近期，关于人工智能可能对人类构成存亡威胁的警告引发关注。1Anthropic公司研究员雅各布·考克松在该公司工作仅四个月后离职，并声称"建造AI的人真诚地相信它可能在这个十年末杀死我们所有人"。1同公司资深成员埃文·胡宾格表示，这种情况在未来十年发生的概率"超过10%"。1
在具体的灭亡场景评估中，科学家们列举了从最模糊到最具体的五个风险路径。1其中包括AI生物武器的威胁——斯坦福大学研究人员已用遗传语言AI模型合成了16种新病毒，成本需数十万美元。1为了对比参照，COVID-19导致的人类死亡率低于1%，而历史上最致命的黑死病在13世纪曾杀死欧洲超过三分之一的人口。1此外，核武库虽已下降，但现存储备仍足以杀死人类的一半。1这些分析表明，AI引发的灾难性风险需要认真对待。
Concerns about artificial intelligence posing an existential threat to humanity have gained prominence following high-profile departures from leading AI safety organizations. Jacob Coxon, a researcher at Anthropic, left the company after four months and claimed that "the people building AI genuinely believe it could kill us all by the end of this decade."1 Evan Hubinger, a senior member at Anthropic, has assessed the probability of such an extinction event occurring within the next ten years at over 10%.1
An analysis of potential AI-driven extinction pathways examines several specific risk scenarios, ranging from unpredictable outcomes to concrete biological and nuclear threats. Stanford University researchers demonstrated one such vulnerability by using generative language AI models to synthesize 16 novel viruses at a cost of hundreds of thousands of dollars.1 For context, COVID-19 resulted in human deaths below 1%, while the Black Death in the 13th century killed more than one-third of Europe's population.1 Meanwhile, nuclear arsenals have declined in number but retain sufficient destructive capacity to eliminate approximately half of humanity.1
评论
还没有评论，欢迎留下第一条。