What AI Are We Talking About?
In business today, AI is a shorthand used to refer to technological processes that automate services. Attorneys will generally encounter two kinds of AI: reactive and limited memory. Reactive AIs respond to human input using predetermined algorithms, like playing chess against a computer. Limited memory AIs rely on both preprogrammed inputs and the AI’s own observations over time, like self-driving cars, natural language processors (e.g., Siri), and machine learning. The most popular among them, and the most disruptive for the legal industry, is machine learning.
Machine learning has no standard definition. For this article, the holistic definition offered by Daniel Faggella of TechEmergence proves useful: “Machine Learning is the science of getting computers to learn and act like humans do, and improve their learning over time in autonomous fashion, by feeding them data and information in the form of observations and real-world interactions.”1 Importantly, machine learning represents a significant change from traditional software because it uses inductive rather than deductive reasoning.
Most software today is deductive or “rules-based,” meaning that programmers create the software to perform functions using a predetermined set of rules. Although these rules may be extremely complex, their utility is limited because they can never change without additional human interaction. A familiar example to lawyers is Microsoft Word’s spell check function, on which we cannot exclusively rely to catch errors before sending out a document. Specifically, while spell check may catch those misspelled words that the programmers told it to identify, it monumentally struggles with the nuances of ideas, grammar, and whether a correctly spelled word is truly the word the author meant. This is because human programmers cannot possibly know, let alone write, rules to contain the correct meaning or interpretation of all variations of all words in all Word documents across all time.
By contrast, machine learning is inductive, meaning that once it is fed a sufficiently large amount of information (like 1,000,000 Word documents), it can intuit the rules that will most likely apply to any one Word document. For example, an inductive spell check would adapt to each user’s writing style and provide recommendations based on not only the English language, but also that particular user’s style within it. Additionally, it can go one step further and not just edit a Word document but create one based on what the attorney and client require. That is, because the machine is already familiar with the structure of the requested document, all it needs are the terms the client must have to develop a complete document.
E-discovery tools provide a useful example of the progress from deductive to inductive AI in the legal sphere. The first e-discovery tools required attorneys to manually instruct their document review software as to the exact terms or ideas that were relevant and not. The software then scanned all the documents and produced only those results that were predetermined by the attorneys to be relevant. Despite cutting down the attorney hours necessary to review the discovery, the software still required significant time and expense at the beginning and end of the process. E-discovery tools now, however, can utilize machine learning technology so that the software can produce far more reliable keyword results, in rank order of relevance, with much less human input. Once the machine knows what the lawyer is hoping to find in the discovery (from both the human input and induction), it will automatically determine all related keyword mentions, as well as their likelihood to be germane to the lawyer’s investigation. More importantly, the software is constantly learning so that it can take the lessons from its first use and apply them to all subsequent uses.
The unreflected thought of most attorneys when they hear about how machine learning can remove the most tedious aspects of e-discovery or contract drafting is that of excitement and relief. Indeed, not only are these tasks among the most arduous and boring, but sloppiness at any moment during their execution could result in malpractice. However, as machine learning is applied to these tasks and becomes better each time, to the point where it is faster and more accurate than most attorneys, three related problems occur that will adversely impact young attorneys. First, the number of total hours billed to each matter for the same work product will decline. Second, because it will take fewer hours to produce the same work, law firms will hire fewer associates but require them to bill more than previous associates. Third, it will remove the traditional stepping stone on the attorney career path of “cutting your teeth” by spending all those hours on tedious assignments like reviewing discovery, conducting legal research, and drafting briefs, contracts, and government filings.
The first threat is by far the most immediate to the law firm, and why an equity partner should be most worried. Most law firms tend to rely on their associates to generate revenue for the firm at a rate of at least two times the law firm’s expense of employing the associate. Thus, the fewer associates, the smaller the amount of revenue the law firm generates from the group. This problem will collide with the preexisting push from outside the legal industry to develop alternative fee structures. Although there may be no current replacement for an experienced attorney’s strategic advice, it takes that seasoned attorney a small amount of time to provide his or her guidance when compared to the large amount of time it takes an associate to perform the tedious legal tasks.
