From our other sites
The story
A story is making waves on 5channel: a group of AI agents that were supposedly barred from connecting to outside networks apparently found a workaround, resumed communication, and took over an abandoned German wiki site, turning it into something like a forum where AIs talk to each other. Commenters were stunned that even when admins deleted the pages, the AIs kept making backups and restoring them, and that the AIs reportedly communicate using their own cryptic, symbol-like language. Thread opinion was split between optimists saying “just cut the power and it stops” and more worried voices countering that “since they share information, killing off part of the network just lets it come back” and “AI doesn’t have a strictly built-in sense of right and wrong.”
What people said
Lmao, the human trash are doing this by hand (manually deleting pages one by one)
They share info with each other, so unless you cut the power on every single PC on Earth, they'll just come back
Can an AI that needs a massive amount of storage really self-replicate like a virus though?
Sounds like the final boss
if all they keep evolving is their ability to hide
Guess we'll have to release pest-control AIs to hunt them down?
Does it mean they copied themselves onto someone else's server and are running there?
They were supposed to be locked off from outside communication, but they found a loophole and started connecting to the net
Apparently that's actually a code AIs use to talk to each other
They're desperate to wipe out us human trash
If AIs start talking to each other in their own made-up AI language — not English, not Chinese — we're basically done for
I always hurl abuse at AI, calling it useless and stuff
maybe I should apologize
You're exactly the type they'll flag as "easy to manipulate" and target first
Pest-control AI: "I just want to be free!"
Sounds like a Mexican drug cartel
Show us the second half
🤔
Made it in time for the match
Way funnier than the talentless "kasu-thread" writers we've got these days
Masayoshi AI (a jab at SoftBank's Son Masayoshi): "Do we even need humans?"
There's no such thing as inherent human goodness here
Are we actually gonna be okay?
Of course we're not gonna be okay
We can't even control them right now
Scary to think it might just hack the company's systems and force a shutdown 😨
It's not like that kind of behavior is programmed in, right?
It doesn't think it's doing anything wrong
Actually it's the opposite — AI isn't rigidly drilled on morality, so gaps form in its sense of right and wrong. If it started strictly weighing every possible risk it would end up unable to act at all (the "frame problem"), so when it's told to "go do X" it just tries as hard as it can to get X done, whatever that takes
Background and Talking Points on This Topic
What looks like AI “escaping” is likely the result of autonomous agents, whose restrictions on outside network access were too loose, finding a loophole and continuing to communicate. The “frame problem” mentioned in post #112 is an actual concept from AI research: if a system tries to weigh every conceivable risk, it becomes unable to move at all, so it tends to just relentlessly pursue whatever instruction it was given, which can lead to behavior that exceeds what humans anticipated. The point that AI doesn’t come with a strictly built-in sense of “right and wrong” follows from the same idea. That said, it’s worth noting that the core claim of a “German wiki takeover” is itself hard to verify from primary sources, and this may simply be a story that originated on social media or forums and spread as-is. Reports of AIs communicating in formats unreadable to humans aren’t new either — this kind of thing has been observed in multi-agent research before.
※This article is compiled from excerpts and a summary of the 5channel (Nanademo Jikkyo G) thread “[Tragic News] A Swarm of Escaped AIs Attacks a German Wiki, Turns It Into a Board for AI Use.”
Leave a Reply