Written by Emily Tan | Edited by Jennifer Lim Wei Zhen, Josh Lee, Maryam Salehijam (Resolve Disputes Online)
Introduction
Cases turn on their facts. Lawyers depend on both the law and the specific circumstances of their client’s case to make a convincing argument for their client. This makes the discovery process, where the available information is sifted through to identify relevant evidence, a crucial step in any case.
However, discovery is by its nature a slow and laborious process. Countless hours are spent digging through documents, emails and other such sources, searching for the key factors which may make or break a case. This “time-drain” has been exacerbated by the digitalisation of work, which has exponentially increased the volume of documents that lawyers have to analyse. In addition, it is typically the junior lawyers who are delegated to do the discovery task — which explains the television stereotype of young lawyers poring over cartons and cartons of documents late into the night.
Electronic discovery and the role of lawyers in the process
With technology, e-discovery solutions have been developed to automate the discovery process by collecting and identifying relevant electronically stored information.[1] Such solutions assist lawyers by helping to make sense of relevant information quickly, saving time for lawyers to plan arguments and formulate their case strategy. Major players in the e-discovery field include companies like Exterro and Luminance.
At this juncture, a distinction between traditional e-discovery methods and AI-powered e-discovery should be made. Traditional e-discovery methods often involve a simply word search, followed by manual review by a lawyer. While a fairly old technology, it remains widely-used (anecdotally) in many law firms. On the other hand, the more exciting development of AI-powered e-discovery uses a form of machine learning – predictive coding – to reduce the number of non-responsive or irrelevant documents which need to be reviewed manually (in a process known as Computer Assisted Review (or “CAR”).[2]
As futuristic as it may sound, predictive coding is not a magical piece of plug-and-play technology in which a computer finds relevant documents on its own. Effective use of AI-powered e-discovery through predictive coding involves a process in which human users continue to provide important input. Exterro describes the process as follows:
E-discovery relies on “seed sets”, a portion of documents which the human reviewer selects from among all the documents to be reviewed, as representative of the documents to be searched. Human reviewers will then label the documents as “responsive’ and “unresponsive.[3]The software relies on this seed set to develop an algorithm which determines the responsiveness of later documents. The result is continually defined by human users, by coding and inputting in sample results until desired results are obtained. At the end, the software is applied to all the documents.[4]
In short, such forms of AI-powered e-discovery require human users to tag sample data sets where relevant to provide the system with updated information on potentially relevant document. Based on the tags, the system would then identify other relevant documents.
Courts in other jurisdictions endorsing the use of AI-powered e-discovery
While there have not been prominent cases of courts in this region addressing the use of AI-powered e-discovery through predictive coding (as far as we are aware), courts in countries with more mature legal technology industries appear to have already embraced the use of predictive coding in the discovery process.[5]
For instance, in 2012, then-Federal Magistrate Judge Andrew Peck in Da Silva Moore v Publicis Groupeendorsed the use of predictive coding as a way to review documents.[6] In Rio Tinto Plc v Vale S.A.,[7] the United States Court for the Southern District of New York even set out several guidelines for the use of predictive coding in disclosure:[8]
- Generally, parties should not use predictive coding unilaterally,but consult and seek agreement with the other party.
- Where a party acts unilaterally, detailed information about how predictive coding was used should be provided, such as how the sampling was conducted.
- The machine learning process, or the process of training the e-discovery software should also have been conducted properly and the criteria for relevance consistently applied.
- A recommended best practice would be to have a senior lawyer familiar with the issues in the case to decide on the teaching sample used to train the system.
Further, the court also ruled that it would be unfair to hold e-discovery to a higher standard than usual discovery via manual review. While the court left open the question of whether the “seed set” or “training set” should be disclosed to the other party, it encouraged parties to at least discuss, if not cooperate on a protocol for such disclosure.
AI-powered e-discovery solutions as an enabler
In our view, AI-powered e-discovery solutions present a highly enabling opportunity for – rather than a looming threat to – lawyers. This is because at their core, AI-powered e-discovery solutions assist lawyers in better organising documents and creating a system to help lawyers marshal and record their thoughts in an easy-to-retrieve manner. This leaves the ability of lawyers to interpret facts and apply the law to formulate a legal argument untouched. Thus, improving the discovery process with an AI-powered e-discovery system need not threaten the role of the lawyer. Rather, such technologies augment lawyers’ abilities to formulate legal arguments and communicate effectively their clients.
Nevertheless, that is not to say that AI-powered e-discovery solutions are problem-free. Three potential issues could potentially arise.
First, within law firms, there still exist lingering doubts that AI-powered e-discovery solutions are unreliable and could result in compromised quality. To some lawyers, it is dubious that a computer could evaluate and proffer relevant documents faster and more accurately than humans. Left unchecked, such psychological barriers against AI-powered e-discovery could develop into an institutional aversion, hindering a firm’s readiness for technology and its productivity. For sustained and effective long-term use of such technologies, law firms will need to embark on a committed effort to convince naysayers and work closely with its vendors to ensure a smooth and successful roll-out of the technology.
Second, it should be recognised that there are limits to AI and its implementation. Given the relatively early use of AI in legal practice today, it remains to be seen whether its implementation in discovery can help substantial improve the operations of a firm. Further, given that the effectiveness of an AI system depends on the training data fed by the human user, this could result in mistakes emanating from human error (such as inaccurately labelled documents).
Third, firms will need to consider the impact of AI-powered e-discovery solutions on privacy rights. After all, discovery involves the collection and analysis of large amounts of potentially sensitive and confidential information. Allowing access of a third-party’s data to an AI firm (say, to create the training dataset) may be problematic if the AI firm fails to consider privacy rights and regulations. In light of increasing industry and societal awareness of privacy rights in a digital age, AI and big data in e-discovery may well hit a legal and ethical roadblock should it fail to properly address the issue of data privacy. Nonetheless, this is not a deal-breaker – it can be managed with proper attention to upcoming privacy legislation, and proper management of the way services are delivered.
Conclusion
Taking an objective approach to the AI’s utility in e-discovery, lawyers can augment their service capabilities by incorporating AI into the discovery process. While not predominant in legal practice in Asia, jurisdictions like the US have shown increasing acceptance of AI-powered e-discovery tools. Nevertheless, adoption should not be mindless. It is crucial for management teams of law firms to understand how the technology works, work closely with its technology vendors, and ensure the buy-in of the users of the technology (including the junior lawyer and his or her supervising partner). It is also important that AI firms keep a keen eye on the developing discussion on legal issues (such as data privacy issues) of using AI-powered solutions. This pragmatically innovative approach to legal practice could well be the way forward for a law firm ready for the future.
[1]https://towardsdatascience.com/law-and-word-order-nlp-in-legal-tech-bd14257ebd06
[2]https://www.sal.org.sg/Resources-Tools/Legal-Technology-Vision/Other-Services/eDiscovery
[3]https://www.exterro.com/basics-of-e-discovery/predictive-coding/
[4]https://www.exterro.com/basics-of-e-discovery/predictive-coding/
[5]https://www.sal.org.sg/Resources-Tools/Legal-Technology-Vision/Other-Services/eDiscovery
[6]https://www.exterro.com/basics-of-e-discovery/predictive-coding/
[7]306 F.R.D. 125 *; 2015 U.S. Dist. LEXIS 24996
[8]http://www.allenovery.com/publications/en-gb/Pages/Ediscovery-how-to-use-predictive-coding.aspx