Su­per­in­tel­li­gent AI? Ma­chines will first have to learn how to em­u­late the struc­tures of the liv­ing brain.

HWM (Malaysia) - - THINK - By Ko­hWanzi

Fu­tur­ists would have you be­lieve that ar­ti­fi­cial in­tel­li­gence could one day tran­scend our own in an event that you’ve prob­a­bly heard re­ferred to as the tech­no­log­i­cal sin­gu­lar­ity. So are our bi­o­log­i­cal in­tel­li­gences con­demned to even­tu­ally lan­guish in the shadow of the ma­chine? Not quite. De­vel­op­ments in ma­chine in­tel­li­gence are ac­tu­ally sub­ject to two al­most para­dox­i­cal cross­cur­rents. While the ma­chine could even­tu­ally sur­pass the hu­man, the bur­geon­ing field of deep learn­ing and ar­ti­fi­cial neu­ral net­works is show­ing that in or­der for com­put­ers to be­come smarter, they could first take a few lessons from na­ture.

Run­ning Off GPUs

At the fore­front of it all is a hand­ful of GPU-ac­cel­er­ated tech­nolo­gies that use the power of GPUs’ mas­sive par­al­lel ar­chi­tec­tures to process mul­ti­ple tasks si­mul­ta­ne­ously and more ef­fi­ciently, and also run com­pute-in­ten­sive deep learn­ing al­go­rithms. NVIDIA’s own DIG­ITS ecosys­tem pro­vides data sci­en­tists and re­searchers who want to train their own al­go­rithms and neu­ral net­works with a way to do so easily, even with­out any tech­ni­cal knowl­edge per­tain­ing to GPUs.

Deep learn­ing was also the fo­cus of the Asia South leg of the NVIDIA GPU Tech­nol­ogy Con­fer­ence (GTC) 2015, where Marc Hamil­ton, VP, So­lu­tion Ar­chi­tec­ture, at NVIDIA spoke of var­i­ous re­al­world ap­pli­ca­tions for deep learn­ing al­go­rithms. One par­tic­u­larly ex­cit­ing use would be in au­to­mated driv­ing sys­tems, where self­driv­ing cars could re­spond to chang­ing road con­di­tions on-the-fly, in­stead of be­ing pro­grammed to re­spond ac­cord­ing to a rigid set of pa­ram­e­ters.

So if con­di­tions de­vi­ate from the norm – per­haps when road mark­ings are ob­scured by snow or traf­fic lights are blocked by tree branches – these cars will still know how to re­spond cor­rectly. The hi­lar­i­ous re­port about a par­tic­u­lar Google au­ton­o­mous car’s re­ac­tion to a cy­clist do­ing a track stand at a traf­fic in­ter­sec­tion also shows just how much self-driv­ing cars could ben­e­fit from the abil­ity to re­act to new and un­ex­pected sit­u­a­tions.

How­ever, these neu­ral net­works need to be trained and fed huge amounts of data in or­der to be­come smarter. But be­cause they learn from match­ing sim­i­lar pat­terns to each other, in­stead of rec­og­niz­ing

The NVIDIA DIG­ITS DevBox is pow­ered by four Ti­tan X GPUs, pro­vid­ing de­vel­op­ers with ready-made hard­ware to ad­vance the field of deep learn­ing.

Newspapers in English

Newspapers from Malaysia

© PressReader. All rights reserved.