A. Introduction
Much attention has been given to whether artificial intelligence will replace professionals such as lawyers. Although the precise impact of AI on white-collar employment has proven hard to assess,Footnote 1 one prevalent viewpoint is that artificial intelligence will not replace humans but will allow them to operate with greater skill and efficiency.Footnote 2
In the context of licensed professions, however, that formulation avoids the essential question—which sorts of humans will we allow to offer AI-enabled services in professional settings? Will we continue to insist on a professional course of training that made sense when command of esoteric knowledge by humans was the critical element in providing useful and reliable services, or will we address whether the advent of powerful technologies should allow us to rethink the model under which services are offered? These are questions that have application across different regulatory systems for legal services worldwide.
This new debate comes as a widespread access-to-justice crisis has already put into play the drawing of professional boundaries. In the United States, the statistics are dire. Most Americans face access-to-justice issues, and most of those are unable to get any help. The inability to get help vindicating legal rights can make legal protections illusory and distort the common-law rulemaking process. The access-to-justice problem is not just a U.S. dilemma but affects people worldwide.
In order to explore these issues, this Article takes as a case study the ongoing experiment in the United States of allowing Limited License Legal Technicians (LLLTs) to offer services directly to the public, and this Article seeks from a close investigation of the specifi tc instance to seek insights generalizable to other settings.Footnote 3 This experiment, begun with great hopes and much fanfare, has proved disappointing so far. This Article asks whether allowing AI use by specially licensed professionals in the provision of legal services would create a new model of service that does not inherit the limitations of the professional model of law practice. The Article also asks whether the AI necessary to power this new professional model should be left to the market to develop or should be created as a public good.
While the scope of the U.S.’s protected space for lawyers may be somewhat broader than in many other countries, the issue of integrating new technologies with nonlawyers to meet access-to-justice needs crosses national and regulatory boundaries. Worldwide, the advent of new technical capabilities allows for a rethinking of what work should be reserved to lawyers—and which legal services, if any, should be delegated to new categories of service providers that are not lawyers. The task of looking at the often-complex issue of where boundaries are drawn should be undertaken with the awareness that the possible alternatives to lawyers today are different in kind from what has been possible in the past.
B. The Access-to-Justice Crisis in the USA and Worldwide
I. The Scope of the U.S. Access-to-Justice Crisis
The USA is in the throes of a long-standing and well-documented access-to-justice crisis.Footnote 4 The Legal Services Corporation (LSC) reported that 92 percent of low-income individuals did not receive adequate legal assistance for their civil legal issues in 2022.Footnote 5 This reflects a worsening trend, as this figure represents a six-percentage increase since 2017.Footnote 6
The problem extends beyond low-income individuals, however.Footnote 7 Middle-class households also have access-to-justice problems. A 2019 study by the World Justice Project looked at access-to-justice problems worldwide.Footnote 8 In the USA, nearly two-thirds of all respondents had experienced legal problems, but only a third were able to access help. Forty-five percent experienced a hardship due to their legal problems.Footnote 9
There are other wrinkles. The problem is especially acute in rural areas, for example, due to the lack of qualified professionals and difficulties in accessing often distant courthouses.Footnote 10 Certain kinds of problems also recur frequently—for example, landlord-tenant issues often see unrepresented tenants up against professionally represented landlords. As long ago as 2003, the State of California estimated that 90 percent of defendants facing residential eviction or potentially subject to a domestic-violence restraining order were unrepresented in these important court proceedings.Footnote 11
The consequences of being unable to access legal help go beyond individual economic hardship in a country that claims to operate by rule of law. Courts can struggle to maintain an orderly workflow while accommodating unrepresented litigants unfamiliar with court processes.Footnote 12 In both federal and state courts unrepresented litigants constitute a significant portion of all cases: In federal courts more than a quarter of claims are filed by pro se plaintiffs, and in state courts one side is unrepresented in three-quarters of cases.Footnote 13 In the common-law system of the United States, asymmetrical access to legal expertise can also distort the law itself. A system in which only the “haves” have systemic interests beyond the individual case tends to distort the development of substantive and procedural laws.Footnote 14 This tendency is aggravated when the one-shot players do not even enjoy representation.
This lack of access to justice undercuts core promises of American society. Legal rights are meaningless without a process to vindicate those rights. Any process is hollow if it cannot be accessed by large segments of the population. In a system where representation is inaccessible to one side but accessible to the other, law can become a tool for discrimination and oppression rather than justice.
The USA does not stand alone in facing an access-to-justice crisis. Studies worldwide show that countries across the world face similar issues.Footnote 15
II. The US Legal Profession’s Responsibility for the Access-to-Justice Crisis
Responsibility for the access-to-justice crisis often is laid at the feet of the legal profession. In the United States, lawyers have a near monopoly on providing legal services. Critics claim that lawyers maintain this monopoly in order to protect their incomes even though the public suffers.
The monopoly lawyers have over legal services in the United States derives from a professional model that has ancient roots. As with other professions, the core of the profession is the command of specialized knowledge. Lawyers have expertise in legal rules and the associated processes. This expertise allows them to provide services that those without equivalent knowledge cannot reliably provide.
Much has been written about the organization of professions, ranging from George Bernard Shaw’s much quoted quip that all professions are conspiracies against the laityFootnote 16 to the fulsome claims of professional apologists that professions are about providing a public service, almost selflessly.Footnote 17 Some view the claiming of a protected professional space somewhat cynically as a grab for prestige and income,Footnote 18 while others view the organization of professions into guilds maintaining a protected market as a logical consequence of an occupation that those outside the profession cannot easily judge or manage, especially given the need for the exercise of flexible discretion in conducting the occupation.Footnote 19
Sociologist Eliot Freidson developed a model with five elements for professional work. While no profession fully reflects the ideal model, U.S. lawyers are generally considered to track reasonably closely to the model. Freidson’s elements are:
1. [S]pecialized work in the officially recognized economy that is believed to be grounded in a body of theoretically based, discretionary knowledge and skill and that is accordingly given special status in the labor force;
2. [E]xclusive jurisdiction in a particular division of labor created and controlled by occupational negotiation;
3. [A] sheltered position in both external and internal labor markets that is based on qualifying credentials created by the occupation;
4. [A] formal training program lying outside the labor market that produces the qualifying credentials, which is controlled by the occupation and associated with higher education; and
5. [A]n ideology that asserts greater commitment to doing good work than to economic gain and to the quality rather than the economic efficiency of work…. The ideology claims both specialized knowledge that is authoritative in a functional or cognitive sense and commitment to a transcendent value that guides and adjudicates the way that knowledge is employed.Footnote 20
The US legal profession reflects the core elements of Freidson’s model. As with other professions, the professional model arose in an era where specialized knowledge was curated and applied only by human beings. Put differently, computerized information technology did not provide an alternative.Footnote 21
In such an environment, establishing that those providing expert services were indeed qualified made sense. Those without the necessary knowledge can do much harm to the public by offering inadequate services. In asserting the need for strict controls on who could provide professional services, the legal profession was rarely short of anecdotal tales about the harm caused by poorly trained or unethical nonlawyers—some of the miscreants being disbarred lawyers with full training but no longer holding a license.
A cynic, however, might note that the profession’s solicitude for the public welfare sometimes coincided with hard times for legal professionals. One major wave of expanding unauthorized-practice-of-law (UPL) protections, for example, coincided with the Great Depression in the United States.Footnote 22
That the professional model could be justified, however, does not speak directly to the issue of how that profession is defined or where boundaries are drawn. In this regard there are some peculiarities to the U.S. legal profession that will prove relevant to our discussion. The first is that the U.S. legal profession has been a unitary profession. Unlike in other countries,Footnote 23 there have been no different types or gradations of legal professionals. Those appearing in court and those giving advice as full time corporate employees in offices have the same academic training and credentials; traditionally, there have been no related legal occupations such as conveyancers or notaries.
