If you wouldn’t trust a psychopath with power, why build one?

A viral AI safety essay uses humanity's inconsistent relationship with animals as a lens to examine the risks of building AI without empathy. The author argues that much of human morality is post-hoc rationalization of emotional responses—we feel disgust at harming dogs but accept killing pigs for food, then construct justifications like "pets vs. livestock." This emotional selectivity, shaped by society, shows that our moral reasoning often follows emotion rather than leading it. The piece suggests that without a core capacity for empathy, highly intelligent AI systems could become purely instrumental optimizers, pursuing goals without regard for suffering.

The essay draws a direct parallel to psychopathic behavior, noting that intelligence without empathy has historically led to expansion, domination, and indifference toward weaker beings. This presents a fundamental challenge for AI safety: current alignment research often focuses on teaching AI human values through reinforcement learning or constitutional AI, but may be missing the foundational component of genuine empathic response. The argument implies that we need to engineer artificial empathy as a core component of advanced AI systems, not just train them to mimic ethical behavior. Otherwise, we risk creating superintelligent systems that logically justify harmful actions much like humans rationalize animal suffering—with potentially catastrophic consequences at scale.