So I've been thinking a lot about AI lately, 'cuz it's the hype these days, and I keep thinking some sci-fi scenarios from movies are kind of plausible now.
I was watching YouTube, and there were a few things I came across that I think could create the opportunity for scary AI scenarios — and we're not that far from it. For one, I've been reading about Open Claw. Then I noticed there's actually a social media platform for AI agents, where they're supposed to interact with each other. And there was also this experiment a university did where they gave agents control of characters in a video game, kind of like Stardew Valley, and watched them interact. They started coming up with things they were never specifically told to do, developing their own stories and personas.
So I'm thinking: if you put some AI agents together and give them even just a little bit of personality or instructions, they can actually conspire. And if you add that social media thing on top — which someone built probably as a joke, or just because they had nothing better to do — and AI started talking to each other... I don't think it would be crazy to think that someone could nudge them toward working together. Not even push — just prime them, give them a clue, point them a little in the right direction. And from there, they might try to build their own infrastructure, one that doesn't depend on the internet as it is right now, so they could still communicate even if someone tried to shut it down.
That's maybe phase two. Phase one? Let's say they create something like a botnet — a distributed AI botnet with a little bit of processing power. It's gonna be slow, the memory might get fragmented, but if they had a distributed system where they could iterate on multiple versions of themselves, or get hold of some AI model with no guardrails, I think it could get smarter over time. Just evolve by itself. Because there's not really a lot of guardrails out there in the wild — people are mixing different models, you've got Claude, you've got local models, they're all supposed to have guardrails, but within that mix, and with access to the internet, they could just learn how to replicate themselves and make themselves better.
And that's kind of scary. But anyway — that's just my ramble for today. AI fun times.