Anthropic AI attended a multi-faith religion confab after hosting one with Christian leaders

[thewayne]

In early April, Anthropic hosted a two-day gathering of 15 Christian leaders to discuss Claude's "morality and spiritual development" at its HQ. Ignoring the concept of coding morals into software and the fact that different groups of the same religion who ostensibly follow the same core book have different interpretations, you do have the problem of this being just one core religion. And Anthropic took some heat for just meeting with representatives of Christian religions,

In early May, Anthropic and OpenAI, the makers of ChatGPT, joined a conference that included scholars from the "...New York Board of Rabbis, the Hindu Temple Society of North America, the Church of Jesus Christ of Latter-day Saints, the U.S.-based Sikh Coalition, and the Greek Orthodox Archdiocese of America...". I note the absence of Hindu, Buddhist, atheists, Native American and Shamanistic traditions, Druidic and Wiccan traditions, and any number of other faiths. Heck, having the Church of Satan present would have certainly livened things up! But this was an invitational gathering called the Faith-AI Covenant, so they're just being a little representative, not remotely all-inclusive.

LLMs have a 'rule book' that are supposed to dictate some of their behavior. In the case of Anthropic's Claude, it's called a constitution. As of a month ago, it was 29,000 words. That's probably pretty complex. Considering how self-contradictory religious texts can be, do you want to code that into a rule book?

The companies making chatbots have a big problem. People using them have been talked into committing homicide, suicide, experimenting with drugs to the point of fatal overdoses, etc. While they say their programs are designed to be protective, their behavior shows that it is anything but.

Martin Luther King Jr. had a great quote: 'Our technology has so out-stripped our morals that we now have guided missiles and un-guided men.' It's relatively easy to guide a missile with radar, GPS, and terrain-matching optics/computers. But to code morals into a computer, based on religion - especially with several competing religions contributing?

I think you're going to end up with an almost HAL-9000 scenario, or any number of other scenarios where the program is paralyzed by the contradictions and mismatches within and between the various faiths. Ethics and morality are tricky codes, and they don't have to come from religion, and if you try to dictate them from religion, it's a great way to get the Crusades and any number of other horrible things.

I don't know what the answer is, but I don't think this is it.

Gizmodo article from last month on the Christian gathering:
https://gizmodo.com/how-do-we-make-sure-that-claude-behaves-itself-anthropic-invited-15-christians-for-a-summit-2000743766

Washington Post article re: Christian gathering, much more in-depth:
https://www.washingtonpost.com/technology/2026/04/11/anthropic-christians-claude-morals/

Gizmodo article on latest multi-faith gathering:
https://gizmodo.com/anthropic-has-added-several-more-religions-on-its-quest-to-inject-perfect-morals-into-claude-2000756740

Comments

[kaishin108]

This is awful. How about some non-secular groups. Or Humanists and Pagans. Sigh. (Silently swears).

[thewayne]

We can't include THEM weirdos! /s  Yeah, I know.  They're just grabbing at the biggies for appearance's sake.

[mtbc]

It's even worse than that, given that they're not reasoning systems. At least HAL-9000 was doing its best in an impossible situation. LLMs simply aren't constructed in a manner that lends itself to reliably staying within bounds and they're inevitably going to occasionally echo just about anything from their training set.

[disneydream06]

Well we all know how much I care for AI and religion isn't far behind, :(
Hugs, Jon

[ranunculus]

And ye harm none, do as you will.

[thewayne]

...As long as you don't wake the children or frighten the horses. ;)

[ranunculus]

Does waking the children harm them? Harm their parents when they have to deal with a nap deprived kid? How about the horse? Would spooking them lead to harm?

[garote]

Or vice—versa

[garote]

Indeed. Wake me when I can have an argument with the thing and “change” it’s “mind”, not just ask things and be told.

Oh wait, IT’S NOT EVEN AN IT. It barely qualifies as a single piece of software, let alone something you can label with a name. And yet marketing bastards have given it a name anyway, to immediately sell the enormous lie.

I mean, I could call my bag of Scrabble tiles “Mephisto”, and pick letters out until it spells something. Was there actually an entity hearing what I asked? Perhaps I could convince you if you believed in a spirit world. Impressively, tech marketers have achieved nearly the same level of outright quackery in a single move.

[richardf8]

This is yet another example of the same damn dumb anthropomorphism of AI that gets people romantically involved with it. AIs don't need religion. The pope, OTOH, seems to have latched on to the idea that coping with and ethically deploying AI is something religion can speak to. Now the tech bros just need to catch up with the Catholic Church.

[silveradept]

You cannot impose morals upon an entity that uses randomness as much as an LLM does, because to do that would mean it was no longer random, but instead at least partially deterministic, and that ruins the bubble idea of AI as the supreme intelligence that it's supposed to be. They're trying to get people to believe in something that's not possible, because they don't want to be liable for when the random rolls a 1.

[thewayne]

Very cogent point, my friend!  I would like to study how LLMs and neural nets work to get a better understanding  On the rare occasion that I use a chatbot, I use it as a search summary, and it does a fairly good job, but I recognize the limitations and don't employ it beyond that.  I started working on a book on Prolog and was enjoying it, but for some reason I couldn't get the examples to work on my Mac!  I'd like to get back to that some day.