A consortium involving six faith-based universities is developing benchmarks to examine how AI systems represent religions, Deseret News reports. At a G20 Interfaith Forum in Salt Lake City, Latter-day Saint leader Gerrit W. Gong also called for a proposed “Humanity’s First Exam” to test AI’s moral reasoning, not only its ability to solve problems.
Watch Desk analysis
What happened
Gong said the proposed benchmark should help assess whether AI systems are trustworthy and guided by human-centred values. The consortium, called CEFE-AI, includes faculty from Baylor, BYU, Notre Dame, Pepperdine, Thessaloniki Aristotle and Yeshiva universities. Researchers told Deseret News that its first studies found systematic religious bias in AI.
The report does not set out the studies’ methods or provide a specification for Gong’s proposed exam. It says the consortium does not intend to make AI overtly religious, promote a faith or adjudicate competing religious claims. Its stated goal is to improve how AI represents all religions.
Why it matters
People ask AI systems questions about faith and belief, as well as using them for translation, education and other services. The consortium’s work puts a concrete question on the table: how could developers test whether systems treat different religious perspectives accurately and respectfully?
A benchmark can make a problem easier to measure, but its usefulness will depend on what it tests, who helps design it and whether results lead to changes. For now, the report describes early research and a proposed evaluation, not a finished standard or evidence that AI systems have been corrected.
Our read
The useful part of “Humanity’s First Exam” is the push to test more than whether a model can produce a convincing answer. But “moral reasoning” is a broad ambition, not yet a clear testing method. The consortium’s stated aim of representing religions without privileging one is a sensible starting point; the details will decide whether the benchmark can earn trust beyond its founders.
What to watch
- Whether CEFE-AI publishes its first studies, including the methods and examples behind its findings.
- How the proposed benchmark defines and measures accurate, respectful treatment of religious perspectives.
- Whether AI developers use the evaluations and report what changes follow.
Discussion spark: Should AI companies be expected to test how their models represent religious beliefs, and who should have a say in designing those tests?
Sources and evidence
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.