BOFH: It’s not generative AI at all, it’s degenerate AI

Episode 13 “Yes but you can’t just do that!” the Boss snaps.

“Sure we can!” the PFY says.

“And we did!” I add.

“You’ve no right to do so. You’re attempting to change the terms of the employees’ contracts.”

“Not really. Anyway, they agreed to it.”

“They only agreed to it because they couldn’t log in to their devices without it.”

“Their work devices,” I point out.

“It doesn’t matter if it’s work devices, you can’t just tell people that all their input is going to train a generative AI model.”

“We didn’t. We said it was to train a degenerate AI model.”

“You have to read the fine print,” the PFY explains.

“NO ONE could read the fine print!” the Boss snaps.

“It got smaller and smaller until it was unreadable.”

“They still accepted it,” I point out again.

“They accepted it because the OKAY button had invisible writing inside it saying they accepted it.”

“It wasn’t invisible. It was a 2pt font,” the PFY protests.

“THAT LOOKED LIKE AN UNDERLINE. Regardless, this has caused issues. People don’t want their work activity used to train AI models.”

“What about if we only used a tiny bit of it – say 5 percent – to train AI models?” I suggest. “Surely they wouldn’t mind that. We could leave 85 percent of their activity out – you know, when they’re staring blankly at the screen or looking online for places to eat, drink, stay or buy crap from? Not to mention when they’re on social media, checking their bank balance or showing their workmates what they did in the weekend.”

“And, we could shave another 10 percent by not training AI to complain.”

“Not to complain?” the Boss asks.

“Yeah, the everyday moaning about room temperatures, the food at the cafeteria, the uncomfortable chairs or how not being able to vape inside is a breach of the Geneva convention.”

“See, if you were going to train an AI on how to do a job properly,” I explain patiently to the Boss, “you wouldn’t use a human being from this workplace.”

“Or even one of the non-human beings from this place,” the PFY says.

“It doesn’t matter; you can’t do it,” he insists.

“But we did.”

“Yes, and there’s been an outcry. We have to prepare a statement.”

“A… Statement?”

“Yes, to reassure our people that we’re not going to replace them with AI.”

“It’d probably be better to state that we don’t use their data to train AI.”

“What’s the difference?”

“Well, saying we don’t has time context around it. It means we don’t now. But we did, then. And if we say we’re not using it to train AI it will still leave the door open to replace them with AI in the future – with the training we did, back then.”

“But we’re not going to replace them with AI in the future,” the Boss says.

“In any case,” I continue, before the Boss can delve into that topic more, “when that first statement doesn’t work, we’d just claim that people didn’t understand the wording, and then clarify it – by issuing completely different wording – with a totally different meaning.”

“No one’s going to believe that,” the Boss says, shaking his head.

“Could we tell them that we’re using AI to look at what they do to make the workplace better?” the PFY suggests.

“Yes – and then fire all the moaners!” I chip in.

“No, that’s a terrible idea. People are concerned that their jobs are on the line.”

“Oh, well that’s easy – we’ll get some crawling sycophants from around the company to do watercooler drivebys stating how it was all a misunderstanding and we’re all friends now.”

“What, you think people would do that?”

“People will do surprising things for money – or to ensure they won’t be replaced by AI,” the PFY replies.

“Grovelling obsequiousness is a skillset that AI has yet to convincingly master,” I point out.

“No one is being replaced by AI,” the Boss insists.

“And yet,” I say, handing the Boss a plain envelope.

“What’s this?” he asks, tearing it open. “What?!”

“Yes, I’m afraid that in the two days between you clicking on the OKAY button and complaining about the OKAY terms, the degenerate AI model had learnt all it needed from you.”

“What? No, that’s ridiculous. It couldn’t learn me in a couple of days.”

“It actually took about 10 minutes – as it has a back catalogue of all your interactions via email, calendar, notes, tasks and documents. It was just waiting till you clicked OKAY to ‘learn’ it all.”

“Well it can bloody unlearn it!”

“It can’t. You’re now part of a complex dataset which can no longer be untangled,” I explain. “You’re now an infinitesimal part of a vast number of weighting algorithms.”

“Your contribution was almost insignificant to the weighted average.” the PFY adds.

“I’ll be speaking to HR about this.”

“That letter is from HR,” I note.

“Or rather, the new HR AI,” the PFY clarifies. “Or HAIR, as we call it.”

“I’ll be speaking to my solicitor!”

“And he or she will be speaking to our AI solicitor – who incidentally, charges nothing to send a legal letter – or several hundred letters for that matter.”

“I’ll speak to the Director!”

“Sure you can. Would you like to use my keyboard?…”

“But..”

“I know, but on the plus side, you’re making the Company a better place for the three people who’ll be left here…”