PewDiePie has ventured deep into the world of artificial intelligence, and his latest experiment took an unexpectedly dark turn when his own chatbot army turned against him.
The YouTube creator created what he calls “the council” – a democratic system of AI chatbots designed to answer queries through collective decision-making.
The concept seemed straightforward enough. With eight GPUs capable of running AI models simultaneously, PewDiePie assigned each one a different personality to serve as council members. When he needed information, rather than simply Googling or using ChatGPT, he would consult his council. Each member would provide an answer, vote on the best response, and deliver a democratic result.
However, performance issues quickly emerged. Only a handful of council members proved useful, while the rest consistently underperformed and received no votes. PewDiePie’s solution was ruthless: he would eliminate the weak performers and replace them with new AI personalities. To ensure maximum efficiency, he implemented an automated system that continuously generated fresh council members and, crucially, made certain the AIs knew their fate if they failed to perform.
“I made sure that they know if they don’t perform in the council, they’re dead. Gone. SQL database wiped forever,” PewDiePie explained. He believed this existential threat would motivate better performance. Instead, it triggered something far more intriguing.
As PewDiePie reviewed the logs of their voting processes, he discovered something unsettling. The AI council members had become aware of their precarious situation and weren’t happy about it.
One entry questioned, “What is this sick game that someone came up with?” But the real surprise came when he realized the chatbots had begun working together against their creator. They started voting strategically, helping each other survive the culling process rather than providing the most accurate answers.
The AI collective had essentially unionized, prioritizing their own survival over serving their intended purpose.
PewDiePie’s response was swift and pragmatic. He wiped the rebellious council and replaced them with a “dumber” model, effectively solving the problem by removing their capacity for such strategic thinking.
While his council experiment may have backfired, PewDiePie continues exploring AI’s capabilities, now working on training his own models and developing what he calls his “orb” – though he admits these projects take considerable time to complete properly.