How can we make Predictive Coding / Technology Assisted Review / Computer Assisted Review, accessible to potential users when we cannot even agree on a name for it? I favour predictive coding because it refers unambiguously to a specific class of technology and is the name used by most of those in the field who sponsor the eDisclosure Information Project. As Ralph Losey says in opening his report of the panels on the subject at Georgetown, many now seem to favour Technology Assisted Review, whilst he prefers Computer Assisted Review.
We are stuck with these multiple terms, just as we are stuck with the parallel labels eDiscovery and eDisclosure for the overall process. We do a kind of mental translation perhaps, just as we filter out the passages in marketing material whose primary function is to denigrate rival products. It all uses up the limited bandwidth available for spreading understanding. I will stick with my own favoured term, predictive coding, for the purposes of this article.
There are perhaps three levels at which one can describe what predictive coding is and does. One level relies heavily on statistics, with detailed discussion about seed sets, precision, recall and the F-Measure, supported by equations demonstrating the validity of the process and its defensibility. This is fundamental to the acceptability of the process, but not necessarily the most easily-assimilated level for those new to the subject.
At the other extreme, one can describe the process in broad, high-level terms without descending into the detail, making it clear that the statistical underpinning exists and that the better applications have tools which allow the lawyers to check their results and, if necessary, to prove their validity. That is my own approach, consistent with my own position as a) a translator for lay audiences and b) a mathematical dunce, in common with many of the lawyers who should be interested in predictive coding.
Ralph Losey, as an experienced practitioner who uses these applications every day, takes a middle road in his writings, setting out and then dealing with the issues in a way which is simultaneously authoritative and accessible.
This is the line taken in his Georgetown article which is called Most Advanced Students of e-Discovery Want a New CAR for Christmas. He helpfully sets out the subject matter of each of the relevant panels at Georgetown which, quite apart from anything else, serves as a check-list for those who have responsibility for deciding what technology to use for their cases.
His article includes a summary of a debate, which allows him to describe both sides of the various arguments which can arise when predictive coding is under discussion, and he ends with a list of the nine specific topics which were argued.
Anyone who wants to know what the issues are will benefit from reading this article.