Thoughts on AI
I haven’t posted about all the ongoing AI issues before now because — well, frankly, because it’s hard to take the inchoate mass of screaming inside my head and boil it down to anything resembling useful words. But I also believe it’s my duty to try, so here we go.
In case you couldn’t tell from the above, my feelings on AI in the current sense we’re using the term are not positive.
To keep this from getting too unwieldy, I’m going to try and boil it down to bullet points.
- First: we’re all clear on the fact that “AI” is not an actual mind, right? ChatGPT and its ilk can achieve an uncanny valley imitation of being a mind, but they have zero comprehension of their output. It’s just algorithms trained on a mass of data, i.e. machine learning (ML). The day we have actual computer sentiences, I will be concerned for their personhood and how we treat them. We’re not there, so I’m not.
- Speaking of that mass of data: you’ve probably seen it mentioned, but it bears repeating, that the data in question may often have been obtained from questionable sources. I say “may” because e.g. OpenAI, the company behind ChatGPT, Dall-E, and others, has been the opposite of transparent about their material and process. We can be fairly certain that a lot of copyrighted works went into the pot, without permission from or payment to the copyright holders.
- (But isn’t that how artists and writers and so forth learn anyway? See the first bullet point, re: these algorithms not being people. We’re talking about companies profiting off other people’s work, without compensation.)
- Less often mentioned that I’ve seen, but still relevant: it isn’t just about grabbing a bunch of data. You also need people to process that data, and it seems like many of the people who have done that work/are doing that work get paid sweatshop wages at best. The training of the algorithms can be unethical, too.
- Also relevant: though I’m told some ML setups are better on this front now, some of them are massive energy hogs, on par with crypto for the damage they’re doing to the environment. Another way in which they are unethical.
- Who owns the output of these things, anyway? Anybody? Bueller? Jury’s out on that one. It’s a mess of a question we don’t have a firm answer to yet.
- Related to but not the same as the above: bear in mind that when we stop using humans to do a thing and outsource it to an algorithm, we are handing power over that thing to a single company, or at best a small group of competing companies, who now own the proprietary means of dealing with a task. Does that sound like a bright future?
- Especially since there is high risk of society’s biases being implicitly baked into the algorithm — but since the whole process is proprietary and secret, we have very little way to identify and counteract it. Sorry, our Health Care AI (TM) has decided to deny you treatment for your illness, because its data says that isn’t worth the effort. (This is probably already happening in some corners, and you may not even know it. The most recent shouting is about art and writing, but ML has made huge inroads elsewhere without nearly as much news coverage.)
- Or sometimes the bias is explicit: ChatGPT will refuse to write fanfic about same-sex characters in a romantic relationship because it would be “disrespectful to the LGBTQ” community. But apparently it’s fine to write the same kind of story about opposite-sex characters? Smells like homophobia to me, not respect.
- Also, this is cannibalizing jobs at a phenomenal rate. If you don’t believe me, ask a freelance writer how much work is available to them now, compared to this time last year. Companies have decided they can just outsource a lot of that work to a program, instead of paying people.
- But don’t think that means the program is doing great work! In many cases the output is bad . . . but hey, it’s cheap, and these days capitalism would far rather have “cheap” than “good.”
- As a result, much of the remaining work consists of “please clean up this thing we had an AI write.” Which is sometimes more obnoxious than just writing the damn thing from scratch — but do you think the company’s paying extra for that work? Hell, no; you’re only cleaning it up, after all. If anything, that deserves less pay, right?
- Plus there are all the AI-generated submissions swamping markets right now, because scammers are out there telling the internet about their latest get-rich-quick scheme, which involves using AI to generate a story/article/etc. and then sell it.
- Oh yeah, the internet. You thought it was a cesspool already? It’s going to drown under the weight of machine-generated misinformation and spam.
- And hey, do you think Hollywood produces too many derivative, cookie-cutter rehashes of stories you’ve already seen? Just wait and see what happens if the Writers’ Guild strike fails to make the studios bend. The occasional “gravity drowned” bit of serendipity notwithstanding, a program that is fed on a bunch of X is going to produce more X. More stale, paint-by-numbers executions of the same tired plot structures, just with shinier FX.
- Bringing this home to my own corner of the world . . . when it comes to art, fundamentally, I care about it as a form of human expression. I do not merely want One Unit of Story. I want the ideas, the characters, the words, produced by an actual human brain, because that human thought they would entertain someone. AI cannot do that, because what we call AI right now cannot think. It cannot express. It can only remix and regurgitate. Even the most banal writer who puts a stupid cliche in their story puts it there because they like it, not because some statistical model says it should go there. I care about that; I care about it a lot. And I hope other people do, too — that they want more than just algorithimically generated models of what a story or a piece of art should look like.
As many people have said, we can’t put this particular genie back in the bottle and somehow un-develop AI. We can, however, push back on the social front, against normalizing the use of such things, against proprietary processes whose inner workings and biases we are not allowed to know, against “cheap” driving out “good,” against discarding the human element and the livelihood of thousands of people along with it. We can want better. We can demand better, and no, it’s not enough to assume that ~somewhere down the road~ this will all produce the utopia we were promised, not when it’s doing increasing amounts of harm now. We can tell the tech companies hopping on the ML bandwagon not because there’s an actual beneficial use for it in their field but just because it’s the new hotness that hey, that isn’t the way to go. We can tell our elected representatives that this shit needs to be regulated, and where it has transgressed, it needs to be prosecuted. We can change the path we’re on.
I hope some of you will help. Better Without AI is a place to start.
I’m closing comments on this post, because it already took more energy than I really have to spare to write it up, and I can’t spare the energy to moderate the responses I’m already likely to get. I know there are aspects I missed, facets I summarized too broadly, developments I’m not aware of, and so forth. The above is more “primal scream” than “comprehensive analysis of the current state of affairs.” But it’s what I’m capable of right now, and so for now, that will have to be enough.