Rob Wiblin and Dean W. Ball clash over whether the Vatican's first AI encyclical permanently rules out future AI consciousness
The Vatican's first AI encyclical is titled Magnifica Humanitas.
I think this is a completely insane reading of the encyclical given the broader context. The document makes categorical claims about all aspects of AI cognition. When the Pope says “AI cannot,” the emphasis is not on the present tense but on the categorical subject of the sentence: “AI.” He is saying “AI can never.” You know this because the document opens with a statement about how they want to avoid making claims about AI too grounded in the present, because of how fast AI changes. They deliberately commit themselves to evergreen statements about AI and then say that. I think you and other safety people are reading the document wishfully because you sense a new ally, which would not be the first time AI safety people have viewed a new ally through rose-colored glasses, only to be screwed by that ally later on.
Rob Wiblin@robertwiblinI see some hand-wringing that the Pope is skeptical AIs can suffer. But the glass is actually very much half-full: 1. The Encyclical deliberately avoided closing the door on future AI consciousness. 2. At the launch Cardinal Czerny (in pre-approved remarks in front of Leo) said it's "a serious question, one that deserves attention and further study". He was probably the key theological advisor on this topic for MH. 3. The Church was clearly fine with Olah at the launch suggesting he thinks AI consciousness is possible and a legit concern. Indeed they selected Olah out of many people they could have invited and let him speak his mind. The Catholic Church knows very well how to shut the door on these questions on theological grounds if it wants to (as some evangelical groups have). But they very much haven't done that and it seems like they probably won't any time soon, at least not under Leo anyway. 1/
Screengrabs from @JohnClarkLevin here: https://johnclarklevin.substack.com/p/pope-leos-first-ai-encyclical-summary
Rob Wiblin@robertwiblinI see some hand-wringing that the Pope is skeptical AIs can suffer. But the glass is actually very much half-full: 1. The Encyclical deliberately avoided closing the door on future AI consciousness. 2. At the launch Cardinal Czerny (in pre-approved remarks in front of Leo) said it's "a serious question, one that deserves attention and further study". He was probably the key theological advisor on this topic for MH. 3. The Church was clearly fine with Olah at the launch suggesting he thinks AI consciousness is possible and a legit concern. Indeed they selected Olah out of many people they could have invited and let him speak his mind. The Catholic Church knows very well how to shut the door on these questions on theological grounds if it wants to (as some evangelical groups have). But they very much haven't done that and it seems like they probably won't any time soon, at least not under Leo anyway. 1/
Nah that's not it - I'd be happy to coordinate with the church on points where we agree even if they were totally wrong about AI consciousness, and would just shrug my shoulders about it.
It's more I'm relying on @JohnClarkLevin having a better ability to interpret encyclicals and church tea leaves than me (he's a very engaged Catholic).
Dean W. Ball@deanwballI think this is a completely insane reading of the encyclical given the broader context. The document makes categorical claims about all aspects of AI cognition. When the Pope says “AI cannot,” the emphasis is not on the present tense but on the categorical subject of the sentence: “AI.” He is saying “AI can never.” You know this because the document opens with a statement about how they want to avoid making claims about AI too grounded in the present, because of how fast AI changes. They deliberately commit themselves to evergreen statements about AI and then say that. I think you and other safety people are reading the document wishfully because you sense a new ally, which would not be the first time AI safety people have viewed a new ally through rose-colored glasses, only to be screwed by that ally later on.
@deanwball @JohnClarkLevin (Unfortunately John-Clark has been locked out of Twitter for live-tweeting too fast, so we'll have to wait a minute to hear his reaction to all this.)
Rob Wiblin@robertwiblinNah that's not it - I'd be happy to coordinate with the church on points where we agree even if they were totally wrong about AI consciousness, and would just shrug my shoulders about it. It's more I'm relying on @JohnClarkLevin having a better ability to interpret encyclicals and church tea leaves than me (he's a very engaged Catholic).
