Artificial Intelligence

 

As digitalization penetrates nearly every facet of our lives, the legal industry faces some monumental challenges. The massive amounts of data that lawyers need to search, navigate and absorb—including but not limited to legal evidence, litigation data, legal source materials, background research documents, practice area guidance and more—add complexity to even the most mundane tasks.

The sheer volume of information is unprecedented, and the size of the task is “machine scale,” which is to say that it is well beyond human capabilities. To keep pace, we need increasingly powerful and sophisticated algorithms to mine the data, organize it and identify meaningful patterns. These are big challenges, but the payoff can be significant. When AI is deployed appropriately with proper oversight, it helps us make connections we couldn't see before, leading us to new legal and business insights, and providing quick and accurate answers to the questions we have as we try to solve legal problems. But teaching machines how to interpret “legalese” is nearly as challenging as the task it is trying to solve.

The Promise and Pain of Teaching AI

AI needs help from human experts. AI technologies are not entirely autonomous. They need a representative body of data from which to learn – the more the better. But first that data needs to be normalized, structured and accurately labeled or tagged. AI works by finding patterns in data, but “dirty,” inconsistent data can get in the way, and it can teach your system to “learn” the wrong things. Normalizing data typically requires a lot of data analysis before machine learning and other techniques can be successful. In fact, effective AI still relies on regular intervention by human data scientists and legal domain experts who iteratively “teach” machines to make better decisions.

Teaching AI legalese is hard. There is a lot of nuance in legal language, much of which comes in the form of rhetorical persuasion and argument. It can also be hard for machines to distinguish between straight-up “facts” and “legal facts.” Many legal terms are in Latin, and there are countless domain-specific terms and plenty of jargon and abbreviations. Language and terminology varies from one practice area to another. Finally, discerning the specific context in which specific words or passages appear is a huge challenge for AI. With the help of new tools like BERT, we are making big strides in this area, but there is plenty of room for us to get better.

Teaching AI to generate meaningful and relevant insights from data is an area with tremendous promise. We are already making extensive use of natural language processing (NLP) and machine learning (ML) in legal technology to extract useful information from unstructured text. Legal analytics, for example, mines litigation data from sources like PACER so lawyers can identify patterns in the behavior of judges, parties, opposing attorneys and expert witnesses to make better strategic decisions in cases similar to their own. Other technologies can find and extract the specific language and case citations that judges rely upon the most, allowing lawyers to insert them into their own arguments to make them more appealing and persuasive.

The Next Steps for AI

The next AI frontier in the legal realm will probably be highly sophisticated chat bots. These are not the rules-based bots you see performing basic customer services functions on many websites. More advanced legal chatbots leverage NLP and ML to engage users in question-answer dialogues. The same AI models we use to extract text can be used to help us identify answers to legal research questions that attorneys are likely to find useful. We begin by finding answers, and then we label them and generate questions that fit the answers. So far this is being accomplished mostly through a supervised learning process. The key will be to get to the point where machines can also generate the questions with little or no assistance.

Adding a voice layer to human-machine interactions magnifies the challenges. Siri and Alexa demonstrate the potential utility of voice recognition, but there are lots of difficulties in the legal domain—and little tolerance for error. The voice layer of a chatbot is essentially a “translation” challenge. Spoken case citations, abbreviations, proper names, case names and, of course, the speaker's particular dialect or accent are all difficult problems to solve. That said, we are not far from having intelligent agents that can understand spoken legal language, provide accurate answers to an attorney's questions and personalize that experience—by deciphering intention, anticipating lines of inquiry and even making proactive suggestions based on the individual's behavior over time.

AI solutions that succeed in the long-term will need to carefully address ethical, as well as technical, challenges. Privacy issues loom large any time you are dealing with data. The contractual and regulatory compliance requirements of law firms and legal departments are complex, and any business offering powerful technology solutions must understand that protecting the confidentiality of clients' data is a fundamental, existential responsibility. Technology companies with long-term experience and proven expertise in safely handling customer data will necessarily have an edge.

Another ethical challenge related to AI is avoiding bias. Is the training data itself manifesting a set of behaviors that we wish to avoid? Is there bias in the data? Where did the data come from? Is my team sufficiently diverse and aware of their own potential for unconscious bias? The more diverse the team in terms of background, ethnicity and gender, and the more sensitive they are to the issue of bias, the less likely they are to introduce it, and the more likely they are to recognize it.

Where AI Fits Into Legal

AI is poised to supplement lawyers, not replace them. Practicing law involves tremendous domain expertise, and lawyers must be able to maintain their grasp on large amounts of information, keep up on changes, and understand how those changes apply and should be interpreted in new contexts. While machines are much better and efficient at processing information and identifying patterns, it's best to think of AI as an enabling technology that provides robust assistance to support attorneys. Gradually, more intelligent tools and interfaces are becoming more accurate, and providing information that allows lawyers to focus on higher-value tasks such as advocacy, decision-making, and interaction with clients and witnesses. AI technology, like legal analytics and chatbots, can supplement those higher-value activities by anticipating what lawyers need to know and do, helping them contend with “machine scale” data volumes so they can produce better legal and business outcomes.

What does the future look like where AI speaks “legalese?” Will there ever come a time when lawyers will be able to ask machines complex legal questions, have them compile and analyze the information, visually display the relevant data and recommend a course of action? It's quite likely. Soon, advances in deep learning and natural language understanding technologies will enable machines to analyze and comprehend what is said in a meeting or conversation, helping us capture testimony during depositions, and record and encrypt other matter-related conversations.

We also envision voice applications assisting in case research. After “interviewing” the attorney to define and refine the search parameters, machines will be able to autonomously conduct extensive legal research and case law analysis, intuitively organize all data and documents related to the matter for easy review and retrieval, create “smart” legal documents that are “aware” of their contents and relationships with other documents so attorneys can formulate case strategy for pleadings.

What's further away is a machine that has true general intelligence. We can teach machines to do specific jobs, but they don't yet have a deep understanding of content and context. General intelligence requires machines to understand inferences and decisioning in a sophisticated way, like humans. When that happens, machines will be able to analyze case theory and strategies and use predictive modeling to determine the best legal approach for a given matter and illuminate potential outcomes.

Teaching AI to speak legalese is still a work in progress, but it is clearly visible on the horizon. However, no matter how fluent machines become in speaking like a lawyer, it will always be the voice of the attorney that clients and judges will want to hear.

 

Serena Wellen is a Senior Director of Product Management for LexisNexis and a former trial attorney. Her portfolio includes artificial intelligence, analytics and visualization capabilities for Lexis Advance as well as litigation research products for LexisNexis. She lives and works in San Francisco, CA.