The End of “Search” is Upon Us!

We start 2024 with a bold statement for eDiscovery: The use of large language models (LLMs) and generative AI (GAI) will kill the use of keyword search in eDiscovery.

“No way!”, you say? I say yes, and I’ll discuss the many reasons here why I think keyword search will be replaced by LLMs/GAI, not just in eDiscovery, but even eventually in society. Here are the reasons why LLMs/GAI will kill the use of keyword search in eDiscovery.

The Ubiquity of LLMs and Generative AI in Legal Tech and eDiscovery

You probably already know that generative AI has exploded on the scene in the business world and society. According to the latest available data, ChatGPT currently has around 180.5 million users and OpenAI’s website generated 1.6 billion visits in December 2023. Not only that, but the use of generative AI has extended to web browsing and searching as well. Microsoft Bing has added a GPT-4 Chat capability to its search engine. And Google is expected to do the same with its new Gemini AI model. Searching on the Web is quickly evolving to leverage these exciting new AI capabilities, changing the way we interact with them from a form of keyword searching to a “chat” to find what we need.

The explosion of generative AI within our consciousness has extended to legal technology as well. One example of that is illustrated in the agenda for this year’s Legalweek conference in New York. Based on session descriptions, there are at least 38 educational sessions where AI and/or generative AI are either the primary focus or a topic to be discussed. That’s a lot. Conversely, “keyword search” isn’t mentioned at all. The word “search” is mentioned three times, but one of those is in conjunction with AI.

Still, lawyers and other legal professionals are learning how to conduct research using keywords in Westlaw and LexisNexis, and that’s not going to change anytime soon, right? Wrong. LexisNexis launched their new generative AI solution Lexis+ AI last October, and Thomson Reuters launched their AI-Assisted Research on Westlaw Precision in November. The approach to legal research is evolving away from a keyword search oriented process to a chat-based process, which means – between general web searching and legal research – new lawyers will be learning to find the information they need using the query-based approach favored by generative AI chatbots.

That means when they start using eDiscovery platforms, they will prefer the query-based approach in those platforms as well. One of the biggest challenges to the adoption of predictive coding has been that many lawyers got their technology start in the industry through keyword searching in Westlaw and LexisNexis and they carried that preference over to their use of eDiscovery platforms. That will no longer be an issue – they will already be used to a chat-based interface when they start to use eDiscovery platforms, which are also embracing LLMs/GAI and the same type of interface approach – rapidly.

How rapidly? In eDiscovery Today’s 2024 State of the Industry report released this January, almost three-quarters of 444 respondents (73.2%) said they expect LLMs/GAI to at least create new workflows in 2024. Some of those new workflows could be replacements for legacy keyword search based workflows. Nearly a quarter of respondents (23.9%) expect LLMs/GAI to have a transformative effect – this year!

Three Advantages of LLMs/GAI Over Keyword Searching

We wouldn’t be able to make the bold statement that the use of LLMs/GAI will kill the use of keyword search in eDiscovery if there weren’t advantages to that approach. Here are three advantages of LLMs/GAI over keyword searching:

1. It’s Intuitive: To conduct keyword searching properly, you need to consider variations such as synonyms, misspellings, Boolean logic, fuzzy parameters and more – it can get very complicated. With LLMs and generative AI, the requests are in query or chat form – straightforward and simple. You may have to rephrase your request to pinpoint the information you seek, but you also often must reconfigure your searches to do the same. It’s much easier when the requests are simple chat commands.
2. It’s Expansive: The algorithm doesn’t just limit the results of your request to documents containing a specific search term. You’re searching for concepts and the algorithm is often smart enough to retrieve documents that may be important using related terms you may not have considered. With keyword search, what you request is what you get – nothing more.
3. It’s Informative: When you conduct a keyword search, you’ll get a list of documents responsive to that search, but that’s it. There’s typically no ranking of documents in terms of potential importance. With LLMs, you can get a list of documents with a confidence level and even a brief description on why the algorithm rated the document the way it did, accelerating the iterative process of understanding your document collection.

Of course, one general caveat to LLMs/GAI is the concern about hallucinations. However, the application of the technology to targeted use cases (plus the rapid improvement in the technology itself) should minimize hallucinations for eDiscovery purposes. The risk will be no worse (and eventually perhaps a lot better) than the risk of executing a poorly constructed keyword search, which happens regularly in eDiscovery today.

Conclusion

The use of LLMs/GAI will kill the use of keyword search in eDiscovery. When will this happen? I’m not sure, but it may happen sooner than you think. Even if LLMs/GAI don’t totally kill the use of keyword search, the use of keyword searching in eDiscovery will change – dramatically. It’s progress. Embrace it! We’ve done just that at Arctera, as we’ve monikered “UPSTREAM eDISCOVERY” as a way to classify more about your data without having to run a keyword search, leveraging AI Classification models that enrich your data before you even begin your investigation.

Arctera eDiscovery is a powerful, integrated, cloud-based solution that enables you to easily collect, review, and produce electronically stored information for legal and investigative needs. It delivers high levels of performance and efficiency to give compliance professionals greater peace of mind.

Learn how Arctera Data Compliance can support your eDiscovery needs.