This year, I went to the 2023 American Association of Law Libraries Annual Meeting with a specific goal in mind. I wanted to write a short memo on the state of AI in legal technology, focusing on the products that are coming down the pike. I was able to catch up with a number of legal research vendors about new technology, specifically legal AI products that are in development. Now that I’ve sent that report to my managers, I’m happy to share my observations with GLL as well.
It is clear that recent advances in Artificial Intelligence (AI), powered by the technology behind ChatGPT4, will deeply affect the legal industry. AI develops in fits and starts, and the leap from the tagging model most legal research databases currently use, to a Large Language Model (LLM) neural network AI, is an exponential improvement in processing, synthesizing, and understanding written language. Even the developers were surprised by the complexity and ability of this next generation of LLM AI.
Although it was clear that this generation of AI will change the legal industry, it wasn’t clear how that would happen, and in what ways. Because these technologies are so new, most of the hands-on experience my professional colleagues and presenters had was with free web-based AI products, which are not suitable for legal work. These free products are prone to “hallucinations” (aka “making up sources”) and it’s unclear how they use the data fed into their portals, leading to data governance issues. These drawbacks can be controlled for by developers, and by sales contracts. However, it will only be controlled for by pay-for-use products specifically designed for the legal market. As always, it is good to keep in mind that old adage “if something online is free, then you are the product.”
While I don’t think the free AI online is useful, the developing products from major legal research platforms show great promise. These paid products control for issues like hallucinations, and provide citations supporting their work so a researcher can confirm the accuracy and context of the materials the AI is pulling from. Issues surrounding data governance (what the company does with your uploaded material and search history) can be controlled by contract, and the legal vendors understand that this is a concern for most legal clients.
Over the course of this year, most major legal research platforms will create programs using this GPT-enabled, LLM AI technology, with plans to roll them out this fall. Where will this rush to generate new AI technology will take us? We’ll see at the next AALL meeting in Chicago!
For now, after reviewing the products, I’m excited about the future of AI legal technology. However, my recommendation to my own court was to take a measured approach toward adoption. The courts don’t need to invest in these new technologies now, while they’re still in the development stage. I estimate that we’ll want access to these products after they’ve penetrated the legal market and new hires are accustomed to using them, which I expect will take about three to five years. I think the technology will probably be wrapped into the existing subscriptions we have by the end of that multi-year period, and most of the bugs, risks, and implications will be evident by then.
However, it is critical that we learn about them now, and understand the impact they are having on the legal field. From what I’ve seen, I believe the most significant uses of this new technology will be in document review, drafting, and processing. It really is a massive time saver in drafting materials. AI will save time on legal research, but still won’t replace current tech completely, or solve all current problems. It will always need a human hand to guide it, and a human brain to sort through the context and make sure additional facts, laws, secondary sources, and other information aren’t lurking outside of the database that the AI exists to curate.
There are some downsides to this technology. This evolution will drive up the prices for legal products, as these LLMs are very expensive to build and run, using a huge number of computer chips and a lot of energy. In most firm settings, I think this will be offset by savings on labor costs, as AI may replace a number of associate attorney positions, consolidating the work to a fewer number of bodies. In the courts, I don’t think the labor savings will be as significant, as AI cannot replace the bulk of our court drafting, reading, and analysis. It may be helpful for the drafting process, in particular, which would be helpful for our judicial branch’s ever-increasing workload.
2023, Current state of the market and the major players
CoCounsel from Casetext
Pablo Arrendondo and Jake Heller, the founders of Casetext, saw a preview of what ChatGPT 4 could do last year. Immediately after that presentation, they pivoted their entire company to be the first to use it for legal applications. Butthey had this early access to the technology, they are at the forefront of this wave of AI products. You may remember seeing the news earlier this year when CoCounsel (using GPT-4) passed the bar! While their AI is excellent, their database (Casetext) is not as robust as any of the other online legal researchdatabases, so their technology is currently pulling primarily from black letter law. (Three Geeks and a Law Blog Podcast, Revolutionizing Legal Practice). The reception to this product among practitioners has been uniformly positive, and in fact, it’s running very slow at the moment because they’re scaling faster
than they can get computer chips.
