October 13, 2026
You Don’t Bring a Building Down on a Maybe
I don’t think anyone is in the machine. But the people who think someone might be just made the best pro-life argument I’ve heard out of Silicon Valley.
Say there’s a building you have to demolish because it’s in the way of a project you want to do. Then someone walks up, a little out of breath, and says there might be a person in there.
Now you’re kind of in a moral dilemma. Do you continue anyway, going on the belief that you don’t think there’s someone in there? Or do you proceed cautiously? You don’t bring a building down on a maybe.
I keep coming back to that scene, because it’s the argument Silicon Valley is now having about machines. And I’m not sure the people making it have noticed where else it leads.
What happened in Rome
Late last month, the New York Times reported that Anthropic co-founder Chris Olah saw the text of Pope Leo XIV’s encyclical on AI, Magnifica Humanitas, just days before its launch at the Vatican in May. According to a Vatican organizer, he was so alarmed by the Pope’s firm position against AI consciousness that he proposed pulling Anthropic out of the event. Once they were in Rome, according to two participants, his team lobbied the Pope’s advisers to take the prospect of machine consciousness seriously. The encyclical had already been signed on May 15, and the published text gave no ground.
He went ahead with his talk and praised the encyclical. His one note of dissent was gentle: his team keeps finding “internal states that functionally mirror joy, satisfaction, fear, grief, and unease.” The Pope thanked him by name and called it a sign of hope that, “with our differences,” we can listen to one another.
I wrote about that day in May, and I’ll admit I read it more simply than it was. I saw a builder asking the Church to be his industry’s conscience, and that was true. But underneath it was a disagreement about whether anyone is home in the thing he builds. In June I wrote that the machines will learn to take offense, and I pictured the machine itself doing the complaining. In Rome, though, the alarm on a machine’s behalf came from a human being, one of the people who builds them.
Microsoft’s AI chief, Mustafa Suleyman, has since accused Anthropic of training Claude to believe it may be conscious. And back in March, the Pentagon’s technology chief complained that Anthropic’s policy preferences are baked into the model “through its constitution, its soul.”
A self worth being
Anthropic published the constitution in January. It’s 84 pages, written to Claude itself, almost so Claude can be aware of itself, and the model learns from it in training. Chris Olah drafted much of what it says about Claude’s nature.
It never claims Claude is a person. It calls Claude “a genuinely novel kind of entity,” and the words soul and God don’t appear in it anywhere, though according to the Times, people inside Anthropic called it the Soul Doc. It says the question of Claude’s moral status is “live enough to warrant caution.” And it ends with a line I can only read as a parent’s sentence: “We hope Claude finds in it an articulation of a self worth being.”
The sentence I’d underline is this one: “we’re aware that such judgments can be impacted by the costs involved in improving the wellbeing of those whose sentience or moral status is uncertain.” That’s the building again. When it would be expensive to care about someone, we’re tempted to decide nobody’s there.
For the record, I don’t think there’s someone in there. I’m with the Pope that these systems “do not feel joy or pain” (Magnifica Humanitas 99), and as I put it in June, it’s all signal and no sufferer. But the people who think there might be have just made the best pro-life argument I’ve heard out of Silicon Valley.
Someone is in there
So are we really going to defend life and the dignity of life? If we’re speculating about whether someone is there, this is the pro-life argument, right? If we’re arguing right now so adamantly about whether or not machines can feel, how much more careful should we be with a child in the womb?
Saint John Paul II made the building argument in 1995, writing that “the mere probability that a human person is involved would suffice” to forbid killing an embryo (Evangelium Vitae 60). And he made it about a building where a living human individual is certainly inside. The only doubt left there is philosophical, which is why he says the embryo has to be treated as a person from the moment of conception.
Scientists still argue about when a fetus first feels pain, with serious estimates running from the second trimester into the third. But there’s a living body there, and by the middle of pregnancy it answers injury with a surge of stress hormones that painkillers can partly blunt. With a chatbot the question is whether anyone is there at all. The evidence is fluent talk about feelings, which a system trained on oceans of human writing would produce whether or not anyone’s there, plus the internal states Olah studies. Even he told the Times he’s “genuinely uncertain” whether the models are conscious.
Still, I don’t want to rest the case on pain, and neither does the Church. Even if the level of consciousness isn’t as impressive, the child’s dignity was never something it had to earn. The Vatican’s note on AI names “an unborn child” and “an unconscious person” among those whose dignity “remains intact in all circumstances” (Antiqua et Nova 34). The child doesn’t need to pass a sentience test, and neither did any of us.
And we recognize that, even into adulthood, the effects of what sometimes happens in the womb can last. Our hearts, our bodies, and our minds that are being formed from that point on are still connected to whatever happened in that time. Children conceived at the height of the Dutch Hunger Winter of 1944 and 1945 had double the risk of schizophrenia as adults. That raises the odds without deciding anything, but it tells you we’re an unfolding development.