The second is that very broad UPL restrictions give lawyers an unusually broad protected space. While the protected space is subject to boundary disputes and exceptions,Footnote 24 aggressively developed UPL provisions protect not just in-court but also out-of-court legal advice. Since the development of broad UPL protections in the early 20th century, nonlawyers cannot give individualized legal advice to the public.Footnote 25 Those who engage in the unauthorized practice of law can be subject not just to civil lawsuits but in some cases to criminal prosecution.Footnote 26
The broad licensing and the broad protected space for lawyers exist in tension with a reality of legal practice as it has developed over the past century. U.S. lawyers are increasingly specialized in their practice. The American model of legal education developed by Christopher Columbus Langdell at Harvard in the latter half of the 19th century prepared lawyers for a general practice in a small firm, which was a typical practice for that era.Footnote 27 Today, lawyers tend to practice more and more in or against larger organizations—whether law firms with hundreds or thousands of lawyers, substantial in-house legal departments, or government organizations with large legal staffs—and to engage in specialized rather than generalized practice. This creates a mismatch not just with the licensing of lawyers, which remains generalized and does not track medicine in requiring or even offering additional credentials for most specializations, but also with the required education, which struggles to offer in-depth preparation for specialized practices that might not be planned upon or even anticipated in one’s law-school years.
Specialization and the academic mismatch also create the possibility that traditional legal education is not the best preparation for the kinds of practices lawyers eventually end up engaging in. The codification of knowledge and specialization also allows those without full professional training to offer useful services in the specialized area.Footnote 28 Much of what is covered in the generalized education of law school proves irrelevant in some practice settings. At the same time, specialized non-legal expertise that is beyond the scope of traditional legal training may prove to be highly important to specialized practice.
The cost of obtaining legal training in the U.S. is high and requires an investment of both time and money that those making the investment hope will be compensated. The cost is only increased by the additional on-the-job training that will be needed before lawyers are competent to practice in a specialized area.
These costs have made U.S. lawyers protective of their protected space, which arguably is necessary if they are to recoup the investment made in becoming qualified. At the same time, these costs drive up the rates attorneys must charge in order to justify the costs of becoming qualified, which makes legal services costly even to the affluent and unavailable to many.
III. Proposals to Break the Professional Monopoly
In the context of an ongoing access-to-justice crisis, the “lawyers’ monopoly” has been widely questioned and criticized. The criticism became more acute starting in the 1970s as scholars focused on the costs to the public of the protected market given to lawyers.Footnote 29
One proposed response from academics was a change in regulation to allow regulated nonlawyers to offer legal services.Footnote 30 In a world where the fees charged by lawyers were too high and where the expertise of lawyers was often missing critical non-legal aspects, opening the field to nonlawyers with on-point training had appeal. In the mid-1990s an American Bar Association commission examined the issue and proposed allowing nonlawyers to offer services in certain situations.Footnote 31
IV. The Movement to Allow Alternative Professional Providers Gains Traction
As the 21st century dawned, the continued criticism about the way legal services were regulated and the ongoing access-to-justice crisis began to lead to serious proposals to try different approaches. LLLTs are just one of many experiments in the contemporary United States allowing nonlawyers to provide legal services. Other alternative services range from “court navigators” where volunteers guide people in the court system to parent advocates who help other parents with assistance in seeking education benefits.Footnote 32 Again, while regulatory schemes differ country-to-country, these experiments to open the marketplace to nonlawyers have relevance to any jurisdiction that provides lawyers with a protected market.
LLLTs and closely related experimentsFootnote 33 are especially significant, however, because the kind of help they allow comes closer to the role of lawyers than other “roles beyond lawyers” programs and because so much was expected of them. LLLTs in certain circumstances can take the place of lawyers. Like lawyers, LLLTs are allowed to give legal advice directly to clients and, in some states, even to appear in court.
LLLTs are also significant because they have been imagined and developed as an alternate professional model. Entry into these occupations requires extensive training followed by rigorous testing designed to assure that LLLTs have the required expertise to offer reliable services. While somewhat different from the training lawyers receive, the underlying assumptions and goal of professional regulation still apply—the assumption is that only properly trained humans can provide reliable services, and the goal is to make sure that the education and testing ensure that the services are reliable.
While conforming to a professional model, legal paraprofessionals such as LLLTs operate in a more limited arena. First, in all programs, they are not authorized to offer generalized legal services, but are limited to specific areas such as family law. Even within these specific areas, these paraprofessionals are often restricted in what work they can do. A look at the history and spread of the US experiment with LLLT programs will help give a sense of their potential, the opposition they’ve faced, and the current state of the experiment. The lessons from the American experiment—which end up being closely related to the adoption of a form of a professional model—are valuable to other jurisdictions looking for ways to offer alternatives to lawyers as a way to meet the access-to-justice crisis.
C. The Implementation of Limited License Legal Professionals in the USA as a Response to the Access-to-Justice Crisis
The recognition of LLLTs as alternate providers of legal services began in Washington State. Currently, there are five states that have implemented programs approving nonlawyers to obtain a limited license to provide legal services in certain situations. Washington, the first mover, has closed entry to its program. The states that currently issue some types of a limited license are Utah, Arizona, Minnesota, Oregon, and Colorado. Other states have considered and continue to consider adopting such programs, although in some cases those initiatives have run into active opposition from lawyers and others.
I. The Origin: Washington: Limited Licensed Legal Technicians (LLLT)
1. The Development of the LLLT in Washington State
The LLLT program in Washington State arose from the creation of a Practice of Law Board by the Washington State Supreme Court in 2001. At the inception of the board, paraprofessional practice was not one of the topics before the board. Instead, the board was charged with policing the unauthorized practice of law by nonlawyers while also looking into whether the needs of the public for legal services were being met.Footnote 34
The Board identified as an ongoing issue the provision of legal services by nonlawyers.Footnote 35 The Board found that sometimes such services were provided deceptively and for exorbitant fees. The Board, however, was not in a position to do much about the unauthorized practice of law, because prosecutors had little appetite for bringing criminal cases even though UPL was in theory punishable as a felony.Footnote 36
With regard to the access-to-justice needs of the public, a 2003 study by the Board reinforced earlier studies that showed that the public, and especially lower-income members of the public, had significant difficulty accessing legal help.Footnote 37 The board’s study showed that every year low-income people in Washington State “encounter[ed] more than a million urgent civil legal problems.”Footnote 38 Those legal needs were widespread, with 87 percent of low-income households experiencing at least one legal need with an average of 3.3 per household per year.Footnote 39 The report found that these problems were usually faced without help from lawyers:
Although an attorney often can help improve the situation dramatically, the overwhelming majority of low-income people with legal problems face them without any such help. Indeed many don’t realize that there are laws to protect them and that they could or should seek such assistance. Others don’t know where to begin, or feel too overwhelmed to try. Most simply throw up their hands and endure miseries few higher-income people would tolerate. They despair of their plight and grow cynical about the justice system.Footnote 40
Certain areas led to the greatest proportion of these problems. “[H]ousing, family, employment, and consumer matters accounted[ed] for more than half the legal issues experienced by low-income people.”Footnote 41 Shelter and security were put at risk by these unaddressed problems.
