💾 Archived View for rawtext.club › ~winter › gemlog › 2023 › 3-26.gmi captured on 2023-07-22 at 17:02:28. Gemini links have been rewritten to link to archived content
⬅️ Previous capture (2023-04-19)
-=-=-=-=-=-=-
The last year in general, and the last couple of months generally, have been bad from a mental health perspective. I've taken some real, tangible steps to healing from what went on when I was younger. I'm seeing a therapist; I need to schedule our next appointment. But it's been helping.
What hasn't helped has been watching text and image generators get kind of laughably bad, then a little bit better, a little bit better. As Judea Pearl put it in an Atlantic article a few years ago:
All the impressive achievements of deep learning amount to just curve fitting.
How a Pioneer of Machine Learning Became One of Its Sharpest Critics
And I know that's true. I know the basics of how the generators work, having studied this sort of thing a couple of decades ago. But that doesn't help when I see people's reactions to it: ChatGPT getting incorporated into every major corporate product; opinion pieces about personhood for chatbots; people posting AI images for lols, for engagement, slowly poisoning the shared consensus as to what's actually real.
Opinion: Is it time to start considering personhood rights for AI chatbots?
Just today on Twitter I see a library surrounded by trees, looking like an abandoned building; I see the pope wearing a puffy white parka and a gaudy, expensive looking cross (lol; lmao). Lev Grossman, author of the Magicians Series, writes about his own despair about the future, figuring that publishing companies will use text generators to generate books, and that people will pay for them, and authors, real authors who have spent their lives reading and writing and likely unknowingly contributing to the opaque datasets used for training, will be left to deal with the fallout, competing fiercely for scraps in what is already a very poorly-compensated profession.
Lev Grossman on AI and authors
And I feel bad myself, even though this stage of the "AI" hype cycle feels startlingly close to the hype cycle around self driving cars (where we were told that within five years, all cars would be self-driving), the hype cycle around Google Glass (whoops!), the hype cycle around e-readers (remember how they'd destroy print books forever?). I feel bad because the poisoning of our online lives is accelerating, and likely will never go away. It used to be you needed a Russian or Romanian troll farm to whip up hatred on Facebook. Now you just need a ChatGPT license and a little creativity. But even if people aren't using it to foment war and genocide, the general feel is still not good. People sit in Discord channels and on subreddits, posting image after image. People are falling in love with an AI bot based on an earlier version of ChatGPT.
Forget ChatGPT! People are falling madly in love with this AI bot -- here's why
We could be doing anything with our lives, and we're choosing to do this.
It used to be when we were lonely and went online, we were able to find other people. I don't mind saying I've fallen in love several times, over the wire; at least it was with real people, and there's the chance that somewhere far south-west, or east, of here, I might be remembered fondly as well. But now the lights in the network - our human network - are going dark. There's less variety in what we're doing online (how many people just scroll Twitter, Instagram, Facebook?), and soon there'll be fewer people to talk to as well.
Generators are filling that void. Generators are writing copy and filling the web. Generators are generating pictures of Donald Trump being arrested, going viral on social media. Yes, I realize these outcomes aren't any different from a photoshop, or paying an impoverished freelancer $5 or $10 a page. But generators have now made this easy, frighteningly easy. Put down $50. $100. Look at all those credits! What can you generate that might catch people's eye?
Fake Trump arrest photos: How to spot an AI-generated image
I don't want to sound like a luddite, because there is still some part of me that believes in the deep, transformational power of computers. I get so much joy working on my life-long software projects. I see people writing basically anonymously in the dark corners of gemini. People still update phlogs, even with an audience of basically none. In Cambridge, someone has made a solar powered web server on which you can leave ephemeral messages. Good people are still doing good things every day. Making beauty in a world that doesn't seem to want it.
But I don't know what to do because it feels like the future is going to be a firehose of fake news, false images, and digital sludge. Everything 2016 and after just feels like a warmup. Right now you can spot the tells, but will that be the case forever? I suppose there's a case to be made that we're past the point of the cleanest training set, that the web of the future will be so full of generated data that the tells will be baked, that generator will never surpass what it can currently do.
Or maybe not. Maybe the generators will keep getting better. Maybe they'll convincingly figure out eyes and fingers. I can see us spending our days doling out dollars for credits, making big tiddy goth gfs, feeding those images into Replika6, picking a voice, having sexy calls, losing entire years of our lives.
What's clear is that we, the collective we, are not going to do anything to ensure what we interact with is real. There should probably be laws, markers, filters. All of this should be opt-in. Datasets should be published, open access. But if we can't collectively agree to tackle our most pressing existential crises, such as climate change, I don't feel any confidence we'll be able to take steps to allow people to understand if what we're interacting with is real. Whether the person at the other end of the email or chat is a person; that the image was taken by a photographer; that the writing was done by someone who wants to create beautiful poems. I want to know that my interactions are actual. That I'm not just sucking from the digital feedbag, spending my days, wasting my life.
On Twitter tonight, I posted something about this, and was told, geez, that image is obviously fake. Look at this detail here; look at that. Come on, winter (perjorative). But that's not the point. Take another look. That's not why I feel this tightening sense of dread.