Knowledge truly is power. However, knowledge when it comes to e-discovery means knowing where the information is and how to use or access it. Electronic discovery remains a complicated and often costly issue.
This blog will take an objective look at Ediscovery trends litigation support and without warning will veer off into personal observations and themes.
Thursday, May 8, 2014
"Predictive" Coding and the Naked Emperor
I've always been suspicious of the claims of the "predictive coding" zealots. Every year it seems that there is a new buzzword in the field of Ediscovery. Technology Assisted Review - Check! Information Governance - Check! Cloud Computing - Check. Big Data - Check.
My friend John Martin did a magnificent job of distilling some of my concerns in his blog entry that can be found right here.
The Emperor has No Clothes - and PC Can't See Image-Only Documents
There are several parallels between predictive coding (AKA technology assisted review) and Hans Christian Andersons' tale, "The Emperor's New Clothes." In the story, two weavers tell the emperor they will make him a suit of clothes that will be invisible to those people who are unfit for their position, stupid, or incompetent. None of the emperor's subjects want to admit to those deficiencies so the emperor parades around with no clothes on until a child states the obvious - the emperor has no clothes.
In the case of predictive coding, its advocates have touted the efficacy of their approaches in white papers, blogs, and social media postings, and have practically created a separate industry to host conferences promoting the wonders of predictive coding. Few people want to ruin the moment or buck the trend by pointing out what is obvious when one considers the technology underlying predictive coding - it is completely dependent on having text to analyze. It will absolutely fail to analyze documents for which there is no text, and will do a miserable job where the text is of poor quality.
This might be just an esoteric debating point if virtually all documents had associated text. However, in some industries like oil & gas, half or more of some collections will be engineering drawings and schematics that were output to image-only PDF for distribution and use by those who don't have the software licenses needed to view the documents in their original file formats.
In practically all industries it is common practice to develop documents in one application like Word and then, once finalized, distribute them as image-only PDF so they can be viewed on a variety of devices and so recipients can't easily change the content. In one collection we analyzed, only 20% of the PDFs had associated text. Even if predictive coding were 100% effective, the most it could classify would be 20% because it literally cannot "see" the 80% without text. If in fact predictive coding has a recall rate of 70-80% of what it can see, that would mean that predictive coding would have identified 14 to 16% of the total PDFs (70% x 20% = 14% or 80% of 20% = 16%). By contrast, BR's visual classification technology classified 100% of them.
PDFs will potentially be among the most relevant file types in a collection because that is the format used to distribute information within and among groups of people within an organization, and among organizations. Note that even if in some unique e-discovery settings predictive coding is acceptable, the text-restriction failing of predictive coding will be fatal for broader information governance purposes.
So... if you're going to use predictive coding, at the very least measure what PC doesn't "see." If you're planning on using PC for information governance purposes, make sure that the organization doesn't mind not classifying a potentially significant percentage of its documents.