AI developers seeking guidance from religious thinkers face questions about whose values their chatbots will reflect, according to an Associated Press report published on October 25. The report from Los Angeles examines consultations intended to shape AI behaviour, alongside warnings that combining beliefs can obscure differences between traditions.
Anthropic described this work in a May 19 announcement, saying it had met scholars, clergy, philosophers and ethicists from more than 15 religious and cross-cultural groups. The company said these early discussions could inform Claude’s constitution, training values and behaviour evaluations; it did not present them as a completed solution.
University of Michigan anthropologist Webb Keane told AP that treating religion as a collection of detachable moral principles risks distorting traditions. His concern is that doctrine, ritual and relationships matter too, and that some religions do not primarily understand themselves as systems of ethical rules.
Meghan Sullivan, a University of Notre Dame philosopher who attended an Anthropic meeting, offered a more positive account to AP. She described substantive engagement: company staff explained the models before participants discussed moral questions, including whether children should use chatbots for literacy education.
Anthropic said its aim was to consider religious, secular and political perspectives without adopting one tradition’s worldview. It also described an experiment giving Claude a tool that reminded it of ethical commitments during tasks. Internal evaluations showed less misaligned behaviour, the company said, while acknowledging uncertainty about whether reminders or pauses explained the effect.
The distinction between consultation and demonstrated improvement remains central. AP reported concerns from legal scholar Paolo Carozza that blending traditions could erase their defining characteristics. Anthropic’s May statement outlined plans to broaden discussions to lawyers, psychologists, writers and civic institutions, test ideas against research and share findings.
Comments