Discussion

Trying to manufacture permission

I wanna expand on something I wrote a few months ago:

One of the features of “AI” is the diffusion of responsibility: “AI” systems are being put in all kinds of processes and when they fuck up (and they always fuck up) it was just the “AI”, or “someone should have checked things”. “AI” companies want to sell machines to solve every issue but give no warranties, take no responsibilities and the same dynamic often extends to organizations using “AI”: You get the support-chatbot to promise you a full refund and when you claim that you get a half-assed “oh but that was just the bot, those tell bullshit all the time”. That’s where human in the loop setups come into play: What if the company can just hire one sucker to “check” all the “AI” slop and when things fall apart that one person has to take the blame.

Diffusion of Responsibility

I think this analysis still holds up. But of course it focuses on the perspective of big tech corporations and “AI” labs: Their idea of careless innovation where you can release all kinds of products – regardless of just how harmful they are – to the public if that release makes number go up. (That number being the valuations of those companies, the stock price or whatever. Literally whatever. Shit no longer needs to make sense or everyone would have laughed Elon Musk – who is a Nazi btw. – and SpaceX off the stage when their S-1 dropped.)

In my day job I get more and more statements to me beginning with “I asked ChatGPT and …”. And not just by the subset of colleagues who do think that you can use LLMs to meaningfully gather information and find solutions but by people who do know better. Because they will add some form of “I know that ChatGPT is just a synthetic token extruder” or “I know that AI’s often make stuff up” to their statement:

“I know that AI often makes shit up but still I just quickly asked it about this problem we were talking about and it said this …”

That is a really fucking weird statement, isn’t it? Because what does it say? So the person is aware of the properties of the tool they chose to use, knows the issues that tool has, references those issues and then presents the problematic result. That is fucking weird, isn’t it?

“I know you shouldn’t use a hammer to put a screw into the wall but I used a hammer and hammered this screw into the wall (and now the wall looks like shit and the screw doesn’t hold shit properly.” kinda has the same vibe. And if someone said that they’d either be 8 and we’d have to have a talk about tools and what’s safe to use for a child or we’d send them to the hospital because they must have a concussion or something.

Why say those words? Well it’s about getting permission to do something you know is bad. Is disrespectful.

Where I work there is a clear “AI” policy: The person you are sending something to must never be the first human to read something. You generate it, you check it. Only then can you pass it on. If that was what people would do, they wouldn’t need to add the “I just used ChatGPT even though it’s shit” addendum: They’d generate something, check it, find it lacking, generate again, and again, until something accidentally checks out. Then they’d release it on other people. But they don’t want to do that. They want permission not to do the labour of checking the outputs. Because who would? Checking AI output is super not fun. Especially if you just want to enter a conversation with a solution/and idea to gain some social capital. Or (again pretentiously quoting myself): ““AI” is that: Don’t put much if anything in and get something that might be passable to sell out.”

Especially when people know better, it’s the disrespect that I find the most annoying. I don’t care too much which tools you use for your process if you get results. I don’t need to tell you how to write notes or debate which programming language you use to build a small program. It really doesn’t matter to me. So use an LLM that cooks your cognitive abilities. But I will not have the fact checking of that system outsourced to me.

Because that’s what’s happening here: Those people are putting their responsibility of making sure their work stacks up on me. Not in a “I need a second pair of eyes/some feedback. Does this make sense?”-mentoring kind of way (I am always happy to do that) but as a “Fuck you”.

This is not a super deep insight, but it has brought a new policy I will enforce for interactions with me.

Don’t insult me by saying you used an “AI” while stating how flawed these systems are. If you bring synthetic content my way, you must have checked it yourself and can stop telling me that you used morally indefensible tools to come to that point. Because I will of course judge you for doing so.

Liked it? Take a second to support tante on Patreon!

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.