Gerrit Gong proposes ‘Humanity’s First Exam’ to test AI’s moral reasoning

Elder Gerrit W. Gong proposed “Humanity’s First Exam” at the G20 Interfaith Forum in Salt Lake City on October 17, 2026, calling for trustworthy systems guided by human values—not simply models that reason more powerfully.

Gong, a member of the Quorum of the Twelve Apostles of The Church of Jesus Christ of Latter-day Saints, framed the proposal as a counterpoint to Humanity’s Last Exam. That existing test contains 2,500 questions across subjects including mathematics, physics, chemistry and the humanities. His proposed benchmark would instead examine moral reasoning and whether AI is grounded in a human-centered moral compass.

Accuracy, respect—and knowing when to stop

Questions about faith give the proposal a concrete focus. Gong said people already ask AI about religion and belief, making accurate, honest and respectful answers a matter of common interest. He also argued that models should acknowledge their limitations and know when not to answer personal questions. His standard therefore includes restraint, not just the ability to produce a fluent response.

He credited focused work with reducing stark anti-Islamic and anti-Semitic bias. But he said models still show religious bias beyond initial inquiries and leave out faith-based perspectives when answering important life questions. Those omissions, he argued, can spread and compound bias at scale.

Gong’s longer-term concern was conditional: if AI becomes capable of improving itself, its reasoning could advance while its ethical grounding remains incomplete. He warned that the gap between reasoning ability and moral judgment is widening, making early choices about how to guide AI consequential.

AI will not fulfill its full potential until we make it as morally good as we make it powerful.

Elder Gerrit W. Gong, G20 Interfaith Forum keynote

Religious representation without religious instruction

Gong connected that ambition to work by the Consortium for Evaluating Faith and Ethics in AI, or CEFE-AI. The collaboration is developing faith-related benchmarks and includes faculty from six universities: Baylor, BYU, Notre Dame, Pepperdine, Aristotle University of Thessaloniki and Yeshiva. Consortium researchers said their initial studies found systematic religious bias in AI.

He drew a boundary around that work. The consortium does not aim to make AI overtly religious, persuade users to adopt a religion, favor one faith or settle competing religious claims, he said. Its goal, which he also assigned to Humanity’s First Exam, is to improve how AI represents all religions.

Gong described faith traditions as potential contributors to benchmark design and as influences on AI companies’ choices about human flourishing, families and faith. That role was not an argument against using the technology. He said his church already uses AI for humanitarian and educational services, translation, communication, and pastoral and community support. He paired those applications with advice to church members: use AI to enhance human effort, not replace it.