Beyond Bias and Compliance: Towards Individual Agency and Plurality of Ethics in AI

Data labeling is foundational to AI ethics: The paper argues that how data is labeled—not just bias reduction or compliance frameworks—directly shapes whether AI systems behave ethically. This suggests companies need to carefully examine their labeling practices as a core ethical responsibility, not an afterthought.

Current approaches miss the core of ethics: The paper critiques both "bias and compliance" focused solutions as incomplete, arguing that real ethics requires applying moral principles to specific situations. This means responsible AI requires deeper value judgments, not just technical fixes or regulatory checkboxes.

Users need actual control, not just trust: The authors emphasize that trust in AI is "unjustified unless users experience the direct value of ethical AI." Without giving users real control over how algorithms affect them, companies cannot build genuine trust—only the appearance of it, which fuels public distrust and apathy toward AI ethics efforts.

"Who decides what is good?" is the critical question: The paper identifies three foundational questions for ethical AI, with particular emphasis on who gets to define what "good" means in relation to system design. This highlights that ethical AI requires addressing power dynamics and representation in decision-making, not just technical solutions.

Plurality of values should be preserved: Rather than imposing a single ethical framework, the paper advocates for AI systems that allow multiple value systems and individual expression to coexist, arguing this is essential for AI to reflect genuine moral character rather than enforced uniformity.