I put your take, the encyclical, JCL's blog post, my tweet into Claude and here was its reaction FWIW:
"Reading the actual text against both interpretations, I think you're closer to right than Dean, but his reading has more force than the response acknowledges, and the honest answer involves real ambiguity.
Dean's strongest point: paragraph 99's language genuinely is unqualified. "Do not undergo experiences, do not possess a body, do not feel joy or pain… do not have a moral conscience… do not understand what they produce." There's no "current systems" hedge in the operative sentences. If the Church wanted to leave the door visibly ajar, the easy move was a single qualifier — "today's systems," "as currently designed" — and that move was conspicuously not made. Combined with the metaphysical-sounding grounds ("they lack the affective, relational and spiritual perspective"), the surface text reads like a categorical claim about what AI is, not a status report on GPT-5.
That said, four things push me toward your reading:
First, the grounds given are a mix. "Do not possess a body" and "do not mature through relationships" are not in-principle impossibilities — embodied systems exist, and "maturing through relationships" is a description, not a metaphysical barrier. The categorical-sounding bits ("spiritual perspective") sit alongside contingent-sounding bits. If the Pope wanted a clean Thomistic case — only rational souls directly created by God can have inner life, full stop — that case is available in the tradition and he didn't make it. He chose mixed, somewhat empirical grounds, which is a meaningful choice.
Second, the meta-text matters. Paragraph 97 explicitly limits scope: "I limit myself to recalling a few essential elements for a moral and social discernment." Paragraph 98 frames current systems as "cultivated" rather than "built," with their internal workings unknown — an open epistemic posture. The encyclical also flags that "any statement regarding AI risks becoming quickly outdated." This is not the register of a Church closing a question; it's the register of a Church reserving the right to revise.
Third, the dog that didn't bark. The Vatican has a well-established machinery for closing theological questions — dogmatic definitions, conciliar statements, CDF declarations. None of that was used. An encyclical is magisterial but not the genre you reach for to settle whether a class of entities can have inner experience. And as Levin notes, Antiqua et Nova a year earlier engaged the consciousness question more directly and was also hedged; the trajectory is toward "needs further study," not toward closure.
Fourth, the launch staging. If Czerny really said in pre-approved remarks that this "deserves attention and further study," and Olah was platformed to advance a "AI consciousness is a live concern" view without contradiction — that's the Church telegraphing where it actually is, more reliably than parsing individual paragraphs. Vatican communications don't accidentally let key advisors freelance on doctrinally settled questions in front of the Pope.
So I'd put it this way: the surface text leans Dean's direction, but the structural choices — what genre, what advisors, what was conspicuously not said, what was endorsed at launch — all lean your direction. The Church is making the strongest claim it can comfortably make about current systems while declining to bind future judgment. That's not the same as "open-minded" in the way an AI researcher would mean it; the underlying Thomistic anthropology probably does think machine consciousness is impossible. But they're deliberately not cashing that out into doctrine, and the gap between "what we suspect" and "what we'll define" is exactly the space you can work in.
One thing worth flagging if this becomes a longer back-and-forth with Dean: he's not crazy to read it categorically, and the response that just asserts "no it's clearly open" without engaging the actual language risks looking like motivated reading. The stronger move is to concede that the text sounds categorical, then point to the genre, the structural choices, and the launch context as evidence the Church is being more careful than the prose suggests. That's harder to dismiss."
Rob Wiblin@robertwiblinNah that's not it - I'd be happy to coordinate with the church on points where we agree even if they were totally wrong about AI consciousness, and would just shrug my shoulders about it. It's more I'm relying on @JohnClarkLevin having a better ability to interpret encyclicals and church tea leaves than me (he's a very engaged Catholic).
I put your take, the encyclical, JCL's blog post, my tweet into Claude and here was its reaction FWIW:
"Reading the actual text against both interpretations, I think you're closer to right than Dean, but his reading has more force than the response acknowledges, and the honest answer involves real ambiguity.
