Machines v. Lawyers

This article originally appeared in the City Journal on June 4, 2014.

By John O. McGinnis

Law schools are in crisis, facing their most substantial decline in enrollment in decades, if not in the history of legal education. Applications have fallen over 40 percent since 2004. The legal workplace is troubled, too. Benjamin Barton, of the University of Tennessee College of Law, has shown that attorneys in “small law,” such as solo practitioners, have been hurting for a decade. Attorney job growth has been flat; partner incomes at large firms have recently recovered from the economic downturn, but the going rate for associates, even at the best firms, has stagnated since 2007.

Some observers, not implausibly, blame the recession for these developments. But the plight of legal education and of the attorney workplace is also a harbinger of a looming transformation in the legal profession. Law is, in effect, an information technology—a code that regulates social life. And as the machinery of information technology grows exponentially in power, the legal profession faces a great disruption not unlike that already experienced by journalism, which has seen employment drop by about a third and the market value of newspapers devastated. The effects on law will take longer to play themselves out, but they will likely be even greater because of the central role that lawyers play in public life.

The growing role of machine intelligence will create new competition in the legal profession and reduce the incomes of many lawyers. The job category that the Bureau of Labor Statistics calls “other legal services”—which includes the use of technology to help perform legal tasks—has already been surging, over 7 percent per year from 1999 to 2010. As a consequence, the law-school crisis will deepen, forcing some schools to close and others to reduce tuitions. While lawyers and law professors may mourn the loss of more lucrative professional opportunities, consumers of modest means will enjoy access to previously cost-prohibitive services.

A decline in the clout of law schools and lawyers could have potentially broader political effects. For the last half-century, many law professors and lawyers have pressed for more government intervention in the economy. This isn’t surprising. Lawyers in the modern regulatory state reap rewards from big government because their expertise is needed to understand and comply with (or exploit) complicated and ever-changing rules. In contrast, the entrepreneurs and innovators driving our computational revolution benefit more from a stable regulatory regime and limited government. As they replace lawyers in influence, they’re likely to shape a politics more friendly to markets and less so to regulation.

The easiest way to grasp the transformative power of machine intelligence for law is to consider another kind of “law”: Moore’s Law, named after Intel cofounder Gordon Moore, which famously observes that the number of transistors that can fit onto a computer chip doubles every 18 months to two years. The power of such dramatic growth is hard to overstate. The computational capacity in a cell phone today is 1,000 times greater and 1 million times less expensive than all the computing might that was housed at MIT in 1965. Projecting forward, the computing power 30 years from now could exceed today’s by 1 million times.

Yes, Moore’s Law will run out when the size of transistors cannot shrink further, something that’s predicted to happen in the 2020s. But Ray Kurzweil, a leading technologist and now Google’s director of research, has shown that Moore’s Law is actually part of a more general growth in computation that has been gaining force for more than 100 years. Electromechanical methods began the push for enhanced computation; they were replaced by vacuum tubes, which were surpassed by transistors, which gave way to today’s integrated circuits. Other methods under research today, from carbon nanotechnology to optical computing, could become new platforms for continued growth. (See “The Next Age of Invention,” Winter 2014.)

The dramatic increase in hardware capability is only part of the change in computational capacity. Software improvements, while less steady, provide another force multiplier for the power of computation. Computers now also interconnect among themselves and with human intelligence, sharing information more seamlessly—increasing effective computational capacity yet more.

Greater computational power in hardware, software, and connectivity relentlessly improves artificial intelligence. The most recent public triumph of AI came in the form of Watson, the IBM machine that, in 2011, beat the best Jeopardy champions by exploiting advances in all three areas. The computer disentangled humor, recognized puns, and resolved ambiguity. Watson represents substantial progress over Big Blue, the machine that beat the world chess champion in 1997, succeeding in a less precisely rule-governed world than chess. Indeed, Watson’s world more closely resembles the chaotic one that we inhabit, of which law is definitely a part.

Machine intelligence is not only increasing its capacity. It’s also expanding its reach by entering new domains. And once it enters, it uses the exponential increase in computer power to improve until it dominates. I remember when I could beat a computer at chess. Now my smartphone regularly humiliates me at it. Cars provide a second striking example of this process. In 2004, no computer-controlled vehicle drove farther than 11 miles on a challenge course through the desert. But Google is already testing self-driving cars, and Volvo will put 100 on Gothenburg’s streets in 2017. By the middle of the next decade, driverless cars will be regularly transporting passengers.

