There’s a common myth that data is mined, like coal or oil. In this way of thinking data is an untapped natural resource, the foundation for a new industrial revolution led by the likes of Amazon and Google.
But that’s not right. Data isn’t mined, it’s made. Sometimes data sets are literally produced by actors performing for researchers and their machines. Other times, data sets are the product of our own performances, every time we wake up Alexa, or upload a video to YouTube.
After words is about these performances which we’ve been thinking of as a kind of computational theatre. The piece is about how we speak to and perform for machine listening systems and about what’s happening to language as a result.
Most of the sounds and many of the voices in Afterworks come from existing data sets. The script was written with and against an autoregressive language model trained on a data set with 175 billion parameters, and it was performed by voice actors, all of whom work in critical data studies. The result is a strange set of semi-fictional tales of computational scripting, instruction, production and performance staged at ACCA in eight channel audio. In its more pessimistic, or maybe realistic moments, the work gestures at a near future in which language has been fully operationalised where every word we speak, has a computational effect and residue.