You may not think about it this way, but every legal practice uses Artificial Intelligence (AI) and Natural Language Processing (NLP). Spellcheck is an AI tool using NLP to understand and improve written communication. When released, spellcheck was a cutting-edge tool born out of Stanford’s Artificial Intelligence Lab. So why do we view some tools as AI, and regard others as commonplace? Why do we raise our expectations of what AI can do, but at the same time grow skeptical of AI’s potential?
It turns out that what we consider AI is continuously in flux. So much so, that AI researchers coined the term “The AI Effect” to explain humanity’s collective tendency to move the goalpost. The AI Effect describes how after widespread adoption of a technology tool, users suddenly stop attributing intelligence to it, and it becomes an essential part of “normal” computing.
Machine Learning and Deep Learning
Machine learning and deep learning are two prominent words in the AI lexicon, and both relate to NLP and applications for legal practice.
Machine learning is the application of human-designed representations and features to create a predictive model. In this process, a machine learning expert tediously engineers features for the computer to learn—over many iterations. After training, the model can interpret new data and make predictions. Think of features and representations as rules that describe a concept. For example, phone numbers have characteristics we can easily identify. We know the area code goes first and to use a plus sign and country code for international calls.
Deep learning is a subset and the next evolution of machine learning. A developer instead directs a computer to discover patterns inherent in data. Deep learning implements multiple layers of intermediate representations to automatically generate features in data. With our phone number example, a deep learning algorithm could discover the different formats and components of phone numbers automatically. This approach, however, requires a larger amount of data to drive the discovery of patterns.
Natural Language Processing
NLP relies heavily on machine learning or deep learning, and each approach has benefits depending on the use case. Machine learning-based NLP holds a prominent place in legal practice as many firms have leveraged the technology by adopting e-discovery tools.
At a fundamental level, attorneys and NLP practitioners both deal with language. The legal profession is language-obsessed. Each word is purposeful, designed for communicating symbols and context unique to the law. NLP is fascinating due to its interdisciplinary nature and application. Developers need subject matter experts to assist in the development and fine-tuning of NLP tools. With increased awareness and growing interest, I hope that every lawyer will embrace AI and NLP as an essential tool for legal practice and contribute to the growth of machines that “think like a lawyer”—or help lawyers think.
It has taken decades of research to develop machines that understand and generate language reliably, particularly because language is imprecise and incomplete. Communicating with language is efficient because we fill in the blanks. We are able to assume relationships and definitions based on our collective experience communicating over our lifetimes.
Take, for example, the headline, “Juvenile court to try shooting defendant.” The phrase has multiple meanings. Does it mean that a young court is trying to shoot a defendant? Unlikely. Most would interpret the sentence to say that a court for juveniles will conduct a trial to determine the fate of a defendant, who is a minor, for an incident involving firearms. To correctly parse legal documents, machine learning experts have to identify, test, and hone models specifically for legal language. This is an impossible task without help from lawyers. Feature selection in machine learning is time-intensive and specialized. However, recent changes are now making NLP more accessible for lawyers and developers.
Intelligent Machines Have Arrived
The explosion of developments in AI technology and NLP this decade is a result of three factors: advances in GPU computational power, data availability, and the democratization of AI.
What is a GPU? The initialism stands for “graphical processing unit.” A GPU is similar to the CPU in your laptop. Recently, GPU manufacturers, most notably Nvidia and Google, recognized the need to develop chips for AI. Self-driving cars and a large percentage of what we consider as AI can be quickly developed on this newly designed hardware.
Second, massive datasets have become readily available. Our lives and activities have become data. Before smartphones and online shopping, it was not easy to amass a dataset for study. Data collection was rigorous and meant time spent in the field counting or sending out surveys. Today, every click, transaction, and movement you make is recorded by an internet connected device and stored into a dataset that will be used to improve profits and services. Data is valuable, and the legal profession can benefit from implementing an ethical data strategy to collect information not protected by attorney-client privilege.
Additionally, vast amounts of data are now publicly available to all on the internet. While writing this article, Google announced a dataset search engine to assist developers in finding data. If you want to know the historical price for, let’s say, Bitcoin, all you need to do is search for the term, and you will find hundreds of datasets available for download and analysis.
AI tools, data availability and GPU hardware have lowered the barriers to entry. It is now possible to create an AI startup with standard tools, frameworks, and public data. Google, Amazon, IBM, and Microsoft have released an extensive collection of tools tailor-made for AI, analytics and data science. This proprietary technology is worth billions of dollars but is available for free as an add-on for their cloud services.
In sum, with limited resources and a few years of technical expertise, it is now possible to construct impactful, creative tools for the legal profession that leverage the technological expertise of companies like Google or Amazon.
Current Tools Lead to Future Profits
More disruption is to come, notably from outside the legal profession. Atrium, a law firm founded by Silicon Valley veterans, is looking to challenge the traditional law firm business model. Atrium applies machine learning to save time and provides predictable and transparent pricing. Atrium raised $65 million on September 10th, 2018 and is a prime example of success when law and technology join forces.
Lawyers are capable of innovating as well and are in a better position to apply tools to practice. Attorneys do not question the value of the AI technologies that check spelling and grammar, or e-discovery products that sift through documents. Innovative lawyers are using machine learning today, and they are not keeping it a secret. Leading the charge in the democratization and dissemination of open-source machine learning-based NLP for the legal market is LexPredict. Directed by Michael J. Bommarito and Daniel M. Katz, LexPredict has become a market leader in legal analytics and NLP tailored for legal workflows.
Lawyers are not programmers, and programmers are not lawyers. A cross-pollination of disciplines is required to push innovation in a meaningful way. While machine learning-based NLP dominates legal practice today, consider the possibilities of deep learning NLP to support firm operations and enrich client interactions. However, deep learning requires vast amounts of data to achieve results, more data than most law firms generate. Law firms will need to think creatively outside the scope of legal practice to bolster other parts of practice unaffected by the ethical rules. Virtual assistants could automate client interactions while providing marketing analytics to gauge impacts in spending, to name a few.
In a decade, a once-obscure revolution, whose battles were fought among computer science academics, has crept into our everyday lives and have invaded your legal practice. The battles between algorithms continue, and the winners leap from the lab to our lives. Let’s break away from the AI Effect and approach challenges today. It is now time to recognize AI and NLP as essential tools and think about applications for every component in the business of law.
About the Author
Vahid Wafapoor is a data researcher & analyst for Legistorm. After completing law school, Vahid learned how to code and launched Cultura Analytics, a startup focused on immigration form automation. Contact him on Twitter @CulturaAnalytic.