Right. Coming back to this, as promised. Resources, yes, but I also want to suggest some strategies. Adding a cut because this will get looooooooong.
First, I think it's helpful (and perhaps soothing) to understand that you are very likely to lose this battle, but that doesn't mean you will lose the war! It's very difficult for someone to back away from something they've sponsored in their workplace without losing face. They'll resist it hard, and the more you push them, the more they'll dig in.
It's even harder for them to back away when the entire society (it seems like) is yelling at them that they have to do this. You've probably noticed that pro-AI rhetoric is heated, deeply coercive, often ignorant, and freakin' everywhere. The higher you go up a workplace hierarchy, I'm sorry to say, the worse this is. If you can, spare a little grace for your reporting chain: they're feeling beleaguered and confused -- and likely threatened by your resistance, which is far less defensible but still something you're stuck reckoning with.
So I think the best way to think about your objective here is building a safe, face-saving retreat route away from AI for them.
Second, ethics arguments are useless for this. I know. I'm sorry. I fucking teach ethics sometimes, and it burns me how people think it's magic pixie dust that ultimately means it's fine to do anything they want to do, just, y'know, ethically... when so often the ethical thing is not to do that thing at all. But in our retreat-route frame, ethics arguments cut off all retreat. To back away from AI, the powers that be have to admit that their AI idea was Wrong and Bad... and they don't want to do that for psychological self-protection and saving-face reasons.
So forget about ethics. You can only use ethics as an argument if you think your customer/client/student/patient/whoever-you-serve base is going to walk away from the organization if it AIs up -- which is starting to be possible, but not the odds-on bet for most organizations just now.
Okay, so how might you build that face-saving retreat route for the powers that be? Pick and choose from, or combine some or all of, the following tactics:
Most new initiatives in most organizations have some kind of "did it work?" mechanism, formal or in-. Put this into service for whatever daft AI thing the powers that be decide they want to do! Help them determine (ideally quantify) and (crucially) write down desired outcomes. Help them list (again, in writing) any red flags that will cause the AI effort to be discontinued.
In fact, try to build on the defending-from-bad-outcomes frame. One way is to bring up cost questions -- chatbot firms are moving to "token-based billing" (it's all over the news and Reddit just now), so what happens if their AI project gets pricey? Try to pin them down to an actual number for "how pricey is too pricey." After all, gotta protect the almighty budget, doncherknow.
If you're lucky, this might even win you the war immediately, if the powers that be don't see success in the offing and retreat gracefully from the idea of shoving AI into whatever.
If that doesn't pan out, though, you've still put yourself in a stronger position to win the war eventually, because you've created a retreat route in the form of "we assessed AI Project against the criteria we set up at project start, and it didn't meet expectations so we're axing it." This is an incredibly ordinary thing for the powers that be to say, so zero face lost!
If you have to start an AI thing, you can likely talk them into starting small. "Pilot projects" are common and ordinary and unthreatening. If you get traction with this, try to nudge them toward a project you know will fail hard and visibly, if you can do so without the blame getting cast on you or your colleagues. (In other words, avoid inviting the "you must not be prompting it right!" accusation.)
I talked about assessment first because assessment is absolutely key to the pilot-project tactic. You NEED to show up to the meeting with suggested success and failure criteria for the pilot project they want (or the one you suggest because you know it will fail), and ideally those criteria should be observable and measurable in numbers.
Error rates are really good here, especially if they're something the org normally tracks anyway. Bullshit machines gonna bullshit; make sure you catch them at it! Speed rates might work if the organization measures speed correctly. "Lines of code per unit of time," for example, is a shitty speed rate that will not help you; the bullshit machines barf out thousands of lines of vomitous code in seconds. "New-feature speed to production" may be better if and only if code review time, QC time, regressions, vulnerabilities (and the time to fix them), and similar get factored in.
Productivity measures, especially subjective ones, are least useful. There's documented evidence that many programmers believe gen-AI coding tools make them more productive when exactly the opposite is happening. Use extreme caution here. Prefer, or at least include, quality measures if you can -- the bullshit machines usually shit themselves very obviously on quality.
The retreat route here is probably obvious, but for the sake of completeness: "pilot project didn't work, so we're axing the initiative" is also a completely ordinary thing for the powers that be to say.
Socratic inquiry based on horror stories
This one's my fave, because I'm Machiavellian like that.
Ask helpful "what if?" questions aimed at preventing bad outcomes, based on things we already know about bullshit-machine behavior that are relevant to whatever product or service the powers that be want to shove AI into. Ideally the bad outcome is existential -- something that shut an organization down or embroiled it in a nasty lawsuit or something else the powers that be will recoil from in horror. A publicly-embarrassing, covered-in-the-news data breach, project failure, or project cancellation also works great.
For this to work, you need "things we already know." I think the easiest way to get some in advance is via AI-detractor communication outlets, including but not limited to:
my own little sideblog @generatingfail
If you are subscribed, 404 Media's AI tag and chatbots tag (even if you're not, you might get headlines that could be leads)
Mystery AI Hype Theater 3000 podcast (has transcripts, so you can search!)
The thing is, though, you cannot show the powers that be any of the above except for 404 Media! You cannot! They will simply say "well, those fools are biased Luddites" and you will lose. All of the above outlets, however, cite/link to their sources. (I'm a librarian. I'm fastidious about this.) Bring those sources instead, okay? Promise me?
Here's your script for this, once you have a contextually-appropriate horror story or two (two is better; three is better still) in hand. Furrow your brow in puzzlement, then say "I heard from a friend that {the AI thing they want to use} does {bad thing from your source} a lot, and it's caused {really bad problem the powers that be do not want}. Can we prevent that somehow? What's our contingency plan for it?"
(It me, I'm your friend, I'm happy to be your friend that you heard this from! Even if you didn't hear it from me!)
If they ask for evidence, pull out your laptop or phone and do whatever search will get you to your source fastest (including if that's a search through your browser cache). Try to have more than one source and/or more than one incident available, in case they ask if the one you showed them was a one-off. (Another way to address this objection is "let's try it!" and you just demonstrate the problem live.)
Again, if you're very lucky, this will kill the project a-borning, because it's a retreat route: the powers that be can say "gosh, that would be Really Bad, let's not!" or even just "let's delay!" or "let's do some testing!" or "let's do a pilot project!" (which you can suggest at this point!) and you win.
If you're not lucky, you've still set yourself up to win eventually, because you've primed the powers that be to look for a specific failure mode in the AI initiative. Once they find it, it becomes a retreat route for them.
Uh, I don't really have one, except "good luck, and I hope this helps."
Okay, maybe this: I hope other people will comment and/or reblog to add ideas and refinements and successful cases of resistance. If you do, I thank you in advance!