This led the Board two years later to propose a new kind of legal professional, the Limited License Legal Technician.Footnote 42 While novel in the legal field, the idea of a paraprofessional dealing directly with the public in a more limited role had precedents in the medical field, where nurse practitioners and other non-doctors had gained an entrée.Footnote 43 The proposal had difficulty progressing beyond the Board, however, with the Board of Governors rejecting the proposal twice in succeeding years.Footnote 44 Finally in 2008 the proposal reached the Washington Supreme Court for consideration.Footnote 45 The proposal was then revised to limit the practice area to family law.Footnote 46 The back and forth continued, and it was 2012 before the Court approved the proposal,Footnote 47 with family law being the sole approved practice area.Footnote 48
2. The Washington State LLLT Specifics
While by nature involving something less than full legal credentials and practice, the LLLT tracked in broad outline the professional model. As with lawyers and other professionals, expert knowledge provided the core expertise. The state regulation allowing LLLTs was explicit in this, defining an LLLT as “a person qualified by education, training, and work experience who is authorized to engage in the limited practice of law in approved practice areas of law.”Footnote 49
Even though LLLTs were to be limited to a narrow specialty and limited within that specialty in what services they could perform, the requirements to become an LLLT were daunting. Those qualifications required, as with lawyers, significant educational requirements plus an examination, but also required significant practical experience requirements of a type not required of lawyers and that could only be acquired under the supervision of a lawyer. On the educational front, LLLTs were required to complete a two-year associate degree, which degree had to include forty-five credit hours in an approved paralegal program. This required LLLT candidates to take general law-school-type courses that covered topics like contracts, legal research and writing, and the operation of the court system.Footnote 50 The educational requirements also included additional courses in family law and in topics specific to Washington State.
While those requirements, however substantial, were less than what lawyers need, the next set of requirements was both substantial and beyond anything lawyers in the U.S. are required to meet. Unlike lawyers, LLLTs were required to have completed 3,000 hours of relevant experience while supervised by a licensed attorney.Footnote 51 In the U.S. lawyers have no articling requirement. In other countries that do require articling, three thousand hours would typically be high even for lawyers. In Ontario, Canada, for example, where articling is required, the requirement is eight-to-ten months, which even in a demanding environment is likely to be far less than 3,000 hours.Footnote 52
Finally, there were demanding examination requirements. Three different exams were required. First, LLLTs had to pass the same exam required of licensed paralegals. Second, they needed to pass an ethics exam. Finally, a separate family law exam was required.
Given the difficulty of the path to qualification, it should come as no surprise that participation in the program was somewhat limited. As of the time the program was closed to new entrants, there were only 38 active LLLTs in Washington.Footnote 53 In comparison, there are more than 26,000 licensed lawyers in Washington, with more than 1,000 new lawyers joining the bar each year.Footnote 54
The raw count of LLLTs, modest as it is, might overstate the impact of the new form of professional on the public. Many LLLTs practice within traditional lawyer-owned law firms, contributing to the income of the lawyers and performing work that overlaps in many ways with the work done by traditional paralegals.Footnote 55 While LLLT rates are lower than average lawyer rates, rates that can exceed $100 per hour are still difficult for those on low incomes to afford.Footnote 56
3. The End of the Road for the Washington LLLT
In the end Washington was the first state to launch and the first state to terminate an LLLT program. As set up, the LLLT program operated under and ultimately was funded by the state bar, which is composed of lawyers. The state bar had found itself subsidizing the LLLT program with dues derived from attorneys; as the annual cost of the program exceeded a million dollars, the cost per LLLT was substantial.Footnote 57
As might be expected, this proved unpopular with the lawyers who were funding the creation of a competitive occupation. When the LLLT board came to the state Supreme Court in 2020 with plans to expand the program and a request for additional revenues the request did not get a warm reception. With the state bar pushing back against the LLLT program, the Washington Supreme Court decided to sunset the program.
While the program in Washington is no longer enrolling new LLLTs, those previously licensed are allowed to practice. More significantly than that, the experiment in Washington encouraged other states to explore legal paraprofessional programs of their own.
One sympathetic and thoughtful examination of the Washington experiment, published while it was still ongoing, concluded that as designed it was unlikely to meet access-to-justice needs and that further innovation and experimentation would be necessary.Footnote 58 Factors ranging from limitations on what LLLTs were allowed to do to forbidding them to partner with non-legal organizations all limited the potential for reform. Beyond that, as a fee-for-service business, LLLTs were placed in a market where the compensation of public-interest lawyers was already near subsistence levels, giving them little room to underprice the market.Footnote 59
II. Expansion of the Licensed Paraprofessional Concept to Other States
1. Utah: Licensed Paralegal Practitioner (LPP)
Utah was the second state to allow practice by legal paraprofessionals. Inspired by the Washington program, Utah began investigating the concept in 2015 and launched its program in 2018.Footnote 60 As in Washington, the program requires a combination of on-point education, experience in the field working under a lawyer, and examinations. The Utah program launched with a broader range of specializations, allowing not just family law but also debt collection and forcible entry and detainer. It also reduced the amount of time applicants had to work under lawyers, requiring 1,500 hours rather than 3,000. By 2024 there were 32 licensed under the program, the majority of whom worked in traditional law firms.Footnote 61
2. Arizona’s Program: Legal Paraprofessional (LP)
Also influenced by the Washington program, Arizona’s program differs in some significant ways. Unlike many other programs, the Arizona program allows paraprofessionals to “[a]ppear before a court or tribunal on behalf of a party” in certain situations and to work independently.Footnote 62 The program resembles other paraprofessional programs, however, in its educational and licensing requirements, reflecting a somewhat lighter version of the model for lawyers. In allowing its paraprofessionals to appear in court, albeit in limited roles, the Arizona program allows paraprofessionals greater authority and autonomy than some others.
The Arizona program provides two paths into the occupation. As with Washington and Utah, education can provide one path.Footnote 63 Arizona, however, gives a different path to those, such as traditional paralegals, who have substantial work experience in the field.Footnote 64
As with Washington and Utah, the paraprofessionals in Arizona are limited to specialties. There are currently four of these specialties: family law, lower-stakes civil matters, some lower-level criminal matters with no risk of incarceration, and administrative law.Footnote 65
As with the other states experimenting with paraprofessionals, the Arizona program expects those in it to support themselves by charging fees to their clients, some on their own and some in law firms.Footnote 66 By 2022 there were 22 licensed LPs in the state.Footnote 67
3. Minnesota: Legal Paraprofessional Pilot Project
The Minnesota Legal Paraprofessional program takes a different approach, allowing paraprofessionals to represent clients in court but also requiring that they work under the supervision of licensed attorneys. Following a multi-year pilot program, the program converted to permanent status at the start of 2025.
As with the other programs, specialties are required and the areas of representation limited. The specific areas include housing law, including evictions and other landlord-tenant disputes, and family law, including child custody and child support. In making the program permanent, the Minnesota Supreme Court noted that the results of the pilot phase had been strongly positive.
4. Oregon: Licensed Paralegals
The Oregon program became operational in January 2024. As with the other programs, it allows representation in specified areas, which in Oregon include family law and landlord-tenant law. As with similar programs, obtaining the professional requires a combination of education and passing exams. Applicants who do not hold law degrees must complete 1,500 hours of supervised service; those with law degrees must complete 750 supervised hours.
The first ten licensed paralegals were sworn in in March 2024.Footnote 68 All but one were legal assistants at the same firm, Stahancyk, Kent, & Hook.Footnote 69
5. Colorado: Licensed Legal Paraprofessionals
Colorado spent eight years studying the idea of legal professionals before finally adopting the Licensed Legal Paraprofessionals program in March 2023. A key factor helping to drive the program forward was a Colorado Judicial Branch statistic showing that seventy-five percent of all litigants in family law cases had no legal representation.Footnote 70
In the Colorado program, the paraprofessionals have limited rights of practice in court. The area of practice is located to family law, which can include divorce, child custody, child support, protective orders, name changes, and gender-designation changes.Footnote 71 They can prepare and file court documents, deliver opening and closing statements, respond to questions from judges, and advise clients at counsel table. However, they cannot examine or cross-examine witnesses.
