LAS VEGAS — Stephen Allen, Global Head of Legal Service Delivery at Hogan Lovells, summed up the Artificial Intelligence zeitgeist at this year’s International Legal Technology Association conference (ILTACON 2017):
“Naughty vendors, naughty firms, and naughty press have made the really useful stuff that AI does seem very unsexy.”
What Allen meant was that the media hype and exaggerated expectations for AI have poisoned the waters, to the point that anything short of massive disruption and fundamental change in the legal industry brought on by AI might be seen as a disappointment. In the meantime, however, the things that AI does really well — such as automating certain high-volume processes involving lots of data — seem dull by comparison.
That’s too bad, because many of those unsexy applications of advanced analytics, machine learning, and expert systems were very much on display at this year’s ILTACON. Indeed, these applications are having a big impact on legal practice even if they don’t meet “Robot Lawyer”-level hype expectations.
Here are some of the examples we saw on display, with the caveat that not all of them really constitute AI by some definitions. Some of these are built off of rules-based expert systems, some on analytics, and some really are leveraging they type of machine learning that most would consider to be AI.
Katie DeBord of Bryan Cave described the systematic understand-experiment-apply process that is embedded in her firm’s technology initiative called TechX (further described in the Spring issue of Legal Tech Link). That approach has led the firm to use Bayesian classification to analyze unstructured time-entry narratives in a tool called Tasker. This assists the firm’s ability to price matters accurately by better understanding the levels of effort associated with specific tasks in past matters.
Allen of Hogan Lovells and Neil Cameron of Neil Cameron Consulting Group picked up on the theme of the “less sexy” applications for AI. Both see some of the biggest returns on AI in simply analyzing a firm’s own activity, pure Business of Law applications.
Cameron focused on the startling levels of write-offs that occur when lawyers under-budget matters because they underestimate costs, completion times and risks. He described the application of a technique called Reference Class Forecasting, which allows users to predict the outcome of planned actions based on a reference class of similar actions. The algorithms that do this forecasting take into account the various complexity factors a given matter might present, and they train forecasting systems on the relationship between those factors and fees in past cases. He pointed out that this process of identifying complexity factors can be done in collaboration with the client, with the result that clients then become “co-conspirators” and more likely to accept both budgets and deviations based on the inputs that both parties have agreed to.
Jeff Sharer described the development of the Akerman Data Law Center, a client-facing system that provides victims of corporate data breaches with analysis and recommendations based on regulations in force in 50 states. The question-and-answer interface is built on partnerships with Neota Logic’s expert system platform; and Thomson Reuters Legal Managed Services provided the original research and the ongoing updates of the regulations data.
A panel consisting of Anna Moca of White and Case, Amy Monaghan of Perkins Coie, Jonathan Talbot of DLA Piper, and Julian Tsisin, in-house at Google, tackled some of the process issues, describing how AI projects can get off the ground and start delivering results.
On the front end, all were unanimous in their focus on identifying a business problem situated inside some business process. Several of the panelists are using Kira or other machine-learning systems in document review, and finding good success there, but this session underscored that AI success isn’t about magic — it’s about process, planning and change management.
In David Curle’s next blog, he will examine the major themes and lessons that emerged from the various sessions on AI at ILTACON.