Skip to main content

EnforceMintz — Tech Corner: Q&A with Mintz’s E-Discovery Pro Regarding Artificial Intelligence

Artificial intelligence (AI) is a phrase that we hear in many different areas of our e-data and discovery practice. But, as it relates to e-discovery, AI is not necessarily a new development. To understand how AI has already been in use in this area — and where we expect the use of AI to go — we sat down for a conversation with the Managing Director of Mintz’s E-Data Consulting Group, John Koss.

What is Mintz’s E-Data Consulting Group, and how does this group work with AI?

The E-Data Group at Mintz is a group of lawyers and technologists who focus exclusively on legal technology innovations that can be purposed to help our clients take advantage of the efficiencies afforded by advances in AI. Because we are a fully integrated consulting affiliate at Mintz, we are able to identify and commercialize such AI technology in a variety of disciplines and industries for the common benefit of all of our clients. These services are in high demand, and we have been recognized globally and nationally as leaders in the field of applying AI to legal questions.

AI is certainly a buzzword we have heard a lot recently. How do you specifically define “AI”?

In short, AI refers to computer systems or applications that can perform functions that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages. In the e-discovery context, AI can also be trained to predict document responsiveness primarily through analyzing text contained in documents identified by humans as responsive to a given subject matter and comparing those inputs to text across the full dataset to find similar additional responsive material. We refer to applications that are able to learn and adapt by using algorithms and statistical models to analyze and draw inferences from patterns in data as “machine learning.”

Is AI really that new to e-discovery?

No, it’s not. We have been using AI in e-discovery for more than a decade. For example, for many years, we have been using AI applications to analyze and search text-based legal data, such as emails, cell phone records, and instant messages. This type of AI-assisted analysis in e-discovery is commonly referred to under the general umbrella term of technology-assisted review or “TAR.” TAR can be anything from basic metadata and text analytics to a machine learning application that can identify a set of responsive documents. Once a machine learning tool is trained, we score and rescore the data continuously to efficiently analyze and accurately elevate responsive material while suppressing less relevant data. We can then use statistics and data science to validate the application’s success at identifying substantially all the relevant data. This is the bulk of the work that our team of project managers and data analysts focus on in today’s modern practice.

In your view, what are the benefits of AI use in e-discovery?

To me, the most obvious benefit is efficiency. Pre-AI, we would need large teams of lawyers to analyze the hundreds of thousands, if not millions, of documents for potential responsiveness to a subpoena or document request, and we would basically review them linearly, first to last. Now, with AI technology, we can remove huge volumes of non-responsive data, in some cases 95 percent or more of the collected data, without necessarily reviewing any or all of the “sludge.” We can better detect programmatically privileged documents or documents containing personally identifiable information that may require redaction, or that should not be produced. Moreover, lawyers might need to review only a small percentage of those documents to verify the accuracy of a review performed using AI technology.

More refined and focused document review can now be accomplished with far fewer lawyers. By using these kinds of AI applications, lawyers can use the time they save on document review to work on other important aspects of the matter, such as analyzing key documents, conducting an internal investigation, or preparing witnesses for depositions.

Ultimately, this AI-driven efficiency is beneficial to all parties. In investigations, document review is much faster, and clients are happy because they’ve saved money. Additionally, AI review ensures a greater degree of accuracy than strictly relying on human review and can be validated using statistics.

What issues or challenges have already been confronted with respect to the use of AI in e-discovery?

A number of key cases have shaped how AI is used in e-discovery. As far back as 2012, federal and state courts have recognized that TAR review is as accurate as human review and that TAR should be a judicially permitted methodology for a party to comply with its discovery obligations. While concerns persist about how a party can ensure accuracy under TAR, courts have continued to acknowledge that the same accuracy issues are present with human review and can also be addressed with sufficiency challenges post-discovery. More recently, courts have examined situations where one party or the other preferred using TAR to control costs and have held that each party should have full discretion whether or not to use TAR. Cases today continue to hone how AI can be used in e-discovery, either by addressing proper validation standards or by addressing disclosure requirements if a party does decide to use AI during e-discovery.

Where do you see the use of AI in e-discovery heading?

When people talk about AI today, they are most frequently referring to generative AI, which is different from the more traditional AI applications we just discussed. Generative AI starts with a prompt in the form of a text, an image, a video, or any input that the AI system can process. Various algorithms then return new content in response to the prompt, including answers to questions posed or visual or audio “responses” to the prompt.

Two recent advances have caused generative AI to really go mainstream: transformers and large language models. Transformers are a type of machine learning that has made it possible to train ever-larger models on billions of pages of text. These resulting large language models — with billions or trillions of parameters — can then be used by multimodal AI applications to generate responses to prompts with extreme depth and specificity (e.g., show me what a beach from Massachusetts looks like or write me a poem about a snowy holiday scene).

Right now, one of the most popular generative AI tools is, of course, ChatGPT, which creates seemingly responsive content based on the large language model on which it was trained. In the future, e-discovery lawyers will be able to use similarly trained applications to generate document templates, predict legal outcomes, obtain information about their datasets, and identify specific languages, emotions, or themes in their own subjective datasets. For e-discovery, these outcomes have the potential to save time, money, and effort for all parties involved.

New uses of generative AI will likewise pose new risks. For example, we are already seeing instances of generative AI “hallucinating” or fabricating cases or responses — so, as always, the use of AI will require vigilance and ongoing quality control by lawyers utilizing the technology to ensure that the ultimate work product outcome retains the same level of quality and confidence that clients expect and deserve.

Subscribe To Viewpoints


Samantha advises clients on regulatory and enforcement matters. She has deep experience handling violations of the federal ant-kickback statute and FCA investigations for clinical laboratories and hospitals.

John B. Koss

Managing Director, E-Data Consulting Group

John B. Koss counsels Mintz clients on e-discovery and large data matters. He handles commercial litigation, investigations, and governmental inquiries, relying on his experience in practice and discovery project management. Previously, John led the Boston office of a major legal services firm.

Trust D. Kupupika