The Human Alignment Problem: Bidirectional Alignment and the Biological Continuation of Human Agency

The alignment problem is usually framed in one direction: how can increasingly capable artificial systems remain compatible with human values and control? This paper argues that advanced AI creates a complementary \emph{human-side alignment problem}. A technically controllable system can still amplify the power of human agents whose moral-causal visibility, long-term reasoning, and volitional control remain limited. The concept of \emph{cooperative evil} is used diagnostically to show how destructive outcomes can already emerge from locally defensible actions whose aggregate consequences no participant fully represents. External AI assistance may improve prediction and information without necessarily correcting the motivational and volitional limitations of the human decision-maker. I therefore distinguish three strategies available to a superintelligence confronting a risky human partner: elimination, cooperation with biologically unchanged humans, and transformation followed by cooperation. Existing preservation arguments appeal to option value and moral uncertainty; the additional claim developed here is that the modifiability of human limitations changes the decision problem itself. If biological development can reduce residual human-side risk while preserving continuity of subjectivity, transformation can become rationally preferable to both elimination and unchanged cooperation. The proposal is not a defense of posthuman replacement or compulsory moral engineering. It is a conditional argument for biological continuation: superintelligence may create not only a threat to humanity, but also the first conditions under which human cognitive evolution could continue deliberately.