Researchers propose Constructive Alignment to reframe AI alignment as governance of evolving human preferences

Researchers Max Kanwal and Caryn Tran propose Constructive Alignment, a paradigm that reframes AI alignment as a control problem over evolving human preference trajectories rather than the optimization of static preferences.

The paper challenges the conventional approach to alignment, which treats human preferences as fixed targets to be inferred and optimized. The authors cite empirical evidence showing preferences are layered, dynamic, and constructed through interaction—particularly with adaptive technologies like AI systems.

The framework models preferences as layered state variables that evolve under interaction with AI systems. It formalizes this view using a control-theoretic approach in which system actions and interaction design jointly influence both world states and human evaluative states.

The authors argue that alignment should focus on regulating how AI systems influence the evolution of human preferences, ensuring that value trajectories remain coherent, reflectively endorsed, epistemically grounded, bounded against manipulation, and empowering under uncertainty.

The paper positions alignment as a problem of governing long-term value formation rather than simply satisfying static preferences, emphasizing the role of AI systems in shaping what people attend to, value, and endorse over time.

The work draws on behavioral economics, psychology, and constructivist social theory to ground its theoretical contributions. It was published as part of the AAAI-26 Workshop on Machine Ethics and consists of 23 pages with one figure.