Five key areas of law now face encroachment by this machine intelligence. Some invasions are imminent, and others more distant but no less likely. The area ripest for computational transformation is discovery. As a young lawyer, I spent lots of time rifling through documents to determine which were relevant to an opponent’s request for information. That was the tedious, if lucrative, lot of the junior litigation associate and an important profit center for the litigation group at a big firm. These days, “predictive coding” is removing that labor-intensive task from lawyers. In predictive coding, a small number of lawyers can swiftly sample a large set of documents and construct algorithms—with the help of computer technicians—to decide which documents are relevant. Computers can sort better than people because fatigue, boredom, and distraction reduce human accuracy, while machine intelligence works nonstop, with no lag in attention or need for caffeine or sleep.

“E-discovery” has already become the hottest new phenomenon in litigation. Job growth in this legal area, unlike most others, is strong. One graduate of Northwestern Law School now specializes in head-hunting for professionals who can strengthen law firms’ e-discovery capabilities. And courts now recognize that e-discovery can curb litigation costs and make justice more affordable. For instance, the Federal Circuit Court of Appeals, which specializes in patent cases, has issued a standing order that encourages the use of e-discovery.

Private firms are also beginning to specialize in these sophisticated services. With a combination of computational and legal knowledge, they can innovate more readily than lawyers who are left to their own devices. Last year, Modus raised $10 million to continue its data-driven competition with law firms in e-discovery. Such innovation will render e-discovery more accurate and less expensive, making use of such methods routine.

More than 100 years ago, a jurist wrote: “Every practitioner knows that when a hard case arises, the law books are ransacked from the time of the Norman Conquest and the court blindly applies any absolute precedent that may have been found by diligent counsel.” Even if he exaggerated, searching for the right cases for precedents remains an important legal skill. Yet just as computers have largely replaced humans in making complex calculations, so machine intelligence will supplant lawyers’ legal search function—a second key area to be disrupted.

Until now, computerized legal search has depended on typing in the right specific keywords. If I searched for “boat,” for instance, I couldn’t bring up cases concerning ships, despite their semantic equivalence. If I searched for “assumption of risk,” I wouldn’t find cases that may have employed the same concept without using the same words. IBM’s Watson suggests that such limitations will eventually disappear. Just as Watson deployed pattern recognition to capture concepts rather than mere words, so machine intelligence will exploit pattern recognition to search for semantic meanings and legal concepts. Computers will also use network analysis to assess the strength of precedent by considering the degree to which other cases and briefs rely on certain decisions. Some search engines, such as Ravel Law, already graphically display how much a particular precedent affected the subsequent course of law. As search progresses, then, machine intelligence not only will identify precedents; it will also guide a lawyer’s judgment about where, when, and how to cite them.

Search is also becoming ever more affordable, even as its efficiency increases. Lexis and Westlaw still charge for their superior legal search engines, but free search is now available from FindLaw and Google Scholar, among others, and these sites offer more than adequate assistance for many purposes. Such cost reduction exemplifies the Silicon Valley slogan that information “wants to be free.” Lawyers have traditionally enjoyed leverage over the laity, partly because of their superior access to information. Low-cost legal knowledge poses a threat to that power.

Today’s search capabilities still require people to identify the legal issues at stake in a given matter, but search engines will eventually do this by themselves, and then go on to suggest the case law that is likely to prove relevant to the matter. The rise of computation in legal search is not a one-time disruption but a continuing revolution.

A third area, legal forms, will also be revolutionized. Since the Middle Ages, lawyers have used form templates to reduce costs. Machine intelligence will allow consumers to shape these forms themselves by providing data online, thus dispensing with the legal middleman. For instance, clients of Legal Zoom can enter information about their assets and intentions for their estate. A computer program can then draft up a will. Trust and estate planning is ripe for this kind of mechanization because most people have relatively simple needs that can be met with few variations.

As computers and software become more powerful, computer-generated forms will take on ever-wider scope. Already one firm, Kiiac, is focusing on contracts. Kiiac’s idea is to evaluate different versions of agreements and determine, through the accumulation of data, the best way to write provisions. With the growing interconnectedness of data, machines can relate specific contracts to relevant court decisions, creating a dynamic for continual improvement of legal forms. A company like Kiiac shows that contracts can essentially become a computer code, rather than simply words on a page. It may then become easier for firms to integrate their legal obligations into their basic operations. Lower-level employees won’t need to consult so much with corporate counsel for contract execution; the code will guide them. The cost savings could be considerable.

Even government is beginning to recognize the advantages of automating legal forms. Nevada’s secretary of state has pioneered online registration for small businesses, which can comply with regulations by following the steps of simple computer programs. Of course, at first, lawyers will remain heavily involved in marking up the drafts of transactional documents that machines create. Even at this stage, though, the savings add up. Some Silicon Valley law firms have already come up with programs that sharply reduce the time needed to create incorporation documents for start-ups. Within two decades, I predict, it will be a rare occurrence when computer-based services do not generate the first draft of a transactional document.

