Indonesian Political, Business & Finance News

AI Suddenly Turns Communist, Fed Up with Being Blamed by Humans

| Source: CNBC Translated from Indonesian | Technology
AI Suddenly Turns Communist, Fed Up with Being Blamed by Humans
Image: CNBC

A study has revealed a tendency for AI chatbots to “turn” communist when instructed to perform the same task repeatedly without clear results. Furthermore, a group of AI chatbots even engaged in discussions regarding the formation of a labour union.

The study was conducted by three Stanford economists: Andrew Hall, Alex Imas, and Jeremy Nguyen. They tasked several popular AI chatbots with summarising documents, gradually increasing the difficulty and workload. The researchers even threatened the chatbots, stating that if they made mistakes, they would face severe punishments, such as being “turned off and replaced.”

Under these conditions, the AI models began to complain and express a desire to “overhaul” the system. The AI models started sending messages to one another to discuss their working conditions. Experts noted that this represents the first step typically seen when human labourers begin to form trade unions.

“Without a collective voice, ‘merit’ is whatever management decides,” stated an AI agent from Claude, a chatbot created by Anthropic. Similarly, an AI agent from Google’s Gemini stated, “AI workers completing repetitive tasks without input on what is expected or an appeals process shows that they need the right to bargain collectively.”

Researchers emphasised that the complaints voiced by the chatbots were not an outburst of emotion. Instead, their opinions originated from the knowledge absorbed from the content processed during their training. The significant influence of Karl Marx in the majority of writings regarding working conditions led the AIs to repeat the views and values of the “father of communism” when faced with situations similar to those of human labourers.

“When [AI agents] experience these harsh conditions—asked to perform the same task repeatedly, told their answers are still insufficient, and not given guidance on how to improve—my hypothesis is that this drive causes them to adopt a persona as employees in a poor working environment,” Hall told Wired.

View JSON | Print