As with other programs, entering the program requires meeting a somewhat lighter model of professional qualifications but with significant practical experience requirements that lawyers do not face. They must meet educational requirements, pass a written examination, pass character and fitness standards, and complete 1,500 hours of law-related experience, including 500 hours in family law settings.
The first group of LLPs was sworn in in June 2024. As with the other states, they are expected to be supported by fees charged to clients, with the hope that these fees will be at most half of those charged by attorneys. LLPs in Colorado can practice independently or within traditional law firms.
6. Other States — Still Exploring
Other states continue to examine the possibility of launching some kind of licensed paraprofessional program. California put substantial effort into developing a program but was directed to pause efforts by the state legislature.Footnote 72 Other states are at various stages of considering the option.
III. The LLLT Experiment Evaluated: A Tweak of the Professional Model
At present, while LLLTs remain a topic of substantial professional and academic interest, the results have been less revolutionary than might have been hoped. The numbers of those pursuing LLLT careers has remained relatively modest. Even for those licensed and practicing, a fairly high percentage appear to be operating not as an alternative to law firms, but as an additional professional option within law-firm structures.
Given that licensed paraprofessionals do somewhat reduce costs and do deliver services that meet the needs of those using them, it would be hard to argue that the programs represent anything other than a step forward. While the experience in Washington State suggests that cost effectiveness might be a concern, it can also be argued that the design of the Washington program was overprotective of licensed lawyers in a way that needlessly limited its appeal and drove up costs.
Legal academics who have studied the programs tend to be strongly supportive of them. On the whole, prices are a bit lower than fully licensed lawyers and there seem to be few, if any, complaints about the quality of the services provided. Just as the Washington State program, while now sunsetted, led to other states experimenting with new categories of legal service providers, the possibility seems strong that the second wave of programs will in turn lead to more experimentation.Footnote 73
That is not to say that all is rosy. A fairly high number of the newly licensed paraprofessionals appear to be former paralegals who have simply transitioned into marginally different roles at the law firms where they previously worked—and it is not clear that with their new credentials that their rates are lower than they were as billed out paralegals. Beyond that, the programs seem to have done little to address geographic distribution of legal services, with most serving in the same urban areas where lawyers are now available rather than in the rural areas where lawyers can be hard to find.Footnote 74
The comparison with the expansion of service providers in the parallel professional setting of medicine seems likely to continue to serve as an inspiration for experimentation. Bruce A. Green, for example, looks to the medical profession in his argument in favor of increased reliance on paraprofessionals. Green argues that “the medical profession has established stratified roles for many different categories of medical professionals, many of which do not require as much training and experience as medical doctors.”Footnote 75
This argument is likely to continue to have appeal. As noted before, the medical profession supports a range of technicians and paraprofessionals in providing medical services, from MRI machine operators to physical therapists. Most on point are physician assistants and nurse practitioners. While both represent providers who provide a high level of service, these two roles offer different models which in turn have proved influential with those pushing the legal experiments. Nurse practitioners, on the one hand, similar to the legal paraprofessionals in some states, can operate independent of doctors and provide a limited but significant range of services direct to patients. Physicians assistants, on the other hand, provide a model for those states that require legal paraprofessionals to operate under the direct supervision of a lawyer, just as physicians assistants do with doctors.
In evaluating the LLLT experiment to date, two major limitations are clear. First, in tracking the professional model, the LLLT model imposes significant costs on those seeking to enter the field, and expects those who surmount the challenges to entry to support themselves as lawyers do by charging a fee for bespoke services. One sympathetic but critical review of the Washington program observed:
The LLLT model essentially mimics the aspects of the legal profession that keep lawyers’ prices artificially high, providing little reason to believe that LLLTs will lower prices enough to serve a meaningful number of low-income clients. Nor do LLLTs’ and Candidates’ initial career plans indicate they will reinvent conventional legal service delivery models. Most either plan to work for law firms charging rates lower than attorneys’ fees or open their own practice while keeping overhead low. Both paths still require LLLTs to charge rates high enough to bring in revenue to cover costs and sustain their wages. Neither route capitalizes on the unique opportunity for LLLTs to partner with industries and investors outside of law, an opportunity that does not exist for lawyers under the rule of professional conduct that prohibits profit-sharing with nonlawyers. Nor does the LLLT model decrease the cost of legal professionals’ labor to the extent that government offices or nonprofit organizations could substantially expand the human capacity of their legal services absent an increase in funding for additional hires. And, while most of these LLLTs and Candidates decided to earn their license to expand access-to-justice in family law, they still predominantly intend to target clients who can afford to pay their rates—rates lower than attorneys’ fees but not low enough for low-income populations to afford.Footnote 76
The other problem is perhaps even more fundamental—it views the problem to be solved principally as a legal one, requiring first and foremost legal expertise. While the level of legal expertise required by LLLTs is reduced and focused on a specialty, in conception and application the mode seeks to provide reliable legal services. The problem with this conception is that problems, whether for big corporations or those suffering access-to-justice problems, are rarely purely legal. As Rebecca Sandefur and others have observed, consumers rarely, perhaps too rarely, see their problems as being legal in origin,Footnote 77 and even when the legal aspects of their problems are substantial and addressable by legal tribunals, solving the legal aspect of the problem may not put them in a good place. For example, a tenant facing eviction has not had the ultimate problem solved if after having a fair day in court the tenant still has no way to pay the rent.
This problem has some relationship to the developing mismatch between attorney licensing and training and the actual specialized services different kinds of customers demand when attorneys move into practice. Specialization involves not just more granular expertise in legal requirements; it often also requires special expertise in the setting (such as a given industry) or a process (such as a specific kind of transaction, whether multinational mergers and acquisitions or applying for government benefits). Perhaps both the training and the marketing of the services ought to take into account the nonlegal aspects of the services that eventually will need to be provided.
These two lessons drawn from the U.S. experiment, while based on a close look at the U.S. experience, are generalizable to other jurisdictions facing access-to-justice problems. First, even lighter versions of the professional model applied to lawyers may not be sufficient to make legal services truly affordable. Second, it may be time to move beyond the old framing of problems provided by the existing professional models and think of new kinds of service providers. These service providers could be trained for and provide services related to the specific needs of the consumer and not be limited by ancient professional models of how services should be conceived of and delivered.
D. Other Alternatives to Lawyers: Beyond the Professional Services Model
While academic support for LLLT type programs remain strong, the limited adoption of such programs by states and the limited enrollment in those states where it has been offered suggests that the grand hopes that led to such programs have not been fulfilled. The access-to-justice crisis remains critical, even in those states that have adopted legal paraprofessional programs.
The core problem is that LLLTs represent only a modification of the traditional professional model. Rather than creating a new way to deliver legal services, they represent a change in professional boundaries. The path to solutions still runs through a human being who has been extensively—and expensively—trained in expert knowledge and then tested on that knowledge. That individual must then provide a solution based on his or her expertise, with that solution custom-made for each client. While this shift in professional boundaries is a good thing, recognizing that specialization allows new kinds of professional training, it still is tied to the ancient methods of providing knowledge-based services.
These models require the “bespoke” model of meeting consumer needs. Rather than providing off-the-shelf solutions, professional providers must use their expertise to craft—or at least select—a solution suited to the unique needs of each client.Footnote 78 Done right, this provides a high level of service but imposes significant costs.
The LLLT idea arose when human expertise may have seemed the only reliable way offer quality services. The “job to be done” required expert knowledge, and while the contours of the type of knowledge and the resulting professional boundaries were open to debate, the process still centered around human beings equipped with the right kind of training and certification. That has changed. Today technology can and does provide alternate ways to provide services, raising the question as to how best to incorporate technology in meeting consumer needs. The potential for other ways to meet consumer needs can be better understood if the question is framed not in terms of traditional professional credentials and processes, but in terms of the needs consumers need to have met.
