Guest post: How much of what lawyers do can be automated? A look at new research
Interesting research has recently been published on the potential for automation within legal services. I refer to two studies, one from a team at McKinsey & Co and one from Frank S Levy at MIT and Dana Remus at University of North Carolina School of Law. They are both worth a read but I have attempted to summarise and compare some of the findings.
McKinsey & Co identifies the portion of work that has potential for automation using currently available technology across a broad range of professions. In the legal profession they estimate 69% of time is automatable for paralegals and 23% of time is automatable for lawyers. For more, see McKinsey interactive infographic. (It’s interesting to note the extent to which other professions have potential for automation, for example surgeons share the same 23% estimate.)
Similarly a paper from Frank S Levy at MIT and Dana Remus at University of North Carolina School of Law has reviewed the potential for automation in legal services and puts the figure for lawyers closer to 13%.
Estimates on how much of a lawyer’s time is automatable range from 13%-23%
Why the difference? Well partly because McKinsey’s (interim) findings take a more expansive view of what is automatable, a sentiment that Remus and Levy are keen to keep in check – both cite IBM’s Watson but for opposing reasons (more on this later) . But ultimately, both studies recognize and agree the significant and growing potential of artificial and machine intelligence, although it’s fair to say machines aren’t set to replace lawyers just yet.
Both studies also take a similar approach to answering the question of how much lawyer’s work can be automated? They start by defining a set of core task or activities performed by lawyers. McKinsey use definitions derived from O*Net, a US Dept of Labor Project and Remus and Levy extrapolate their core tasks from The American Bar Association (ABA).
In the case of the McKinsey study they then assess the capabilities required to perform these tasks and subsequently assess the “automatability” of those capabilities through the use of current, leading-edge technology. For example, information retrieving is a cognitive capability, emotional sensing a social capability – the latter being more difficult to automate than the former, for more see: http://www.mckinsey.com/insights/business_technology/four_fundamentals_of_workplace_automation
The tasks identified by McKinsey are:
•Draft legislation or regulations
•Evaluate information related to legal matters in public or personal records
•Identify implications for cases from legal precedents or other legal information
•Interview claimants to get information related to legal proceedings
•Meet with individuals involved in legal processes to provide information and clarity
•Prepare documentation of proceedings
•Prepare legal documents
•Provide legal advice to clients
•Represent the interests of clients in legal proceedings
•Research relevant legal materials to aid decision making
•Supervise activities of other legal personnel
Unfortunately, they do not share the detail behind the specific capabilities required to perform these tasks nor indeed how automatable each task is – although more detail is due to be published this year.
The core tasks used by Levy and Remus extrapolate the ABAs 114 activities into 13 core tasks. They view these tasks in terms of how structured and repetitive they are and whether contingencies are predictable and manageable. This view allows them to estimate how successful machine intelligence will be in automating them relying as it does on either deductive or data driven instruction – the more structured task the more it can be automated. They also refer to the success of current technologies, for example document review and document management technologies being already well advanced. The table below shows their core tasks grouped according to the effectiveness of automation (current or near term) A strong employment effect means they believe computer technology can automate all or most of a task and reduce lawyer time by 85%. A moderate employment effect means successful automation of part of a task and a 19% reduction in lawyer time. A light employment effect refers to a heavily structured, clerical task with minimal lawyer impact – a 5% reduction in lawyer time. Please also note Tier 1 firms>1000 lawyers and Tier 5 <25.
Table 1: Percent of Invoiced Hours Spent on Various Tasks, Grouped by Estimated Extent of Computer Penetration
|Task||Tier One Firms||Tiers Two – Five Firms|
|Strong Employment Effects||4.1%||3.6%|
|Moderate Employment Effects||39.7%||40.4%|
|Case Administration and Management||3.7%||5.6%|
|Legal Analysis and Strategy||28.5%||27.0%|
|Light Employment Effects||56.0%||55.7%|
|Court Appearances and Preparation||13.9%||14.5%|
** Percentages may not sum to 100% due to rounding.
This table also shows the time lawyers in the US spend on each of the tasks shown as a percentage of their total activity (as mentioned sourced from Huron Consulting’s time allocation/billing database.)
