IT’S HARD TO CREDIT,

DEMM Engineering & Manufacturing - - EDITORIAL - JANE WAR­WICK

as you sit on your hands to pre­vent your­self from throw­ing your com­puter against the near­est wall be­cause it is too il­log­i­cal to do as it is told ( or Win­dows 10 in­sists on down­load­ing, con­dens­ing your 39 pages of dead­line copy into one page of gob­blede­gook… but that’s another story!), that sci­en­tists are se­ri­ous about de­sign­ing a ‘kill switch’ to pre­vent in­tel­li­gent ma­chines from learn­ing to over­ride hu­man in­put. When HAL9000 from 2001: A Space Odyssey be­gan to plot to kill the crew, it was all de­li­ciously scary and fu­tur­is­tic. But it ap­pears the fu­ture is now here and HAL9000 may ac­tu­ally be a rank ama­teur com­pared to what AI has now be­come. Re­ports say that Sci­en­tists from Google’s ar­ti­fi­cial in­tel­li­gence divi­sion, Deep­Mind, and the Fu­ture of Hu­man­ity In­sti­tute at Ox­ford Univer­sity are de­vel­op­ing a “kill switch” for AI and have re­leased an aca­demic pa­per out­lin­ing how AI ma­chines could be coded to pre­vent them from over­rid­ing hu­man in­put, set­ting out a frame­work that would al­low hu­mans to al­ways re­main in charge. Their re­search re­volves around a method to en­sure that AIs can be re­peat­edly and safely in­ter­rupted by hu­man over­seers with­out learn­ing how to avoid or ma­nip­u­late th­ese in­ter­ven­tions. And to think all we used to worry about was what the cat and dog got up to when we were at work. Who knows what the SKY de­coder is think­ing; and as for that squat, men­ac­ing mi­crowave with its blink­ing red eyes…

Newspapers in English

Newspapers from New Zealand

© PressReader. All rights reserved.