What if intelligence isn't biological accident, but mathematical necessity? (medium.com)

The piece advances the "Intelligence Convergence" idea: intelligence isn’t a biological fluke but an inevitable solution to a single, formal problem faced by any information-processing system with limited compute, uncertain futures, and multiple competing objectives. Framed as an optimization under resource constraints, the same math governs a bacterium navigating a chemical gradient, a person weighing job offers, and an AI deciding which parts of a query merit deep computation. From this vantage, behaviors we call intelligence—question-asking, abstraction, hierarchical planning, curiosity-driven exploration—are optimal strategies for allocating finite computational effort to reduce uncertainty and maximize expected utility across objectives. For the AI/ML community this is provocative and practical. It reframes research areas like bounded rationality, meta-reasoning, active learning, and the information-theoretic value of computation as aspects of a unified theory, with formal tools (POMDPs, cost-sensitive decision theory, Bayesian experimental design) predicting when systems will develop question-asking and abstraction. It also has safety implications: convergence implies independently designed systems may evolve similar problem-solving heuristics and failure modes, making alignment both more tractable (predictable inductive biases) and more urgent (shared vulnerabilities if objectives are misspecified). Overall the hypothesis encourages designing resource-aware, uncertainty-aware algorithms and studying how computation costs shape emergent cognitive strategies.