TAR, Information Governance, and the Relativity of Wrong

Keynote Address by Jeremy Pickens

Machine learning technologies have had wide success in many areas of information seeking and organization, both outside of the legal domain and now increasingly within it. Supervised machine learning, in particular, has been responsible for massive increases in document review efficiency. Technology Assisted Review, the legal term for what academics call Supervised Machine Learning is the hot topic in electronic discovery today.

TAR's success in e-discovery ties to the fact that the relevance with which the algorithms are being trained is typically well-defined; the investigation is about something specific. This gives supervised learning algorithms the ability to latch onto something concrete. There is a usually a "right" and a "wrong" answer to every investigation.

This may not be as true for Information Governance. TAR has been shown effective at looking backwards rather than forwards. But IG tasks often go beyond historical search. Can supervised learning play a role in such forward-looking processes as monitoring for fraud or rogue employees? In this keynote, I will make the argument that both the problem of and solutions to Information Governance should be viewed along a spectrum, a gamut of suitability, a “relativity of wrong". Even when technologies are not perfect fits, they can still move the gradient. Detecting and moving along that gradient will provide ample challenges and opportunities for years to come.

Doug Oard

Last modified: Sat May 9 16:24:50 2015