Artificial intelligence (AI) is all over the headlines. If you’ve shied away from reading about technology because you think it’s too complex to figure out, stop worrying. The details of how AI works may be mind-bending, but the upshot is that it can make your life—and your eDiscovery—easier. Chances are, you’re already using more AI than you realize.
What exactly is AI? In short, it’s the capability of computers to simulate human thought and reasoning. Many AI systems incorporate machine learning, which allows them to learn from data, often through human feedback loops. AI systems excel at churning through huge volumes of data at lightning speed. In fact, all that electronically stored information (ESI) that your eDiscovery team is drowning in is exactly what an AI system needs to learn and do its job.
AI in eDiscovery
There are multiple AI applications in eDiscovery. We recently wrote a blog about natural language processing, a form of AI that determines the meaning of words and phrases and analyzes which terms commonly occur together. NLP and AI-based analytics tools allow eDiscovery professionals to greatly winnow their data sets when collecting and preserving ESI, using techniques like email threading, concept clustering, and near-deduplication. This doesn’t just save the people on your team precious time: it also has the power to slash your budget for data collection, hosting, and especially review.
Speaking of review, one major AI application in eDiscovery is technology-assisted review (TAR), also known as computer-assisted review (CAR). TAR, as its name suggests, streamlines and expedites the review phase of eDiscovery by prioritizing documents for human review teams. This sorting process allows the review team to focus on those documents that are most likely to be relevant and helpful.
Depending on the tool and the approach, the TAR system may start from a seed set of pre-coded documents, from which it learns which types of documents should be labeled with which codes. The originating seed set must, of course, be accurately coded and representative of the entire data set or it will be less helpful. A more advanced approach uses continuous active learning to eliminate the need for a seed set. Instead, human review teams start their process while the computer watches and learns. Soon, it starts offering documents to the team that it deduces may be relevant. Both approaches require validation by searching the null set—those documents the AI believes are irrelevant—to ensure that nothing critical has been missed.
Don’t Panic: AI Isn’t About to Take Your Job
For all its improvements, AI can’t truly replicate human thought or analysis yet. The machines aren’t imminently planning a hostile takeover of either eDiscovery jobs or Skynet.
For a bit of context, consider that six years ago, AI was struggling to do something that most three-year-olds can handle: identifying cats (and differentiating them from, say, dogs).
Still, it only attained 75 percent accuracy in cat identification. And it’s not like that came easy, either. It took 16,000 computers three days to view 10 million images within YouTube videos to begin to recognize the concept of “cat.”
Surely AI has improved since 2012, right? Absolutely. Improvement, though, doesn’t equal perfection. In fact, an MIT research group “tricked Google’s image-recognition AI . . . into thinking that a baseball was an espresso, a 3D-printed turtle was a firearm, and a cat was guacamole.”
What do cats (or guacamole) have to do with eDiscovery? The point is just this: AI makes eDiscovery easier for humans, but it doesn’t replace humans. By all means, use your advanced technological tools to improve, simplify, and streamline your eDiscovery workflows. But don’t fall into the trap of thinking you can turn off your own brain and let the software do all of the thinking for you.
Use AI to augment, rather than replace, your own intelligence, and you’ll find your eDiscovery running more smoothly and easily than ever before.
Ready to incorporate more intelligent technology into your eDiscovery workflows? We’ve been using it for years, building ever-more-advanced tools that help lawyers focus on the human tasks of lawyering. Please contact us to learn more.