Technology Can Work For or Against You
All technology tools have strengths and weaknesses. In order to properly use the tools, one must understand what they can and cannot do. Importantly, results always need to be validated. It is just as important to test the discarded documents as it is to classify those first identified. Most defense-oriented providers are focused on the culling properties of TAR. They are adept at elimination. Machine learning can be a powerful tool, but it is taught how to decide by humans. We want to assure that defendants don’t allow their implicit bias to teach the system in a manner detrimental to Plaintiffs. Let us help you negotiate an effective ESI Protocol. The Rules are already biased against Plaintiffs – let’s at least obtain some balance in the discovery rules governing your case. Having ILS by your side to assure the agreements you make are beneficial can have a greater impact on the outcome than effective Vior Dire.
Continuous Active Learning (CAL) is great for defendants and symmetrically situated opposing parties. It avoids having a subject matter expert teaching the system and instead, utilizes a regular contract review team whose decisions ultimately determine what gets produced. The system enriches review batches based on similarity with prior responsiveness calls and enables the review to cease before all documents have been “eyes on” reviewed. However, the review is based on instructions from defense counsel and often contains the implicit bias held by the defense team. And what if the same bias is internalized by those low paid contract reviewers? Their decisions will teach the computer to find more like what the reviewers have identified as responsive. With slight tweaking, the machine learning system may be RIGGED against Plaintiffs. As usual, the devil is in the details. With ILS on your side, we’ll help you make the arguments to assure that the process is balanced.
In applying TAR, defendants want the maximum precision. They don’t want to produce any borderline or extraneously responsive documents. Precision is the percentage of retrieved documents that are relevant. Well, what about the other relevant documents that are NOT retrieved? Recall is the percentage of relevant documents that are retrieved – that’s what Plaintiffs want. A high precision number is focusing on a subset and ignoring the larger context of the document population. TAR could retrieve a fraction of the relevant documents and still have a high precision score, even though there were substantially more documents that weren’t found than were scored in the calculation.
When Plaintiffs receive defense productions, particularly in large volumes, the key to using TAR is in classifying and categorizing what has been produced into usable themes or issues. This is ILS’ sweet spot. The analysis. Using Analytics to determine if there are custodians who seem to play a bigger role than their data volume would suggest indicating that there may be missing custodians. Examining the data volume chronologically to determine if there are data gaps around key events; examining social networks with 6 Degrees® Visualization; and conducting a search term optimization exercise based on real data to augment search methodology. And using the Find More Like functionality to parse the database for other documents on the same area of interest. ILS will facilitate this level of investigation. Again, it’s not the tools per se, it’s the way they are applied. For Plaintiff-Only discovery, our focus is on using advanced technology tools to characterize what was produced. It’s a different mindset.