I’m going to approach this as though when tumblr user tanadrin says that they haven’t seen anti-AI rhetoric that doesn’t trade in moral panic, that they’re telling the truth and more importantly that they would would be interested in seeing some. My hope is that you will read this as a reasonable reply, but I’ll be honest upfront that I can’t pretend that this isn’t also personal for me as someone whose career is threatened by generative AI. Personally, I’m not afraid that any LLM will ever surpass my ability to write, but what does scare me is that it doesn’t actually matter. I’m sure I will be automated out whether my artificial replacement can write better than me or not.
This post is kind of long so if watching is more your thing, check out Zoe Bee’s and Philosophy Tube’s video essays, I thought these were both really good at breaking down the problems as well as describing the actual technology.
Also, for clarity, I’m using “AI” and “genAI” as shorthand, but what I’m specifically referring to is Large Language Models (like ChatGpt) or image generation tools (like MidJourney or Dall-E). The term “AI” is used for a lot of extremely useful things that don’t deserve to be included in this.
Also, to get this out of the way, a lot of people point out that genAI is an environmental problem but honestly even if it were completely eco-friendly I’d have serious issues with it.
The number 1 concern that I have with genAI, as I’ve already touched on, is that it is being sold as a way to replace people in creative industries, and it is being purchased on that promise. Last year SAG and the WGA both went on strike because (among other reasons) studios wanted to replace them with AI and this year the Animation Guild is doing the same. News is full of fake images and stories getting sold as the real thing, and when the news is real it’s plagiarised. A journalist at 404 Media did an experiment where he created a website to post AI-powered news stories only to find that all it did was rip off his colleagues. LLMs can’t think of anything new, they just recycle what a human has already done.
As for image generation, there are all the same problems with plagiarism and putting human artists out of work, as well as the overwhelming amount of revenge porn people are creating, not just violating the privacy of random people, but stealing the labour of sex workers to do it.
At this point you might be thinking that these aren’t examples of the technology, but how people use it. That’s a fair rebuttal, every time there’s a new technology there are going to be reports of how people are using it for sex or crimes so let’s not throw the baby out with the bathwater. Cameras shouldn’t be taken off phones just because people use them to take upskirt shots of unwilling participants, after all, people use phone cameras to document police brutality, and to take upskirt shots of people who have consented to them.
But what are LLMs for? As far as I can tell the best use-case is correcting your grammar, which tools like Grammarly already pretty much have covered, so there is no need for a billion-dollar industry to do the same thing. I am yet to see a killer use case for image generation, and I would be interested to hear one if you have it. I know that digital artists have plugins at their disposal to tidy up or add effects/filters to images they’ve created, but again, that’s something that already exists and has been used for very good reason by artists working in the field, not something that creates images out of nothing.
Now let’s look at the technology itself and ask some important questions. Why haven’t they programmed the racism out of GPT-3? The answer to that is complicated and the answer is complicated and sort of boils down to the fact that programmers often don’t realise that racism needs to be programmed out of any technology. Meredith Broussard touches on this in her interview for the Black TikTok Strike of 2021 episode of the podcast Sixteenth Minute, and in her book More Than A Glitch, but to be fair I haven’t read that.
Here's another question I have: shouldn’t someone have been responsible for making sure that multiple image generators, including Google’s, did not have child pornography in their training data? Yes, I am aware that people engaging in moral panics often lean on protect-the-children arguments, and there are many nuanced discussions to be had about how to prevent children from being abused and protect those who have been, but I do think it’s worth pointing out that these technologies have been rolled out before the question of “will people generate CSAM with it?” was fully ironed out. Especially considering that AI images are overwhelming the capacity for investigators to stop instances of actual child abuse.
Again, you might say that’s a problem with how it’s being used and not what it is, but I really have to stress that it is able to do this. This is being put out for everyday people to use and there just aren’t enough safeguards that people can’t get around them. If something is going to have this kind of widespread adoption, it really should not be capable of this.
I’ll sum up by saying that I know the kind of moral panic arguments you’re talking about, the whole “oh, it’s evil because it’s not human” isn’t super convincing, but a lot of the pro-AI arguments have about as much backing. There are arguments like “it will get cheaper” but Goldman Sachs released a report earlier this year saying that, basically, there is no reason to believe that. If you only read one of the links in this post, I recommend that one. There are also arguments like “it is inevitable, just use it now” (which is genuinely how some AI tools are marketed), but like, is it? It doesn’t have to be. Are you my mum trying to convince me to stop complaining about a family trip I don’t want to go on or are you a company trying to sell me a technology that is spying on me and making it weirdly hard to find the opt-out button?