Extract from Cassandre Coyer’s article “Generative AI and E-Discovery: 6 Lessons From the Past Year”
Over a year since OpenAI’s human-like chatbot ChatGPT introduced generative artificial intelligence to the world, and by extension, the legal industry, many are still trying to determine how different this new technology really is—and whether it makes so-called traditional AI obsolete.
During Friday’s “An E-Discovery AI Primer on Distinguishing Machine Learning from ChatGPT and Other Tools” webinar from ACEDS, panelists unpacked some of the misconceptions that e-discovery professionals may still have about generative AI from what it means to future jobs and what lawyers’ duty of competence actually encompasses to the many technical definitions in between.
Below are six takeaways from the speakers:
1. Understanding Discriminative vs. Generative AI
For most legal practitioners, ChatGPT is the poster child of generative AI and what it can do. But there are many more buckets under the large AI umbrella that are already used just as often, if not more, in e-discovery contexts.
Traditional AI, also called discriminative AI, is usually used to classify, predict or rank information—technology-assisted review (TAR) is one example. Generative AI is used to, as its name suggests, generate new content, explained Maura R. Grossman, research professor at the School of Computer Science at the University of Waterloo.
The difference between these techniques comes down to how the models were trained. For instance, to build a model for TAR, one would use supervised machine learning. This is where a seed set is manually labeled relevant or not relevant. Asnew documents are suggested, they will be labeled as one or the other, until the system can learn which features or characteristics will make a document fall in one bucket or the other.
