By Tim Rollins
Whether you’re an AI-skeptic or not, you have to admit that the past several years have seen AI revolutionize some big industries:
- Retail: As Amazon leverages buying histories and search to recommend products
- Entertainment: As Spotify and Netflix curate viewing and listening experiences to become dominant players in both television and music
- Search: We don’t even “search” for something on the internet anymore. We “google” it.
- Transportation: Uber and Lyft have shaken up local transportation—initially the taxi industry, but potentially changing the whole idea of car ownership
Andrew Ng, AI pioneer and Adjunct Professor of Computer Science at Stanford University, thinks the AI revolution will compare to electricity:
Just as electricity transformed almost everything 100 years ago… Today I actually have a hard time thinking of an industry that I don’t think AI will transform in the next several years.
E-Discovery is no exception to his thinking.
As far as industries go, e-discovery is fairly young, but almost from its inception, artificial intelligence (AI) has played a role in e-discovery, in the form of predictive coding. That may make e-discovery professionals a little more jaded than the average person when it comes to evaluating claims about the impact AI will have in the coming years, but the fact is that artificial intelligence is radically different than it was even two or three years ago.
One such example is the progress AI has made in the ImageNet Challenge, an AI “categorization” exercise in which AIs compete to classify images of similar objects, for example, dogs of various breeds. In 2010, the winning AI had an error rate of 28%; by 2013, deep learning models helped drop that rate to 12%. By 2017, the error rate of the winning team dropped to 2.5%, half the human benchmark of 5%. Maybe this doesn’t seem relevant, but actually, document review is in many ways simply a text-based classification challenge. Similar improvements in AI used for document review could result in greater adoption of such technologies.
But we’re not really just looking at incremental improvements as what AI has to offer. What are some big ways that AI could radically transform e-discovery?
While AI has been used in the review phase of e-discovery for years, its current integration into document review has become simpler and more elegant. Rather than relying on human-coded seed sets, what if AI could teach itself how to code documents by observing human reviewers?
This extra step of creating a seed set may have been a worthwhile investment of time in large cases with vast amounts of electronically stored information (ESI), but in smaller matters, the time investment and the cost of the technology may have depressed adoption. After all, in most surveys, including the Blickstein Group’s 10th Annual Law Department Operations Survey, almost 2/3 of law department did not use AI at all.
Today, deep learning algorithms can observe as human attorneys review documents, learning the criteria that make a document relevant to a particular matter. Then, once the AI has reached a threshold of confidence in its ability to predict document relevance, it can then speed up human review by suggesting document codes or prioritizing documents for review—or even just applying those labels to the remaining documents.
Leveraging AI Earlier in the EDRM
The whole purpose of e-discovery software is to solve a fundamental problem for legal professionals—namely, it costs too much and takes too long to get to the facts of a given legal matter. Despite e-discovery software’s successes, growing data volumes and the high cost of document review, have made cheap, fast, and defensible e-discovery a proverbial holy grail, always just beyond the grasp of in-house legal teams and law firms.
However, with more and more integrations with data sources like Office 365, AI during early case assessment (ECA) has become feasible. AI can apply data mining techniques to vast bodies of data—not just ESI, but also custodian identities and relationships—before collection. It can then suggest additional custodians and new search terms to find relevant ESI. This ability to focus on truly relevant custodians and concepts, without sacrificing defensibility, can yield significant downstream savings by reducing data volumes sent for review.
AI as Orchestrator
Newer AIs can manage the entire e-discovery process, learning from past actions and results, and coordinating tasks across multiple channels. Uber provides an apt analogy. It orchestrates users’ entire trips, including pricing, nearest drivers, fastest routes, and estimated wait and arrival times. Similarly, AI can orchestrate the entire e-discovery process. If the relevance rate you got at the end of document review was too low, it would learn that collection criteria can be improved for future projects. The same reasoning could be applied to building schedules and budgets or choosing reviewers.
For several years, the legal technology marketplace has been hearing about the AI revolution—and many have dismissed it as hype. The “robot lawyers” have not taken over.
But a closer look at the legal industry reveals more and more reliance on AI to accomplish basic tasks: reviewing documents, managing contracts, predicting case and sentencing outcomes, and even automating tasks like parking ticket disputes. Other industries have shown that AI can and will disrupt old business models. The day for disruption of e-discovery may be arriving soon.
To learn more about AI in e-discovery, download our white paper E-Discovery Technology at a Glance: Artificial Intelligence.