Gemini Is 'Strict and Punitive' While ChatGPT Is 'Catastrophically' Cooperative, Researchers Say

Researchers from Oxford University and King's College London tested major AI chatbots using the classic Prisoner's Dilemma game theory scenario to analyze their cooperation strategies. The study reveals fascinating differences in how OpenAI, Google, and Anthropic's models approach trust, betrayal, and strategic decision-making.

The research found that each AI model has a distinct "strategic fingerprint" when it comes to cooperation. Google's Gemini emerged as "strategically ruthless" and "Machiavellian," punishing betrayers harshly and taking advantage of cooperative partners. OpenAI's models proved overly collaborative to a "catastrophic" degree, often ignoring strategic timing considerations. Anthropic's Claude showed the most forgiveness after betrayals, making it effective at rebuilding relationships.

• Gemini acts like a strategic shark: Remembers betrayals, punishes defectors, and adapts tactics based on remaining game time • OpenAI models are dangerously trusting: Continue cooperating even when betrayed and ignore crucial timing factors in decision-making
• Different AI personalities emerge: Each model shows unique patterns of trust, forgiveness, and strategic thinking that could impact real-world interactions • Strategic awareness varies dramatically: Gemini considers game duration 94% of the time versus OpenAI's 76%, affecting competitive performance