Limits of AI Moral Enhancement
Section: Articles
Abstract
Artificial intelligence moral enhancement is a sub-topic subordinate to the "human moral enhancement" trend. It is hoped that human moral cognition can be improved through the "artificial moral advisor (AMA)", which is capable of "objective" moral representation and strong information acquisition. However, since AMA cannot meet the "ideal observer (IO)" standard after all, it will inevitably fall into the embarrassment of expertism. In addition, researchers who resort to the construction of "moral machines" also presuppose an IO perspective by constructing learning algorithmic frameworks for automatic ethical classification and decision making. Through the experimental study of Guarini's Moral Case Classifier (MCC), we have found that the hypothetical function without pre-classification could not be fitted, which means that the generalist AMA that fits the IO cannot be built. In view of this, we propose another different "AI moral enhancement" scheme - an "AMA as The Other", and hope that the artificial general intelligence (AGI) will become the basis for its realization.
This article addresses the conceptual and ethical limits of AI-driven moral enhancement and situates the debate within broader concerns about autonomy and technological intervention.