Martha Louks, Director of Technology Services - McDermott Discovery, McDermott Will & Emery LLP
In recent years, Artificial Intelligence (AI), or machine learning, has been incorporated into an increasing number of applications, but how exactly can these tools be used in real cases to change the way we do e-discovery and amplify the impact attorneys can have in the litigation process? In our experience, strategic and creative uses of AI have been extremely successful in reducing the costs associated with e-discovery, as well as helping counsel find the most important information in a large set of documents very quickly. Simply put, we are able to be more effective and efficient in e-discovery and litigation preparation by using Artificial Intelligence technology.
The challenge of big data has contributed to rising discovery costs for our clients. Document review is often the most expensive part of litigation, and these days even cases with a “small amount” of data invariably have a greater-than-expected number of documents. AI technology is uniquely poised to address the barriers to meaningful analysis presented by large numbers of documents. We have found that there is tremendous value to using some form of data analytics or AI on virtually every case, instead of simply adding more reviewers to sift through the data.
How AI Has Changed the E-Discovery Paradigm
No need to review every document: In “old-fashioned” document review, attorneys would manually read the documents in a case to assess for responsiveness, privilege, and issues. This type of manual review required substantial hours, and, given that the majority of documents reviewed were ultimately non-responsive, this meant that a lot of attorney time was spent looking at documents that were not relevant to the case and would never be produced.
With the advent of Technology Assisted Review (TAR), also known as Predictive Coding, there was now a way to use machine learning to sort ever-larger data sets into buckets of responsive and non-responsive documents. TAR is an iterative process in which reviewers provide example documents to the system, teaching it to classify documents for responsiveness, privilege or issues using input from subject matter experts. TAR is an effective, defensible way to reduce e-discovery costs, because practitioners can comfortably rely on the accuracy of the system to identify relevant documents without having to read everything. Depending on the size of the starting data set, this may mean that hundreds of thousands or even millions of documents could be excluded from manual review.
AI technology is uniquely poised to address the barriers to meaningful analysis presented by large numbers of documents
More efficient TAR workflows: TAR applications have also evolved to support a Continuous Active Learning (CAL) protocol, which classifies documents more quickly than “traditional” TAR workflows. CAL provides more presumptively responsive documents to the team training the system, and this process combined with advanced algorithms and enriched data mined by AI, allows us to stabilize the TAR model more quickly. The prevailing sentiment used to be that TAR was best suited for very large data sets, e.g. 250,000 documents or more. However, we have seen that newer AI tools and a CAL protocol are highly effective with data sets as small as 20,000 documents.
How AI Makes Us Better Litigators
Early Case Assessment (ECA): The volume of data in many cases now exceeds the ability of humans to determine what is in the data through manual review alone. Even if 100 reviewers are going through the documents, it is difficult to combine the disparate knowledge of 100 different people into cohesive ideas about the content of hundreds of thousands or millions of documents. Using AI, we can see what is in the documents very early in the case. AI is able to analyze the documents and extract a wealth of information that can help attorneys find important or interesting information without having to read a large number of documents. AI will identify such information as:
1.Concepts and topics discussed in the documents
2.Entities mentioned in the documents, like people, organizations, etc.
3.Social network visualization showing who is talking to whom
4.Sentiment of email conversations, i.e. is the communication positive, negative or neutral,
5.Spikes in communication after business hours, on weekends or during particular days or weeks
6.Fraud signals, such as rationalization
This list provides a good representation of what is possible, but there are many other ways that these tools are able to bubble up the most important features of a set of documents.
By leveraging AI, the speed with which we are able to find the important data has increased dramatically, and we can simultaneously ignore large swaths of irrelevant data that we would otherwise have to review if we didn’t use AI technology. Our legal teams are able to gain an early understanding of facts in the documents, and they also can see what isn’t in the data. For example, they may learn that a previously-unknown individual was involved in a lot of conversations about a particularly relevant topic, or they may find that a person believed to be important was not involved in key events. This information can be used to negotiate the scope of discovery to be as effective and efficient as possible.
Using technology in ECA gives attorneys an opportunity use the patterns and other data points detected by AI in a strategic way that makes the most sense forthe particulars of a case. Data-driven analysis can support arguments to reduce the scope of discovery or, depending on what is learned, it may even inform settlement negotiations.
How AI is Important for Our Clients
Because we use advanced technologies, we are able to realize significant savings in the e-discovery process by reducing the amount of attorney time needed to get the work done. Using AI effectively can save up to 85 percent throughout the discovery phase of a case, compared to traditional methods. In addition to reducing costs, we find that thoughtful, consistent use of technology gives practitioners the opportunity to focus their efforts on higher-level analysis instead of spending time shuffling through an unnecessarily large number of documents. This means the hours spent on a case have the highest possible value for the cost. Information learned through data analysis and AI is an opportunity to make better and more strategic decisions at key points in the discovery process, intelligently negotiating the scope of discovery, and ultimately attaining the best results for our clients - all while reducing discovery costs as much as possible.