Ken Liu: The Real AI Risk Is Humans Behaving Like Machines

The question preoccupying most AI commentary is who loses their job, and when. In a ChinaTalk interview published May 8, 2026, science fiction author and translator Ken Liu argues that question misses the deeper threat. The real danger, he says, is not machines replacing humans but systems that train humans to behave like machines. He locates this pattern historically, pointing to the assembly line and the contemporary call center as examples of how capitalism has repeatedly pressured human beings toward standardization. AI is the newest mechanism, not a categorically different one.

Liu also introduces a concept that does not yet have a product category: the "egolet." He defines it as an AI capable of capturing the part of a person where they say no, all the parts they have denied, not just what they have said aloud. A genuine egolet requires total user control over both training data and hardware. That condition rules out essentially every consumer AI assistant currently on the market, which runs on someone else's servers and learns from data someone else controls.

On the question of intelligence and wisdom, Liu draws on Laozi and Zhuangzi to mark a distinction that AI benchmarks do not capture. Large language models are intelligent, he argues, but cannot be wise: language captures only shadows of truth, and everything that truly matters lies beyond it. Having intelligence completely divorced from consciousness, will, intention, and subjectivity is, in his framing, philosophically novel territory that we are still learning to map. He adds a companion point for creators: AI slop will not stop humans from making art that matters, comparing the current moment to photography's mechanical reproduction era.

The honest caveat is that these are a writer and translator's arguments, rooted in Daoist philosophy and science fiction's long conversation with technology. The interview does not engage with AI researchers who would push back on the wisdom claim, and the piece gives you a reframe rather than empirical evidence. But the reframe is pointed: if the real risk is behavioral conditioning toward standardization, the relevant governance question is not just job displacement but what AI systems are quietly optimizing humans to become.