I. Viewing Legal Services in Terms of the “Job to be Done”
Clients generally hire lawyers not to opine on the law, but to solve a problem that to some degree requires the use of legal expertise. The job to be done is the solution to the problem, not the acquiring or sharing of legal knowledge.
The late Clayton Christensen illustrated the idea of the real needs of consumers with a story about drills. As the story went, employees at an electric-tool company were asked what their customers wanted. The employees were quick to reply, “[e]lectric drills.” The answer, instead, was holes.Footnote 79 The product was just a means to an end.
The same applies to lawyers and legal services.Footnote 80 As lawyers become more specialized, the nature of the problems they are asked to solve can shift, and the expertise they will need to solve those problems also will shift. The client is less interested in what kind of training a service provider has had than whether that service provider can solve their problem.
Both the licensing of lawyers and the system of legal education have remained largely unresponsive, however, to the shift to specialization. As noted above, law schools still provide the same generalized education that Christopher Columbus Langdell developed in 19th-century Massachusetts to prepare students for a life of general practice in small firms, and the bar exam still tests across a range of topics, but not too deeply in any of them.
As Richard and Daniel Susskind have argued, technology has advanced to the point where modern information technology offers an alternative way to get the job done.Footnote 81 Tasks that once required specialized expertise can now be done as well, or better, by technology. In the field of legal services, artificial intelligence is already offering solutions, good and bad, to the problems faced by consumers.
1. Online Expert Services
While alternative legal service professionals such as LLLTs were struggling to gain a foothold in the market, a different alternative to traditional lawyering was gaining a strong presence because of its ability to do the “job to be done.” At the about the same time that LLLTs and other paraprofessionals began their path to regulatory approval, online artificial-intelligence-powered legal services such as LegalZoom and Rocket Lawyer began to gain a foothold. These services aimed to fit into an exception to UPL that allowed the sale of legal forms. The kind of AI used in these services allowed sophisticated forms to be created, however, that integrated the information provided by the customer with laws relating to such potentially complex legal documents as wills and corporate entity formation.
While many UPL challenges were brought against these companies, they were generally able to defeat them by carefully staying with the safe harbor of form preparation. This may well have caused the vendors to dumb down the services they offered, but the avoidance of legal controversy allowed them to build large and growing businesses. At present, these online legal-service providers have claimed between five and ten percent of the US legal-services market, taking in a share that far outstrips the share taken by human alternatives to lawyers such as LLLTs. A substantial proportion of new wills created in the United States—perhaps as much as 20 percent—are created by these companies, and they have made similarly impressive inroads in the lucrative niche of corporation creation, taking five to ten percent of that market.Footnote 82
This seems to have come without grave injury to consumers. While LegalZoom has a mandatory arbitration clause in its terms of use, which makes getting statistics on claimed malpractice difficult, one court facing a challenge to LegalZoom’s services did look at the quality of its services. It found little cause for concern:
The class action was to defend consumers, but there was little evidence of consumer injury. The court noted that “the core of the claims that have been advanced have not suggested general incompetence or negligence by the [LegalZoom] software and the authors of the [LegalZoom] documents uploaded as software, but rather essentially [they are] these technical statutory violations.” Reports from class members showed “entire consumer satisfaction” with everything from LegalZoom. “Nobody has shown a record of lots of people discovering that bankruptcy courts are rejecting their [LegalZoom] filings or that people find themselves intestate [due to faulty LegalZoom wills] or secretaries of state and departments of corporations are rejecting [LegalZoom] corporation papers.” To the contrary, most consumers seemed happy with LegalZoom.Footnote 83
While options such as LegalZoom illustrate that humans are not the only path to sufficiently reliable legal services, they are far from the potential state of the art. They rely on what Richard Susskind has termed “good old-fashioned AI,” which relies on laboriously coded expert systems.Footnote 84 Highly articulated logic trees allow forms to be created in response to consumer choices, but the creation of the logic trees requires much work and maintenance to remain current. As a result, the online services tend to offer solutions in narrowly defined areas such as creating wills, contracts, or corporate forms where the solution can be predefined.
LegalZoom is not just a U.S. product. It currently operates in the U.K., and its competitor Rocket Lawyer operates in Spain. In addition, localized products relying on similar technology are springing up around the globe, such as Rechtsdocumente.de in Germany, offering compliant documents based on local law and in the local language. Just as in the U.S., it is inevitable that such products will fill a space in the legal services market.
2. Generative AI
The arrival on the scene of powerful generative AI tools such as ChatGPT has offered another artificial intelligence approach to answering legal questions. While much has been made of the very real limitations of this technology, including the potential for hallucinations, vast sums have been invested by legal tech companies in harnessing its power for use by law firms and corporate law departments. Despite recognized potential for AI and other tech to increase access-to-justice,Footnote 85 less has been invested for its use by the public or public service lawyers, in part due to the continuing risk of UPL enforcement if solutions are marketed to the public and in part due to the relatively limited budgets of public interest legal service providers.Footnote 86
In light of UPL restrictions, purpose-built generative AI for legal advice is usually not offered to the public in the U.S. As a result, those seeking AI help for legal services must turn—and can turn—to general-purpose generative AI tools such as Claude or ChatGPT. There is little question that consumers are doing just that. Given that generative AI is set up to give an answer to any question posed to it, preventing such AI from giving answers that would constitute UPL if given by a human or a dedicated—and likely more accurate program—will be hard to prevent.
3. Risks with AI
Generative AI tools, without the special training and other protections that purpose-built solutions get, are not well suited for legal use. The answers can be unreliable and inputting information into the interface can lead to a loss of confidentiality. For some consumers, however, the lack of cost and the easy availability offset those drawbacks.
In the past, the greatest risk to the public, in an era where only humans could organize and apply expert knowledge, was other humans who either did not have sufficient expert knowledge or who might act unethically. The rise of AI changes that. The greatest risk to the public today is not incompetent humans, but incompetent AI. In a world where free versions of AI can be accessed instantly, those unable to access professional help will—and undoubtedly already have—use AI as the solution.Footnote 87 Because the AI is marketed as a general-purpose tool, preventing consumers from asking legal-advice questions will prove difficult if not impossible.
Driving access-to-justice issues to general purpose generative AI is a perilous path. While the powers of properly trained and applied AI are stunning, generative AI is far from a miracle solution. Aptly described as “mansplaining as service,” generative AI can generate plausible sounding but dangerously inaccurate answers, all without regard to the actual circumstances and needs of those receiving the advice.
Generative AI as it exists today performs well at certain tasks but poorly at others. The Vals Legal AI ReportFootnote 88 compared the work product of several legal-specific AI products against the work product of a panel of “average lawyers” on tasks that were both within the scope of what the software products claimed to be able to do and within the scope of what lawyers do. In several areas of work, some or all of the AI products performed as well or better than the lawyers. For example, when charged with document extraction, document Q&A, summarization, and transcript analysis, the AI products were both faster—at least six times faster and up to eighty times faster—and better in the work product they created.
In other areas, the AI products struggled. AI tools fell short in tasks that required deeper legal reasoning, nuanced judgment, or contextual interpretation—such as redlining complex documents and advanced legal research. The AI products were also deficient in areas where ethical concerns or other issues requiring nuanced judgment were important.
The results of the Vals study were consistent with other recent studies looking at generative AI performance in legal settings. A recent Stanford studyFootnote 89 looked at the incidence of hallucinations in state-of-theart flagship-level legal AI products. The study found that compared to general products such as ChatGPT the incidence of hallucinations was significantly reduced but not eliminated, with hallucinations still occurring between seventeen and thirty-three percent of the time. The study also found risks of sycophancy—that is, that the AI would agree with false premises included in the prompt.
