Most attempts right now for the Church to understand AI and how it should be used are just too broad to be useful. There is a good reason for this - most people in the Church aren't AI experts because hardly anyone is an AI expert. What we now call AI is very new.

So when it comes to evaluating whether or not AI should be used by a person to help with their job, or by a Church institution with their mission, it seems like there are only two options: Use AI, or don't.

The former is dangerous - just turning on the AI faucet is a fast track to all kinds of bad things. But so is the latter. If we put our heads in the sand about AI, we're not executing good stewardship, and we're not engaging in the promotion of that Social Doctrine which the Church holds as a key part of its ministry.

Almost no ethical judgments about AI should be made at the top level. Work is atomized into many tasks, and just like a good worker brings their judgment to each scenario they encounter - every discrete task that could be done by AI needs to be examined on its own against the moral standards and prudential principles we have discerned.

I am calling this idea Atomic Ethics. It requires two things, that we submit to a nuanced understanding of labor and its many components, and that we develop for ourselves - derived from the top level of Church teaching impressed on us by the encyclicals and other guidance - a set of specific domain-relevant principles with which to judge each part. If we take up this burden, we reach a future where AI will not replace secretaries, it will replace some of what they do, and might even change what they come to be called.

It is impossible for anyone to lay down a set of rules for AI usage without deep knowledge of the particular area they will be applied in. This is why, whilst most working in the Church today have moderate to little technological expertise, their particular hard-won experience and wisdom make them indispensable to the questions of AI in a way that a purely technological expert could never be.

There is another reason why these ethics need to be particularized to different cases by the many people of the Church, and it speaks directly to what I think is the greatest danger from AI. The worst thing that AI can do is not take someone's job, or commit communications errors - it's that it can atrophy the parts of the human intellect and spirit that come from formation. Even if - an unlikely outcome - AI could be made to completely take over all the tasks that a human does, this would not be a win, even on pragmatic resource-management grounds, because the formation that humans receive in getting to that point of competency would be completely nullified. The next generation of humans would have no way to learn all of the ineffable skills and sympathies that happen naturally in the course of growing into a role.

Whoever follows after would be like the people of Britain in the dark ages, living among the ruins of a greater civilization, wondering: who were the giants who made these wonders?

So the task we have is this: to determine what parts of labor can be removed from humans to free them for more valuable work, without removing those parts that form them or which only they can accomplish in an honest way. (And remembering that even the dullest work can be formative in its way.)

This is why we need to apply our ethics on the most atomized scale feasible. Assuming that we need to use AI at all (and in most cases, given finite Church resources, it's just good stewardship to at least evaluate it) it cannot just be thrown at an entire job unless we want to end up with a Church and a society filled with robots talking to each other over the addled gruntings of diminished humans.

Every task that is done in the Church has a hidden task behind it. Someone asking about how to get a marriage annulled is seeking more than just a factual answer. Even when someone asks about Mass times (the perennial example question in the Church website world) there is a hidden element to this: why are they asking? Are they new? Parish hopping?

To an AI system merely tuned to productivity, these tasks might be done at great speed, without any understanding of what is really important, or really serving the human person. Simply providing an answer or a result without any friction might completely stop the train that the person is on - ending their journey abruptly at a technical answer that gives them a next step too large for them to take.

Each task that you evaluate for AI should pass through a kind of sieve - the larger and more weighty the task or its hidden effect on the person, the less likely it is to pass through.

We cannot avoid AI. It is here already. But we still do have the time to shape it and its role, before it shapes us.

-- Sid