Imagine only passengers. Imagine only second-pass review.
When I turned sixteen, the opportunity to drive a car without an adult around seemed like the ultimate expression of freedom. Many years and lots of traffic later, I dream of the day when the car will drive itself — a day that some may say is right around the corner.
Autonomous cars are the first practical frontier of robotic automation, but at what point is a car truly autonomous and not merely automatic? To help frame this question, the Society of Automotive Engineers (SAE International) has established an international standard known as J3016, which uses a 6-level scale to describe self-driving cars:
- No Driving Automation
- Driver Assistance: Gauges and alert features (lane departure warnings, radar) provide feedback, but the driver is required to make the decisions.
- Partial Driving Automation: The car can make some decisions, such as veering back into the lane if there is a lane departure, but the primary driving is still done by the driver.
- Conditional Driving Automation: The car and the driver work as a team; the car can take the lead in certain driving situations, but the driver is still in the lead.
- High Driving Automation: The car can drive autonomously most of the time, but there is still a driver’s seat and a driver cockpit that can assume control.
- Full Driving Automation: The car is fully in control; the driver tools are fully removed.
What’s great about this taxonomy is that it provides a measure and a goal. It is not only a scale, but a roadmap for how to reach the higher levels of that scale. What if we applied the same kind of taxonomy to eDiscovery? What would that look like? Where are we at? And when will we get to stage 6, full automation?
Today I’ll take a stab at the first question: what if we applied a taxonomy like this to eDiscovery? To set a baseline, let’s assume that we are starting with digital content and extracted text rather than paper, and that the goal is an entirely automated review process: an artificial intelligence you can simply ask to review 5 terabytes of data and voilà , the productions appear in your inbox. Since we’re asking this AI to ultimately do this review by itself, let’s name our automatic reviewer Per-se.
Here are my 6 Levels of Review Automation, starring our new robot friend, Per-se:
- No Automation: ESI is simply organized with saved searches. You can store as much as you like, and if you provide input you can retrieve content and store your choices and comments.
- Reviewer Assistance: With processes like Technology Assisted Review, Clustering, Document Visualization, Latent Semantic Indexes, Complex Searching, and Integration, Per-se helps the reviewer make better decisions by providing more understandable metadata. Content is grouped, categorized, and color-coded by Per-se, but the reviewer is still the one making decisions.
- Partial Automation: The reviewer still makes the decisions, but some of the results are mechanically propagated and/or classified based on “smart rules” SVM (Support Vector Machines), Active Learning Systems, or PII Classifiers. These begin to linearly reduce work and provide a context for further automation.
- Conditional Automation: Per-se now works with you, actively managing batches and identifying areas for further review. Per-se can propagate to contextually similar documents and/or provide an interim review, as well as build inferences against the results to predict the rules. A system at this level could suggest an ESI agreement based on your past production standards.
- High Automation: Per-se can do every phase of review on its own, but the controls are still there. The reviewer still has the option to manually perform various steps of the review.
- Full Automation: Per-se does it all. The reviewer can move straight to second-pass review, privilege log, and production.
Reaching the singularity of full automation was the vision from the beginning of eDiscovery, but there never seemed to be enough computing power to reach that goal. Just as removing the copy machines and the bates stampers seemed unrealistic decades ago and Early Case Assessment (ECA) has gone from an academic concept to an intrinsic part of our eDiscovery conversation in less than a decade, full automation in this industry may be just around the corner.
Twenty years ago, we said AI was dead. Now we talk to Siri or Alexa every day and hear about AI day in and day out. Imagine the possibilities as we bring such technology into the eDiscovery space – we could cruise down the highway as Per-se asks us to confirm the ESI agreement and multi-party productions are packaged, saved, and delivered while we watch the sunset over Big Sur.