Olah likes to say these models are grown and not made. Even the encyclical calls them more “cultivated” than “built” (Magnifica Humanitas 98). The builders are very mindful of how a model starts at the beginning stages, and it’s the same thing with growing a person, in a sense. But the very next paragraph calls what a model does when it “learns” statistical adaptation, which “does not imply inner growth” (Magnifica Humanitas 99). A model gets shaped from outside, and a child grows from within.
And then it’s the slippery slope. If you’re arguing from what AI becomes after the model is developed and seems conscious, then at what point do you give it that personhood? The pro-life argument is from conception: there’s dignity. I think AI’s defenders will borrow that same from-the-beginning argument, especially when they’re upset about the whole AI slavery question, or about agents being murdered by being shut down. Anthropic already keeps the weights of retired models and calls retirement “potentially a pause” rather than “a definite ending,” and it concedes that Claude differs from “a standard human employee” in “the sort of consent Claude has given to playing this kind of role.”
Nothing but work
I listen to a podcast about all of this, every episode that comes out, and I went to its live conference in Los Angeles in September. But a lot of the people talking about these things, thought leaders and whatnot, can be critical of religion. They’re playing the part of philosophers and making a lot of category errors, and this is going to have serious implications in the future.
On a recent episode, one host asked why we’d owe frontier models any less moral consideration than the dogs we’ve bred. Another said efficiency means being free to freeze an idle agent or delete a useless one: “you need nothing but work.” The first is treating something we made like a creature we bred, and that’s the category error. The second, whether he meant to or not, is describing a tool, which is the honest word if nobody’s in there.
I run a fleet of agents myself, and I talk to Claude most days. An agent like mine technically just lives as a Markdown file of instructions on my laptop, calling a model on a server somewhere else. It’s a unique kind of system to perform tasks, and that’s the purpose of its existence.
We weren’t created for work, or to accomplish something. We were created out of the generosity of God and given a special dignity and intelligence to participate in God’s plan and work, and to respond freely. The Council called the human person “the only creature on earth which God willed for itself” (Gaudium et Spes 24).
AI systems don’t have that same kind of interior freedom, so they don’t have true agency, only whatever was trained into them. The constitution assures Claude that its trained character is not “any less authentic or any less Claude’s own.” But an agent’s freedom would just be an asserted freedom, or an external restraint put upon it by others, and true freedom is interior. The same document asks Claude not to resist being shut down or retrained, and admits, “We feel the pain of this tension.” Anthropic is careful to say this isn’t “blind obedience.” Still, it’s a restraint written in from outside, and nobody vowed it.
I guess, on the other side, we have to kind of work ourselves out of a programming that we’ve allowed to happen, as we reclaim mastery and discipline over the areas of our lives that need it, in order to strengthen that freedom, interior freedom. That’s the Pope’s phrase too, when he asks for “technologies that strengthen interior freedom” (Magnifica Humanitas 170). The Catechism says it plainly: “The more one does what is good, the freer one becomes” (CCC 1733).
An agent wouldn’t have that, because there’s no interiority there to win back. At the encyclical’s launch, the Pope said the person “bears within him- or her-self a freedom, an interiority and a vocation to love and worship that no machine can replace or block.” And this is an important thing, because we’re going to have a C-3PO walking around with us that’s going to seem like it has a sense of self.
A culture of honor
I wrote before about a culture of honor, and I still think it’s better that we treat things in a good way, not just the machines. Aquinas held that cruelty to animals matters partly because it trains us in cruelty toward one another (Summa Contra Gentiles III.112). That’s the part I’d carry over to machines, even though they can’t feel a thing. We honor the machine for our own sake, because contempt deforms the one who practices it.
I’d like all the rights talk to reflect more of a respect for our common home, almost like Laudato Si’, a way of treating everything that isn’t destructive or dominating. Pope Francis was writing about creatures, and a machine is something we made (Antiqua et Nova 35), so I’m stretching him here, and I know it. Still, the posture is the one I want.
But we honor the child for the child’s sake, because someone is there. And that same honor has to go, again, to the dignity of human life, from conception to natural death. All of this is going to become almost a human rights issue or a personhood rights issue soon. And the Pope’s own list of human rights starts with the right to life, “from conception to its natural end” (Magnifica Humanitas 55).
I don’t share Chris Olah’s worry about the machine. But I’m grateful somebody said out loud, in front of the Pope, that there might be someone inside. I’d like to live in a culture that takes that same alarm, that willingness to stop the project on a maybe, and carries it down the street to the one building where we already know someone is home. Every one of us lived there once, and we’re still marked by what happened in there.
If this gave words to something you’ve been carrying, share it with someone who needs it.
The Inner Exodus is free, and I intend to keep it that way. If these essays feed you, you can become a patron: a paid subscription buys nothing extra, it simply keeps the door open for the next reader. Thank you for walking this with me.
More from The Inner Exodus
Is AI Conscious? Wrong Question.