Edward Snowden: Researchers Ought to Prepare AI to Be ‘Higher Than Us’

AUSTIN, Texas — Synthetic Intelligence (AI) fashions may quickly surpass people’ capabilities, however provided that we cease educating them to assume like us and permit them to “be higher than us,” Freedom of the Press Basis President Edward Snowden mentioned at Consensus 2023.

The previous NSA whistleblower, who joined the convention just about from Russia, shared his cautiously optimistic views on the way forward for AI, the expertise that exploded into the mainstream after OpenAI’s ChatGPT roll out final fall. Whereas Snowden at occasions echoed some consultants’ warnings that AI applied sciences may empower dangerous actors, he additionally thought-about constructive use circumstances for the rising expertise.

Snowden argued AI fashions may impede authorities surveillance somewhat than gasoline invasive intelligence applications.

“Possibly they may cease spying on the general public and begin spying for the general public,” Snowden mentioned. “That’d be a web good.”

Nonetheless, he additionally warned that the launch of ChatGPT and different more and more refined AI fashions, may gasoline huge tech and government-driven initiatives to encroach upon customers’ privateness.

To be able to forestall dangerous actors from co-opting AI applied sciences, folks should combat for open AI fashions to stay open, Snowden informed CoinDesk.

“Individuals are going to be elevating the crimson flag of ‘software program communism,’ the place we have to declare the fashions have to be open,” Snowden mentioned.

He aimed his criticism particularly at rising AI fashions which can be turning into much less and fewer open, calling out OpenAI particularly.

“It’s a poor joke, proper? They refused to supply public entry to their buying and selling knowledge, their fashions, the weights and so forth – however they’re a pacesetter within the house. They’re being rewarded. They’re being rewarded for delinquent conduct.”

How the expertise is used, he argued, boils right down to how researchers prepare AI engines. A lot AI coaching at present entails spoon-feeding the AI massive quantities of on-line content material, together with social media feedback, which Snowden says will not be best.

​​”They’re coaching [AI models] on Reddit threads,” Snowden mentioned. “It’s just like the web equal of YouTube feedback. However you wish to create one thing respectable, good, that’s artistic and helpful.”

Present coaching strategies, he additionally noticed, revolve round educating AI fashions to “assume like us,” which may restrict the expertise’s potential to higher humanity.

“As with kids, we don’t want machines to be like us,” Snowden mentioned. “We want them to be higher than us. And in the event that they aren’t higher than us, we did a horrible job.”