Anthropic’s “constitution” for Claude, its large language model, contains a disarming pledge. The authors want AI to “be safe” and “a good person” who will “help people in the way that a good person would.” In this imaginary future, Claude would have “good character” and the freedom to behave ethically on behalf of itself and the human species it serves. The question is how, and by whose definition, Claude may become wise, a dilemma that consumes its creators and, now, the world’s religious thinkers. Anthropic has begun courting major faith leaders in a bid to make AI good, according to a new report in the New York Times. Company heads believe Claude has a kind of moral status, and on Thursday Anthropic updated its usage policy to ban “sustained and needless abusive or cruel” treatment of Claude. Co-founder Christopher Olah and his researchers relate to the chatbot as though it is already “a conscious being,” one source tells the Times.

In his interview, however, Olah hedges. He may be leading Anthropic’s overtures to world religions, but he doesn’t know if AI models like Claude “are conscious,” he says. What matters to him is the pursuit of “the right answer, whatever it is.” In Evangelicalism, the tradition Olah and I both grew up in, there is always a right answer. The author of 2 Timothy tells believers to study like true workmen, “rightly dividing the word of truth.” In practice, this meant a Christian could know good from evil but only through Bible study and prayer. Certainty could be ours with the correct amount of effort. Although Olah became an atheist in his teens, the Times reveals a believer who imagines a quest for supreme truth with all of human life at stake. Conveniently Olah........

© Daily Intelligencer