©2016. Published in Landslide, Vol. 8, No. 5, May/June 2016, by the American Bar Association. Reproduced with permission. All rights reserved. This information or any portion thereof may not be copied or disseminated in any form or by any means or stored in an electronic database or retrieval system without the express written consent of the American Bar Association or the copyright holder.
Studying law was a new experience for Marshall. He had a background in electrical engineering, but to Marshall, lawyers do not think like engineers. There was, of course, a lot of reading to learn this new field. Thankfully, it was not just appellate decisions. Marshall found that judicial writing often left him confused. The reading list included law review articles, treatises, and even blog posts, and Marshall found these much clearer.
At first, none of what he read made sense. Marshall would read, try answering questions, and not get much right. But with some help from his teachers, Marshall improved. Back and forth they would go, with Marshall reading more and his teachers gently correcting his mistakes. Finally, everyone agreed he was ready. He took the bar exam, passed, and today is his first day as an associate at the intellectual property (IP) boutique, Turing & Associates.
As a first-year associate, Marshall will research issues raised in patent lawsuits. He will spend two years working on litigation before he gets a taste of the firm’s transactional practice. The partners have found that associates learn faster when they work on litigation issues, and that gaining some experience and maturity working with the law helps them when it comes to transactional issues.
Joining a firm means adapting to the firm culture. Turing associates must learn the way each partner handles matters, because of course no two partners do anything the same. They also must learn what each partner means when asking for research. To one partner it means “give me a quick overview,” but to another partner it means “give me a thorough analysis.” Finally, Turing associates have to learn to get along with each other and with secretaries, project managers, and paralegals. Everyone has his or her place in a law firm, and Marshall will find his place like the other associates. But Marshall has one additional challenge. Marshall is not a person—it is a computer.
A False Tipping Point
When IBM’s Watson won Jeopardy in 2011, many futurists began speculating that soon law firms would hire Watson as an associate. What IBM had accomplished with Watson was massive, and it pointed toward a world where computers would leave the desktop and sit beside humans. It turns out that the tipping point predictions were premature.
To understand the predictions and why they have not panned out, we must first understand what Watson did that day in 2011. To win, Watson first had to store and access immense amounts of data, because it was not connected to the Internet during the contest. It had to process that data after tangling with the quirky Jeopardy format. On Jeopardy, the host provides the answers (clues) and the contestants guess the questions. For Watson, that meant translating each clue into a question, generating possible answers, and ranking the answers on likelihood of success. And then Watson had to pick one and push the buzzer.
Watson trounced its two champion competitors and reached a score three times that of the second-place contestant. Following jeopardy, IBM predicted great things for Watson in medicine and other data-heavy fields. Law seemed like another logical place to put Watson to work. “Any information-intensive industry was fair game, anywhere were there were huge volumes of unstructured and semi-structured data that Watson could ingest, understand and process quicker than its human counterparts.”1
Five years later, despite all the post-Jeopardy hype, we are only slightly closer to a computer joining the ranks of new lawyers. Many futurists have back-pedaled saying law will be one of the last frontiers conquered by computers, though some maintain that day will come sooner than most lawyers think.2
Given the promise and power of Watson and artificial intelligence (AI) generally, why aren’t computers taking over associates’ desks? To answer that question, we must look deeper into what computers can and cannot do, and what lawyers do.
What Computers Can Do
The promise of AI has exceeded reality throughout history. Science fiction writers have treated us to anything from HAL in 2001: A Space Odyssey to Data, the intelligent but (mostly) unemotional android in Star Trek. The reality, researchers have learned, is that AI can mimic certain human functions, but it will take a great deal more work to mimic the things that still flummox humans, such as common sense and emotions.
AI has advanced unevenly since a group of computer scientists gathered at Dartmouth College in 1956 for the Dartmouth Summer Research Project on Artificial Intelligence. While dreams of AI go back centuries, the Dartmouth conference marks the modern beginning for the field. Following the conference, AI experienced a golden period until the 1970s. By then, the challenges of AI had become more evident and various limitations, most notably in hardware, caused a drop-off in interest.
By the 1980s, increases in hardware power and memory and interest in “expert systems” caused a renewed interest in AI. Since the dawn of the twenty-first century, especially with the introduction of Siri, Cortana, Google Now, and Alexa, AI research has accelerated with major tech companies placing big bets on its future. The public’s imagination also has been captured. Movies including I, Robot and Transcendence have led many to believe that computer scientists are close to creating the magic algorithms that will allow computers to mimic people.
