By Kristen M. Hallows, Bricker & Eckler LLP
Fastcase CEO Ed Walters has had enough with the magic and the unicorns and the hype surrounding artificial intelligence, or AI. He urged attendees at the American Association of Law Libraries (AALL) session, “Powered by AI, Built in the Law Library,” to think of AI like pivot tables in Excel: they’re just tools. They’re not magic, but they can be to those who don’t understand them.
He began by sharing a few hilarious examples of the limitations of AI. Is it a Shar-Pei, or is it soft serve? AI doesn’t know! It can’t differentiate between the two. And, whatever you do, don’t expect appealing names for paint colors from AI. Stoner Blue might seem appropriate for your teenager’s room, but can you imagine taking home a color sample by the name of Bank Butt? How about a light brown named Turdly?
So, AI is good at some things and not good at others. When it works, we stop calling it AI. You may not identify it as such, but AI is “baked into” some very common tools law firms and libraries probably use every day, such as spellcheck and Google Translate.
Ed refers to the first wave of AI, where we are currently, as “read only” AI. What’s coming is the second wave, which he calls “read/write” AI. It’s a much cooler phase in which we get to go from consumer AI to maker AI. Maker AI presents a new suite of tools that information professionals can use to provide more customized and actionable information to attorneys and firm administrators. Whereas traditional legal research services offer the same data to all users, maker AI lets information professionals create their own datasets and extract results unique to them. These results can provide insights to help structure alternative fee arrangements or to help inform litigation strategy or settlement decisions.
Take the Fastcase AI Sandbox. The AI Sandbox was designed to empower people. It’s a set of secure servers with datasets and metadata from Fastcase, coupled with an extensive suite of AI tools. Law firms or law schools can combine the Fastcase data with in-house data. Once you have your desired dataset, you can query it and get results out. For example, you can load a set of judicial opinions and get personality insights out–a judge’s preferences or tendencies. Using Docket Alarm’s new tool, you can create your own analytics on a subset of documents, such as mandamus petitions in Texas. Upload your own data and crunch it! And you can build your own apps with Neota Logic, rules-driven software with built-in decision tree logic.
Legal information professionals can drive this new read/write AI. Law librarians can build things with AI now, not just create reports, and some librarians are already doing it.
For confidentiality reasons, the panelist couldn’t show us exactly what they’re working on, but DLA Piper has think tanks for certain areas, such as litigation. For an initial project to build in-house solutions using machine learning and natural language processing (NLP), the librarians were tasked with obtaining the documents to feed the machine. The firm wanted to identify good sample documents, extract metadata from those documents, and create analytics. At first, library staff members were supplying documents to a partner involved in the project, but not really participating at a higher level, so they simply asked to attend meetings with data scientists so they could listen and learn. Eventually, they became fully integrated in the project, and their information professional skills are being utilized in metadata validation (because no machine is perfect).
Let’s face it, teaching the machine is laborious and time consuming. Data scientists need content experts (i.e., librarians) because while the attorneys have the expertise, they don’t have the time. The information professionals are able to serve as intermediaries, helping the data scientists better understand the legal documents and the utility of the data contained within them. They can also assist with validating results, and with providing feedback on search mechanisms for accessing the data and feedback on a mechanism for users to report system problems.
BYU uses Fastcase data to produce federal judge analytics; a tool called Kibana provides the data visualization. Who are the top five district court judges in terms of caseload? Which judge hears the most cases under NOS code 195? BYU can answer these and many other kinds of requests, and it’s excellent exposure for students to have access to AI tools.
Similar to Alexa, BakerHostetler’s Bankruptcy Bot thrives on questions. As an example, you could type, “Do you have Colliers’ e-book?” A link to the e-book collection would be provided; one day, Katherine Lowry, Director of Practice Services at BakerHostetler, would like the bot to be able to take the user directly to the item requested. The bot can also assist in finding attorneys at the firm who have experience in a certain area. Katherine and her team participate on legal innovation projects as part of the firm’s IncuBaker initiative. The team has come up with different chatbot versions, moving from a simple Q&A chatbot to a decision tree bot, as they work to develop a version that will best meet the firm’s information needs. Eventually, they plan to come up with a series of interlinking bots (bankruptcy, securities, tax).
Legal researchers at BakerHostetler also serve as subject matter experts or content specialists on project teams, and procure data to put into their “data lake.” This data lake contains internal firm information as well as third party information, such as details about products (i.e. books, forms) produced by outside vendors.
One interesting question from the audience had to do with the difference between training and programming. In supervised learning, a person indicates whether a question was answered correctly, and the machine “learns” accordingly. It’s an iterative process, as opposed to simply telling a computer what to do.
Ed closed by stating that prepackaged analytics are the new floor; using your own data is the new ceiling, which provides the competitive advantage. When embarking on an AI project, data scientists working for vendors can be helpful; they have tons of them! The key is for information professionals to get involved and start working on AI, for example, with APIs. Law firms can now provide information insights to clients as a new service or product, and information professionals can drive these information insights.
Read/Write: Artificial Intelligence Libraries—How Information Professionals Can Put AI Tools into Practice,” by Ed Walters, AALL Spectrum, September/October 2017