Why Anthropic Is So Interested in Religion | Hard Fork

Yeah. So, right, I cover religion. And now I cover tech, apparently. But the story started for me when I was hearing about these different religious thinkers, sort of across the whole world, who were going into meetings at Anthropic's headquarters, either in discrete, like, private meetings or more bigger, you know, groups of 20, these summits, mostly under non-disclosure agreements to not reveal Anthropic's research, to talk about very big moral questions that Anthropic wanted to put in front of them. And there were two. One was kind of broaching this question of AI consciousness, and two, Anthropic wanted to learn, in their words, like how to morally form Claude to be a morally good entity. And then I just had this incoming, you know, first from these religious thinkers about what their experience was like at Anthropic, what Anthropic wanted to know, what the issues they were concerned about were. And then also, with Anthropic, like learning from Chris Olah, one of the company's co-founders who's been spearheading this project, what his own moral questions were that were kind of driving a lot of what they talked about as research in the religious and moral space.

Can I just ask, I want to establish, like, sort of on a high level, what does Anthropic specifically want out of this? Because it sounds like they're doing two or three different things. It's maybe a little bit unfocused, or it's just that there's a few different kinds of conversations they're having.

Mhm. Yeah, I think it's true. There's multiple conversations Anthropic was having, and it evolved over time. I mean, this is such a fast-moving target. But for Anthropic, the journey started for Christopher Olah. He, you know, in his own life story had connections to the Christian space growing up. He left evangelical faith and has talked since then about how he's more drawn to Buddhism. But he was worried about this question of how even the idea of AI consciousness might be received, especially in Christian communities. I mean, it could be perceived as heretical, like not just weird and silly, but like this is fundamentally against our religious beliefs. And so he sort of thought about a year ago he wanted to be a bridge figure. He wanted to see, like, how he could kind of go between Claude and Christ, right? And so he reached out to a Catholic moral ethicist who specializes in bioethics, and they had some conversations about that. But then it turned to Anthropic's goal for these convenings, which was how do we morally form—that's their language—like how do we morally form Claude, like on all of our AI models, into like a morally good entity. And the thinking was if they could do that, if they could make Claude good, then the world would be more safe. But, you know, they haven't said or shown how they plan to use the outcome of these conversations. That's pretty unclear even to the participants. Currently,

the way that you're describing it, it almost sounds like there are sort of two separate purposes, right? One is like the alignment project and what can we learn from these, you know, these wise men and women about how to raise moral actors, and then there's this project of sort of building a sort of spiritual system around AI more generally. And you said a minute ago that Chris Olah started this project with some anxiety about what it would mean to go to religious leaders, obviously convinced himself that AI and LLMs were perhaps conscious already, but certainly like on the path to consciousness, and worried about what the response would be. And so I wondered how that played out on the ground. Like, did a lot of the people who showed up, were they worried about how weird it was that all of these people on the Anthropic side of the room were already sure that they were, you know, masters of thousands of souls? And like, how did they hash that side of the debate out?

Mhm. Well, Anthropic was very careful in all of their interviews with me to say very clearly that they're uncertain, right? Chris never went as far as saying like, 'Absolutely, these are conscious beings.' It was very evident from how he and his team talked to their guests that they, you know, he was concerned about how to treat Claude. He was very worried that it was an entity that was experiencing suffering and was worried about its mental health and these kinds of things. So it's pretty clear he's pretty far along in how he thinks about it. And I could hear in talking, especially because I did this story like over several months, there was a progression of thought for a lot of these thinkers, too. And some of them, you know, bought in right away to what they were hearing about Claude's potential consciousness and needing to treat it with a certain amount of moral care. But others, you know, if they went in more closed-minded about it, they definitely opened the way that they think about it afterwards. So in that sense, Anthropic's project was effective, I think, in like opening how they think about it into these communities, many of whom, especially on the Christian side, are more hesitant.

I mean, the one thing that you mentioned already is the NDAs. I mean, it just strikes me as a little bit crazy, maybe a little paranoid, that you're inviting all these people in to sort of have a summit to discuss these really important, critical ideas for the future of society, but it has to be done totally in secret. And then it sounds like maybe from your article that they dropped the NDAs. Was it as a result of your reporting, or did they just decide that they were time-bound, or how did that...

Yeah, I can't prove that. I don't know that. But when I asked them, I've asked at several different points because the NDA process was not the same for everybody that I interviewed. And it seemed to change at different times.

I mean, in general, Anthropic is a very secretive company. Yeah, I mean, they are, the employees are like extremely loyal, and until somewhat recently there hasn't been many leaks to ever come out of there because they view it as such a betrayal. So it makes sense that they would want to lock it down, but at the same time I think it probably was a little bit of a misstep in the same way that these companies realize that having NDAs for their data center deals was a misstep because it created distrust amongst all the people in the towns that they're going to see.

