It is difficult to read the news today without running across an article saying that AI will change everything. Is this also true of the legal profession? How will the practice of law and the provision of legal services change, if at all, in response to AI technology?
Some claim that we’re in the midst of an “AI apocalypse:” that every level of society will be massively disrupted. Although much of this press is hype and clickbait, the reality within the practice of law is that AI is indeed beginning to have a substantial impact. Ultimately, AI will be more disruptive to the legal profession than the move in the last century from typewriters to word processors. For this reason, organizations that don’t anticipate the changes being driven by AI today are likely to be left behind.
What is the source of this disruption? Simply put, AI software can analyze words: It can automatically classify and search for paragraphs, and compare documents and highlight changes. It can also learn over time from humans performing a task: AI/machine learning can use historical data to get started, then learn from human decisions going forward to continually improve its accuracy.
Multiple use cases
The most fertile area for AI in law to date is in handling boilerplate document completion and repetitive transactions. An example is the chatbot DoNotPay, called “the world’s first robot lawyer.” Originally intended to help users to fight parking tickets, it has today helped users contest more than 160,000 tickets across London and New York, all for free. DoNotPay’s capabilities have now extended to new use cases: It helps people sue Equifax in small-claims court and helps them identify when airplane ticket prices drop and obtain cash back. According to TechCrunch, the company is now “pushing out 1,000 new bots that can assist people in filling out transactional legal forms in all 50 U.S. states and the U.K.”
Another example: JPMorgan’s learning system, called COIN, short for COntract INtelligence. COIN analyzes a document in seconds, with fewer errors than humans, resulting in considerable cost savings. To date, it has replaced hundreds of thousands of human lawyer-hours in interpreting commercial loan agreements.
Artificial intelligence in context
Bill Fenwick is co-founder and currently Partner Emeritus of Fenwick and West LLP, which is based in Silicon Valley and has offices worldwide. Starting the firm that handled Apple’s incorporation and Facebook’s IPO, Fenwick has been a legal technology thought leader since the 1960’s.
Speaking with Bill last month, he told us that AI will impact many aspects of the legal profession. These include defining what constitutes the practice of law: lawyers’ rules of conduct, ethics, and malpractice, the duty to investigate, corporate law, taxation, and intellectual property. Says Fenwick: “With AI, it will be quite possible to instantaneously predict the likely outcome of most disputes. What will it do to the function of judges and the judiciary? What will it do to government revenue systems? We know now, regardless of what’s happening with the current administration, globalization is here to stay, and that involves hundreds of different countries’ legal, social, economic and revenue systems.”
A similar perspective comes from James A. Sherer, a partner in the New York office of BakerHostetler LLP, “…AI is already impacting current attorney practice in four discrete areas: (1) document review in e-discovery (“predictive coding” or technology-assisted review), (2) contract due diligence review in corporate transactions, (3) third-party legal research products in multiple practice areas, and (4) time entry and matter analysis.”
Consider (2): due diligence review. In the past, an attorney reviewing documentation for even a reasonably simple contract might need to consult previous similar contracts, searching for comparable language in the new document, as well as anomalous language or terminology. It’s likely the attorney may have access to tools capable of conducting some word and phrase searching through these different contracts, at least partially highlighting distinctions and similarities, but it will be a human doing almost all of the analysis. This could take anywhere from a few hours to several weeks, depending on the complexity of the documents and the issues being faced.
With AI, however, a document can be reviewed in a matter of seconds, recognizing important language and issues and also finding and identifying those distinctions and similarities based on the previous contractual language from which the machine has been trained. In addition, the machine can distinguish between language involving the different areas requiring analysis, e.g., liability and indemnification, taxation, intellectual property, and insurance issues.
Note that AI is not doing legal work for the attorney but is “pre-filtering:” identifying factual and contextual analysis that gives the attorney information and clues as to where issues may arise in downstream negotiations or activities. AI software, having been trained on previously vetted documents and their nominal language, can compare the current document with all that it was previously trained on and determine where anomalies occur, thereby reducing the attorney’s analysis time.
This “AI augmentation” arrangement―where the human and machine work hand-in-hand―is a proven pattern: It powered some of the earliest successful AI/machine vision systems in medicine. There, AI identified suspicious locations (possibly containing cancerous cells) on pathology slides, which drastically reduced the workload for humans, who only had to examine suspicious slides and slide regions, instead of reviewing them all. According to one speaker at the time, this was like “going from reading an encyclopedia Britannica every day to just reading a few pages.” Going forward, most legal use cases for AI will have a human in the loop in this way.
Keeping humans in the loop helps to mitigate the biggest challenge to AI in law. Says Fenwick: “the biggest problem with AI and the judicial system is the difficulty of creating the Trust (of citizens and institutions in its reliability) that will make the use of AI acceptable for dispute resolution.”
