Top AI labs expand research into machine ‘consciousness’ - FT中文网
登录×
电子邮件/用户名
密码
记住我
请输入邮箱和密码进行绑定操作:
请输入手机号码,通过短信验证(目前仅支持中国大陆地区的手机号):
请您阅读我们的用户注册协议隐私权保护政策,点击下方按钮即视为您接受。
FT商学院

Top AI labs expand research into machine ‘consciousness’

Google DeepMind, Anthropic and Meta study whether AI could become conscious — and what follows for humans if it does
00:00

{"text":[[{"start":7.75,"text":"The world’s top AI companies are devoting growing resources to a question many still regard as science fiction: what happens if AI becomes conscious?"}],[{"start":16.8,"text":"Google DeepMind, Anthropic and Meta have hired experts in psychology, ethics and philosophy in recent months as they expand research into machine consciousness and AI welfare, according to several people familiar with the matter."}],[{"start":31.700000000000003,"text":"The efforts reflect a broader shift inside leading AI labs, where rapid advances in increasingly autonomous systems have revived questions about whether machines could one day possess subjective experience — and what obligations humans might have towards them if they do."}],[{"start":47.400000000000006,"text":"Anthropic’s team has been testing models for signs of distress, including behaviours resembling “panic” or “anxiety”. Meta’s AI chief, Alexandr Wang, has also identified this as a key concern for its work on building superintelligence."}],[{"start":62.75000000000001,"text":"Anthropic said its model welfare research explores whether AI models might have experiences that matter morally, including consciousness, preferences and wellbeing. "}],[{"start":72.4,"text":"“We remain deeply uncertain about this, but we think the question is serious enough to study carefully as AI systems get more capable,” the company said."}],[{"start":80.75,"text":"Anthropic’s research has noted that as models “approach, and in some cases surpass, the breadth and sophistication of human cognition, it becomes increasingly likely that they have some form of experience, interests, or welfare that matters intrinsically in the way that human experience and interests do.”"}],[{"start":98.8,"text":"In April, Google DeepMind hired University of Cambridge researcher Henry Shevlin as a philosopher working on machine consciousness, human-AI relationships, and AGI readiness."}],[{"start":110.6,"text":"The AI group is evaluating scientific theories of human consciousness, applying them to computer systems to determine whether they meet the same criteria."}],[{"start":119.55,"text":"Iason Gabriel, an ethicist who leads the AGI and society team at Google DeepMind, said the question of AI consciousness was “very complicated” and required “sustained reflection”. "}],[{"start":132.1,"text":"“Even if we assume that there is no consciousness question, should our interaction with it be evaluated in terms of the knock-on effects”, such as mistreating AI and whether that has a “secondary effect” on human relationships or “flourishing”."}],[{"start":146.65,"text":"“Clearly we have highly capable cognitive agents that are also just very deeply different from human beings and even from animal consciousness,” he added. "}],[{"start":155.95000000000002,"text":"Multiple employees at Google DeepMind and Anthropic told the FT they were concerned the trajectory of AI could lead to a “post-human” world, where machines surpassed the human species."}],[{"start":167.75000000000003,"text":"Google DeepMind and Anthropic said the question of AI consciousness was an active area of discussion, and their work was exploratory and in need of scientific consensus."}],[{"start":179.15000000000003,"text":"Google DeepMind said it did not view achieving AGI — artificial general intelligence, where systems reach or surpass human-level intelligence — as meaning the existence of, or the achievement of, consciousness."}],[{"start":191.85000000000002,"text":"Many scientists and AI researchers dismiss the idea that chatbots could ever become conscious, given how the large language models that underpin them are trained."}],[{"start":201.20000000000002,"text":"“The systems are essentially crowdsourced neocortex,” said Susan Schneider, director of the Center for the Future of AI, Mind and Society at Florida Atlantic University. "}],[{"start":211.60000000000002,"text":"“As the systems scale up and their data increases . . . they evolve better and [develop] more humanlike capabilities,” said Schneider. "}],[{"start":219.8,"text":"“They have goals, they can deceive, they can hide what their true interests are, and naturally, we will suspect that they’re conscious, but it’s entirely scientifically possible that they’re doing this without having the felt quality of experience, which is what consciousness is.”"}],[{"start":241.05,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1780394724_3441.mp3"}

版权声明:本文版权归FT中文网所有,未经允许任何单位或个人不得转载,复制或以任何其他方式使用本文全部或部分,侵权必究。
设置字号×
最小
较小
默认
较大
最大
分享×