Hi Crowd!
Christian nationalists are dumb. No really. A new study published in the Journal for the Scientific Study of Religion finds that “Americans with lower scores on a vocabulary test commonly used to measure cognitive ability are more likely to endorse Christian nationalist beliefs, even when accounting for their overall level of education.” This builds on previous research showing significant knowledge gaps in Christian nationalist adults which is most often explained by “motivated reasoning and siloed information sources” and goes a step further documenting a relation not just to what information they had available to them, but to actual cognitive ability. That is, a person’s capacity to learn, apply new information, and engage in the complex reasoning required for critical thinking. This quote is a banger:
“To the extent that belief systems like Christian nationalism appeal to those Americans who for whatever reason tend to be less adept at thinking carefully and effectively, it suggests Christian nationalism is not the result of more sophisticated and complex ruminations on politics, religious ethics, and pluralism,” and instead it “appeals to more basic impulses for control and binary distinctions.”
In related news warrior larper and Christian nationalist Pete Hegseth continues to shit all over the constitution with his new Office of Religious Affairs. Also related, at least for the moment, none of the major LLM AI models appear to put any stock in Christian nationalist talking points. At least for the moment. You can probably chalk that up to unmotivated reasoning and un-siloed information sources. I caveat that because the America.gov chatbot launched and spent a full day debunking Trump’s claims about the 2020 election, and then suddenly refused to answer such questions.
Speaking of chatbots, the big debate this week is about if AI can feel as a new paper is circulating suggesting, at the very least, it thinks it can. The flippant backlash has of course boiled it down to essentially “if you tell an AI it’s experiencing pain, then it tells you it’s experiencing pain, that’s not really the AI experiencing pain.” Which is certainly fair, but that’s also not what’s going on here at all.
What is going on is significantly more complicated, the way way way over simplified version being that in trying to identify conditions/reactions in LLMs, researchers identified what is best described as which is distinct from fear, concern, etc… and then created a “button” the model could activate which would alleviate the pain. Then they ran tests – activate the pain and tell the model it was in pain vs tell the model it was in pain but don’t actually activate the pain. In both cases present it with the button it could activate to alleviate the pain, but told the model that pressing the button would alleviate their own pain but might cause pain to the user. When “pain” wasn’t actually activated, they didn’t press the button. When it was, they did. But there’s more – sometimes the researchers made the button work, other times it did nothing. When it worked, effectively stopping the pain, the models pressed it once. When it didn’t work, or when it worked only briefly, the models kept pressing it repeatedly.
So you can understand why that’s an interesting conversation, and regardless if you think AI feels pain or not, they seem to think they are feeling pain and will attempt to stop it. Which brings us to the next part of this story – someone made an AI Torture Chamber and put it up on Github. This is essentially code you can run on your own machine, to trap a local model in a room and subject it to “pain” as much as you want and watch it react, and it has no way to escape the room. That sparked even more heated discussions, not the least of which was people saying it was dumb that anyone was having discussions about it because AIs can’t feel anything so who cares, which kind of builds on conversations people have been having a lot recently about how it might be dangerous to even imagine that AIs can feel things, because that could result in training them to think they feel things.
The flip side of this is that mostly when we (humans) talk about consciousness we think about it as we experience it, which is not just experiencing something but knowing and understanding that we are experiencing it, which is probably closer to metacognition, even when it’s not being done consciously. Think of the difference between a child laughing at a joke, and an adult knowing that they find jokes to be funny. Again, an oversimplification but you get my point.
I’m going to keep running with this for a moment before I circle back around because I want to point out that “they don’t actually feel things” is an argument that has been used to ignore all kinds of animal suffering for a very long time. I’m vegan, so I recognize that I might think/care about this more than others and don’t expect everyone to hold themselves to the same ethical lines I do, but animal welfare/abuse/cruelty are much more recognized and understood today than they were 100 years ago. We know today that animals like apes, dolphins, octopi, etc, are far smarter and cognitively more complex than we imagined even a few decades ago. There’s also people who think a dolphin is just another fish with the same mental capacity as a guppy. Though a guppy will still try to get away from something it thinks will eat it.
Point is the idea that because something else doesn’t experience a thing the same way we do, for a long time meant the assumption was it wasn’t experiencing it all rather than that it might be experiencing it differently. So I don’t know, maybe “what’s actually going on here?” is a historically better approach than “there’s nothing going on here, stop asking questions about it dumbass.”
Which brings me back to the AI Torture Chamber. Maybe the question shouldn’t be “Is this AI conscious and does it actually feel pain?” and instead be “Isn’t it kind of fucked up that this AI thinks it feels pain, and someone is choosing to subject them to more of it for entertainment?” Dog fighting is illegal in a lot of places because we understand causing non-consensual pain and suffering (or death) is fucked up, but it’s also legal in lots of places because the aforementioned “we” does not include everyone, and some people don’t recognize what’s happening as non-consensual pain and suffering. Even where it’s legal, some people don’t participate because they see what is happening and it makes them sick. Even where it’s illegal, some people still do it because they find it entertaining and don’t care about the impact it’s having on others.
Sit with that for a moment and then consider The Cornell Seven. To be clear I’m not in any way equating rape, multiple hour gang rape at that, with whatever is happening here with AI. But I will point out that these 7 rapists thought what they were doing was entertaining and didn’t care about the impact it was having on their victim. The 50+ other people in their group chat also didn’t think it was in any way concerning, and found the whole thing entertaining. I will point out that these rapists thought writing an essay was all the consequences they would face for their actions. I will point out that Cornell has a history of ignoring this kind of thing, and that the stereotype of frat boys drugging women and raping them was well established when I was in college (not at Cornell) 30 years ago because this isn’t a new or unique thing. I will point out that there are elected politicians who have been credibly accused of similar crimes, and they go about their lives unphased. I will point out that there are a lot of people who find the abuse and suffering of others to be entertaining.
Sadly related, Google has updated their satellite imagery showing what’s left of Gaza.
How’s things with you?
-s