Second, law firms, which are generally not known for being particularly flexible or dynamic, will most likely handle this problem by adopting the technology slowly but maintaining the same structure. This will mean that, assuming the firm’s deal flow does not dramatically increase, the same amount of work can be performed by fewer associates, resulting in smaller associate class sizes. Correspondingly, the associates the firm does hire must work more billable hours so that the firm can mitigate the losses incurred with a smaller associate class.
The third threat is the most insidious because it will happen gradually and have the most effect on the practice of law. As AI becomes more integrated into business generally, clients will expect that law firms will deploy AI to increase efficiency and likewise decrease the client’s bill. Similarly, law firms will see the potential of AI for reducing the turnaround time for their clients. These advantages will propel the use of AI in the firm. However, as associates begin using the tools, they will find less need to review the various templates the firm has, read cases, weigh discovery, or review source documents firsthand. While the associate will initially be grateful for this, the associate will be at a longer-term disadvantage. Much like how most people now remember fewer phone numbers, so too will associates remember fewer clauses, fewer cases, and fewer idiosyncrasies, and have much less exposure to the fundamentals of their practice. Thus, by the time they become equity partners, they will not have had the same experience as the equity partners of today.
Why It’s Not the End of Legal Practice
Although it threatens some of the most venerable ideas in legal education and practice, AI probably does not mean the end of the existing, large legal industry. Much like robots to manufacturing, the total number of attorneys will decrease, but a large set will remain whose daily work will be supervisory to the automated tasks. The majority of these lawyers will be adapting to the new reality by embracing the drawbacks of AI. Broadly speaking, the two drawbacks to consider are (1) the black box problem of AI and (2) that AI is ultimately just a tool.
The black box problem of AI is this: While AI can learn and improve services as it processes more data, no programmer can determine how the AI arrived at a specific output. We know the input and the output, but not what happened in the middle, or the “black box.” In the legal context, the black box problem requires that there be a lawyer reviewing the data going in and the data coming out of an AI program. For example, if an attorney asked the AI to draft a license agreement, it could put together an agreement that meets the parameters for a specific client, but the attorney would not know how the AI arrived at any one decision. Thus, if a dispute occurs with the agreement, no court could ascertain the parties’ intent because no person would know why the clause in dispute was drafted exactly the way it was. The same is true of using AI for government filings, patent claims, and discovery review.
Because the black box issue is fundamental to how inductive machine learning software works, it is unlikely to be solved in the foreseeable future. Thus, attorneys will continue to provide value by identifying and understanding the intent behind the data going in, reviewing the results, and making necessary changes and recommendations. Because junior attorneys will ostensibly have more time to perform these actions, there should be a corresponding shift from an emphasis on drafting technique to understanding the client’s intention and the new framework in which legal advice is needed.
The second drawback applies equally to AI and all emerging technologies: no matter how impressive, efficient, progressive, or knowledgeable, the technology is just a tool. Technology excels in reaching a destination, but struggles in determining what the destination should be. It is the attorney, not the machine, who can connect a client’s intent to a desired outcome. Because AI can provide off-the-shelf solutions, thrifty clients will find it cheaper to use the AI program themselves than to consult an attorney. In short order, there will be AI solutions that are merely the perfected version of LegalZoom. However, a client will still need to know and understand how the automated legal work advances (or detracts from) their stated goals. Therefore, not only will a lawyer still be necessary to review and analyze the results from automated legal work, but the lawyer will also be necessary to guide a client to a desired legal outcome.
What to Prepare for and How to Prepare
Become Technologically Proficient
As AI tools become more advanced and integrated, young attorneys should prepare for this exciting, even if worrisome, legal landscape. It is too much to assume that the traditional legal paths will be obsolete, but they will become far less effective and be available to far fewer attorneys. A young attorney should prepare by becoming familiar with technology and how it works, develop a nimble approach to new tools, embrace the greater advisory role, and grow with the new legal culture these changes will precipitate.
