AI

At work, a big part of my professional development now involves wrestling with the problems of AI and how that will change what I need to help people with when they do research. I watched a talk last week called “The AI Dilemma” from the Center for Humane Technology, an organization that I have followed since the wake of the Cambridge Analytica scandal. I think it is worth your time watching, so if you are folding laundry or doing another low-attention household task, now is the perfect time to press play.

In the talk, they framed the deployment of AI much like the deployment of social media — a fight that they say “we [humans] lost” — and I found that to be a very accurate way to put what happened. I also found it to be a great way to improve AI literacy, as most people do not realize that social media algorithms and generative AI are two different implementations of a broader AI bucket.

Social media has helped disseminate information about polytheism and paganism, and it did provide a space for many early communities to connect … before the attention algorithms started working against us to twist and fragment people apart — or, social media started with us Millennials having fun poking friends on college-access-only Facebook and seems to have ended with democratic crises and deeply fractured and polarized societies too dysfunctional to tackle the thorny and hard issues of our times. My analysis of paganism and polytheism is that much of the positive impact on the dissemination of our information is just vibes, as the shift onto social media withered and destroyed many in-person networks that were not already sufficiently large. We would have grown more slowly in a different type of Internet environment, but it might have been better for us. I say this as someone with an avid Zoom social life.

Working against the “drive to the bottom” of social media attention algorithms requires a lot of mental effort, ranging from the low effort of keeping an Instagram feed on-point to be restorative (my private Instagram shows me cats and inspirational stories about women and breathing exercises) to the medium effort of keeping YouTube productive (such as the example I like to give of searching for decluttering content and ending up watching videos of extreme minimalists agonizing over whether they need to declutter their final underwear) to the high effort of the herd-of-cats TikTok algorithm (like a friend of mine who created an account for art content and ended up being served a mix of 4B and BookTok). None of us can be on 100% of the time, and these tools have been trained specifically to be responsive to the feedback that we give them to keep us attentive, typically through more and more extreme and siloed content. The CHT’s presentation accurately described the harms.

The new types of AI may be similarly disruptive. The CHT frames generative AI as doing for intimacy what the social media AI did with our attention. (You shunt people into silos and destroy everyone’s trust for others, and then you provide an AI tool to each person that can be our special friend. As the presenters say, which AI becomes our close companion? What gets to know us so well that it can play our psychology like an instrument?) Currently, there are no safeguards, and there is no mindful approach to actually fitting the capabilities of the technology to moral and ethical frameworks. Even the applications of it are being done without looping in all stakeholders to ensure that the product is good, as is happening in AI literature review tools, which IMO are terrible unless your field is less than 15 years old because it does not understand analog resources, but tons of grad students are using them now anyway. This goes beyond the frustrations in pagan and polytheist spaces about copyright and the data that the AI was trained on or the shift in creatives’ practices towards closed-access works and away from freely available content.

This is just as difficult to talk about as social media in some ways. This type of AI could be groundbreaking in decoding cetacean communication (we may be able to talk to sperm whales soon!!) because it has emergent capabilities based on data it has already trained on. It is decoding the Herculaneum texts. It has been groundbreaking in medical imaging. It can produce captions and screen reader glosses. But, overall, if we do approach it in the same way that we approached social media — which seems to be how we are doing things again — these are the CHT’s warnings:

I think it’s worth everyone’s time to take a look at that presentation and to think about how this will impact you in the next few years. I also think it is worth it for each of us — after all, the unexamined life is not worth living — to look at the negative impacts graphic they made of social media and to think about our own actions there and what we might do differently.

---

Not everything is in our power, but we most certainly can work on the things which are.

---

In the religious sphere, this impacts us in another way.

Many of us have, over the past few years, experimented with using AI tools for things related to religion. (Sometimes, it has been fun. I, too, have had ChatGPT write sonnets about Sosipatra.) Earlier this spring, I spent about an hour querying an AI tool — Bing, if you are curious — because I wanted to feed it several leading questions about the Chaldean Oracles to prime the AI to generate some options for chants to sing that were thematically attuned to the style of the Chaldean Oracles and which could work in my daily rituals.

I asked it for both household worship chants and chants for specific deities.

Most of what it produced was inappropriate.

Out of all of the results it generated, one chant was passable, and I made minor edits to it to ensure that I could use it in the setting I needed. This prevented me from needing to spend several years learning Ancient Greek (hey, I’m not even done with the Modern Greek Duolingo yet), which was a benefit to me. However, it involved wading through a probability graph that had set me up to fail.

Others have commented that AI output often seems soulless, as if it is missing something. I talked to a polytheist friend soon after generating the output. During the conversation, we agreed that the AI isn’t “missing” something. It is trained on a dataset that includes a lot of deeply impious information attempting to describe the Gods. Most of the open web data and published data that it consumed believes that Gods are stand-ins for natural phenomena or that they have a one-to-one correspondence to specific domains, like a creative writer I once spoke to who said she “already had a Goddess of Love” in her work and thus didn’t need another one. This is not how polytheism actually works.

So, despite being primed with the Chaldean Oracles, the chants it produced — just as all of the stuff it produces — come out of what it originally ate, from an unruly disorder. It was, said my friend, amazing that I was even able to generate one usable chant given how deep in the mud the AI is. And that’s not even getting into the ethics of how the AI was trained and on what.

That meant less to me given that the Chaldean Oracles were written very long ago, but it means a lot more in the context of most users’ religious queries, where they are asking for examples of rituals they can do for the New or Full Moon or for a specific deity and are unaware that there are entire books dedicated to lunar worship written by experienced polytheists and pagans, books that ground each ritual script in enough preliminaries to help a person along and that are a connection — however tenuously parasocial — with an actual person. The AI might even suggest practices that are inappropriate for a devotee to start on their own without training from an initiated specialist and/or elder. The only ones who stand to gain from using these AI tools for such queries are experienced persons who already know enough to evaluate and act on the information they are presented with, yet AI is now integrated into search tools for everyone.

In the spirit of being proactive about AI, that our conversations need to move from being reactive to being proactive. The Center for Humane Technology’s podcast and other content are a good place to start for building that background awareness.

---

And finally, this — which also came up in the conversation with a friend —: AI art will never be the same as creating the art oneself or getting it from a polytheistic artist. Or, in the words of Proclus in Book 4 of his Parmenides commentary at almost-852:

Suppose someone had seen Athena herself such as she is portrayed by Homer […]. If, after encountering this vision, a man should wish to paint a picture of the Athena he had seen and should do so; and another man who had seen the Athena of Phidias, presumably in the same posture, should also want to put her figure into a picture and should do so, their pictures would seem to superficial observers to differ not at all; but the one made by the artist who had seen the goddess will make a special impression, whereas that which copies the mental statue will carry only a frigid likeness, since it is the picture of a lifeless object.