LEGAL TECH, CIVIL PROCEDURE, AND THE FUTURE OF ADVERSARIALISM.

• Author: Engstrom, David Freeman

INTRODUCTION 1002 I. THE LEGAL TECH LANDSCAPE 1008 A. Flavors of Legal Tech 1009 B. Technical Limits and the Trajectory Puzzle 1018 C. Implications 1030 1. Legal Tech and the Legal Profession 1031 2. Legal Tech and Rule of Law 1035 3. Legal Tech and Distribution 1037 II. LEGAL TECH AND CIVIL PROCEDURE: THREE CASE STUDIES. 1041 A. Predictive Coding, Proportionality, and Plausibility Pleading 1042 1. The New World of Discovery 1043 2. Proportionality's Retreat in a Frictionless World 1051 3. Re-Centering Twombly and Iqbal 1056 B. Predictive Analytics and Forum Selection 1059 1. Forum-Shopping in Federal Courts and the Promise of Predictive Analytics 1060 2. Will Predictive Forum Selection "Work"? 1062 3. The Future of Forum Selection and Civil Procedure 1068 C. From Borrowed Wits to Borrowed Bits: Legal Tech and the Work Product Doctrine 1071 1. Information and Adversarialism: Reframing Legal Tech's Distributive Costs 1072 2. Hickman's Work Product Bargain 1077 3. Work Product for a Digital Age 1080 III. LEGAL TECH AND "OUR ADVERSARIALISM" 1086 A. An IP for Civil Procedure 1087 B. Legal Tech and the German (Dis)Advantage 1092 CONCLUSION 1099 INTRODUCTION

"Legal tech," most agree, is transforming litigation and law practice, and its steady advance has tapped a rich vein of anxiety about the future of the legal profession. (1) Is law like a driverless car, or is it irreducibly complex and grounded in dynamic human judgment? How to square online dispute resolution and automated legal advice with rules governing unauthorized practice of law? Can BigLaw survive? Much of this has a profession-centered and even defensive quality in its narrow focus on what legal tech portends for the professional authority and profitability of lawyers. Much of it is also profoundly futurist--full of prophecies of "robolawyers," (2) "robojudges," (3) or even an eventual state of "legal singularity," (4) when machines can perfectly predict the outcomes of cases before they are filed.

Lost in this rush to foretell the future of lawyers and their robotic replacements is what should be an equally or even more important concern: what effect will legal tech's continued advance have on core features of our civil justice system and, in particular, the procedural rules that structure it? And how, in turn, can or should those rules be adapted to further the ends of justice? This Article seeks to enrich--and, in places, reorient--budding debate about what many see as a coming revolution in legal tech. Simply put, if law and the legal profession will look different ten or fifteen years from now, then civil procedure and the inner workings and structure of the adversarial system will look different as well. Indeed, though virtually unmentioned in a lively but high-altitude new literature on legal tech's potential implications, it is the rules of civil procedure and related doctrines that will serve as the front-line regulators of the new legal tech tools and critically shape their evolution in the near-to medium-term. As a result, judges, rulemakers, and legislators should begin to think about whether, and if so how, to adapt civil procedure to new litigation realities as legal tech continues its move to the center of the civil justice system.

We aim to spark concrete thinking about this mediating role for civil procedure by focusing on the near future--not out at a hazy horizon dotted with robojudges and robolawyers--and then asking how legal tech will change litigation and, in turn, how procedure can or should adapt in response. (5) The core of our argument proceeds from the premise that legal tech's proliferation is likely to alter two foundational aspects of any litigation system: the distribution of litigation costs and the distribution of information. In a nutshell, there is good reason to believe that the concern about high and asymmetric litigation costs that has fueled several decades' worth of litigation reforms will progressively fade as new and powerful e-discovery tools propagate. By contrast, it is plausible that increasing uptake of legal tech tools, including e-discovery tools but also tools that perform legal research and analytics and predict case outcomes, will worryingly widen information asymmetries within the system, between judges and litigants, and also between litigants and litigants--particularly litigation's "haves" as against its "have nots."

