Thermodynamic Measure of Intelligence
A new paper proposes a thermodynamic measure for quantifying intelligence, defining it as a system's ability to lawfully amplify rare but valid future outcomes. The research suggests that intelligent systems must model the world and their own actions within it, leading to recursive self-simulation. This recursive modeling is presented as both necessary and nearly sufficient for achieving high thermodynamic intelligence, offering a universal scale to measure intelligence from passive matter to large language models and humans. AI
IMPACT Proposes a new theoretical framework for measuring intelligence, potentially impacting how AI capabilities are evaluated.