Machine intelligence will learn to automate simple briefs and memos, too—the fourth area—though this may be a more distant prospect. Legal forms are generally easier to systematize than legal memos or briefs, depending, as they do, on more preset formulas. Still, consider Quill, a program written by a recent start-up, Narrative Science. It takes information in the form of the basic box scores of games and statistics and generates reports on sporting events. It can also produce business stories using similar inputs. While earlier programs churned out stories obviously written by machine, Quill’s read, according to the New York Times, as if a human being wrote them—albeit not an accomplished wordsmith.

In their early stages, machine-made briefs and memos will serve only as rough drafts, even for the simplest matters. Nevertheless, an experienced lawyer could then easily shape a computer-generated draft into a more polished product. As with other advances in machine intelligence, moreover, once programs start being useful, they get more effective over time. That’s been true of everything from word processing to speech-to-text programs.

Finally, legal analytics will displace lawyers’ hunches. As Professor Dan Katz of Michigan State has noted, “moneyball” is coming to law. The Michael Lewis bestseller Moneyball recounts how armchair data analysts propelled the Oakland A’s to victory—not only over major-league baseball opponents but also over the old-time scouts who relied mostly on instinct to choose players. Moneyball dramatized an early example of using “big data” to guide decision making. Lawyers make judgments about litigation prospects whenever they advise their clients on bringing lawsuits, settling them, or going to trial. Until now, experience and intuition limited their guidance. Not for much longer.

A fairly simple model, incorporating Supreme Court decision making from previous rulings, predicted more accurately the Court’s decisions over one term than did legal experts. Analysts can now buy a data set of patent cases compiled by a new company, Lex Machina, to forecast outcomes in patent litigation. More generally, legal-management consultants analyze big data to help companies decide when to litigate, when to settle, and how to manage litigation costs. Companies such as Huron Legal employ analysts with expertise in data analysis. As with e-discovery, these firms will probably innovate faster than law firms in applying the new science to practical legal problems.

Legal analytics will always be imperfect, providing likelihoods rather than certainties. Yet it does not need to be perfect—it can displace lawyers simply by making better predictions than they do. Lawyers are vulnerable on this front. After all, computers have far greater power to evaluate data—and they don’t feel. Studies have shown that lawyers’ overconfidence often produces poor advice. Legal analytics will also reduce the number of cases that go to trial by providing better estimates of cases’ value. Cases will settle earlier, and trial lawyers will have less to do.

Discovering information, finding precedents, drafting documents and briefs, and predicting the outcomes of lawsuits—these tasks encompass the bulk of legal practice. The rise of machine intelligence will therefore disrupt and transform the legal profession.

A relatively small number of very talented lawyers will benefit from the coming changes. These superstars will prosper by using the new technology to extend their reach and influence. For instance, the best lawyers will need fewer associates; they can use computers to enhance the value that they offer their clients. Already, the ratio of associates to partners in big law firms appears to be declining. In complex cases, lawyers will continue to add value to machine intelligence through uniquely human judgment. Even now, when computers regularly beat the best chess grandmaster, a good chess player and a good computer combined can often beat the best computers. Thus, for important cases and transactions, good lawyers will still add substantial value, even if computers do more of the work.

Lawyers practicing in highly specialized areas subject to rapid legal change—such as Dodd-Frank regulation—may also flourish, at least initially. Machine intelligence succeeds through pattern recognition; in narrow, fast-changing areas, it has less data and thus fewer opportunities to identify promising correlations. In such areas, lawyers will have room to craft intuitively appealing arguments to regulators and courts. And the technological acceleration of our age may create the need for new kinds of legal talent. Judge Richard Posner has called for lawyers schooled in science to help devise and implement legal frameworks to address new kinds of catastrophic risks, such as those from nanotechnology and biotechnology.

The biggest winners may be lawyers who can use machine intelligence to create an automated large-scale practice. The Walton family, it’s worth recalling, got rich by effectively automating large-scale retail. More generally, there may be jobs for a new category of engineer-lawyers—those who can program machines to create legal value.

But the large number of journeyman lawyers—such as those who do routine wills, vet house closings, write standard contracts, or review documents on a contractual basis—face a bleak future. They will have far less to contribute to legal analysis, and they will face relentless evaluation from clients using new data-driven metrics. Journeyman lawyers may still be able to earn their keep by persuading angry and irrational clients to act in their self-interest. Machines won’t be able to create the necessary emotional bonds to perform this important service. Some lawyers may even do better to immerse themselves as much in the modern version of Dale Carnegie as in the modern version of Blackstone. Legal therapists, though, will generally not earn as much as the lawyer who is adding lots of analytic value.

