. . . [T]here are many human activities that cannot be formally described. It is those aspects of human behavior that computers cannot be programmed to simulate.
That view is supported by a new study, “Can Robots Be Lawyers?”, a draft of which was posted last week on the Social Science Research Network by Dana Remus, a professor at the University of North Carolina School of Law, and Frank S. Levy, an M.I.T. labor economist. In the study, they explored which aspects of a lawyer’s job could be automated.T he research suggested that, for now, even the most advanced A.I. technology would at best make only modest inroads into the legal profession. Based on their analysis of actual billed hours, the researchers examined the work that lawyers do in broad general categories. They then analyzed how much of each category might be displaced by existing A.I. and automation technologies.
As it turns out, being a lawyer involves performing a range of tasks, from reading and analyzing documents, to counseling, appearing in court and persuading juries. Indeed, reading documents accounts for a relatively modest portion of a lawyer’s activities.
The researchers noted that many of the tasks that lawyers perform fall well within what Polanyi defined as human behavior that cannot be easily codified. “When a task is less structured, as many tasks are,” the researchers wrote, “it will often be impossible to anticipate all possible contingencies.”
The paper itself, after careful analysis of the tasks lawyers perform, and the kinds of skills they require, as well as the extent to which they can reasonably be automated, comes to a less dramatic conclusion about the impact of technology than many have embraced, that here would only be a 13% reduction of legal work if there were full use of technologies.
Key is an astonishing analysis of a huge database of hours billed by law firms in a billing companies database (pp. 33-34), leading to a table showing the percentages of time billed by activity, with the activities divided into categories of extent of likely employment impact of technology.
The conclusion (reporting only tier 1 first here, although all tiers give very similar results) is that only 4.1% of work is subject to “strong employment effects” of technology (document review), with 39.7% subject to moderate effects and 56% to light potential effects (such as court appearances) (p. 35). Interestingly, the percentages of time do not change much with length of tenure at the firm, with the exception that document review is very sensitive to tenure, varying from 8.5% for first two years to 1.1% for partners (p. 39). (A side implication may be that labor is not nearly divided enough in the firms, or maybe that the routine work is already all handled by paralegals and computers.)
The paper, by analogizing to data from other fields, then estimates the likely savings of time in these areas from full technology deployment (pp. 39-43). Areas with light employment effects are analogized to the medical context, specifically impact on clinician productivity of electronic medical records (p. 43). Those with medium level effects are analogized to exceptions processing at a bank (p. 42). The one with high effects (document review) concedes a high 85% impact (p. 41). The bottom line is then the net 13% reduction in lawyer employment, far less than many of the more dramatic articles cited in the paper appear to posit. Applying the assumption that these technologies will take 5 years to deploy fully, the paper goes on to calculate an overall low impact at any point (p. 48).
The final part of the paper discusses the impact of technology upon the legal ethics and deregulation issue, pointing out that replacing the legal analysis activities that are being projected to be replaced by prediction could be both dangerous and less productive than it might seem.
More specifically, the paper asks what will be lost if software replaces lawyers in estate planning and tax. It answers: counseling, robust understanding, respect for clients’ interests, access to reasons, and interaction with the legal system (pp. 64-65).
There is an interesting and complex relationship between this list, and some of access innovators’ thinking about how to define the practice of law, with my own analysis focusing on certainty versus exercise of judgment. Many of these things come from substantive expertise, with many of us believing that the needed skill may sometimes be better found in one without a law degree, as summarized by Lippman, then CJ, at the White House, “Sometimes an expert non-lawyer is better than a lawyer non-expert.”
Obviously this paper has very large implications for our access to justice strategy, which has long included as a critical component the use of technology to increase the efficiency of access to justice services. See. e,g, the recent resolution by the Conference of Chief Justices on 100% access, which explicitly lists technology and many other innovations, themselves leveraged by technology.
Some thoughts on the access distinction:
- As a general matter, much of the legal aid access to justice caseload is routine, making much more of the work potentially automate-able. Indeed, the huge numbers of those served by LSC grantees through technology shows the extent to which this already occurring (almost 18 million in 2014, LSC Fact book at 29. Obviously this does not apply to class actions or complex cases.
2.Access to justice innovation, largely unhindered by fears of reducing billable hours, has placed a large emphasis on self-help services and assisted pro se, with technology playing a huge part in this.
3.We have also moved to make much better use of nonlawyers, including not necessarily supervised by lawyers. In this context technology may be the most useful, because it allows the human context to be provided by less expensively trained and paid people, with some of the expertise internalized in the support software.
4.We have promoted discrete task representation which again shifts as much of the work as possible to the client, with the lawyer doing only what the client wants (and can pay for). This again can be done far more easily and safely when the client can use software for things like preliminary document drafting, data gathering, and even outlining of oral presentation.
Perhaps ultimately even more significantly, with initial impact on the access the justice end, but ultimately on the whole legal system, is the potential of technology to change how the dispute resolution and compliance systems themselves work, thus reducing the total need for lawyers.
In the long term, the real savings (and such savings should be seen as positive, rather than negative) will come from changes in the underlying ways in which relationships are formed and in the ways that disputes are resolved when those relationships break down. An analogy might be from the time when it seemed as if horse driven carts would be replaced by engine driven carts. There were surely plenty of theories as to why this would not happen (horses are more intelligent, they navigate varied pathways better, they are more reliable, etc), but the building of high quality roads and other changes made these distinctions irrelevant.
Thus the technical capacities discussed in the paper will not only assist in the specific tasks that the law currently requires, but they are likely to mean that dispute resolution will take fundamentally different forms. We are beginning to see hints of this with moves to simplify court procedures, reduce the numbers of forms and hearings, involve the courts more in the enforcement of the judgements, something that it now left almost entirely to the litigants and lawyers.
To take the enforcement example, all kinds of effort is now required to get money out of the losing party. Linking of banks, government, business databases, could make that largely automatic. Similarly, much of the discovery process could be automated, in the sense that any electronic information required to be provided to the court could be gathered automatically.
It may well be that it will be easier to pilot these changes in low income areas of practice, in which most people cannot afford lawyers, both because there will be less vested interest resistance, and because even the complex tasks may be simpler.
Indeed, this approach is highlighted as follows at the end of the paper:
We have also argued that the existing literature focuses too narrowly on employment impacts, ignoring an important set of broader questions. The broader inquiry starts with the ways in which computers approach particular tasks differently than humans, and then asks how those differences may change legal practice and through it, the law itself. These questions are critical to a meaningful normative and regulatory approach to new technologies, and will only become more pressing as legal technologies continue to advance.