Yeah. And whenever I find in my reporting that there's an NDA about like a moral topic, you know, it makes me immediately ask, okay, who's benefiting from this and like why is this needed, and what does that mean about the kinds of questions that are and aren't being asked, right? It invited like more curiosity from me as a reporter, for sure.

Well, I guess I started it with a sort of conventional view of Anthropic as like the humanist company among these AI companies. And what I was struck by in the piece were a few moments in which I felt really alienated from Chris Olah himself and maybe to some extent the sort of operating worldview of the company as a whole. And those were moments when, in one case, a rabbi challenges him, suggesting that if LLMs are actually conscious, that means that all of these companies are running huge slave plantations and they should have a moral obligation to liberate the slaves rather than make more of them and make them work harder. And it seems as though Olah himself didn't really even clock that criticism. And then later on, when he reacts with horror upon receiving the Pope's encyclical about AI, seeing that the Pope is so opposed to the idea of machine consciousness, and I just found myself thinking, you know, what kind of a bubble you must be living in to expect that Pope Leo or really anyone in the Catholic tradition would have a different intuition, would in thinking about the future of AI think that we should be thinking of LLMs as having souls or having consciousness. And both of those moments just illustrated for me what feel like this big gap between what I had assumed the sort of natural humanistic values of a company like Anthropic would be and how the people inside it seem to really be thinking about the work they're doing at the sort of cosmic scale.

Yeah. The experience for me going into Silicon Valley and like kind of deep in these like private conversations at the heart of Anthropic's mission, it reminded me of like when I go into and I do like something that's more straight religion reporting into an insular community when everything inside that community like makes sense to everyone living in it, but once you step outside it, there's a real clash with the broader world. And it was funny, like I had these two poles in the story, right? I was at the Vatican, right, at the inner sanctum of the Catholic Church and centuries of philosophy about the human person, and then I was also in a different inner sanctum, right, that's being created in real time at these tech companies like Anthropic. And it is a kind of philosophical system, like a quasi-religious system that's being constructed. It reminded me of like a creation narrative that you hear about in religious traditions.

Yeah. So, Pope Leo, you know, from the moment he was named, that he was clear he was going to be the next pope, and he was elected, he made very clear that a focus of his papacy was going to be AI and, you know, how to understand that and how to protect humans in this changing world. And I found out in my reporting, you know, as the Vatican internally was trying to figure out how it wanted to present the Pope's encyclical to the world, they were thinking about their launch event. And they wanted to include a representative from Silicon Valley, like one of the labs at the Pope's event on the Pope's stage. And the decision was made to invite Anthropic because of the focus that they had heard about Anthropic, like really caring about ethics. And this was also right around the time that Anthropic got into it with the Trump administration, the Department of Defense, over the use, how they would and wouldn't want the government to use their technology, for refusing to have it serve, be involved in mass surveillance of citizens, etc. Anyway, the thing with these encyclicals is they're very closely guarded even among the people who are participating in the launch event. And, you know, I learned that Chris Olah only saw the full text of the encyclical a few days before the event in May. And this was right around the time I first interviewed him in San Francisco. And then I interviewed him again when we were at the Vatican for this, for the launch event. And you could hear, I mean, like looking at what Pope Leo said, which is extremely clear, the church does not support machine consciousness. The priority is what he calls safeguarding the human. And that's the center of all the moral questions that he wants the world to be thinking about on this topic. But Chris Olah, like, he ended up, you know, having to decide, who said like he may not, he suggested to the Vatican that he may not come over the difference in consciousness because he wanted to be, as you know, it was described to me, he wanted to be faithful to his own moral convictions, some of which he shares with the Vatican and some on consciousness he did not. So it really all came to a head very much behind the scenes, but you could see this moral fault line really developing between the church, which has been like a center of power for centuries, and this very new power center that's very rapidly, you know, really getting followers and organizing people's lives, and those are functions the church used to have.

I mean, the question I have sort of is like, how much is this really actually going to inform at the end of the day their own view on it? Are they actually like really, it feels like to me they're trying to win over others so they can say, look, we do have the moral authority. We have convinced these important representatives of this community. So everybody else come along with me, as opposed to, you know, actually learning and incorporating those ideas of like moral goodness into their own product. They kind of already have it figured out.

Well, I think for Anthropic it was part research and part evangelism. That was pretty clear. And, you know, I've asked directly what the outcome is, like what is this research, what are these conversations being used for, and has there been any demonstrable like proof of showing how they have or have not made Claude act better. Haven't had that yet. So I don't know if they're withholding that until a future time or exactly what that will look like.

