超级智能:进路及其应对

Superintelligence:Approach and Response

目前人工智能发展中超级智能的现实性与日俱增,这从实现超级智能的三条进路,即通用智能、具身智能和交互智能可以看出,对于人类来说,超级智能本身有失控的风险。因此,作为应对方案的对齐工作显得尤为重要。在对齐理解中,存在两种对齐概念的普遍误解亟待纠错。其中,价值化误解将讨论引入到“价值嵌入陷阱”里,技术化误解实质是技术主义,不仅遮蔽真正的问题甚至还会带来更大的危害。当超级智能来临,我们不仅要考虑人类的未来,还要考虑机器的命运。合约伦理学是一个很好的选择。人与超级智能的“签约”不仅是建立边界的约束关系,更是在共存中相互塑造、互相理解的协同关系。在这种关系中,人类不仅通过超级智能的镜像照见自我,也能更好地理解和塑造自我。

The reality of superintelligence in the current development of AI is increasing gradually,as confirmed by the three approaches to superintelligence-generalized intelligence,embodied intelligence,and interactive intelligence-and there is a risk of superintelligence itself away from the human control.Therefore,alignment as a coping solution is of particular importance.There are two common misunderstandings of the concept of alignment we need to be urgently corrected in the alignment understanding.The value-based misunderstanding leads the discussion into the“value embedded trap”;the technological misunderstanding is essentially technological,which not only obscures the real problem but also brings even greater harm.When superintelligence comes,we should not only consider the future of human beings,but also the fate of machines.Contractual ethics is a good choice.The“contract”between humans and superintelligence is not just a relationship of boundaries and constraints,but also a synergistic relationship of mutual molding and understanding in coexistence.In this relationship,human beings not only see themselves through the mirror of superintelligence,but also understand and shape themselves better.