those who develop systems. For example ‘networks are often “black boxes”, in which the (decision making) processes taking place can no longer be understood and for which there are no explanatory mechanisms’ (van den Hoven van Genderen, 2018: 50–51). Perhaps the rules should require that systems make clear to lawyers what they are doing, it having been suggested that ‘where algorithms do not provide causal accounts, the ethics of decision-making become opaque’ (Devins et al., 2017: 398). And/or, it may be that such requirements should relate to the provision of services by lawyers and IT service providers in a holistic manner, ensuring that the provision of technological legal services is safeguarded by an adequate combination of legal and IT expertise, working in conjunction. This could be in accordance with agreed methodologies that safeguard the competent delivery of legal services, including professional obligations related to legal professional privilege. Rob van den Hoven van Genderen (2018: 51) suggests that a mechanism which allows some degree of transparency regarding how artificial intelligence systems work could become a legal requirement.

In relation to the suggestion that conduct rules could be amended to deal with the impact of legal technology, two particular issues arise as regards the position of solicitors, by far the largest of the legal professions, ⁵⁵ in England and Wales. First, the SRA intends to introduce distinct codes for solicitors and for firms, ⁵⁶ so one issue would be which provisions should ideally be in which code. For example, it might be that provisions concerning the interaction between a legal service provider and its IT service providers should be in the code relating to firms, with requirements relating to the competence of solicitors in relation to IT being in both codes. The second issue presents the greater problem, however. It is that the SRA intends to take a shorter, sharper, less prescriptive approach to regulation (Solicitors Regulation Authority, 2018: 21, 23). In its consultation response to the SRA, The Law Society suggested that:

[t]he codes are shorter and simpler and the overarching Principles have been reduced from 10 to 6, losing the principle ‘provide a proper standard of service’ amongst others. This is both a standards and client protection issue. Furthermore the language in the codes is so lacking in specificity that firms will spend more time trying to establish what will comprise compliance; there will also clearly be a wide margin of discretion for the regulator to decide what constitutes compliance (The Law Society, 2016: 22).

It could be argued that, so far as developing technology is concerned, the SRA’s less prescriptive approach to regulation does make sense, given that ‘increases in the power of computing are exponential rather than linear’ (McGinnis and Pearce, 2014: 3046). Thus, requirements regulating the nature of the relationship between robots and human lawyers could easily be obsolete as soon as (or even before) they came into force. This could result in regulations concerning the operation of older technologies rapidly becoming ineffective as new technologies develop. Alternatively, the creation of such regulations could hinder the development and/or implementation of new technologies since ‘unnecessary regulation [can] chill additional innovation’ (American Bar Association, 2016:41). The likelihood is that the limits of legal advice privilege as it does or does not exist in the context of varying degrees of robot/human legal adviser interaction will need to be explored by the courts. It may be that regulatory body action will only be catalysed, if at all, if future judicial decisions make clear that developments in the technology underlying the provision of legal services by solicitors operating in regulated law firms threatens the existence of the privilege to the detriment of the public.

Conclusion

This article makes an original and distinctive contribution to discourse in this area through offering novel perspectives on and solutions to the previously unexplored question of whether common law legal advice privilege can and should attach to communications between a client and a robot. In answering this question, the following areas have been investigated.

First, whether the rationale underlying the existence of legal advice privilege encompasses communications between clients and robots. Applying the decision of the Supreme Court in Prudential, it is clear that the rationale underlying legal advice privilege is capable of applying to communications between a client and a robot giving legal advice. Allowing a client to arrange its affairs in accordance with relevant legal provisions is strongly in the public interest and, in this respect, complete candour is promoted by the ability to keep communications between client and legal adviser confidential. This rationale is equally applicable to legal advice given by a human non-legal professional or by a robot.

Secondly, whether legal advice privilege at common law may be applicable to communications between clients and robots. Again, by analogy with the decision in Prudential, which concerned the position of accountants, it seems to be clear that the courts would be unlikely to extend the privilege in this way and would regard such a decision as one for Parliament, unless it is possible for a robot to be classified at common law as a member of one of the legal professions and to give advice that qualifies as legal advice for the purposes of the privilege.

Thirdly, whether robots may be capable of giving advice that qualifies as legal advice for the purposes of legal advice privilege. The parameters of legal advice for the purposes of legal advice privilege are not framed by human reasoning, analysis and application. They are framed by context and underlying rationale. As concluded above, it seems clear that the rationale underlying legal advice privilege is capable of applying to communications between a client and a robot giving legal advice. In terms of relevant legal context, the judgments in Three Rivers and other authorities referred to therein suggest that this requires that the communications must relate to private or public law rights, liabilities, obligations or remedies and must have been made in circumstances in which it was reasonable to seek the ‘special professional knowledge and skills of a lawyer’, the communications being directly related to the performance of the lawyer’s ‘professional duty as a legal adviser’. No matter how unsophisticated the software, the client is clearly consulting the automated system in the relevant legal context, for the purposes of receiving legal advice. If the output of the automated process enabled the client to regulate their affairs in accordance with the law, then the underlying policy justification for legal advice privilege would be met and the output could amount to legal advice even if it did not patently take the form of advice.

