Mat Velloso’s joke about that (https://twitter.com/matvelloso/status/1065778379612282885) certainly strikes a chord. AI has become a very popular marketing term for a whole range of technologies that have been gradually improving for years.

There are a lot of technology areas — collaborative filtering, spelling correction, semantic vectors / word embeddings, image classification, speech recognition, the list goes on — that use machine learning in many ways. Back in the early 2000s I wouldn’t have said “I’m working on AI” when working in any of these areas. Partly because AI had a bad reputation for soaking up millions of dollars and producing easily-breakable brittle results, and partly because it sounded misleadingly grandiose.

Like most other researchers and engineers, in the past few years I’ve given up denying and accepted a seat on the bandwagon. Lots of other people I meet are quite sure that I work on AI, and it has strong positive associations, so while I’m happy to explain more of the details and the various decades of work that have gone into so many different areas, there isn’t much to be gained by saying “I don’t really work on AI”.