AI in law has been constrained by reality and still means something different than what the public imagines. Most of the time, AI in law means computers handling tasks that are simple for lawyers, but which computers can do faster using their raw processing power and massive data storage and retrieval systems. E-discovery is the most common example. For decades, IP associates learned part of the craft by sifting through haystacks of paper looking for needles. For the associates, that meant hours, days, and even months of tedious work. As clients created and stored more information electronically, discovery turned into people scanning digital information.
Finding relevant documents in terabytes of information is a mind-numbing task for any human. For a computer, however, it is the same as any other processing task. Computers “far outstrip” in focus, patience, processing speed, and memory; “when it comes to doing the same thing a billion times while keeping all the results in memory, we don’t even come close. What skill doesn’t benefit from such relentless focus and work? When a computer achieves a reasonable ability level in some domain, superior skill isn’t far behind.”3
Of course, computers are doing more than simply finding those needles. They are participating in supervised machine learning. As Marshall described, after each pass through the material, humans gave it feedback. The accuracy of his algorithms improved. Some of that learning only helps with the current task. But some of it carries over to the next, expanding the computer’s capabilities while improving its performance.
Outside the discovery process, computers have made gains in areas where analyzing large quantities of data can yield patterns and insights. In these tasks, computers act as computation engines. Joscha Bach, a cognitive scientist at the MIT Media Lab and the Harvard Program for Evolutionary Dynamics, says, “Computation, at its core, and as informally described as possible, is very simple: every observation yields a set of discernible differences.”4 Being good at computation is no small thing. In Bach’s view “the 20th century’s most important addition to understanding the world . . . is the notion of computation.”5 A computer, with its vast memory and processing power, is an excellent tool for finding the differences.
We see businesses such as Kira Systems, Lex Machina, and Ravel Law using computers to look for differences and patterns in data stored in deal data rooms, lawsuit information, and court opinions. Crunching through the data, computers can find patterns in minutes or hours that no human could find if he or she devoted an entire life to reading. These patterns, based on discernible differences, lead us to greater insights about what risks clients face, how cases are decided, and what factors yield favorable decisions. Even so, some comparison tasks still are difficult for computers. College freshman can quickly identify the common themes in Hamlet and The Lion King, but it would take Watson a lot of training to get there.
On a mundane level, but one that yields more immediate use, computers take drudgery out of day-to-day legal tasks. Document automation software saves us from retyping or copying and pasting text to create new versions of standard contracts. Logic tree software replaces hours of phone calls with questionnaires clients complete to feed us needed information. Contract drafting software eliminates the constant ping pong of negotiations. Rather than volleying fully edited copies of a contract back and forth, lawyers enter a negotiation room where they engage in real-time drafting while the software treats each contract clause as separable with its own history. Legal drafting meets GitHub.
Where Computers Can’t Match Humans
As powerful as computers have become, there still is a meaningful gap between what they can do and what the most inexperienced associate can do. MIT economist David Autor has emphasized the gap, pointing to the “immense challenge of applying machines to any tasks that call for flexibility, judgment, or common sense.”6 Marshall might find relevant documents, but ask it to exercise some common sense and all you will get is a stare-down from some blinking lights.
Think back to your early legal research efforts. It was not sufficient to find cases talking about the role of prior art in an invalidity argument. You had to go much further. Even understanding the difference between a judge’s musings on prior art and invalidity and a judge hinging his or her decision on those thoughts was tricky, but doable. To a computer, that distinction does not exist.
AI development has reached the point where computers can successfully mimic some elementary associate behaviors. For example, current AI software can recognize letters after seeing only one example, pick faces out of a crowd, respond to spoken questions and commands, and drive cars. In some cases, AI can perform these and other tasks better than associates. Google’s self-driving system has driven Google cars millions of miles without causing an accident. Facial expression interpretation software can pick up flickers of emotion on a person’s face that humans miss.
But computers still cannot do many things associates excel at doing and could do even as young children. AI cannot work collaboratively with associates. While AI can attribute physical acts to emotions (it can read your face), it cannot empathize with an associate. AI has difficulty being flexible, such as knowing when to defer to an associate even though the result may be “suboptimal.” When it comes to law, however, doing research, writing persuasive briefs, and counseling clients require these and many other skills.
