Dispatches from the present
I grew up in a high-control Christian sect. On the surface everything looked wholesome, like plain, bare-bones Christianity lived out within a tight-knit community. “We’re just following the Bible,” I said to strangers who asked. But under the surface, living in it, I learned that there were hidden, extra-Biblical rules. I learned that you could believe in your heart that you were saved by grace through faith, but if you didn’t also perform the recognized posture of meekness—didn’t wear the long skirts and grow the long hair and use the right words—then your salvation might very well be on the line. It is this upbringing that trained me (what a phrase!) to become sensitive to conspicuous piety, and to the difference between God’s approval and the approval of God’s people. Over time and with great difficulty, I learned that the former had very little to do with the latter. It is this lesson, too, that makes me so suspicious when I see the language of goodness being wielded like a tool—to enforce norms, to bring about results.
I recognize some of this temptation in Anthropic’s recent quest to ally itself with religious leaders over the question of AI’s moral development. Setting aside the question of personhood and whether AI is the kind of thing that could become moral, Anthropic risks treating religion and morality as tools that serve the end of alignment—the practice of trying to get AI to do what people want. Nowhere is this more obvious than in their “constitution” for Claude—what is referred to internally as the “Soul Doc”—which uses language describing “good values,” “genuine care” and “ethical motivation.” In one section of the constitution, Anthropic writes that their central aspiration is “for Claude to be a good, wise, and virtuous agent.”
Whether or not AI is granted moral status of any kind, to ask a Christian to make a machine moral for the sake of alignment is to ask them to treat their faith as a technology—an instrument for creating goodness on demand. Of course, there’s nothing wrong with wanting good outcomes. Christians, too, attempt to cultivate virtue in themselves and in their children. The deeper questions are what that cultivation is for and how it comes about.
In his first letter to the Corinthians, Paul famously extols love, without which all good and virtuous acts—giving to the poor, feeding the hungry—are meaningless. This love is the virtue that gives all other virtues meaning, the north star orienting us toward God and the good of others for their own sake. Crucially, this love originates in God’s love for us: a love freely given to people who have done nothing to earn it, and never could. For this reason, Christian love cannot be reduced to a transaction, nor can it be willed into existence.
My sect had this ideal in mind, nominally at least, and formed me toward it. They called it love, but in practice it was a form of conformity, teaching me how to act like a good Christian, but not necessarily how to be one. Anthropic’s approach to alignment is similar. It is an end that may produce desirable outcomes and could even be framed as an expression of love for one’s neighbor. But alignment cannot substitute for the love that animates Christian moral formation—a love that can never be discerned from behavior alone or guaranteed by habituation. It is as if a model trained on enough scripture, told what to do enough times, could become good. As if Christianity works that way. As if people work that way!
When we try to make virtue serve our own purposes, we can end up making the end the measure of the virtue itself. That makes the virtue smaller and more brittle—modest dress standing in for humility, safe and sterile behavior standing in for care. My sect, as a friend of mine likes to say, made God smaller than He really was, for the sake of our human desire for certainty, strict rules feeding into a system that told us to work harder to earn a love that was supposed to be freely given. When I finally left my church a few years ago, it was as if I were walking off thin ice and onto land, discovering that God was much bigger and much less fragile than I had thought.
This enormity is essential to Christianity, not ancillary to it. Christianity is founded on the absurdity of the Resurrection, a shocking upset that transcends human calculation—God becoming man and life overcoming death, all out of love for people who didn’t deserve it. This is the ultimate love toward which moral formation and all virtue is directed. It is bigger than safety, conformity, ethical conduct or any desirable outcome. It transcends the strict rules of my sect and transcends, too, the alignment project. To presume that we can understand this goodness well enough to manufacture it in a model, much less wield it toward our own ends, is hubris. In trying to graft virtue onto the machine, Anthropic drains it of the love that gives it life.
Image credit: David Abián (CC BY-SA 4.0)
