Wednesday, 19 April 2017

implicit bias

Academics looking forensically into the melt-down and subsequent transformation of the chatbot Tay to an obscenity-spewing misanthrope after being exposed to the ravages of internet for only a brief time, finding that the way artificial intelligences learn language—basically digesting the whole of on-line text exchanges—can magnify unconscious word associations and reflect them back at us in a very unflattering way. The tokens and markers of speech that the machines repeat and rely on for guidance contains a lot of latent prejudice but also shows that in context it truly is communication and language that drives consciousness and informs culture.