Listen, I’ve spent enough time in bars to know that getting people to cooperate is about as easy as convincing my landlord that the rent check is “in the mail.” But at least drunk people eventually figure out how to share the last bottle of bourbon. AI, as it turns out, can’t even manage that basic courtesy.
So here’s the deal: Meta - you know, Facebook’s midlife crisis rebrand - just announced they’re planning to populate their platforms with AI-generated users. Because apparently, the current mess of MLM schemes and your aunt’s conspiracy theories isn’t quite dystopian enough.
Connor Hayes, Meta’s VP of whatever-the-hell generative AI is supposed to be, told The Financial Times they’re going all in on artificial users. Complete with bios, profile pictures, and the ability to generate content. Because what Facebook really needed was more fake content, right?
The funny part? They already tried this. Remember “Liv,” the “proud Black queer momma of 2 & truth-teller”? That AI-generated persona had all the authenticity of my attempts at sobriety. Meta had to quietly delete these profiles after they failed to engage real users. Turns out people can smell synthetic bullshit from a mile away. Who knew?
But here’s where it gets interesting - and trust me, I needed an extra shot of Wild Turkey to process this part. While Meta’s fumbling around like a freshman at their first keg party, some actual scientists decided to do something useful with AI social personas.
They called it GovSim, probably because “AI Kindergarten” didn’t sound academic enough. The experiment was based on some Nobel-prize winning research by Elinor Ostrom, who proved that real humans can actually figure out how to share stuff without turning everything into a Mad Max scenario.
The researchers took 15 different language models - including the fancy ones from OpenAI, Google, and Anthropic - and gave them some simple sharing scenarios. Share a lake for fishing. Share land for grazing sheep. Share the right to pollute (because apparently that’s a thing now).
And guess what? These supposedly brilliant AI minds failed 43 out of 45 times. That’s worse than my success rate at remembering where I parked my car after last call.
The only silver lining? The smarter the AI, the better it did at playing nice with others. Though that’s like saying the expensive bourbon gives you a slightly less horrible hangover. You’re still going to regret it in the morning.
Here’s what keeps me up at night (besides the usual demons): Meta wants to fill their platform with these digital dimwits that can’t even grasp the basic concept of sharing. The same AI that would probably hoard all the fish in a virtual lake is going to be generating content next to pictures of your kid’s birthday party.
The whole thing reminds me of that time I tried to teach my cat to fetch. Sure, theoretically it could work, but in reality, you’re just setting yourself up for disappointment and scratches.
So what’s the takeaway here? While the lab coats are proving that AI can’t play nice in the sandbox, Meta’s planning to turn their platform into a digital daycare for artificial idiots. And they’re doing this voluntarily, which makes about as much sense as ordering a vodka martini at a whiskey bar.
Maybe instead of trying to make AI act more human, we should focus on making humans act more… well, human. But what do I know? I’m just a guy who thinks the best social network is still the local dive bar, where at least the artificial intelligence is limited to whatever’s in the jukebox.
Now if you’ll excuse me, my bottle of Buffalo Trace isn’t going to empty itself.
Signing off from the only place where the conversations are still genuine (even if the memories aren’t), Chinaski
P.S. The real kicker? While writing this, I tried to share my bourbon with my laptop. It didn’t go well. Still better cooperation than those AI models though.
Source: AI Social Media Users Are Not Always a Totally Dumb Idea