Lawyers can be a tough crowd. Especially when it comes to technology, many lawyers are reluctant to put their confidence in workflows and solutions that seem to operate in a black box. Perfect often becomes the enemy of good. Technology-assisted review is a prime example—although the use of predictive coding and other advanced analytics have been widely accepted in the U.S. courts, many legal teams continue to question their efficacy and often default to more expensive and time-intensive traditional methods. In a recent survey, 73% of general counsel said their legal teams are not using artificial intelligence (AI) for any function and only 33% said they agree attorneys currently have adequate technological knowledge and capabilities.
While it may seem like a safe play to delay adoption until AI and other advanced tools reach wider acceptance or lawyers gain more technical proficiency, the reality is that the next generation of e-discovery is arriving. Ready or not, matters are quickly and consistently becoming more complex. In this new landscape of exponentially growing data volumes and diverse emerging data sources, machine learning and AI will play an important role in prioritizing documents and providing statistical proof of error and accuracy within datasets, to help quickly surface key facts in litigation, investigations, and other legal or regulatory matters.
Beyond a lack of technological confidence, another factor contributing to lagging adoption is that many legal teams haven’t fully experienced the benefits of AI. Our teams are beginning to see an opportunity to shift this dynamic, and demonstrate clearer, more impactful, and more accessible results through the use of portable AI models.
Portable AI is an emerging tool that our teams are beginning to use to deliver practical improvements in e-discovery. It is essentially the next step in predictive coding, in which an e-discovery platform can re-use predictive coding models developed on prior matters or build custom models designed for specific topics or issues. This allows legal teams to take previous learning or pre-built algorithms and apply them to a new matter, use them to improve another model, or set them as the foundation for a model being built from scratch, without needing to do any new training. The process is an evolution in how e-discovery workflows can be standardized to accelerate data analysis and quickly surface insights from a large dataset.
Because this approach provides a fully functioning learning model that can be applied right at the outset of a matter, it provides significant time and cost savings. When a model is pre-trained to surface the specific types of information of interest, legal teams can get to the data they need much faster and more reliably than when they must train, re-train and test a model. These results make AI much more tangible for skeptical lawyers who have yet to utilize predictive coding or other analytics in a meaningful way.
As this type of AI continues to advance, we’ll see an increasing range of applications. In the near term, there are several specific use cases in which it can be leveraged. These include investigations, in which the team has an idea of what they are looking for, but need AI to quickly surface patterns or behaviors of interest across a large dataset. Similarly, portable models can be utilized for sentiment and behavior analysis to help counsel find evidence of specific concerning or illegal behaviors such as discrimination, harassment, or workplace bullying.
Portable AI models may also be used for data cleanup, to automate the process of removing “junk” data such as newsletters, event invites, and out-of-office notifications from a dataset at the outset of a matter. This will allow the e-discovery team to get up and running more quickly with a clean and more targeted document population. In more advanced applications, portable models can be designed to detect certain sentiments in a dataset, including those that may signal sexual harassment or toxic work environment activity, which may be useful in proactive compliance monitoring as well as investigations. If an investigation is kicking off, the team could apply a portable detection model to the dataset as a first step to quickly find documents of interest that can help reveal important clues and information about the issue or people under investigation.
Privilege review is another useful application for corporate legal teams that have years’ worth of privileged documents stored. An AI model can be trained to understand the parameters that the organization uses to define privilege. Then, using the portable functionality, the model can be applied to quickly identify and parse out privileged documents in new cases. Some of the emerging functionality we’re seeing among technology providers includes prebuilt models that provide privileged document detection, which may allow some organizations to skip the training process altogether. In theory, this application could also work for an organization that deals with many alike cases, such as employment investigations, to give the team a quick look at the types of documents that are typically relevant in that type of matter.
While I believe we’ll continue to see more advancements in this area and more lawyers willing to embrace AI due to increasing, widespread validation of the benefits, this is not a silver bullet to solve every e-discovery challenge. As with any predictive model, whether the team spends time training it or not, legal teams must examine—via keyword search, sampling, or other methods—the documents that the AI is not bringing back as relevant before they are discarded. This important step in quality control will confirm whether the model is working as intended and/or provide guidance on any additional training that is needed. Legal teams may also still need the support of technical experts who know how to leverage this new technology in a defensible way and control for potential oversights.
A healthy dose of caution is warranted when it comes to the use of AI in any setting. Capabilities that fully live up to all the hype that has surrounded machine learning and AI are still a long way off. But that fact shouldn’t stand in the way of steady progress and tapping into the actual benefits that are available now. For cases that require pattern detection or behavioral analysis, and/or legal teams that have repeat matters with similar content, portable AI models are a practical way to re-use pre-trained machine learning and explore the benefits in a relatively low risk, high return manner.
About the Author
Jessica de Brignac is a Senior Director within the Technology segment at FTI Consulting. She has more than 15 years of experience in advising clients in areas of data management, analytics, e-discovery, blockchain, information governance, and emerging technology. She leads projects with large, global corporations and top law firms through managing and supporting discovery activities involving complex litigation, class actions, HSR second requests, bankruptcy, SEC investigations, and federal investigations spanning a variety of industries.