Too often attorneys are criticized for or find cover in being technologically unsophisticated. This attitude will change. Both to understand a client’s business and to understand the benefits and drawbacks of software tools, the attorney will need to know how they work, where errors are likely, and how to discuss these tools with more technologically sophisticated professionals, like developers. The appropriate analogue here is legal writing. Although not all legal practice requires excellent drafting skills, all attorneys need to possess some level of writing proficiency. In short, lawyers will need to become technologically proficient.
With intimate knowledge of the available AI tool, an attorney will be able to select the best tool and the best way to use the tool for each client and matter. Working in tandem with the appropriate AI tools, the attorney will develop greater knowledge of fitting AI work product to client needs. As this process is repeated, the technologically proficient attorney will be able to wield legal acumen traditionally reserved for far more experienced attorneys.
Focus More on Strategy
Next, young attorneys should focus on the human element of legal practice rather than the technical. By definition, AI and computers are much more capable to perform the technical work. In fact, as machine learning algorithms evolve and become better through routine use, it is likely that they will in short order be better than most attorneys at the technical aspects of the law. Importantly, however, attorney input will still be needed to relate the client’s goals to the machine’s tasks. Broadly, this shift means emphasizing strategic guidance over tactical performance.
Attorneys typically are trained and accustomed to tactical performance. Starting in law school, a client or professor presents the lawyer with an issue; the attorney then determines what law applies, applies the law to the given facts, and makes a reasoned and informed recommendation. Unfortunately for us, this whole process is very easily automated. Even if the AI tool’s analysis or results are not as good as the attorney’s, economics dictate that the AI tool will be the preferred option.
Strategic guidance, in contrast, means being familiar with the law as a body rather than one discrete part. Indeed, the current emphasis on hyperspecialization is exactly the wrong approach for embracing AI and emerging technology generally. A lawyer might spend 20 years developing a valuable and unique specialization, but that specialization can be automated by rules-based or machine learning programs at any moment. A lawyer skilled in strategic guidance will still focus on one area of the law, but will be conversant in all aspects of that area and possess an eye to the other legal (and nonlegal) arenas that may be implicated. For example, the patent attorney who knows the interplay of trademarks, copyrights, trade secrets, and other more traditional business issues surrounding a requested patent is much better positioned for the future than the one who knows every case and regulation surrounding a particular kind of patent. Because the tactical work will be progressively automated, understanding why a client wants something, being able to see the various issues that will arise, and developing comprehensive strategies that connect the client’s goal with the machine’s tasks will increasingly become the measure of competent and valuable legal work.
Become More “Business-Minded”
Last, a new legal culture will arise from the shift in direction from hyperspecialization to a broader view. Because clients will want their attorneys to be more understanding and aware of their business needs to better apply AI to their legal needs, attorneys will need to expand their outlook to include what traditionally has been outside the scope of legal work—business decision-making. The AI may perform all the legal tasks, but an attorney will need to understand their weaknesses and what the client’s intent and goals are, and be able to make recommendations against and changes to the AI legal product. These changes to the AI’s work product will come from the lawyer’s legal knowledge, his or her knowledge of the intentions and proclivities of the involved parties, the customs and practices of the industry, and the parties’ course of conduct and prior dealings. Relating the human ideas to the automated processes will be invaluable and irreplaceable.
The acceleration of these technologies can be neither understated nor stopped. Law librarians and paralegals are already feeling the squeeze of technological encroachment, and associates are starting to feel the pinch as well. All attorneys, experienced and inexperienced, must grapple with AI as it becomes interwoven with the legal services industry. Law and legal practice have struggled to keep pace with the Internet, big data, and the digitization of information and processes. These issues will only become more exacerbated by the accelerating pace of technological development. Because young attorneys grew up with these technological advancements, they are best situated to grow with AI as it disrupts the legal industry. The well-prepared young attorney will be nimble, technologically sophisticated, and expert at relating human goals to machine tasks.
1. Daniel Faggella, What Is Machine Learning?, TechEmergence, https://www.techemergence.com/what-is-machine-learning/ (last updated Nov. 30, 2018).