Fourthly, whether a robot is capable of being a legal adviser for the purposes of legal advice privilege. Extrapolating the judgement in Prudential, if a fully autonomous robot can be regarded as a qualified lawyer, then legal advice privilege could attach to communications between client and robot. It is also clear from Prudential that, despite recognising the irrationality of restricting legal advice privilege to advice from qualified lawyers, at common law the privilege will not be extended to cover communications with other professionals. So, if a fully automated robot cannot be regarded as a qualified lawyer, despite being programmed to give legal advice, then legal advice privilege will not attach to communications with the client. At present, legal technology does not appear to be capable of producing a robot that possesses the combination of knowledge, skills and ethical awareness that would enable it to qualify as a member of one of the legal professions. In addition, in order for a robot to qualify as a solicitor, statutory intervention would be required either to amend section 1 of the Solicitors Act 1974 to encompass machines as well as persons or to give robots the status of persons. Moreover, if robots were to be admitted to membership of a legal profession, this would presumably require amendment of relevant professional conduct rules.

Fifthly, we considered, if legal advice privilege is not applicable to communications between clients and robots, whether statutory intervention extending the privilege to such communications would be desirable. Prospective users of legal services may choose to instruct a non-lawyer for what is essentially legal advice in order to take advantage of specialist expertise with (or without) the knowledge that communications are not protected by legal advice privilege—for example, instructing an accountant for fiscal legal advice. Consumers using free or low-cost fully automated legal advice services are likewise making a choice, trading receiving a low or no-cost service against the protections that come with instructing a lawyer, including the benefit of legal advice privilege. Again, the choice may or may not be an informed choice. There seems to be no reason why consumers should not be entitled to choose to access free or cheap unprivileged automated legal advice. Moreover, attempting to draft statutory provisions identifying the types of automated advice to which privilege would attach would appear to give rise to issues of uncertainty similar to those that Lord Neuberger identified in Prudential when he indicated that if legal advice privilege was extended beyond lawyers there was a risk of uncertainty regarding which professions would be encompassed. Thus, attaching legal advice privilege to legal advice provided by robots is arguably unnecessary, potentially problematic and, unless statute also extended the privilege to human non-legal advisers, would put the latter at an unfair competitive disadvantage.

An alternative would be to require providers of automated legal services to make clear that their services are unprivileged. Whether this would result in consumers having a full understanding of the significance of this fact is unclear. The disclaimers on the relevant websites might not be read at all or, if they were read, might not be understood. Where a human non-legal professional is instructed to give legal advice, they may not even provide the client with information concerning the non-existence of legal advice privilege (the non-legal adviser may not even be aware of the issue). Requiring providers of automated legal advice to give such warnings would logically suggest that similar requirements should be imposed on human non-legal advisers who might well protest at being required to publicise what they could properly regard, post Prudential, as an unfair competitive advantage possessed by lawyers. Again, it could also result in significant uncertainty in identifying which types of automated or non-professional human advice should be classified as legal advice which falls within the ambit of such a warning requirement. It is suggested that whether legal advice privilege should attach to robots that give legal advice without human supervision and/or whether unprivileged services provided by robots should come with an appropriate health warning are both issues that, post Prudential, should be considered by Parliament, if at all, in line with equivalent reforms directed at human non-legal professionals who give legal advice.

Finally, we considered the consequences for legal advice privilege of the use by a human legal adviser of a robot to provide legal advice in circumstances in which the level of supervision by a human legal adviser is at most nominal and where the understanding of relevant algorithms is minimal. Legal advice from a human legal adviser who, relying upon their professional skills and knowledge, utilises technological resources in formulating that advice, is and should be covered by legal advice privilege. This is currently the case where legal databases are used, for example to research the law and should continue to be the case where more sophisticated legal technology is utilised, provided that the lawyer is deploying their professional skill and judgment.

If, however, a robot that was not a member of the legal profession, received client instructions and formulated legal advice with no more than nominal supervision from a human legal adviser, we assert that communications should not be protected by legal advice privilege. In these circumstances the lawyer is not using his or her professional skill and judgment to give advice, but is simply a conduit through whom the advice flows from the robot to the client. To allow privilege to apply in these circumstances would be to give law firms an unfair competitive advantage over other professions using similar technology to give legal advice.

To counter suggestions of unfair competitive advantage in the above circumstances, relevant professional body codes of conduct could be amended to require minimum levels of supervision by a lawyer where technology is very heavily used by a law firm to give legal advice. Codes could also require prescribed levels of cognisance relating to how software works at a holistic level involving IT professional and lawyers so that the supervision that does take place is meaningful. The current trend towards less prescriptive regulation by the SRA makes such changes unlikely. Limited regulatory intervention is not necessarily a bad thing, both because in a rapidly evolving technological environment regulations might become obsolete virtually as soon as they are drafted and because overregulation might have the consequence of stifling desirable technological innovation. It remains for the courts to determine the extent to which variations in the level of robot/human lawyer interaction might threaten the existence of legal advice privilege, with detailed regulatory intervention being catalysed, if at all, depending upon the nature of such judicial determination.

The fundamental question must be whether at some time in the future we reach a point at which the bulk of legal service provision to individuals and corporations is by technology which provides a relatively cheap and reliable service in an environment that does not give rise to legal advice privilege. If and when that point is reached, it may become clear that the presence or absence of the privilege is not a key factor either when potential clients are determining whether to obtain legal advice or when they are determining what information to disclose to their legal advisers. Consequently, it may be that the rationale underlying the very existence of legal advice privilege is eroded to such an extent that the necessity for its continued existence comes into doubt.