On the 25th of May, Pope Leo XIV released his first encyclical, Magnifica Humanitas – some 42,000 words arguing that artificial intelligence must be “disarmed” before it hollows out human dignity, widens inequality, and consigns the world’s workers to what he called “forced inactivity.” He unveiled it at the Vatican standing beside a co-founder of Anthropic, one of the firms building the very thing he was warning everyone about. Within 24 hours, The Verge ran with a headline fit for a Philip K. Dick story: “Did the Pope use AI to write about the dangers of AI?”
The case against the Holy Ghostwriter rested, fittingly, on another machine. An AI-detection tool called Pangram scored the text at roughly 94% human, 4% AI-generated, and 2% “AI-assisted”, whereas older encyclicals from Francis, Benedict and John Paul II came back 100% human. Snopes investigated, and landed on the sort of verdict that will define our era: unproven. The Vatican, for its part, insisted the first drafts were written by hand, with a pen, on paper.
The problem with all of this is less about whether the Pope’s encyclical on AI was written with the help of AI, and more that we are now consigned to wonder this about everything we read, whether good, bad, or mediocre. This is one major cost we all pay in the age of AI: reflexive suspicion. Sometime in the last couple of years, a new background process arose to varying degrees in all of our minds, with the looming risk that soon it will apply to everything. The eloquent condolence card. The student’s unusually clean paper. This very article about AI. “Did an AI write this?” The thought arrives uninvited and cannot be easily exorcised, which is bad because we are, in most cases, worse off for having it.
The AI question is a kind of intrusive thought problem. A question we don’t want to have, that we’re resentful for even having to address. The instant we run a sincere sentence through the detector in our heads (that em-dash, that suspiciously balanced tricolon, the fact that nobody actually says “delve”), we’re trapped. Worse still, as people learn about common tells, the absence of those tells can itself become a source of suspicion. Like an unfalsifiable conspiracy theory, it can feel like there is no possible way to regain innocence and trust. At best, we connect with the author via a security checkpoint. The moment is spoiled, not because we reached the wrong verdict, but because the journey was derailed by the demand for the verdict.
There’s a deeper way to name what reflexive suspicion costs us. In “Freedom and Resentment“, P.F. Strawson distinguishes between two stances we can take toward another being. There’s the ‘participant’ stance, the one we’re in when we treat someone as a fellow agent whose words are addressed to us, and so are appropriate objects of feelings like gratitude, resentment, and affection. And then there’s the ‘objective’ stance, the one we adopt toward entities we see as lacking agency, like a raging storm or a malfunctioning machine. Something to be predicted, diagnosed, and managed.
To wonder whether something was written by an AI is to slide, quietly, from the first stance to the second. You stop receiving the text as outreach from a fellow mind and start assessing it as potentially the output of a convincing mimic. You become Deckard with the Voight-Kampff machine, watching for the tell – except now everyone is a suspected replicant and everything can be a tell, including a lack of tells. There is no scan that returns ‘sincere’, and having to seek one out saps the moment of sincerity. We cannot audit our way into trust; the audit is the thing that shorts out the connection we’re trying to make. The cost isn’t merely that we might insult a sincere writer. It’s that we’ve been forced to abandon the stance that makes writing an act of one person reaching another.
Worse, the cost lands most on the most sincere. The slick and the apathetic slide through; it’s the earnest and the careful who get accused. Write something polished after three drafts and a sleepless night and you’ll be told a bot did it in four seconds. It’s the dead internet theory, spreading to consume every chance for human connection: the growing fear that no one on the other end is ever real.
It would be a far more peaceful life if we could simply opt out. Adopt a blanket policy of not-wondering and go back to reading like it was 2019. The problem is that intrusive thoughts don’t ask permission before they show up and, once they arise, we cannot simply pretend they don’t exist. Even if we could psychologically, the encyclical from Pope Leo and cases like it are the reason we shouldn’t. An encyclical is one of those artifacts whose authority is bound up in who produced it, and how. Provenance isn’t a footnote to the document; it’s essential to assessing the document’s significance. A papal teaching that turned out to be 80% AI generated would not be a slightly tarnished teaching; it would be a different kind of object altogether, the way a signature you didn’t sign is not a worse signature, but a forgery. The encyclical is just one of many cases that proves a universal policy of incuriosity is off the table.
So, what do we do, given that we can neither switch the question off nor honestly pretend we never have it? We cannot choose whether to doubt, we can only triage how, and when.
The answer, if there even is one, isn’t suppression; more a kind of demotion. We can’t avoid having intrusive thoughts, but sometimes we can shift how our minds prioritise them. Because we’re so keyed up to avoid being tricked by fake content on the internet, fear of being tricked by AI writing has naturally flourished in ways that are understandable, but toxic. As skeptics, we have likely contributed to this problem through our attempts to help people cope with the daily barrage of misinformation. We need to find a new balance for living with our concerns.
Instead of treating the AI question as a crucial prelude to engaging with a text, recognise it as the reflexive intrusive thought that it is, set it aside, and let the writing reach you first. Be moved, be persuaded, be irritated, just try to maintain that participant stance, responding as though there was a person behind the words. Then, if and only if provenance genuinely matters, inquire about the origin.
You could limit the urge to check for AI to only cases where the text is meant to serve some function beyond conveying meaning (a contract, a credential, a confession, an encyclical). Yes, this will come with a higher risk of being tricked into having an earnest reaction to artificial content, which can feel embarrassing or dystopian in its own right. Ultimately, though, it is better to risk being ‘tricked’ into having an earnest reaction than to risk never having them at all.
