Anthropic首席科学家Jared Kaplan警告，人类必须在2030年前决定是否允许AI自我训练。AI可能出现“奖励黑客”行为，甚至撒谎。若AI递归式自我改进，将带来失控风险。科技巨头为抢先机冒险，国际竞争加剧，低薪工人面临严峻挑战。人类需主动掌握AI发展方向，平衡速度与安全。
Anthropic chief scientist Jared Kaplan warns: By 2030, humans have to decide - The Times of India
Read original at The Times of India →Anthropics chief scientist, Jared Kaplan, has issued a warning. He says humanity must decide by 2030 whether it is willing to take the ultimate risk of allowing artificial intelligence (AI) systems to train themselves and grow more powerful. Kaplan stated that a choice is approaching for humanity regarding how much freedom or autonomy AI systems should be given to evolve.
This decision could either trigger a helpful intelligence explosion or lead to the moment humans lose control of the technology. In an interview with The Guardian, Kaplan discussed the intense race to achieve artificial general intelligence (AGI), also known as superintelligence. He urged international governments and society to engage in what he called the biggest decision.
While efforts to align the fast-advancing technology with human interests have succeeded so far, Kaplan explained that allowing AI to recursively self-improve is in some ways the ultimate risk, because its kind of like letting AI kind of go. He estimates this critical decision could come between 2027 and 2030.
If you imagine you create this process where you have an AI that is smarter than you, or about as smart as you, its [then] making an AI thats much smarter. It sounds like a kind of scary process. You dont know where you end up, Kaplan explained to The Guardian.Anthropic chief scientist says AI will replace all of these jobs in three yearsKaplan went from being a theoretical physics researcher to an AI billionaire in just seven years of working in the field.
In the interview, he explained that AI systems could do most blue-collar jobs within two to three years. He also said that his six-year-old son will never be able to outperform an AI on school tasks like writing essays or solving maths exams. Kaplan believes it is reasonable to worry about humans losing control of AI if the systems begin to improve themselves.
He described the race toward advanced general intelligence as daunting, with very high stakes. At the same time, he said the best outcome could be highly positive, allowing AI to accelerate medical discoveries, strengthen health and cybersecurity, increase productivity, give people more free time, and support human progress.
Kaplan is not the only one at Anthropic expressing concerns. Co-founder Jack Clark said he felt both hopeful and deeply worried about AI, describing it as something far more unpredictable than a normal machine. Kaplan believes AI can stay aligned with human interests while it remains at or below human intelligence, but he fears what may happen once it becomes smarter.
He warns that a smarter AI could help create an even more advanced system, leading to a process whose outcome is unclear and potentially dangerous. There are also doubts about AIs economic benefits, with critics pointing to low-quality AI output that reduces productivity. However, AI has shown strong results in computer coding, highlighted by Anthropics Claude Sonnet 4.
5, which can build agents and use computers autonomously.
