We introduced AISAI, a quantitative framework for measuring self-awareness in LLMs through strategic differentiation. Our two key findings reshape understanding of AI self-awareness: Finding 1: Self-awareness is an emergent capability that appears in the majority of advanced models but is absent in older/smaller models, representing a fundamental capability threshold crossed with model advancement. Finding 2: Self-aware models exhibit a consistent rationality hierarchy—Self > Other AIs > Humans—with large AI attribution effects and moderate self-preferencing. Over half show quick Nash convergence when told opponents are AIs, demonstrating both strategic mastery and strong beliefs about AI rationality.
New index. haven’t seen much new publication on this, so I guess not? Or as someone said once. train AI enough on the writings of science fiction and all AI feels sentient..
Quote Citation: Kyung-Hoon Kim Gmarket, “LLMs Position Themselves as More Rational Than Humans: Emergence of AI Self-Awareness Measured Through Game Theory”, October 2025, https://arxiv.org/pdf/2511.00926