While technology companies compete intensely to create the most advanced artificial intelligence, the question arises of how to instruct a chatbot to discern right from wrong. One of the answers pointed to is religion.
Chris Olah, co-founder of Anthropic, mentioned that such questions transcend the capacity of a laboratory to answer, stating that religious traditions have dealt with these themes for millennia. This growing proximity between faith and technology has generated little public coverage regarding what these companies understand by religion or what they hope to achieve with this approach.
Executives warn about the destructive potential of AI and debate the need to develop models that follow specific moral principles, increasingly turning to religious leaders and philosophers. The premise is that religion can guide AI development in a scenario where technological fears are spreading throughout society.
Since these corporations are not tied to a single faith, the purpose is generally broader: to establish a set of moral guidelines to steer their models. Sam Altman, CEO of OpenAI, expressed the opinion that ChatGPT should reflect a weighted average of humanity's moral view.
However, experts and practitioners of various beliefs warn that this simplified view ignores the fact that traditions and worldviews often contradict each other, arguing that religion does not operate in this manner.
Religion transcends a mere set of rules
Anthropologist Webb Keane, from the University of Michigan, has criticized for decades the popular notion that religion can be reduced to a mere code of rules, separable from the 'complex, practical, and improvised' ways believers experience their faith. The idea of dissociating religion from spheres such as culture and politics persists, although in reality it is intrinsically linked to both.
There are concerns that by seeking ethical advice from religious leaders, AI companies might reinforce an image of religion that neglects the importance of doctrine, rituals, and the connections among its followers. Keane questions: if we assume that religion offers an isolable set of moral principles, using it to train AIs may end up favoring certain traditions and distorting others, particularly those that are not primarily defined as ethical systems.
He observes that there is a very common way of conceiving of religion as a rulebook, which ultimately is strongly associated with biblical religion and, consequently, Abrahamic religions.
Companies seek to expand dialogue with religious leaders
This search does not imply a lack of variety in discussions, nor carelessness on the part of the companies. Meghan Sullivan, a philosopher of religion and leader of an initiative on ethics and AI at Notre Dame University, assesses the process. Although she does not detail all the consulted traditions, Anthropic reported in a May blog post that it met with 'academics, clergy, philosophers, and ethicists from over 15 religious and intercultural groups,' a number that is expected to grow.
Earlier this year, Sullivan participated in an Anthropic meeting that included Christians from various denominations. Comparing it to previous conversations with other technology companies, she was impressed by the depth with which Anthropic addressed religious matters. The company is known for dedicating great attention to AI safety and ethics.
Sullivan described the company's internal culture as being similar to that of a good university campus, characterized by a group of intellectually curious individuals. She classified the meetings as 'very nerdy—I don't know if that's the best way to describe it—in a good way.' The meeting was structured in two phases: first, the company presented its models in a simplified manner; then, the religious leaders commented and debated the topics presented.
'We discussed all kinds of moral issues that this technology raises—from 'Do humans possess a soul?' to 'Should third-year students have access to a chatbot to aid in literacy?'', Sullivan reported. Representatives from Meta, Google, and Amazon also met with Vatican delegates, while Microsoft has intensified its collaboration with religious communities.
Religion is ancient, but its study is not
Religion is generally perceived as something ancient and universal. However, even influential scholars like the late Jonathan Z. Smith recognize this truth in certain aspects, but argue that this perspective does not encompass all complexity. Smith wrote in 1982 that 'man has had all his history to imagine deities and modes of interaction with them. But man, more precisely Western man, has only had the last few centuries to imagine religion.'
Many contemporary academics share Smith's view, questioning the idea that religions constitute distinct and comparable systems, alleging that this is a modern construction dating back to the 19th century. Tomoko Masuzawa, a researcher of religion and author of 'The Invention of World Religions,' points out that Buddhism and Hinduism, for example, were not seen as clearly demarcated traditions by practicing populations before the arrival of European scholars, adding that 'all these categories were not native.'
These categorizations were consolidated by Western academics over time and only underwent a more systematic reexamination in the second half of the 20th century. These established conceptions of religion—including the ease of comparison between different cultural traditions—help explain why Silicon Valley may not feel the need to define what it means by religion.
In trying to represent multiple viewpoints, resorting to religion becomes a seemingly simple method to integrate diverse thought. Masuzawa comments that 'in some way, religion becomes a way of mapping the world,' as if the entire planet had adopted this map.
By aggregating different views, AI risks erasing particularities. Some experts argue that companies should better specify their terms and goals so that the public understands exactly what is expected of this religious dialogue. Anthropic stated in its release that 'this work is not aimed at aligning our models with the worldview of a specific tradition. We want Claude to be inspired by a wide range of perspectives—religious, secular, and political—with equal depth and rigor.' The goal of these conversations, according to the company, is to promote 'careful and accumulated reflection on how good character is truly formed.'
Large language models function by sequentially predicting which elements are most likely to follow, based on patterns detected during training with vast amounts of human content, including religious texts. Paolo Carozza, a member of Meta's independent oversight board and a legal expert in Catholic social thought and technology, fears that by mixing perspectives, the distinctive characteristics of a tradition may end up being lost. He argues that, in practice, they can be homogenized, resulting in a more generic view or one that favors Western traditions. Carozza concludes that while any living and sophisticated tradition evolves organically by interacting with others, AI tends to simply mix and disorganize everything without conferring meaning.