Lawyers might also look to the deus ex machina of government intervention to save them. A burgeoning regulatory state may continue to complicate law. But this expansion, if it occurs, will ultimately prove no match for the growth in computational power. By separating superstars from the rest of the legal profession, technology will increase income inequality among lawyers—but by delivering lower-priced services, it will decrease consumption inequality among consumers.

The rise of machine intelligence is probably partly to blame for the current crisis of law schools—and will certainly worsen that crisis. While no law school has recently closed, most have lost students and even more have lost revenue, as they discount prices to attract students in a shrinking applicant pool. Financial-monitoring agencies have downgraded the bonds of some schools toward junk status. The job market for law professors, both at the entry and lateral stage, has shrunk.

To match the wide variety of tasks that lawyers will undertake in a world increasingly defined by machines, law schools will need to differentiate themselves in cost and function. No longer can every school aspire to be a junior varsity Yale. Some schools will ask faculty to teach more, even at the expense of legal scholarship, or use adjuncts who write no scholarship, thereby slashing costs. Many schools will substitute videos for some live instruction. They can then redeploy some professors to focus on improving legal writing and problem-solving skills. Negotiation may get more emphasis, as it contains emotional elements that machines cannot easily replicate.

Law schools can seek new revenues by preparing students for the computer revolution in law—providing courses, say, on improving the interface between legal machines and humans. Some schools might also provide shorter courses of study to engineers and computer scientists, who can design the in-house legal analytic tools that many corporations and law firms will require. Here, though, law schools may lose out to business schools, which have traditionally provided a better setting for quantitative analysis.

More fundamental reforms may be necessary to serve an increasingly stratified legal profession. Already some respected legal educators, such as the dean of Northwestern, favor permitting students to take the bar after two years of legal education. More radical proposals—such as making the study of law a mostly undergraduate prospect, as it is in many countries now—would save the cost of going to law school altogether.

The most profound long-term effect of the rise of machine intelligence on the legal world may be a decline in lawyers’ social influence. When Alexis de Tocqueville visited the United States almost 200 years ago, he saw lawyers as the aristocrats of America’s democratic regime. “Lawyers,” he wrote, “form the highest political class and the most cultivated circle of society.” But machine intelligence empowers those involved in computation at the expense of those skilled at rhetoric. To some degree, engineers—the descendants, really, of blacksmiths—are destined to replace the wordsmiths in society’s commanding heights.

In a nation of laws, not men, lawyers’ knowledge made them, paradoxically, the key men. Some Framers, such as Alexander Hamilton, may have even believed that judicial review—the province of lawyers—was a needed aristocratic element in a mixed regime that, according to Aristotle, was better than a pure democracy. Lawyers could be counted on to protect at least some property rights against the passions of the mob. And judges, who would have served merchants and property owners as clients before ascending to the bench, would act as a brake against democratic excesses.

In the twentieth century, lawyers continued to wield power, but the direction of their influence in economic affairs changed. Since the birth of the modern regulatory state and social democracy, lawyers have had incentives to increase and revise legislative mandates; they became the technocrats of regulation and redistribution. The more a nation intervenes in the free market, the more in compliance costs and transfer payments that lawyers can expect to receive. As a result, lawyers don’t tend to be strong proponents of economic liberty or even of a stable rule of law. Their interest frequently lies in legal complexity and the uncertainty it brings.

The decline of lawyers may therefore prove a boon to the rule of law and to market norms. Computational innovators benefit from capitalism’s process of creative destruction; their new applications transform industry after industry. Their success lies with a stable rule of law and relatively light regulation. True, once successful, innovators become incumbents and may seek to use government to hamstring new entrants. But the dynamism of technological acceleration will make it difficult even for big government to hold back waves of new “disruptions.”

The rise of computational innovators may also foster a more data-driven politics. A modern, law-oriented politics often is excessively rhetorical; competing ideals quickly become abstractions. We debate same-sex marriage, for instance, at the federal level in terms of claims about equality, and school funding at the state level in terms of a right to education. The relentless march of computation, by contrast, permits a focus on the actual effects of social policies and encourages experiments to test those effects.

Such a change of emphasis could be a welcome catalyst for political humility. An experimental politics recognizes that no party or faction possesses a monopoly on wisdom and that public deliberation can be improved by systematically evaluating the evidence for contending claims. It de-emphasizes the intuitions that may divide us and orients us toward what we have in common—the facts of the world.

Edmund Burke famously mourned the replacement of the Age of Chivalry by “that of sophisters, calculators, and economists.” For Burke, the first category would probably include many of today’s lawyers. So long as there is law, however, we will need lawyers to offer interpretations of difficult texts and to smooth legal difficulties in the most important transactions. And so long as we have our Constitution, lawyers will have an essential role in the nation’s governance. Nonetheless, in the Age of Computation, the calculators are gaining on the lawyers—at work and in politics.

- John O. McGinnis is a professor of law at Northwestern University.