It’s not about replacing the human that makes the decision, it’s about replacing the write-off hours sending a junior off to slap together drafts. And like that junior, the AI isn’t going to know everything. Unlike that junior, Co-Counsel is going to transparently provide the attorney with a full grasp of the bases of what it does and does not know.” (Patrice, Legal AI Knows What It Doesn’t Know Which Makes It Most Intelligent Artificial Intelligence Of All).
While Casetext had early access to ChatGPT technology (they’re a Stanford CodeX company with strong ties to other innovators in Silicon Valley), other databases were moving forward with more advanced tagging technology. While practitioners were excited to see this product, I believe the industry reaction has been more along the lines of panic. Casetext (and therefore, CoCounsel) has been offered a bid for 650 million dollars by Thomson Reuters, the parent of Westlaw, in order to get this technology and to get a jump on applying it to legal materials. (Ambrogi, As Thomson Reuters Explains Its Acquisition of Casetext, Some Investors Seem Uncertain).
Once CoCounsel has access to all of the secondary sources, indexing, key number systems, and other resources on Westlaw, I believe it will be truly impressive. I saw Pablo Arrendondo speak at AALL 2023, and while he seemed happy about the deal, he was incandescent about the chance to get his hands on West’s data. (AALL 2023, Live Vendor Presentation). I don’t see any reason the merger/buyout won’t proceed, and I’m excited to see what will result from this partnership.
Westlaw recently launched a new platform called “Westlaw Precision” which features a more granular system of tagging information. “Westlaw Precision is tackling the gnarly and perennial problem of language ambiguity by doubling down on taxonomy.” (O’Grady, Westlaw Precision Launches With Promise to Cut Lawyer Research Time in Half).
The librarians at our court recently reviewed Precision to see if its upgrades are worth the purchase price. While we are still coming to a decision on that, I’ll note that the capabilities of Precision as it stands are very good. TR/Westlaw is about to shell out an astonishing amount of money to acquire Casetext/CoCounsel, and plans to integrate a Chat-style AI into Precision in the near future (Fall/Winter 2023).
LexisNexis has been working with an earlier version of neural net technology to create Fact & Issue Finder, and is creating an AI tool called Lexis+ AI. Lexis+ AI is a similar product to CoCounsel, in that it is a chatbot that provides custom answers with documented sources. We had a sneak peek at this technology at AALL, and it seems very good! My understanding is that for now, it will be sold as a separate product. The platform we currently use at my court, Lexis+, has already integrated large language model (LLM) functionality via Microsoft- backed OpenAI’s ChatGPT (but they haven’t disclosed what version). New Development as of August 2nd: “Today LexisNexis is announcing that their soon-to-be-launched AI-powered solutions will be integrated into Microsoft 365. The product integrations include– Lexis+®, Lexis ® Connect, Lexis ® Create, and Lexis ® Create Plugin with Microsoft 365 Copilot. Both companies are committed to providing legal professionals with “intelligent, AI-powered tools, enhanced capabilities, and more efficient, guided workflow experiences” inside the suite of Microsoft products where they currently work: Microsoft Outlook, Word, andTeams. Back in May Lexis announced two AI initiatives a commercial preview and the AI Insider Program as well as plans for incorporating generative AI into their product. Lexis announced its commitment to launching GPT-enabled solutions with the Microsoft Azure Open AI Service which has already been rolled out to customers via Lexis+.” (O’Grady, LexisNexis Announces Integration with Microsoft365 Copilot/Generative AI Solutions).
Bloomberg has not yet committed to a firm timeline for their new legal AI product. Their representatives indicated that something could be coming late this year, and it may be based on the Bloomberg Finance AI, which they say is very impressive. In the meantime, it appears they’re still using the information from this 2019 article “How Bloomberg Law (BLAW) Uses AI and Machine Learning to Prove its Case.”
Editor’s note – This article republished with the author’s permission with first publication in Newsletter of the Government Law Libraries Special Interest Section of the American Association of Law Libraries, Fall 2023. Volume 49 Issue 3 GLL News.