Dean's strongest point: paragraph 99's language genuinely is unqualified. "Do not undergo experiences, do not possess a body, do not feel joy or pain… do not have a moral conscience… do not understand what they produce." There's no "current systems" hedge in the operative sentences. If the Church wanted to leave the door visibly ajar, the easy move was a single qualifier — "today's systems," "as currently designed" — and that move was conspicuously not made. Combined with the metaphysical-sounding grounds ("they lack the affective, relational and spiritual perspective"), the surface text reads like a categorical claim about what AI is, not a status report on GPT-5.
That said, four things push me toward your reading:
First, the grounds given are a mix. "Do not possess a body" and "do not mature through relationships" are not in-principle impossibilities — embodied systems exist, and "maturing through relationships" is a description, not a metaphysical barrier. The categorical-sounding bits ("spiritual perspective") sit alongside contingent-sounding bits. If the Pope wanted a clean Thomistic case — only rational souls directly created by God can have inner life, full stop — that case is available in the tradition and he didn't make it. He chose mixed, somewhat empirical grounds, which is a meaningful choice.
Second, the meta-text matters. Paragraph 97 explicitly limits scope: "I limit myself to recalling a few essential elements for a moral and social discernment." Paragraph 98 frames current systems as "cultivated" rather than "built," with their internal workings unknown — an open epistemic posture. The encyclical also flags that "any statement regarding AI risks becoming quickly outdated." This is not the register of a Church closing a question; it's the register of a Church reserving the right to revise.
Third, the dog that didn't bark. The Vatican has a well-established machinery for closing theological questions — dogmatic definitions, conciliar statements, CDF declarations. None of that was used. An encyclical is magisterial but not the genre you reach for to settle whether a class of entities can have inner experience. And as Levin notes, Antiqua et Nova a year earlier engaged the consciousness question more directly and was also hedged; the trajectory is toward "needs further study," not toward closure.
Fourth, the launch staging. If Czerny really said in pre-approved remarks that this "deserves attention and further study," and Olah was platformed to advance a "AI consciousness is a live concern" view without contradiction — that's the Church telegraphing where it actually is, more reliably than parsing individual paragraphs. Vatican communications don't accidentally let key advisors freelance on doctrinally settled questions in front of the Pope.
So I'd put it this way: the surface text leans Dean's direction, but the structural choices — what genre, what advisors, what was conspicuously not said, what was endorsed at launch — all lean your direction. The Church is making the strongest claim it can comfortably make about current systems while declining to bind future judgment. That's not the same as "open-minded" in the way an AI researcher would mean it; the underlying Thomistic anthropology probably does think machine consciousness is impossible. But they're deliberately not cashing that out into doctrine, and the gap between "what we suspect" and "what we'll define" is exactly the space you can work in.
One thing worth flagging if this becomes a longer back-and-forth with Dean: he's not crazy to read it categorically, and the response that just asserts "no it's clearly open" without engaging the actual language risks looking like motivated reading. The stronger move is to concede that the text sounds categorical, then point to the genre, the structural choices, and the launch context as evidence the Church is being more careful than the prose suggests. That's harder to dismiss."
Dean W. Ball@deanwballI think this is a completely insane reading of the encyclical given the broader context. The document makes categorical claims about all aspects of AI cognition. When the Pope says “AI cannot,” the emphasis is not on the present tense but on the categorical subject of the sentence: “AI.” He is saying “AI can never.” You know this because the document opens with a statement about how they want to avoid making claims about AI too grounded in the present, because of how fast AI changes. They deliberately commit themselves to evergreen statements about AI and then say that. I think you and other safety people are reading the document wishfully because you sense a new ally, which would not be the first time AI safety people have viewed a new ally through rose-colored glasses, only to be screwed by that ally later on.