The optimization assumption pervades theoretical biology, cognitive science, economics, and artificial intelligence: organisms maximize fitness, brains minimize prediction error, agents maximize utility. These frameworks are locally powerful, and modern variants accommodate uncertainty, non-stationarity, and exploration bonuses within their formalism. Yet a deeper assumption remains unchallenged: that the state space, action repertoire, and model class over which the system optimizes are fixed, or at most expand according to pre-specified rules. Living systems violate this assumption routinely—they restructure their own representational and actionable spaces in ways not reducible to optimization over any pre-given space. We propose teleonomic ascent as a formal framework for this phenomenon: adaptive systems undergo directional, open-ended, irreversible movement through configuration space, driven by the continuous detection and resolution of suboptimalities, where each resolution reconfigures the space itself. We distinguish propagation (the mechanism of open-ended reconfiguration) from optimization (convergence within a fixed space) along three formal dimensions: convergence behavior, effective dimensionality dynamics, and reversibility. We introduce the Agential Efficiency Potential as a scalar measure of teleonomic efficiency and its gradient as the local direction of ascent, and we formalize the irreversibility conditions under which resolutions constitute genuine phase transitions. We further develop a taxonomy of ascent strategies—self-benefit optimization, problem-solving, collective alignment, and feedback-decoupled pseudo-altruism—showing that what psychology calls motivation styles and ethics calls moral orientations are formally distinct policy classes with measurable coupling signatures and predictable failure modes. The framework generates testable predictions across biological, cognitive, and collective scales, and suggests that a complete science of adaptive agency requires not only an account of within-configuration dynamics (the domain of existing optimization frameworks) but of between-configuration transitions—the domain of teleonomic ascent.
Topcular et al. (Wed,) studied this question.