The Future of Law and “Intelligent” Technologies: Prophecies, Technologies and Opportunities – Part 2

In the first part of this blog post, we looked at the current benefits we are enjoying from technologies resulting from AI research. We also examined some risks accruing when AI approaches are deployed in legal activities where transparency and justifications are required. In the following lines, we will borrow from a recent study made of the impacts of AI on lawyers employment. We will also try to enumerate potential benefits of AI technologies in our own line of business, legal publishing.

In “Can Robots be Lawyers?” (forthcoming in the Georgetown Journal of Legal Ethics, (Spring 2017), currently available on SSRN), Remus and Levy examined the various facets of lawyering and the potential effects of new AI technologies on each of them. In their paper, profs. Remus and Levy try to assess the argument that automation will soon replace much of the work currently done by lawyers. To analyze the effect of technologies on lawyer work, they use a large data set compiling the distribution of lawyers’ time as reflected in the billing they send to their clients. The time use appearing in this data is arranged by Remus and Levy in thirteen larger categories of tasks, such as Document Management, Legal Writing and Advising Clients, and each category is associated with a percentage of a lawyer’s time dedicated to it. Then, the potential impact of current and planned AI systems is assessed for each category of task. Especially pertinent to what I want to do here, they discuss relevant technologies to estimate their automation potential in the various categories of tasks.

They see a strong automation impact on tasks related to “Document Review”. Predictive coding technologies are transforming the way that discovery work is conducted. They estimate as moderate the effect of automation on “Case Administration” (contract review systems), “Document Drafting” (document assembly systems for internal use, or offered directly to consumers à la LegalZoom), “Due Diligence” (document review systems), “Legal Analysis Strategy” (prediction software) and “Legal Research” (more on this later). Finally, they deem light the impact of automation on all other categories of tasks, taking up, all together, over 55% of lawyers’ the time. They conclude that technology is indeed displacing lawyers, but at a modest pace.

Summarized that way, Remus and Levy’s conclusion is quite anticlimactic for those anticipating a tsunami of technologies and disturbances. There are two reasons for this envisioned moderate rhythm of change. First, they note that a lot of the technologies automating routine work are already in place, and, second, they are mainly interested is estimating the net result of automation on lawyers’ employment in the years to come. For instance, in “Document Review”, a lot of time is to be saved in reviewing by using predictive coding, but some of the gains must be reinvested in classifying an appropriate sample and in training the system. So, if the impacts they list are summarized in the way I have done it, it does not directly reveal how much the technologies change the work of lawyers. Their detailed examination also let see that, on one hand, changes will be especially important in routine work, but, on the other hand, much more limited for complex legal tasks because of the machine learning-based approaches difficulties in processing situations outside the training set on which they learn.

In some fields of importance for the legal community, more ambitious goals could be contemplated. Legal publishing comes to mind. Legal publishing is composed of many tedious tasks. Some of them require a good understanding of the law, but many are routine tasks. Several publishing activities can benefit from AI and consequently impact research activities in law firms.

As natural language processing techniques and related tools will become more widely available, they will lead to improvements in the accuracy and comprehensiveness of search results. Good design will probably require that legal researchers keep access to the more “calculated” mode of searching to ensure the complete transparency of the process. Automatic preparation of short summaries and classification will help to improve legal research processes by facilitating the drilling down of search results will keep usage costs down. Furthermore, AI techniques can help build better citators with limited human intervention, and links between decisions can be tagged to reflect the treatment of cited decisions. Clustering could serve the discovery of neighboring (similar) cases. Altogether, the various improvements just mentioned can lead to significant changes in research processes.

These and other uses of AI can reduce the cost of commercial offerings and extend the scope of free databases, such as CanLII. For instance, machine learning can accelerate the processing of decisions requiring redaction. In specified contexts, search tools can be informed by learning from past users’ feedback or even simply by past users’ interactions. Many advanced filtering features can be provided to users to better exploit their search results based on named-entity recognition and other smart pre-processing of published data.

This said, building software systems takes time. Developing systems, even with AI technologies, as they become more mature and available, will require time and effort. The resulting transformation of legal life will be surprising and sudden only if we compare its pace with that of constitutional changes in Canada.

— Daniel Poulin, Lexum

Comments are closed.