Extract from Zach Warren’s article “Skipping Square One: Where Reusable AI Models Are Finding a Foothold in E-Discovery”
The past several years have seen the explosion of machine learning in law, particularly in e-discovery where technology-assisted review (TAR) gave way to an upgraded TAR 2.0, which itself gave way to continuous active learning (CAL) and TAR 3.0. This use of AI looked to introduce speed and cost efficiency to the discovery process by whittling down documents from the beginning, while still maintaining eyes-on final review.
But the next step for some in e-discovery is asking: Why does every litigation need to be different? One way some are looking to save even more time is through the use of AI model libraries—essentially, reusing algorithmic models in multiple matters, with the ability to store and select between different models in future litigations. Those who have reused models across matters say the algorithms need to be narrowly defined to a specific matter type, but if the model fits, it can help jumpstart the review process with less training time.
At Saul Ewing Arnstein & Lehr, director of litigation support services Richard “Ricky” Brooman said the firm has been reusing AI models primarily for repeatable matters that have similar case structures, such as labor and employment cases. From there, the litigation team will drill down into the type of labor matter, such as departing employees or harassment matters, then determine whether it’s a fit for AI.