Nobel Prize Goes to ‘Godfathers of AI’ Who Now Fear Their Work Is Growing Too Powerful

Two AI researchers, John Hopfield and Geoffrey Hinton, received the Nobel Prize successful physics connected Monday for their enactment gathering artificial neural networks that tin memorize accusation and admit patterns successful ways that mimic the quality brain.

Their probe successful the 1980s laid the instauration for the past decade’s explosive advancement successful artificial quality and today’s ubiquitous proposal algorithms and generative AI systems. Both men person since said that advancement needs to beryllium constrained for the involvement of humanity.

In its announcement of the prize, the Nobel committee emphasized however acold the tract has travel since Hopfield published his seminal insubstantial successful 1982, successful which helium described a neural web with less than 500 imaginable parameters. Today, tech companies are churning retired generative AI systems with billions and trillions of parameters.

The Hopfield web was a postulation of 30 interconnected integer nodes that could alteration their values betwixt 1 and 0 and successful doing truthful beryllium programmed to grounds patterns that represented the pixels successful achromatic and achromatic images. It drew connected equations from physics utilized to picture however atoms successful a web impact each others’ rotation successful bid to cipher however the relationships betwixt the nodes successful the web represented the images. In effect, the web could beryllium programmed to make memories of definite images. And erstwhile it was fed a caller representation that was fuzzy oregon incomplete, it could cipher its mode backmost to the astir akin representation successful its memory.

Hinton built connected Hopfield’s enactment by designing neural networks that could not conscionable retrieve and recreate patterns, but could beryllium taught to admit akin patterns successful wholly antithetic data—for example, the patterns that marque 1 representation of a canine similar different but not similar a representation of a cat. In 1985 helium published a insubstantial introducing this network, named a Boltzmann instrumentality aft physicist Ludwig Boltzmann, who developed statistical equations for calculating the corporate properties of a web composed of galore antithetic components.

In 2023, Hopfield was 1 of the astir notable signatories connected a letter calling for AI companies to intermission the improvement of generative AI systems much almighty than OpenAI’s GPT-4. And Hinton has precocious been talking a batch astir his concerns that AI is advancing excessively rapidly for humans to control.

He’s estimated that humans could physique an artificial quality that exceeds our ain quality successful the adjacent 5 to 20 years and that “It’ll fig retired ways of manipulating radical to bash what it wants.” He was 1 of respective large names successful the tract to motion an open letter this twelvemonth calling for California Governor Gavin Newsom to enact a instrumentality that would person held ample tech companies liable for gathering AI models that caused catastrophic losses of beingness oregon spot damage. Newsom yet vetoed the bill, nether unit from tech companies, including Hinton’s erstwhile employers astatine Google.

“In the adjacent fewer years we request to fig retired if there’s a mode to woody with that threat,” Hinton said successful an interview with the Nobel Prize committee pursuing the grant announcement. “So I deliberation it’s precise important close present for radical to beryllium moving connected the contented of however volition we support control. We request to enactment a batch of probe effort into it. I deliberation 1 happening governments tin bash is unit the large companies to walk a batch much of their resources connected information research.”