AIs are get­ting closer and closer to pass­ing the Tur­ing test. Note that every AI man­aged to fool at least one human judge.

Sci­en­tists at the Uni­ver­sity of Read­ing tested five machines to see if they could pass them­selves off as humans in text-​based con­ver­sa­tions with peo­ple. The test was devised in 1950 by British Math­e­mati­cian Alan Tur­ing, who said that if a machine was indis­tin­guish­able from a human, then it was “think­ing”. One robot, Elbot, came close on Sun­day by reach­ing 5% below the pass mark. One of the machines in the Tur­ing Test No com­puter has passed the test by fool­ing 30% of its human inter­roga­tors No robot has ever passed the Tur­ing Test, which requires the robot to fool 30% of its human inter­roga­tors. Dur­ing the exper­i­ment, five arti­fi­cial con­ver­sa­tional enti­ties (ACEs) com­peted in a series of five-​minute long, unre­stricted con­ver­sa­tional tests.

BBC NEWS | UK | Eng­land | Berk­shire | Test explores if robots can think

The cool part here is that the tech­nol­ogy that goes into pro­grams like Elbot can be used in con­junc­tion with voice recog­ni­tion to make devices that under­stand vocal instruc­tion and can even ask for clar­i­fi­ca­tion when they need it. Put that kind of tech into a car-​based GPS, and you’re well on your way to KITT…