ChatGPT and E-Discovery: Match Made in Heaven or Rocky Roads Ahead?

Previously, I co-authored a two-part post on the advantages and disadvantages of using Technology Assisted Review (“TAR”) in E-Discovery document reviews. These articles can be found here and here. TAR helps attorneys during the review phase of E-Discovery by deploying algorithms that can quickly classify documents based on training provided by expert reviewers. TAR can provide statistics, categorizations, and reporting information that a human reviewer simply could not do in a timely manner.

The main takeaway from the previous articles was that TAR was not created to replace standard reviewing processes and protocols but instead was intended to streamline those processes so that reviews are more accurate, timely, and efficient. Of course, new technologies are being created and utilized every year. The most significant developments lately are the rise of chatbots - software applications that allow for online chat conversation via text or text-to-speech, without any direct contact with a human operator. Currently, the chatbot garnering the most attention is OpenAI’s ChatGPT program. This article will focus on this technology and how it works either for or against the E-Discovery review process.

What Is ChatGPT?

ChatGPT is a chatbot that is powered by an artificial intelligence (AI)-based language model. The GPT stands for generative pre-trained transformer, which in turn is a group of language models being trained on Large Language Model (LLM) sets. These LLMs are the main difference between it and current E-Discovery TAR programs. This tool is capable of answering complicated questions and responding in a conversational manner. It generates its responses based on predicting the most likely sequence of words based on information that has already been inputted. In fact, ChatGPT has already been trained by analyzing over 350 billion words extracted from the Internet and book collections. ChatGPT’s knowledge is so vast, it reportedly was able to pass an MBA entrance test as well as a medical licensing exam.

As we all know, E-Discovery can be a time-consuming, arduous, and costly process. Finding any way to help streamline this process is welcome. Current E-Discovery practices generally involve some form of keyword searching to decrease the document population for human reviewers to oversee. Also, in certain instances, the use of predictive coding or TAR is employed to further narrow that document population. Of course, the use of TAR still requires a human being to input documents into the algorithm which takes time and is prone to human error. Could the use of ChatGPT help solve these problems? As with any new technology there are going to be advantages and disadvantages.

Potential Advantages of Using ChatGPT in E-Discovery

While unproven, there are potential advantages of using ChatGPT in E-Discovery..

• Early case assessment: ChatGPT could potentially be used to quickly go through the relevant document universe and uncover patterns, trends, create timelines, and provide information on the strengths and weaknesses of the case. This information could be useful in determining strategy and perhaps guide settlement negotiations.
• Organization: There is nothing more frustrating than combing through a sea of documents without having a clear picture of the important themes of the case. ChatGPT could potentially be used to sift through mounds of data and organize and cluster documents according to various patterns, themes, and issues. This aids lawyers in understanding the important issues and gets the most relevant documents in front of their eyes.
• Streamlining reviews: By separating the wheat from the chaff so to speak, finding the documents that are most relevant to the case helps focus the review team and prioritizes and categorizes the documents. Using ChatGPT could potentially results in less human hours, which means lower review costs as well as an increase in effectiveness and efficiency.
• Email threading and near-duplicate detection: ChatGPT could potentially be utilized to group all iterations of an email thread together, and with its ability to quickly identify issues, concepts, and themes, can make this process much more efficient. Also, it could potentially group documents that are similar or somewhat similar together as well, helping attorneys to be able to quickly review these sets in a more consistent manner.
• Continuous Active Learning: ChatGPT could potentially also be used to augment current TAR processes, by providing a more accurate way to train the system to locate and identify important documents based on documents identified by the case team.

Potential Disadvantages of Using ChatGPT in E-Discovery

As with any technology, ChatGPT comes with pros and cons. The pros of using ChatGPT in E-Discovery are theoretical and unproven. However, we can foresee some clear potential disadvantages of the technology as well.

• Cost: Currently, ChatGPT is not free, and OpenAI is currently charging based on the number of words and punctuation that you submit, which could become cost prohibitive depending on the volume of material.
• Security concerns: Also currently, ChatGPT is housed at OpenAI and would require the client documents to be sent to OpenAI for processing. This brings up data security issues that many clients would not be willing to risk.
• Lawyers are still needed: While the technology is certainly impressive in its speed and accuracy, it will still need lawyers to oversee the E-Discovery process. Lawyers will still have to make privilege calls and some relevance calls. ChatGPT may be more suited for a first pass of the documents, leaving the second level review to the attorneys.
• Ethical concerns: With new technology, attorneys will have to consider the issues of client confidentiality as well as attorney-client privilege. By inputting client information into ChatGPT, there is the real possibility of waiving attorney-client privilege. As with any new technology, it is imperative that attorneys keep their clients well-informed about any risks and get informed consent.

Conclusion

As with current predictive coding processes, these advanced technological advances are designed to help improve the efficiency of certain E-Discovery tasks. By streamlining the processes, technology cuts down the time needed to review and process information, and ultimately cuts down costs. However, even with these advances, these technologies will certainly not replace the need for lawyers. Ultimately, there is no replacement for human expertise and judgment; these tools are merely meant to aid in the process.