OpenAI Backs Duke University Research to Predict Human Moral Judgments
OpenAI provides a $1 million grant to Duke University researchers to develop algorithms that predict human moral judgments in complex scenarios. The project raises significant questions about whether artificial intelligence can accurately navigate the nuances of ethics.
OpenAI provides a $1 million grant to Duke University researchers for a project focused on predicting human moral judgments. The three-year grant supports professors Walter Sinnott-Armstrong and Jana Borg as they explore how algorithms handle ethical conflicts in fields like medicine, law, and business. The principal investigator declines to discuss the specifics of the ongoing work, which is scheduled to conclude in 2025.
The researchers involved have a history of exploring AI as a "moral GPS" to assist humans in making complex decisions. Their past collaborations include creating a morally-aligned algorithm to allocate kidney donations and studying public willingness to let AI make ethical choices. This new OpenAI-funded project aims to train machine learning models to forecast how people resolve moral dilemmas where competing values clash.
Skepticism surrounds the idea that current technology can truly understand such a nuanced concept as human morality. Previous attempts at ethical AI, like the Allen Institute's Ask Delphi tool, fail spectacularly because machine learning models simply recognize statistical patterns rather than grasping genuine ethical reasoning. Consequently, these systems tend to parrot the dominant cultural viewpoints found in their Western-centric web training data instead of demonstrating authentic moral comprehension.