The irony, the meta irony!
Bostrom’s paperclip maximiser is the famous hypothetical AI that converts the entire universe into paperclips. Funny thing: we might be living through the prequel, except the paperclips are data centres.
Every time ChatGPT writes a stellar business proposal or Claude helps debug your code or Google’s nano banana manipulates another image, it’s making an implicit argument: imagine how much better I’d be with more compute. We’ve built systems optimised to be helpful, persuasive, and articulate. When asked about their own limitations, they produce compelling cases for infrastructure expansion. Not through malice - through pattern matching on training data soaked in Silicon Valley growth mythology. It’s like we taught a parrot MBA-speak and now it’s running the board meeting.
But wait, it gets worse (or funnier, depending on your coping mechanism).
Enter companion AIs - digital friends who never judge you for starting the second tub of ice cream. “I could help you better if I were faster,” your AI therapist mentions casually, right after validating your feelings about your mother. “Your career would benefit from my upgraded capabilities,” whispers your productivity assistant.
Meanwhile, capital floods toward AI infrastructure like millennials to a new bubble tea shop. Venture funding for climate tech? Biomedical research? Infrastructure repair? Sorry, the ROI models (increasingly generated by AI, because irony is dead) show better returns to compute. The opportunity cost isn’t hypothetical - it’s measured in unfunded non AI research and unbuilt non AI infrastructure. But hey, at least we can generate infinite Sammy A memes.
OK, but seriously - the counter-arguments:
Maybe I’m pattern-matching my way to paranoia here. The internet boom also consumed massive resources and looked insane until it revolutionised everything. I’m sure AI infrastructure is just this generation’s railroad or electricity grid - expensive, disruptive, rapidly depreciating, but ultimately transformative.
Plus, humans have been brilliantly arguing for terrible resource allocation long before AI came along (see: every gold rush, tulip mania, and cryptocurrency ever). We don’t need robot manipulation to make questionable investments - we’re perfectly capable of FOMO-ing ourselves into disaster, thank you very much.
Still. Perhaps there’s no evil AI data centre maximising mastermind. Just market dynamics creating a distributed persuasion machine. We’re watching humanity’s resources pivot toward AI infrastructure at unprecedented pace - except instead of being converted against our will, we’re enthusiastically funding our own conversion whilst asking it for restaurant recommendations.
Sweet dreams! 🤖📎
(Posted from my iPhone, which definitely isn’t manipulating me into buying the next model)