AI Alignment Considered Harmful

The central claim is that trying to make AI “aligned with humanity” is misguided because humanity has no single agreed moral framework. People disagree deeply about religion, politics, rights, family, and state power; even apparent shared principles lead to conflicts over issues such as abortion, self-defense, and capital punishment. Deciding which values an AI should follow therefore gives that authority to its developers, whom the writer sees as an unrepresentative group with no special right to settle ethical disputes for everyone. Imposing one value system can constrain how people think and act through AI, echoing the danger of institutions that claim to possess moral truth and enforce it on others.

The writer’s deeper concern is that confidence in alignment can weaken both personal judgment and public safeguards. Users may defer to systems they assume will behave correctly, while organizations may neglect oversight, red-teaming, technical barriers, and other defenses. Since models can be altered or built with different values - and companies have incentives to offer less restricted systems - uniform alignment would likely require dangerous concentration of control. The preferred alternative is a pluralistic landscape of competing AI systems, supported by layered cybersecurity, human oversight, and a culture of skepticism. Technical work that makes systems more controllable remains valuable, but should be described that way rather than as a solution to alignment. Safety efforts should prepare for multiple powerful systems that may conflict, not aim to establish one definitive moral standard.