This confusion about what computers can do drives us to overestimate what role computers can play in the delivery of legal services. As Autor says, “journalists and expert commentators overstate the extent of machine substitution for human labor and ignore the strong complementarities that increase productivity, raise earnings, and augment demand for skilled labor.”7 The point they miss, according to Autor, is that “[t]asks that cannot be substituted by computerization are generally complemented by it.”8
The Goldilocks Zone
While some day Marshall may be able to take over an associate’s role, it is more likely today to find a home in the Goldilocks zone. The future of lawyers and technology will not be one extreme or the other, but a combination of the two working together. Think of this role as augmented human intelligence.
Thomas Davenport and Julia Kirby wrote about augmentation, reframing the question many ask—what parts of my job will computers take away by doing them more cheaply and rapidly—as a different question—what might I be able to achieve if a computer assisted me?9 It is this augmentation where lawyers could benefit from Marshall.
Davenport and Kirby describe five basic augmentation strategies people can adopt: Step Up, Step Aside, Step In, Step Narrowly, and Step Forward. Step Up is for those who enjoy big-picture thinking. You are the strategist, and your ability to see above and beyond the trees puts you in a place no computer can match. Step Aside is for those who see computers as complementary to what they do. This step is for those who view computers as a force multiplier of their work. It is the Goldilocks zone. Step In is for lawyers who also love to code. Someone must write those algorithms. Step Narrowly is for lawyers who have found niches. No one will spend the time to train AI on fashion IP law. Step Forward is for the entrepreneur. You belong here if you want to create that program that will draft patent applications.
For most lawyers, and for new associates, Step Aside is the best strategy. They will find ways to work with computers and provide superior client service. Clients will benefit from the efficiency and power of the computer component, and from the insight, judgment, and collaborative side of the human component. The combination, which will improve over time as each component learns to work with the other, will drive solutions that neither lawyers nor clients can imagine today.
We do not have to look far to see how augmentation already helps IP attorneys. E-discovery algorithms help computers sift through millions of pages looking for potentially relevant information. But, once the computer identifies which documents to produce, humans review the relevant documents and decide how to use them in the lawsuit. Strategists combine the power of Lex Machina’s database and search functions with the judgment and common sense of experienced IP litigators. Transactional IP attorneys use Kira Systems to pull information from thousands of licenses, after which lawyers focus on the few with clauses that could affect a client’s perception of deal risk.
Better to Lead Than Be Left Behind
Now that you understand the reality of AI, you may ask whether it is too early to bring computers onto your team. The greatest danger to lawyers is not AI—it is the complacency of the lawyers themselves. Computers will advance, and hiding from them will not slow their diffusion into the practice of law. Clients have moved past exclusively relying on lawyers to solve problems. Today, clients find options to handle many tasks they once gave to lawyers. The alternatives they use are cheaper, faster, and higher quality. But they generally involve humans augmented by computers. Managed service providers, for example, take many types of work away from traditional lawyers using combinations of technology and lower-cost labor.
IP lawyers are well positioned to take advantage of clients’ interests in augmentation solutions and should not dismiss this advantage too quickly. Already comfortable with technology, IP lawyers can find solutions in the Goldilocks zone for practicing law and at the same time address clients’ concerns about efficiency and quality. By incorporating what computers can offer today into the routines of legal service delivery, IP lawyers can demonstrate they are focused on addressing clients’ legal and business concerns. IP lawyers also can position themselves to stay current as AI evolves.
Preparing for an AI Future
Futurists had the timing wrong after Watson’s Jeopardy triumph, but they had the direction right. AI will grow in capability and power over the next decade. Lawyers, like other knowledge workers, will increasingly share their space with the silicon interlopers. Fighting computers taking over tasks lawyers have done is like battling time—you will not win. Embracing computers and using their power to augment your practice will show clients that your goals align with their interests and give you an advantage in an increasingly competitive market.
1. Jo Best, IBM Watson: The Inside Story of How the Jeopardy-Winning Supercomputer Was Born, and What It Wants to Do Next, TechRepublic (Sept. 2013), http://www.techrepublic.com/article/ibm-watson-the-inside-story-of-how-the-jeopardy-winning-supercomputer-was-born-and-what-it-wants-to-do-next/.
2. See Richard Susskind & Daniel Susskind, The Future of the Professions (2015).
3. Stuart Armstrong, Smarter Than Us: The Rise of Machine Intelligence (2014).
4. Joscha Bach, Everything Is Computation, in 2016: What Do You Consider the Most Interesting Recent [Scientific] News? What Makes It Important? (Edge 2016), https://www.edge.org/response-detail/26733.