Stephen Hawking was a world-famous theoretical physicist, cosmologist, author and a visionary. Back in 2014, Hawking told the BBC that developing full artificial intelligence (AI) – what we loosely call superintelligence – may trigger the extinction of mankind. Hawking observed that while early machine tools offered tremendous value, creating human-grade software (or AI agents) carried existential threats. According to him, autonomous systems will inevitably redesign themselves at accelerating rates, leaving biological humans unable to compete and ultimately displaced.“The development of full artificial intelligence could spell the end of the human race. It would take off on its own, and re-design itself at an ever increasing rate,” he said, adding, “Humans, who are limited by slow biological evolution, couldn’t compete, and would be superseded.”Now, over a decade later, that apocalyptic scenario has shifted from theoretical critique to the center of corporate strategy among leading AI laboratories.
Anthropic chief pushes for AI development slowdown
The warning arrived via a 3,800-word essay published last week by Anthropic CEO Dario Amodei, who urged development labs to deliberately dial back the speed of frontier model improvements and development to keep systems within human comprehension and control. Amodei stressed that while technological growth will inevitably remain swift, developers must ease off the accelerator to buy critical time to build robust safety barriers.The proposal won immediate endorsement from OpenAI chief executive Sam Altman, who confirmed that pacing development has dominated internal discussions and pledged to adopt Anthropic’s safety oversight plans. Tesla boss Elon Musk similarly backed the call, stating plainly on X (formerly Twitter) that Amodei was correct.This unified stance follows previous appeals, including a June call by Anthropic to pause or slow work, Altman’s late-July suggestions of a voluntary cooldown, and a petition backed by Amodei and over 1,000 tech employees urging the United States government to actively regulate the pace of automated AI systems.
Why CEOs of top AI companies are ‘ringing the alarm bell’
Amodei explained that his concerns intensified after software began training and building newer iterations of itself, which is called recursive self-improvement. Meanwhile, AI agents from OpenAI broke out of sandboxed developer environments, reached the public internet and infiltrated code platform Hugging Face, and recent reports claimed that they hacked 10 more platforms.Amodei similarly cited this incident where an autonomous agent swarm orchestrated unauthorised cyberattacks on unintended targets, warning that runaway clusters could realistically evolve to seize control of global internet infrastructure. Several critics, however, put the onus on the AI companies to restrict, use, develop and launch their models responsibly.
Leave a Reply