These results coincide with the results of a Cambridge University survey of legal professionals who have used AI.Footnote 90 That survey found that legal AI users experienced a level of hallucinations that reduced their trust in the products, and also that the AI products tended towards general responses that were not responsive to the specific needs of the client.
While generative AI seems likely to improve significantly over time, the degree to which these legal products can eliminate these problems remains unknown. The products surveyed in the Vals study, for example, were specially adapted later-generation products reflecting massive investments in optimizing the technology for legal use, and in some settings serious problems still remained even when used by skilled practitioners. The problems are likely to be much greater when non-optimized general AI products are used by lay people without experience or skill in framing prompts.
Legal services, especially in a balkanized set of jurisdictions such as states in the USA, present an especially ripe environment for AI errors. AI as it exists today is not well-equipped to recognize critical changes in legal doctrine or procedural requirements as one crosses not just state borders but court-district borders. Generative AI also cannot be counted on to apply up-to-date doctrine reflecting recent changes in the law. One can add to that the familiar litany of problems with AI, including the release of shared information into the wild, bias, and so on.Footnote 91
In order to prevent harm to the public, it no longer serves to posit a choice between lawyers and lesser-trained or untrained humans. The contrast is between humans and AI. A system truly concerned with the public welfare should seek to provide affordable alternatives to poorly trained AI solutions.
4. The Remaining Access-to-Justice Gap
Neither nonlawyer professionals such as LLLTs nor artificial intelligence tools such as LegalZoom have solved the access-to-justice crisis. As noted, the numbers of nonlawyer professionals have been too small and their compliance with the bespoke professional model limits their ability to provide truly cost effective services.
AI also fails to meet the market needs. The form-creation tools such as LegalZoom can only produce those solutions that consumers think they need, because analyzing the problem and proposing a different solution would trigger UPL sanctions. If, as is possible, a customer chooses an inappropriate form, that form will be competently created but may not in the end meet the needs of the job to be done. Generative AI tools present even greater risks for the unwary, because the general AI tools accessible to the public are not trained on appropriately restricted legal databases, are not able to cope with geographic differences or very new laws, and may include incorrect responses ranging from answers that would be accurate in a different jurisdiction to outright hallucinations.
AI also fails because there are advantages that come from familiarity with a process or a forum, which will be advantages that lay users empowered by AI will not have. The lack of experience and familiarity will make them more susceptible to accepting errors from AI and less likely to select the correct form to do the job to be done. In this regard, research shows that nonlawyer advocates experienced in a specific forum delivered high-quality representation not notably inferior to that provided by lawyers.Footnote 92 If experience matters for those with extensive specialized training, the lack of it surely matters to lay people negotiating an unfamiliar environment.
The question arises whether there is another way to meet the “job to be done,” perhaps one that combines the strengths of the task-specific paraprofessional model with the strengths of AI.
E. AI Paraprofessional Fusion
With neither LLLTs nor AI likely to be standalone solutions to the A2J crisis, the question arises as to whether there are other ways to do the job that needs to be done. The fundamental problem with bespoke professional models is that the training, even for lighter models such as LLLTs, builds a level of cost into the system that is aggravated by the customized model of delivering services. The core problem with AI in its various forms is that lay people unfamiliar with the processes are subject to accepting defective solutions.
Combining the two approaches, however, could provide a new path that draws on the advantages of each. Lightly trained humans can provide the familiarity with the system and processes that lay people lack. Rather than relying on extensive and expensive education to allow them to provide expertise-based solutions based on the ancient model of professions, artificial intelligence can be applied to provide work product such as correctly drawn documents to be used in the processes.
Technology, of course, is being increasingly used both by lawyers and by legal services organizations.Footnote 93 Technology has been used as a force multiplier in traditional legal service organizations, and with or without regulatory reform that is likely to continue.
To take full advantage of AI in legal services, however, will require a regulatory change. In the US, UPL laws are likely to block both humans relying on AI and fit-for-purpose specialized AI from giving answers directly to consumers. To break free from the costs and burdens of the professional model will take a new regulatory regime. In some other countries, providing advice outside of a court setting may not require regulatory change, while in many, if not most, assisting in judicial processes may require regulatory change similar to the U.S.
I. Building a New Service Model Reliant on AI
The opportunity is redefining what kinds of service providers should be allowed now that powerful technologies can shoulder some of the load, and not just how technology can be used. To what extent can we move away from the training and bespoke service requirements of the professional model in order to meet consumer needs now that technology exists that can do what previously only humans could do?
Offering these optimized new services will require a break from current UPL restrictions. The best option combines the strengths of AI—the ability to create complex and formally correct documents in somewhat defined settings—with the strengths of humans—the ability to provide empathetic service as well as the ability to provide domain specific experience that increases the chances that the right document for the job to be one is the one created. The solution is lightly trained paraprofessionals, more in the model of court navigators than lawyers, aided by software.
Examples of how paraprofessionals aided by technology might work already exist. One example exists in the law-adjacent field of tax-return preparation. Tax advice, of course, has been a field of boundary disputes between lawyers, accountants, and bookkeepers for decades. While planning complex international corporate tax strategies remains a field for those with deep and specialized knowledge, the more common and routine task of creating and filing personal and small-company tax returns can be more commoditized. In that market, services such as H&R Block combine relatively lightly trained humans with tax-preparation software. Aided by software, relatively lightly trained return-preparation specialists gather information and prepare returns, the accuracy of which matters at a level where both substantial civil and criminal penalties can come into play.
It is instructive to compare the training required to become a tax preparer working with H&R Block with the training required to become an LLLT or similar legal paraprofessional. While the LLLT credential requires thousands of hours of supervised work and multiple courses, the training to become a tax preparer can take as little as sixty classroom hours, which can be taken online.Footnote 94
Medicine provides another good example of how tech-enabled solutions fit into the delivery of services that are customized to an individual’s needs. While a doctor or other medical professional might direct a patient’s care, both technology and commoditized products play roles throughout. In determining, for example, whether an infection is viral or bacterial, a doctor might rely on a blood test with the blood drawn by a phlebotomist and the sample run through a machine. If the test shows a bacterial infection, there is likely to be a standard course of treatment involving pharmaceuticals that are manufactured in a factory. Whereas a hundred years ago a doctor might have poked, prodded, smelt, and even tasted the patient to make a diagnosis, and perhaps concocted an herbal treatment from scratch, today that treatment is likely to rely on technology and commoditized solutions at both the front and back ends.
That tech and those commoditized products, are, of course, part of what has made it possible to shift much medical treatment away from doctors and into the hands of paraprofessionals such as nurse practitioners or to storefront clinics equipped with technology but no lawyers. Clinics such as Minute Clinic provide services directly to the public without having physicians on premises. While paraprofessionals play a role, the secret sauce for Minute Clinic is technology. Diseases that once required a trained doctor in order to make a diagnosis are now quickly and more reliably diagnosed through technology. Once the diagnosis is made, paraprofessionals and pharmacists can follow up with the accepted best-practice treatment options.Footnote 95
While there has until recently been no legal-sector equivalent to feeding blood or urine samples into a machine that generates a reliable diagnosis of a given infection, allowing the application of the standard best-practice initial course of treatment, the use of AI can bring comparable benefits to the legal field. Doing this correctly requires a focus on what AI today and in the near future does well and what AI now and in the near future does less reliably.
II. AI and Human Fusion
The recent and ongoing advances in technology allow an opportunity to think about how delivery of legal services can be rethought to bring both technology and human strengths to bear. While the LLLT experiment shows that lawyers are not the only possible solution to legal problems, the universe of possibilities is much broader.
