Ideal theory in AI ethics

AI ethics research often relies on "ideal theory"—analyzing AI systems as if they operate in perfect conditions—rather than studying how they actually function in messy, real-world contexts with existing power imbalances and resource constraints. This approach can make ethical recommendations impractical or irrelevant for addressing actual harms.

The structural incentives of academia push researchers toward ideal theorizing, including pressure to publish novel findings, disciplinary norms favoring abstract analysis, and the difficulty of accessing real-world AI systems for study. These incentives can inadvertently distance AI ethics from practical problem-solving.

Ideal theory in AI ethics risks overlooking how historical injustices and current inequalities shape AI development and deployment. By assuming a level playing field, researchers may miss how AI systems perpetuate existing discrimination or how marginalized communities bear disproportionate risks.

The paper argues for "nonideal theory" approaches in AI ethics that grapple with real constraints, power dynamics, and implementation challenges. This shift could make AI ethics research more useful for practitioners and more attentive to whose interests are actually served by AI systems.

Moving beyond ideal theory requires changing how the AI ethics research community operates—potentially through new incentive structures, collaboration with affected communities, and greater emphasis on studying deployed systems rather than hypothetical scenarios.