Isolating legal tech's effects on these deep dimensions of the system provides needed analytic traction and grounds a set of concrete judgments about how an array of civil procedure rules and doctrines--among them the plausibility pleading standard set forth in Twombly and Iqbal, the bundle of procedural rules, doctrines, and statutes concerned with forum-shopping, and the work product doctrine--can, or should, adjust in response. When these assorted dynamics are lined up and viewed together, it is not a stretch to say that legal tech will, in time, remake the adversarial system, not by replacing lawyers and judges with robots, but rather by unsettling, and even resetting, several of its procedural cornerstones.

These are big claims, and they demand both a technical grasp of the legal tech toolkit and command of contemporary civil procedure. Given these complexities, we build our argument deliberately, in three steps.

Part I offers a full and quasi-technical canvass of where legal tech currently is and where it is likely to go in the near- to medium-term as natural language processing (NLP) and other machine learning techniques that power the most consequential legal tech tools continue to improve. In so doing, we strike a skeptical note and also go about our labors with a heavy dose of humility. As with any emergent technology, legal tech is a fast-moving field, and any effort to capture its many facets risks becoming antiquated almost as soon as the ink dries. We manage this contingency by surveying the legal tech landscape in three pieces. Section I.A reviews legal tech's flavors and offers some ways to slice and dice them. Section I.B turns to legal tech's technical trajectory. It shows that the frontier is quickly moving beyond e-discovery and digital referencing tools (think Westlaw or Lexis) to tools that automatically gather legal materials, predict case outcomes, and even draft legal documents. However, there are also legitimate questions about how far and how quickly legal tech can advance. Just how much progress can be made on outcome prediction tools given pervasive confidential settlements and the resulting lack of well-labeled data, or the current technical limits of natural language processing (NLP) in extracting and analyzing legal argumentation? Section I.C summarizes some key implications of legal tech--for the legal profession, for the distribution of power within the legal system, and for law itself--as sketched in an emerging academic literature that, while highly abstract, has begun to stake out the poles of a rich debate. A thorough survey of the legal tech landscape provides the raw material for the more focused case studies of procedure to come.

Armed with Part I's extended account of legal tech's pathways of innovation and diffusion, we turn in Part II to offering three concrete cuts at how legal tech's advance will reshape American litigation and how procedural rules might mediate those effects.

Section II.A starts on familiar ground: e-discovery and, more specifically, the "technology-assisted review" (TAR) and "predictive coding" tools that are quickly becoming a fixture of complex litigation practice. Our core claim is that, contrary to the views of some, civil litigation may well see a steady decline in overall discovery costs and, by extension, a narrowing of the litigation cost asymmetries that have motivated decades of litigation reforms, from the Civil Justice Reform Act of 1990 to the 2015 amendments reshuffling Rule 26's proportionality constraint. In a lower-friction world, we predict, battles over proportionality would largely abate or become peripheral. Narrowing litigation cost asymmetries may also alter, or at least destabilize, the normative foundation of a very different and controversial part of civil procedure: the plausibility pleading doctrine set forth in Twombly and Iqbal. That doctrine sits at the intersection of two competing concerns: litigation cost asymmetries, with attendant concerns about undue settlement leverage and the conversion of low- or even negative-value cases into positive-dollar settlements, and information asymmetries in cases where only discovery can dislodge privately held information about wrongdoing. By systematically narrowing litigation cost asymmetries, TAR could undermine the positive foundation of the new plausibility pleading regime.

Section II.B turns to legal tech tools that predict case outcomes. An obvious concern is that continued advances in outcome prediction tools will foster forum shopping, placing pressure on the rules, statutes, and doctrines--venue, removal, Erie doctrine--that seek to limit or shape its pursuit. Here we sound a more skeptical note about legal tech's implications for civil procedure. Current procedural rules and doctrines touching upon forum shopping strike a permissive pose, and so an initial question is whether successful deployment of predictive analytics should change that pose. Further grounds for skepticism are the technical and practical limits of outcome prediction tools, which may not "work" well enough to meaningfully increase forum-shopping in the first place. Even if the technical hurdles can be leapt, two other major obstacles stand in legal tech's way. First is the huge cost of assembling enough docket and document data to obtain sufficiently large and representative samples for contemporary prediction methods. Second are a pair of endogeneity problems that raise profound questions about either the initial or subsequent usefulness of even...