In addition to changing how legal practice is conducted, AI is disrupting the shape of the legal ecosystem itself. The Financial Times reports, “Change [from AI] is being driven not only by demand from clients but also by competition from accounting firms, which have begun to offer legal services and to use technology to do routine work. ‘Lawtech’ startups, often set up by ex-lawyers and so-called because they use technology to streamline or automate routine aspects of legal work, are a threat too.”
So what’s the best way to organize an AI in law project for success? Co-author of this article Pratt has led dozens of AI projects stretching over 30 years and has advised SAP Globalization Services with a project, called Law 2 Action, that is embodying a number of best practices, achieving great early results. The team’s experience illustrates some important principles that apply to most legal AI projects.
Law 2 Action helps product managers to update software based on regulatory changes. Today, SAP product managers in several countries must manually monitor government regulatory authority websites for changes in the law. For instance, in the UK, proposed regulatory changes in England, Ireland, Scotland, and Wales may all be different, and there may be tens of legal changes each day. SAP product managers must translate these updates into configuration files: an onerous process requiring many full-time analysts. SAP’s Law2code pilot was demonstrated to provide substantial work reduction for product managers, using the Romanian Ministry of Finance site as a demonstration.
Some surprising best practices from this and similar projects are as follows:
- Work closely with users. Understanding user needs, even taking a “design thinking” approach, is essential.
- Avoid “looking for solutions under the AI lamppost.” In the Law 2 Action project, substantial value came simply from developing a web scraping tool and integrating it with workflow management, to help product managers to organize their work. There is no AI in this process, but it still provides substantial value.
- Keep humans in the loop. Law 2 Action highlights important or anomalous text passages, which are organized for human analysts, who then make the final decisions on what is important and what is irrelevant. This simple bit of automation augments the knowledge and judgment of those who use it. As companies incorporate or comply with legal changes, this “AI Assistant” mode will dominate for many years to come.
- Don’t (necessarily) start with a data scientist. In the last few years, an explosion of open-source AI frameworks and training resources has substantially lowered the bar to creating a successful AI project. The Law 2 Action project was led by senior software engineers with a rich toolkit of skills, including basic software skills like the ability to build crawlers, websites, and production code. Just as you don’t need to understand the internal combustion engine to competently drive a vehicle, today you no longer need an AI PhD to obtain great results.
- Use the “virtuous cycle of continuous improvement”. Adaptive machine learning, combined with humans in-the-loop means that a system that initially produces a high false-positive rate (highlighting paragraphs that are not indeed of interest) can improve over time. Law 2 Action is designed such that humans provide feedback as to the system’s accuracy, and this information is fed into new machine learning models.
- Start with bad data. The virtuous cycle, above, implies a tipping point: It is more important to develop a valuable system that is used regularly and that improves itself over time than to worry about comprehensive, well-cleaned data at the start.
- Start with metrics. To ensure that the system is continuously improving, and to demonstrate its value (thus maximizing trust), it is essential to develop a metric for success and to continuously back-test it. This catalyzes system acceptance and validates the virtuous cycle approach to machine learning, above.
A framework for understanding the impact of AI on law
Another perspective on AI and law comes from Michele Colucci, a Silicon-valley based legal technology entrepreneur and CEO of Justiquity.com. The following diagram shows her view of how AI and related technology is changing the legal profession:
Colucci says that data and data analysis supports the higher-value levels in this “AI in Law” pyramid. Data can be used to support vertical expertise, which can help a defendant, plaintiff, or lawyer to be more effective. Examples are analyzing past cases to determine the best strategy for a case, such as the most favorable jurisdiction, attorney, court, or judge.
Complex legal reasoning, says Colucci, will always be the purview of humans, with their detailed knowledge of the world that is unlikely to be replicated by computers anytime in the near future. “Good lawyering,” says Colucci, “will always be the secret sauce separating us from the machines.”
So how can technology help? Decision Intelligence―the most advanced layer―provides the solution: A mechanism for combining historical information as captured in data and analyzed by AI, with models of the world provided by humans. Here, AI provides a component of the reasoning process in a decision.
This article has explored just a few of the many use cases for AI in law, along with a number of best practices gained from practical experience pushing this envelope. Today we have moved beyond questions of misleading AI hype to a certainty that this technology is in the midst of disrupting the profession, which is beset with massive complexity. It is up to us to shepherd the technology to provide the greatest benefit. BakerHostetler’s Sherer agrees it’s both needed and inevitable: “We are analog people dealing with an exponential growth curve, so something has to give.”
 Lorien Pratt, personal notes.