A new way of testing the intelligence of potentially deadly AI routines is being backed by speech recognition specialist Nuance, with the system binning Turing's conversational computer tests in favour of a word-swapping Q&A approach.
The Winograd Schema challenge is being funded by Nuance, with $25,000 up for grabs should anyone code an AI clever enough to answer its series of questions. The questions themselves are designed to challenge understanding of sentences, with key words able to be swapped out and completely change the meaning of the question and its answer.
The human thinking behind this is that it's a better option than the Turing test, what with success in the Turing test seeming to come from misunderstanding and trickery these days, rather than any serious leaps of computer intelligence. [Gigaom]