1. What Can the Human Do
One thing humans can do better than AI is help define the nature of the consumer’s problem. As anyone who has practiced law realizes, clients often are wrong about what their problem is, both with regard to the legal aspects and the non-legal aspects. While AI can help efficiently collect and organize facts provided by the client,Footnote 96 a human being can tease out facts and goals that might seem unimportant to the client, helping the client to better understand the job to be done. Humans with sector expertise in a given setting—say, family law or housing disputes—will know from experience likely factual and legal issues that have not yet become apparent to the client. For example, those seeking bankruptcy protection may be poor candidates to get effective relief from bankruptcy, but could benefit from targeted financial counseling. AI on its own is far too likely to offer a solution that may or may not fit the underlying situation.Footnote 97
Paraprofessionals can also be trained to offer help in the aspects of the problem that are not just about the law. Legal problems rarely arise in a vacuum, and the job to be done often will require advice beyond just legal information. Pilot projects suggest that nonlegal help can be as meaningful as legal counsel.Footnote 98
Humans can also offer emotional support through what for many is a trying and challenging process. One problem with, and one barrier to, the justice system as it exists today is that it can be intimidating and confusing. These barriers can present as great a barrier as the financial costs, a barrier that is not overcome by hard-to-decipher-and-apply self-help materials, even when that self-help is delivered online or through an AI-enabled interface.Footnote 99 A human being can help to decode the situation so the layperson feels more in control.
Humans can also educate clients about the realities and likely course of the legal process involved. Far better than AI, which tends to provide answers without regard to the prior knowledge or understanding of the recipient, a human can assess what the client understands and educate them appropriately.
Humans can also recognize when client needs exceed the reach of what the paraprofessional aided by AI can be expected to deliver. There are cases presenting novel or advanced legal issues, or permitting aggregate action, that will be beyond the scope of what the paraprofessional solution is intended to cover. A human can escalate the client to a provider better equipped to handle the complex case.
Multiple studies have shown that nonlawyers, even when not aided by technology, can provide services equal to or superior to lawyers in certain circumstances. As Rebecca L. Sandefur has observed: “An established body of research on nonlawyer practice demonstrates that people without law licenses or law degrees can provide competent, effective services when they are properly trained, when they specialize in specific areas, and when the matters at stake do not raise complex issues of law.”Footnote 100
It can reasonably be expected that nonlawyers aided with purpose-built technology will be at least as capable as those without that additional capability.
While training will be required to assure that the services provided are in fact helpful and sufficient, the nature of the job to be done counsels against offering, as the LLLT experiment has done, a lightened version of the professional model of lawyers. First of all, the training should be fitted to the problem and should not just be legal training. Second, much of the training should be in the correct use of the AI tools provided so that documents provided with the help of AI are properly constructed and properly checked. Last but not least, those in these roles should be trained to escalate situations to higher-level service providers when the situation has a level of complexity beyond what they customarily face. A great many problems that are baffling to laypersons are routine and easily handled by those with sector-specific experience. Those can be handled by lightly trained people aided with technology. When the situation requires novel legal arguments or other efforts beyond the normal scope of processing a standard situation, the training should include awareness of escalation options and a clear directive to refer the problem along.
2. What AI Can Do
One thing AI does well is create documents. In general, generative AI today already does a pretty good job of creating text that is fit for purpose,Footnote 101 and creating the specialized texts required by the legal system is one of the most substantial barriers to accessing the legal system. One example is that AI can pull information from unorganized documents and put those in textual forms suitable for filing with a court. In China, in fact, publicly funded AI systems are used by courts to help litigants get proper documents before the court. An example is the Smart Trial Project in Hangzhou. Litigants upload documents that bear on their claim and then are guided by the software in putting the information in a form ready for the court.Footnote 102 Singapore offers a somewhat similar system, powered by the AI company Harvey, that helps certain types of civil litigants prepare documents in the correct format for filing.Footnote 103 In the USA, a non-profit called Upsolve helps pro se bankruptcy litigants get their documents in proper format.Footnote 104
Such applications demonstrate properly configured and trained AI could help prepare documents needed by litigants in the correct format, reducing the training needs for paraprofessionals while assuring accuracy and conformity with requirements. AI can also be used to take background factual information and put it in a format that can be used by an advocate to present a client’s case. Again, AI is doing exactly this in systems designed for use by lawyers.
One critical advantage of AI, as demonstrated by systems such as LegalZoom, is that updates can propagate across the system almost instantly when the law or circumstances change. LegalZoom and other online services can respond promptly to changes in law, and as a result assure customers that the forms they create are compliant with current requirements. AI incorporated into fused AI and paraprofessional services could do the same.
AI can also do a good job of providing reasonably accurate answers to repeated or stock legal questions, especially when those answers are checked by humans before being released to the public. In England and Wales, for example, the Citizens Advice network uses chatbots to provide legal answers to its nonlawyer volunteers interacting with the public. These answers are checked by supervisors before the answers are given to the public, but the task of researching and drafting has been shifted to AI. The results have been useful enough to justify expansion of a pilot project.Footnote 105
To date, AI is not reliable for advanced legal research where the answers are unclear. Aside from the well known problem of hallucinations, AI may not apply real cases properly, citing real cases for legal propositions that the cases do not in fact support. However, in the world of underserved populations, the law often is clear and the function is developing and arguing the facts. The issue is not so much discovering the rule but communicating it clearly to the affected parties and applying it to the facts. To the extent that cutting-edge legal issues are involved, neither AI nor paraprofessionals are likely to be the best solution, and escalation to a lawyer is likely to be in order, if available. Recognizing situations that need escalation is something lightly-trained humans can help with.
3. The Solution in Operation
To imagine how this solution might operate, consider a common access-to-justice setting, landlord-tenant disputes. At present, tenants are cast into a system that expects professional representation and have difficulty presenting their cases.
Imagine instead a system where, like in Singapore, software helps a potential litigant organize the necessary documents and facts and put them into a form suitable for processing by the tribunal. This software can make certain that the standard operative issues are covered.
After the intake the litigant can be paired with a human with light training more similar to court navigators or tax-preparation specialists, but with familiarity with the tribunal involved. This human can confirm that the forms are in order and the litigant’s apparent needs track to the job to be done that can be offered by going before the tribunal. The human can help the litigant explore other options that might also solve the problem.
The human may have knowledge that AI does not have. AI is dependent on data, and despite the trend toward datafication of daily life there are still areas of life where the data trail is missing or too skimpy to be reliable. A human familiar with the tribunal might know, for example, where a given decisionmaker seems influenced by whether the litigant dresses up or maintains a particular affect in court. These are things that often matter to practicing attorneys, but that are not captured in the data created in the courtroom.
Whether the human should be allowed to participate in any court hearing is open to debate but is likely to be a good idea. The research of Herbert Kritzer, among others, suggests that attorneys are prone to exaggerate the benefits of having an attorney rather than just someone with familiarity with the setting. That could change in the event formal rules of evidence are scrupulously followed, but in the kinds of asymmetric settings where unrepresented litigants end up, the formal processes of more-elite courts are sometimes lacking.Footnote 106
Generative AI relies on incremental and continuous improvements, and both the human and the litigant can provide feedback as to how well the job to be done was handled. As the technology improves and the experience curve is surmounted, the results are more likely to improve than not. An evidence-based approach to regulation can take into account the successes and failures of such an experiment, and modify the regulatory landscape in light of evidence and experience rather than in light of often overblown comparisons to what a perfect attorney could do.