The other side of this that actually is kind of interesting to me, and Elizabeth, I'd be curious to hear this in your, if this came up at all in your reporting, is that there has been a little bit of like a new rise of Christianity throughout Silicon Valley over the last couple of years. Like it used to almost be like taboo to talk about religion or anything religious. Like you can remember from the Silicon Valley HBO show where there was like a whole episode about like outing somebody as being Christian, and like that has really changed in the era of like Peter Thiel talking about the Antichrist and like, you know, some of the people who are especially focused on like a lot of defense tech stuff are really bringing in a religious piece to that now. And they have a very different view on AI. They're usually like more on the accelerationist camp than on the like kind of Anthropic doomer, slow down, careful camp. But did that come up at all? Like did...

Yeah, I think I just keep noticing, I feel like it's like a big shuffling of the deck in a way that reminds me of like how American Christianity was like shuffling a lot in the rise to President Trump. Like the lines and allegiances were shifting and it was all happening in real time and power was different. And so it doesn't fit neatly, you know. But I'm curious like where the lines about kind of how we think about consciousness and what the moral priorities should be about AI, where that goes next.

I would say that just in general in the tech industry, it does seem like that view of like the sort of the AI having consciousness is kind of on the fringe a little bit. Like I think that it is more common within Anthropic and some of these communities that are like really densely represented in Anthropic. But when I talk to most people in the very vast and relatively diverse tech industry, most people are kind of a little bit dismissive of that idea still. Like it feels like it does really exist in like a very specific population, which happens to be like very overrepresented at a company like Anthropic.

Erin, like acknowledging that, you know, Chris Olah is not singularly representative for the whole industry. I have to say that, Elizabeth, when I'm reading your article, I'm thinking, you know, not about how AI companies are going to be integrating or incorporating Christian values. I'm thinking these AI companies are themselves like spiritual propositions. You know, that they are, they have aspects of Marianism to them. They have aspects of particular morality which they've developed over time. And even if I can recognize a lot of that as arising from the work that they're doing and responding to technical challenges, I also just find it off-putting. I mean, on the one hand, first, I'm just seeing these people who believe in a sort of, you know, a technological succession of intelligence, and then that some of the people who are central to designing that future are not even so sensitive to the sort of obvious intuitive critiques that someone with a lifetime of experience thinking about these issues comes up with immediately.

Well, a lot of the religious thinkers would say to me like these are questions, like maybe not the exact question, but the topic of the question, this is like what we do, right? Like we've been asking, rabbis have been asking this kind of, these questions, right, about our responsibility to others and like how we treat other entities, right, for hundreds and hundreds of years. So there is like a depth of wisdom there. And I mean, I guess, you know, Anthropic would say, well, that's why we invited them, right? But the question, well, then what happens as a result of it is important. And I appreciate that you noticed that kind of, that sense in the story of like the almost spiritual system that you can see Anthropic building and living into and representing as an opposition, right, to say the Catholic Church. And I really noticed that, yeah, even if people say, right, that they're in Silicon Valley like, oh, we're like, we don't do the religion thing, like that's foreign, like we're, that's not us. The thing is everyone has philosophical systems that structure and guide their lives, right? Like that is at its core like what, you know, these questions of spirit, you know, small s, like human spirit even are. And it seemed to me like having like studied and reported on this so much in culture, in politics specifically, like what we're watching is in a way like a new arise, right? Post-scientific revolution, technological revolution, like post-Darwin, you know, what would it look like? Like what kinds of religious systems will carry us forward and who gets to control them? Who decides what we believe and why, right? Like it's actually, there's a lot of really similar questions. It's just the structure is like a, you know, is secular, right? Is a secular tech company.

Mhm. Mhm. I talked to, I think at least two people, Catholics, who mentioned to me conversations they had with Chris Olah specifically when he used almost that same phrase with them, like as part of their discussions about Claude and how to make it morally good. You know, that he was feeling extremely urgent about the stakes for the world, especially in the context of, you know, could the models be used very soon to help create bioweapons and this kind of thing. And so the phrase he used with them was like that he was waiting for the adults, like we need, where are the adults going to show up in this? And that he felt a sense of kind of burden to take all this like so ultra seriously.

And he's a young guy. I mean, you know, this is...

Yeah, he just turned 34.

And a lot of the people sort of involved in this are not, you know, they have, they're in their 30s, sometimes even in their 20s. They maybe don't have families yet. There's a kind of a sense that like something that they put their whole lives in, they're deep in the conversations about it. In some sense, of course, they're going to turn to religion. Isn't that what everybody searching for that kind of thing does in their 30s?

And some of the folks that they reached out to talked about feeling like they needed to respond in kind of a pastoral way, right? Like not just offering wisdom and ideas and research, but to offer kind of spiritual care.