Re­searchers: Com­mon sense com­put­ers not too far off

The Washington Times Daily - - Nation -

PITTS­BURGH | Re­searchers are try­ing to plant a dig­i­tal seed for ar­ti­fi­cial in­tel­li­gence by let­ting a mas­sive com­puter sys­tem browse mil­lions of pic­tures and de­cide for it­self what they all mean.

The Carnegie Mellon Univer­sity project is called NEIL, short for Never End­ing Im­age Learn­ing. In mid-July it be­gan search­ing the In­ter­net for im­ages 24/7 and, in tiny steps, is de­cid­ing for it­self how those im­ages re­late to each other. The goal is to recre­ate what we call com­mon sense — the abil­ity to learn things with­out be­ing specif­i­cally taught.

For ex­am­ple, the com­put­ers have fig­ured out that ze­bras tend to be found in sa­van­nahs and that tigers look some­what like ze­bras.

It’s a new ap­proach in the quest to solve com­put­ing’s Holy Grail: get­ting a ma­chine to think on its own us­ing a form of com­mon sense. The project is be­ing funded by Google Inc. and the Depart­ment of De­fense’s Of­fice of Naval Re­search.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.