To summarize, Levy and Remus believe computer technology is the most advanced in the area of document review though much of the work in this area is performed by paralegals such that the technology will have limited employment impacts on lawyers (note by document review technology they refer largely to the use of predictive coding in the discovery process, something that is likely to have an altogether different employment effect in the UK.) They believe Computer technology is the least advanced in the areas of legal writing, advising clients, communications and interactions, factual investigation, negotiations, and court appearances. These tasks combined total 56% of a lawyer’s work. The reason they give for this is that these are tasks that entail largely unstructured work where there is limited room for automation.
Levy and Remus believe that 56% of a lawyers work has limited room for automation.
In contrast to the sentiment of the McKinsey study Levy and Remus believe some tasks are too opaque and complex or require a level of emotional intelligence that means they won’t be automated anytime soon. One such example of this is the gathering of information from an interview where significant amounts of information may be transmitted nonverbally.
McKinsey on the other hand, whilst acknowledging that tasks requiring capabilities such as creativity and sensing emotions are difficult to automate, talk of ‘the magnitude of automation potential’ believing that the speed with which advances in artificial intelligence and machine learning, challenge our assumptions about what is automatable. It is on the question of what tasks and capabilities are candidates for automation that the two studies appear to diverge.
‘It’s no longer the case that only routine, codifiable activities are candidates for automation and that activities requiring “tacit” knowledge or experience that is difficult to translate into task specifications are immune to automation.’
McKinsey point to the fact that the amount of time that workers (not legal specific) spend on activities requiring creativity and emotion sensing capabilities is low. Just 4 percent of the work activities across the US economy require creativity at a median human level of performance. Similarly, only 29 percent of work activities require a median human level of performance in sensing emotion. Are legal tasks and activities really so different?
To further illustrate ‘the magnitude of automation potential’ they refer to Amazon’s fleet of Kiva robots which is ‘equipped with automation technologies that plan, navigate, and coordinate among individual robots to fulfill warehouse orders roughly four times faster than the company’s previous system.’ They also reference IBM’s Watson which ‘can suggest available treatments for specific ailments, drawing on the body of medical research for those diseases.’ Levy and Remus reference IBM’s Watson for different reasons believing that for the foreseeable future, is prohibitively expensive requiring ‘significant investments of human time and energy.’
The discussion on the displacement of lawyers by computers and the effects of technology is not likely to end any time soon. For now I will give the final word to Levy and Remus (mostly because they have concluded their study) They believe most writing on the computerization of legal services overstates the likely impacts. They consider the work of Richard & Daniel Susskind, professors John McGinnis and Russ Pearce and potentially McKinsey authors Michael Chui, James Manyika, and Mehdi Miremadi guilty of:
– a failure to engage with technical details to appreciate the capacities and limits of existing and emerging software;
– an absence of data on how lawyers divide their time among various tasks, only some of which can be automated;
– inadequate consideration of whether algorithmic performance of a task conforms to the values, ideals and challenges of the legal profession.
I look forward to seeing more from the McKinsey team.
Both the recent studies are clear that experience or skill level is not an accurate predictor of automatability and that very few occupations will be automated in their entirety in the near or medium term. Rather, certain activities are more likely to be automated.
For me as advanced as the computer technology becomes, critical factors in the impact of automation will always be the rate and extent of adoption, adaptation and implementation of the technologies. This is determined by the need to innovate – ‘Innovation in Legal Services – Are We There Yet?’ will be the title of my next post.
Peter Nussey is a Director at F5Legal, a Consultancy specializing in Legal Business Transformation, Document Automation and Business & Product Development