Can machines discern right from wrong? Should they? From facial recognition biases to the ethics of lethal autonomous weapons, AI’s moral dilemmas are growing in complexity. OpenAI recently awarded a $1 million grant to a Duke University research team to tackle one of the most fundamental questions in this field: Can AI predict and guide human moral judgment?
News headlines frequently spotlight ethical failures—like biased algorithms leading to discriminatory hiring or surveillance systems threatening privacy. Against this backdrop, Duke’s research offers a proactive approach: using AI to aid in ethical decision-making rather than merely reacting to its missteps.
The study, titled Making Moral AI, is being led by Walter Sinnott-Armstrong, a professor of practical ethics at Duke, and co-investigator Jana Schaich Borg of the university’s Social Science Research Institute. The research conducted through Duke’s Moral Attitudes and Decisions Lab (MADLAB) aims to design AI algorithms that serve as a “moral GPS,” guiding users toward ethical decisions.
The growing ethical challenges associated with AI extend far beyond coding errors. AI systems are already making life-altering decisions in sensitive areas like healthcare, criminal justice, finance, and national security. These systems often function as black boxes—making recommendations based on patterns in data that even their developers may not fully understand.
In each case, AI’s involvement in decision-making introduces a moral dimension. But, traditional methods of programming algorithms fall short of addressing ethical complexity. That’s where Duke’s MADLAB comes in: MADLAB takes an interdisciplinary approach, as it integrates computer science, philosophy, psychology, neuroscience, and game theory. By analyzing patterns in moral judgments, the lab hopes to create tools to support professionals in high-stakes fields—such as healthcare, law enforcement, and governance—where ethical dilemmas are frequent and impactful.
While AI-driven moral guidance could prevent errors like racial biases in predictive policing or ethical oversights in medical resource allocation, it also introduces its own set of challenges. Critics argue that encoding morality into machines risks reducing nuanced human values into rigid, oversimplified rules. Furthermore, who decides which moral framework to adopt? Western ethics might emphasize individual rights, whereas Eastern traditions may prioritize community welfare.
Besides, AI systems trained on biased or incomplete data could perpetuate existing social inequities. As Sinnott-Armstrong notes, “The goal is not to replace human moral reasoning but to enhance it.” This delicate balance—leveraging AI to assist without undermining human autonomy—remains a focal point of the research.
This project arrives at a critical juncture. Ethical failures could have dire consequences with AI increasingly integrated into judicial systems, healthcare diagnostics, and military applications. A well-designed “moral AI” could revolutionize how society approaches complex dilemmas, offering guidance in ambiguous situations like triaging patients during a pandemic or setting equitable bail conditions.
Efforts like the European Union’s AI Act and UNESCO’s Recommendation on the Ethics of Artificial Intelligence are setting foundational guidelines. Duke’s research could complement these initiatives by providing empirical insights into human moral behavior.