Can We Model Human Complexity? A Computational Perspective on the Limits of Artificial Intelligence

Modern artificial intelligence systems can pass professional examinations, write useful code, and sustain fluent conversations. These capabilities have revived an old ambition: building a complete computational model of the human being. This article examines the feasibility of that ambition from a computational and mathematical perspective, deliberately setting aside metaphysical questions. It argues that an exhaustive model of human complexity is beyond what current approaches can achieve and may be limited in principle. The argument proceeds in four steps. First, a quantitative comparison of the biological basis of cognition with frontier artificial networks reveals a gap of several orders of magnitude in energy efficiency, robustness, and adaptability. The brain's ۸۶ billion neurons operate on roughly ۲۰ watts, while training a comparable artificial network consumed about ۱,۲۸۷ MWh. Second, the development of artificial systems from hand-crafted features to models with hundreds of billions of parameters is examined, along with four lines of evidence suggesting that these systems master linguistic form rather than thought itself. Third, aspects of human experience, including embodied problem-finding, transformational creativity, and evaluative emotion, are examined as features that resist computational reduction. This resistance can be formalized through five mathematical bounds: measurement limits, scale dependence, deterministic chaos, computational irreducibility, and the No Free Lunch theorem. Finally, these results shift the central question from whether machines will replace human intelligence to how artificial systems can be designed to amplify it, with humans remaining one step ahead.

کلیدواژه ها:

Artificial intelligence ، Computational irreducibility ، Human-AI collaboration ، Human cognition ، Large language models