Ethereum co-founder Vitalik Buterin said temporarily halting globally accessible computational resources might be a way to “buy more time for humanity” in the face of a potentially dangerous form of AI superintelligence.
In a Jan. 5 blog post following up his November 2023 post advocating the idea of “defensive accelerationism,” or d/acc, Buterin said super-intelligent AI might be as little as five years away, and there’s no telling if the outcome would be positive.
Buterin said a “soft pause” on industrial-scale computer hardware could be an option to slow AI development if this happens, reducing available global computing power by up to 99% for one to two years “to buy more time for humanity to prepare.”
AI superintelligence is a theoretical AI model that is typically defined as being far more intelligent than the most brilliant humans in all fields of expertise.
Many tech executives and researchers have aired concerns about AI, with over 2,600 urging in a March 2023 open letter to halt AI development due to “profound risks to society and humanity.”
Buterin noted that his post introducing d/acc made only “vague appeals to not build risky forms of superintelligence” and he wanted to share his thoughts on addressing the scenario “where AI risk is high.”
Still, Buterin said he’d only push for a hardware soft pause if he were “convinced that we need something more ‘muscular’ than liability rules,” which would mean those who use, deploy or develop AI could be sued for damages caused by the model.
He noted proposals for a hardware pause include finding the location of AI chips and requiring their registration. Buterin proposed that industrial-scale AI hardware could be fitted with a chip that allows it to continue running only if it gets a trio of signatures from major international bodies once a week.
Related: AI took giant strides in 2024, as AGI comes into view
“The signatures would be device-independent (if desired, we could even require a zero-knowledge proof that they were published on a blockchain), so it would be all-or-nothing,” Buterin wrote. “There would be no practical way to authorize one device to keep running without authorizing all other devices.”
The d/acc idea supported by Buterin advocates a careful approach to developing technology, in contrast to effective accelerationism, or e/acc, which pushes for unrestricted and unbridled tech changes.
AI Eye: A bizarre cult is growing around AI-created memecoin ‘religions’