Anthropic's Chief Scientist Jared Kaplan has issued a stark warning that humanity will soon face a monumental decision: whether to allow advanced artificial intelligence systems to train themselves. In a recent interview, Kaplan stated this critical juncture could arrive between 2027 and 2030, a choice he describes as "the ultimate risk" .
Kaplan explained that permitting this recursive self-improvement could trigger an "intelligence explosion," potentially birthing artificial general intelligence (AGI) that surpasses human intellect. While this could accelerate major scientific breakthroughs, he warns it equally risks creating systems that "snowball beyond our control," leaving humans at the mercy of their own creation. "You don't know where you end up," he remarked of the process.
Before co-founding Anthropic in 2021, Kaplan joined OpenAI in 2019. He is a professor of physics educated at Stanford and Harvard, and previously conducted research at Johns Hopkins University and at CERN in Switzerland. His immediate forecast is also jolting. Kaplan predicts that AI will be capable of performing "most white-collar work" within the next two to three years. He suggests that the technology his company is building could render traditional human advantages in tasks like writing and mathematics obsolete for future generations.
These dramatic warnings, however, are not universally shared within the AI community. Yann LeCun, Meta's chief AI scientist, argues that the large language models (LLMs) behind current chatbots have fundamental limitations. He contends they lack true reasoning and a model of the physical world, suggesting the path from today's AI to superintelligent AGI is far from certain or inevitable.
Other experts urge a focus on more immediate, tangible impacts often overshadowed by apocalyptic narratives. Researchers highlight AI's staggering environmental cost, including surging energy and water use from data centers, alongside complex copyright issues and concerns about its effects on human cognition and mental health. Meanwhile, studies from institutions like MIT suggest AI is currently more likely to augment human workers than replace them outright, emphasizing uniquely human skills in empathy, creativity, and judgment.
