The future is here. An article appearing a couple of weeks ago in the CNN Tech online journal reported that self-driving cars were now legal in California. This makes three states (Nevada and Florida also allow self-driving cars) that will soon have driverless vehicles travelling around on public streets. This begs some questions. Who will be liable when a driverless car is involved in an accident? Who needs to carry the insurance? How will police give a ticket to a car with no one at the wheel?
All very interesting questions in the abstract, but this is a Canadian e-Discovery blog. How does robot cars even remotely apply? Well, the same questions could be asked about “reviewerless” discovery reviews. By that, we mean reviews in which a large portion of the heavy lifting is accomplish by “black-box” technology. Regardless of whether you are using keyword searching, predictive coding, or anything in between, the way that the system ultimately makes a decision on whether a record should be seen by a lawyer or not is usually only marginally understood.
Take keyword searching for example. The basic process is word matching – you type in a word and the system gives you a list of records where that word appears. Pretty simple, right? Well, in a effort to make keyword searching fast enough so that you don’t have to wait a week for the results, a lot occurs behind the scenes. Every record needs to be indexed in order for the system to make a list of all unique words. Then every record where each unique word resides needs to be associated with the word. In order to achieve fast results, some words (usually called stop words) are excluded. In addition, all included words are distilled down to numeric values (called word hashes) because numbers are easier for a computer to process than words. Finally, all this information needs to be organized in such a way that both simple and complex searches can be performed quickly. This is called the index.
This process can be fraught with problems. What happens if the system can’t read a record, or worse, can only read part of a record?. What if the word you want to search for is a stop word (it won’t be indexed, so your search will return no results)? What if there is a bug in word hashing or searching programs that result in missed records?
Even a seemly simple thing like keyword searching could result in an “e-discovery accident”. When this happens, who is liable? In Canada, would that be the lawyer responsible for the production or the vendor?.
Don’t get us wrong. We’re not advocating returning the good old days of brute force manual review. The volume of electronic information out there simply precludes that approach. However, we do caution legal professionals to become more familiar with how the various technologies work. Run some tests, or at the very least obtain test results run by others. Talk to the technical people behind the software products (not the sales people who take you out to lunch) to find out how the product achieves what it says it does. Finally, don’t be afraid to get help from a third party to make sure you understand just what that ‘robot reviewer” does. It could help avoid future e-discovery mishaps.