In his groundbreaking 1999 novel The Age of Spiritual Machines, Ray Kurzweil described a future in which Artificial Intelligence exceeds Human Intelligence and machines take over many roles previously delegated to the human realm. Kurzweil establishes a timeline in which computer intelligence initially asserts its superiority over the human mind by performing mundane tasks; but eventually, human intelligence is completely supplanted even for the most highly nuanced, complex, and significant functions.
With the prolific development of technology dedicated to solving problems commonly encountered in the world of litigation document review, it’s worthwhile to take a survey of the legal/technological landscape and ask: “How far away are we from the possibility of computer intelligence replacing human intelligence in the art of legal document review?”
To answer that question, we need to take a look at the current state of legal technology. Currently, the most advanced technological development seems to be the integration of software analytics (really Artificial Intelligence) into document review and so-called Early Case Assessment platforms. For simplicity’s sake, these offerings can be broken down into three generations of products. The three generations can be described roughly as follows:
First Generation: Platforms which serve the end user directly at the end user’s command. These are basic review platforms that house a database and a document viewer. They can be used to store and retrieve information with basic keyword searching components, but they do not have an artificial intelligence component.
Second Generation: Platforms that have integrated software analytics as a feature. These applications can group documents into clusters that the software interprets as having similar content, perhaps making predictive coding choices based on user coding of documents within the cluster. However, Second Generation platforms still rely on the human analysis of the end user to expedite the review process.
Third Generation: Software that can independently interpret a single human-generated request (i.e., a request for production) and codify documents based on its analysis with no human guidance. At this stage, the machine operates accurately without user input, and we’ve reached a state where machine technology is better than humans at the task of document review.
I would argue that we are currently in the Second Generation of technology platforms. Perhaps the lay person would claim that their product is Third Generation and obviates the need for human analysis, but based on actual user experience, it is clear that we are living in age of Second Generation document review technology. It is natural to assume that this technology will continue to develop and improve, so we must be approaching the day when machine analytics will replace human document review, right? After all, the New York Times just published an article entitled “Armies of Expensive Lawyers, Replaced by Cheaper Software,” (March 4, 2011) so the end must be nigh, yes?
I would argue that there is one very good reason we may never see that day: in the world of litigation, it doesn’t really matter how good the software analytic technology is. Litigation, by nature, is a highly contentious, adversarial system. Opposing sides argue over every procedural detail in an attempt to gain the most favorable outcome for their client. If software analytics were used as a replacement for human document review, counsel would need to make an official representation containing language along the lines of: “These documents were produced based on a software analysis of your Request for Production and our client’s corpus of documents; Software XYZ reported that these are the documents that are responsive to the request.”
The problem we encounter is that, in our Second Generation technology, the analytic components of the software are highly product-specific and can vary greatly depending on which product is utilized. This issue would only become more pronounced in a hypothetical Third Generation. Courts have accepted the practice of running search terms on data, but, have not endorsed or required a specific product be used to execute the search terms; it is simply assumed (perhaps misguidedly, but that’s another article) that the search terms are executed using software tools that will soundly produce the desired results.
Since analytic software is product-specific, it becomes all too easy for opposing counsel to challenge the ability of that particular product to produce reliable results. Even if a test could be produced that showed that the software was more “accurate” than human review, to survive a challenge, the court would ultimately have to endorse Software XYZ, thus setting precedent that that specific product is an effective replacement for human document review. I have trouble envisioning a court ever taking such a step.
Software analytics in review and ECA platforms are great tools. They can help expedite human review by grouping documents together, and they can be extremely useful in investigative scenarios, as they leverage the brainpower of a human investigator. However, it’s unlikely this technology will ever obviate the need for humans to review documents in response to a Request for Production. That is not to say that technology will never be able to exceed human ability in this discipline, but rather that the constraints of the litigation arena will necessarily preclude the widespread application of such technology within the document review space.
About the Author
Jim Iseman III currrently serves as Director of Litigation Technology Services for TransPerfect’s Legal Solutions division based out of New York City. Mr. Iseman oversees the execution of eDiscovery, hosting, and legal technology projects. In addition, he acts as a consultant on complex, large-scale eDiscovery matters.
Prior to joining TransPerfect, Mr. Iseman managed technical discovery for complex groundwater litigation matters, including MDL 1358, In re MTBE Product Liability Litigation. He is a graduate of Duke University.