▲ Anthropic
As artificial intelligence (AI) begins to create better versions of itself, accelerating research and development, a Wall Street Journal (WSJ) columnist warned on November 2 that it is driving humans into anxiety, meanwhile.
Ben Cohen reported in his regular column, "The Science of Success," that recursive self-improvement (RSI)—meaning AI becoming smarter on its own—is turning into reality much faster than humans expected.
Consequently, AI industry leaders view RSI as a turning point that will transform not only the tech industry but also the global economy and the destiny of humanity, leaving them feeling uneasy about its pace.
According to data from Anthropic tracking and analyzing 15,000 R&D tasks found in internal Slack messages and documents by mobilizing Claude AI agents, the proportion of tasks where AI directly performs most of the R&D under human supervision (Level 5 out of 6) has surged explosively.
Just a year ago, it stood at a mere 1%, but by August, Claude itself drove a staggering 26% of the next-generation Claude R&D work.
At this rate, reaching the final automated Level 6, where AI excludes humans, may depend only on time, computing power, and human decisions, Cohen pointed out.
Jack Clark, co-founder of Anthropic, revealed in a recent International Monetary Fund (IMF) speech that he was deeply shocked when he returned in February of this year after taking paternity leave in November of last year to find engineers entirely dependent on AI for coding tasks.
He declared, "The moment has arrived where AI reaches the capability to build itself," moving up the timeline for RSI—where AI improves itself—from late 2028 to 2027.
Cohen introduced that mathematicians already warned of "machines improving themselves" 70 years ago.
At a Dartmouth workshop led by mathematician Claude Shannon in 1956, mathematician I. J. Good noted, "The first ultra-intelligent machine is the last invention that man need ever make, provided that the machine is docile enough to tell us how to keep it under control."
The process of AI training and fine-tuning AI is a key to the rise and fall of corporations, but at the same time, it is instilling a fear of losing control among developers themselves.
Just like in science fiction novels, scientists are leaving research labs in succession, warning of humanity's doom.
Cohen emphasized, "Anthropic CEO Dario Amodei also maintains his vigilance, stating that if not properly controlled, it could completely exceed human understanding and capability, so it must be pursued with extreme caution."
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.