Going to admit, I don't think there's any good way to handle AI-generated materials in external articles with a generalized AI ban.

* If the policy is that a person can't link to AI-generated material such that it gets captioned on Pillowfort as an image or text, you're expecting them to have way more dilligence than can reasonably be expected. Sometimes caption images don't even show up in the original article, and this is one of those times where people don't always look closely at the content. 

It's also not doing anything to address the root problem of this material, namely that artists and writers are not getting paid to create it. Even if this was something I thought could reasonably be enforced, it's far too separated from the actual wrongdoers in this equation to be doing any good. None of the places writing these articles are going to have their bottom line affected by Pillowfort's AI Ban policy. The people who should be preventing this kind to exploitation from happening are the government, but realistically speaking, it's often organizations like unions who do the real work in pushing back on this kind of exploitation (eg. the terms negotiated by the WGA).

* And if the policy is that AI-generated articles can be linked to show these kind of AI header images or AI-produced headlines, then the policy is tacitly unfair and by and large targeting people who have no real place in this exploitative dynamic. Saying that people who are mostly just using AI to generate and post images or text they have no intention of selling for a profit are banned, while saying that people who are absolutely involved in this dynamic don't need to be held to the same standard.

I think this is a case of some very exploitative people being nigh untouchable, so instead, for people to feel like they are doing something, they demand punative action on easy targets, despite the fact that this does nothing to actually address the core issue.

I think the main threat from AI is as a means of capitalist exploitation. And this is not because AI's are even adequately good at the things these corporations want to apply them to. They're just an excuse to look good to stockholders for doing more layoffs while their product quality plummets. If AIs do ever become capable of autonomous decision-making on the level of a human (or even like, a dog or a cat), then that's a completely separate issue from what's going on at present. However, I think a lot of people are worried about current or future machines actually being able to outthink us.

But I don't think we need to worry about an AI explosion making exponentially more intelligent AI in any near future. I don't think the process by which programs like ChatGPT are made can scale in any reasonable way. If you look at the gargantuan knowledge base that ChatGPT was trained on, and you look at the conceptual basis by which the people making these programs think they can improve them, it doesn't add up.

ChatGPT is good at sounding smart--much less so at actually being smart. And the idea is to take these incredibly limited programs that have already required enormous resources to create (the combined corpus of a huge percentage of the internet and who knows how much computer hardware) and just...add more resources. The model doesn't scale. 

And not only that, but I just really, really don't think that's how brains work. I don't think that's how they evolved. And I don't think it's realistic to expect to recreate cognition by thinking we can just add more and more neurons and feed them more and more data. I think the idea of depending on the superficial similarity of neural nets to brains with neurons to eventually lead to a deeper similarity if we just keep adding more is a form of magical thinking.

It's like how we invented flight. Nature can inspire, but it's often infeasible to emulate nature precisely, and inefficient as well. Airplanes and helicoptors allow us to fly, but not by exactly imitating the flight patterns of, say, birds to do so. Birds are both more complex than necessary for flight and less efficient than our purposes call for. Airplanes don't need to flap their wings, and helicopters can employ a mechanism that doesn't even exist in any known lifeform to fly.

I don't think making a brain capable of complex thought is just about adding neurons. I think it's about adding a lot of abilities we have that our brains likely implement in unnecessarily complex ways through accidents of evolution. I think we have to be able to break cognition down and understand its component parts, truly and completely, before we can emulate it. I think we can't just make programs that are structured vaguely like brains and expect them to behave truly intelligently any more than we can make machines that are structured vaguely like wing muscles and expect them to fly for us.

I also think that the concepts that underlie the methods by which brains work are many orders of magnitude more complex than the concepts that underlie the methods by which wings work. And I think they will absolutely require a deep understanding of their mechanisms and processes to implement, which the current black box methods of implementation simply aren't compatible with.