Humanity is rapidly approaching a decision that could redefine the relationship between people and artificial intelligence, warns Jared Kaplan, chief scientist at Anthropic. According to Kaplan, the coming years may mark a turning point regarding AI self-training, with major implications for human control and the global labor market, Mediafax reports, citing Futurism.
AI self-training: progress versus loss of control
In an interview with The Guardian, Kaplan said that starting as early as 2027 — and most likely by 2030 — society may be forced to decide whether to allow AI systems to train themselves without direct human oversight.
Read also: Schaeffler bets on humanoid robots and automation by 2035
Such a decision could trigger what he describes as an “intelligence explosion,” potentially leading to artificial general intelligence capable of matching or surpassing human intelligence.
While this scenario could accelerate breakthroughs in science and medicine, Kaplan cautions that it could also push AI development beyond meaningful human control. He describes AI self-training as an “extremely high-stakes decision,” noting that once humans are removed from the loop, oversight becomes uncertain.
Continuous self-improvement, where AI systems refine themselves autonomously, raises fundamental questions about transparency, safety, and whether humans can still understand or guide the behavior of highly advanced systems.
Labor market impact and expert disagreement
Kaplan’s warnings echo concerns raised by other prominent figures in the field, including Geoffrey Hinton and Anthropic CEO Dario Amodei, who have predicted significant disruption to labor markets. Kaplan believes that AI could take over most white-collar jobs within the next two to three years.
Not all experts agree. Researchers such as Yann LeCun argue that current AI architectures are unlikely to achieve true artificial general intelligence. Others question whether AI consistently boosts productivity, pointing to cases where companies replaced workers with AI systems only to reverse those decisions later.
Kaplan acknowledges that AI progress could stall, but maintains that rapid advancement remains the most likely outcome.
Photo: WTR