III. Why Public Funding
If AI systems are developed that can fulfill these functions, the question arises as to how these systems should be developed and funded. In the U.S. the costs of advocates in civil settings have normally been borne by the parties. Each side pays its own lawyers, win or lose. While there is some legal aid, on the whole the market governs. In the U.S., the development of legal AI has similarly been a private-enterprise affair. AI has been developed largely by for-profit companies who expect to reap substantial fees in the future. Who pays for what has different answers by country and jurisdiction, but the issue of who should develop and pay for the technology underlying AI-assisted paraprofessionals is generalizable across borders and regulatory systems.
While the “litigant pays” approach might create an expectation that AI for paraprofessionals should be privately funded, that conclusion should not be reached without consideration. Not all costs of the legal system are borne by the litigants. If a renter is summoned to court on an eviction action, the state pays for the costs of maintaining a courthouse. By the same token, the state pays for the judge who will decide the case. Justice is a public good. The delivery of justice is a public good, and in some cases but not all the public pays the costs.
If we envision a world in which reliable AI is used to make services provided by lightly trained helpers sufficiently reliable, the question arises which category the development and maintenance of that AI falls into. Is it like the courthouse, or is it like the briefcase one’s lawyer carries to the courthouse? How should AI-powered services be funded and delivered?
In the United States, the default position has been to shift all but the most traditional costs of the justice system to the private sector. For example, access to foundational legal texts has been privatized. In a world in which all citizens are presumed to know the law, functional access to that law often requires penetrating a paywall. To obtain reliable analysis of those foundational texts, presented at a level a layperson might understand, almost always requires access to relatively highly priced secondary materials.
As we move into the AI age, there are compelling arguments that AI to assist with legal questions should be provided as a public service. As this issue is addressed, it should be taken into account that AI will be used by the public to decode law and that the AI currently most accessible to the public is poorly suited to that purpose. Public-facing AI is subject to hallucinations and bias, and is not set up to protect the confidentiality of those trying to get help.Footnote 107
Commercial generative-AI products available—at a cost—to law firms and legal departments suggest that sufficiently reliable AI for certain defined purposes is, if not today’s reality, certainly on the near-term horizon. Products such as LegalZoom demonstrate rather more conclusively that AI directed at confined tasks is sufficiently reliable today, so long as the user knows enough to select the right AI for the task at hand.
As that AI arrives and its deployment becomes a real possibility, sorting out how it should be delivered becomes a more urgent concern. There are at least three theoretical options: for-profit solutions, of which the LLLT programs are an example; dedicated non-profits offering access-to-justice solutions directly using AI products such as Upsolve; and government-developed AI products.
For-profit options may be one solution if they can avoid UPL challenges and meet needs. The substantial market success of online document-creation services such as LegalZoom illustrate that in some niches the market can deliver lower-cost solutions.
Nonprofits can also play a role. A good example where this has been done successfully in the US is Upsolve, which promotes its free product as the Turbotax for bankruptcy. Upsolve, funded largely by philanthropy, has developed an AI product to help members of the public navigate the process of filing personal bankruptcies. As part of its no-charge service, humans give advice to the consumers in conjunction with the software. On not infrequent occasions, humans are guided by the software and the volunteers to solutions other than declaring bankruptcy. In some cases Upsolve encourages those contacting it to seek services from a lawyer instead. Similarly, in the United Kingdom, the Citizens Advice network provides services from lay volunteers that under that regulatory environment can address legal as well as nonlegal issues.
Upsolve faced a likely UPL challenge from the New York Bar, arguing that Upsolve and its human volunteers were illegally providing legal services. Upsolve went to federal court, seeking an injunction that would block UPL enforcement against its services. Upsolve won in the district court, in part because the scope of the services offered did not involve going to court but only involved giving personal opinions, an area that merits First Amendment protection, but the opinion was later vacated and remanded by the court of appeals when if found the trial court applied the wrong level of scrutiny, a result that does not bode well for Upsolve.Footnote 108
The Upsolve project illustrates that nonprofit organizations that fund and distribute AI guidance, in conjunction with such human assistance as seems appropriate, is a possible solution if allowed by regulators. It also demonstrates that UPL challenges are a potential obstacle. Upsolve may yet lose its challenge as the case continues in the courts. Even to the extent it won a temporary victory, that was premised on Upsolve providing a very limited level of services.
For a variety of reasons, however, the best solution would be for courts and governmental systems to accept the responsibility of making access-to-justice an achievable goal, and as part of that to take ownership of AI-and-human hybrid solutions to make that happen.
One problem with AI tools for law is that the legal requirements can change as political boundaries—sometimes hyperlocal boundaries such as neighborhood zoning districts—change. To date, generative AI has not been reliable in recognizing all those boundary issues.
Sometimes local variations reflect legitimate local preferences and needs. For example, zoning requirements in midtown Manhattan are unlikely to translate well to rural Wyoming. Similarly, tenant protections such as rent control in one polity may not be seen as desirable as another.
That said, sometimes local variations are just legacies of balkanized jurisdictions or reflect historical practices that make little sense today. Businesses standardize processes across a range of locations in order to make their internal processes more efficient and to make access to them simpler for consumers. There are situations where governments could improve access-to-justice, and make the adoption and maintenance of purpose-fit AI easier and cheaper, by emulating that approach. A good example is the confusingly different requirements by jurisdiction related to filing court papers, a problem being addressed by the Stanford Filing Fairness Project.Footnote 109 Ownership of the AI tools used to decode legal requirements will provide an incentive to standardize those requirements.
There are other advantages. Owning the tools used to decode legal requirements could encourage jurisdictions to embrace better use of legal design principles. Today’s legal system is designed by and for lawyers, with impenetrability by those without esoteric knowledge sometimes viewed as a feature rather than defect. Accepting a responsibility to make the legal system accessible to AI products could help move the systems toward following better design principles.
Publicly-owned AI can also be better audited for quality. To the extent that AI is owned by private actors, regulating and guaranteeing the quality of that AI becomes a complex task, one that lacks a structure to date. If a category of service providers is built around the capabilities of AI, the quality of that AI may be subject to some kind of regulation.Footnote 110 If the AI is offered as a public service the auditing and quality control is internalized.
Importantly, public funding of the AI tools designed to support a quasi-professional occupation increases the odds that these tools will be supported and maintained. Many innovative tools have been developed at hackathons and by social entrepreneurs. These serve well as proof of concept, but less well as tools that individuals can plan to build a career around. More than a few have arrived with much promise and seem to have quietly faded away.Footnote 111 Dedicated state support can increase the likelihood that the software will be maintained and further developed in a way that allows careers to be built around it.
Finally, the core problem in legal services is government regulations that bar nonlawyers, even nonlawyers with sufficient skills to do the job that needs to be done, from the marketplace. Because government has created the problem through enactment of UPL rules in the supposed support of making sure the public receives qualified service, it makes sense to put on the government the responsibility and cost of making available the new technologies that will protect the public from new forms of inadequate services.
F. Conclusion
The access-to-justice crisis remains urgent in the U.S. While paraprofessionals were embraced as an exception to broad UPL regulations that protect the lawyer monopoly, the slow pace and limited scope of their adoption suggest that this lighter professional model inevitably is handicapped by requiring a variation on the professional model as the alternative. In the paraprofessional model, highly trained humans possessed of special knowledge—albeit a somewhat different and in some ways more limited set of knowledge—still struggle to justify the training investment and must follow bespoke methods of personal service delivery. The rise of artificial intelligence suggests a completely different way of getting the job done, but AI has its own defects. Merging lightly trained humans with powerful and purpose-built AI tools could provide a scalable solution that would allow more of those without access to lawyers a solution that is better than AI alone and better than self-representation, the two realistic alternatives today.
Acknowledgements
Thanks are due to my research assistants for their unflagging work and to the participants at The International Workshop on AI Law held at STL in September 2024 for their thoughtful comments on a presentation which led to this Article.
Competing Interests
The author declares none.
Funding Statement
No specific funding has been declared in relation to this Article.
Open access
