[Paper Review] Enfrentando a la Complejidad: Predecir vs. Adaptar

Carlos Gershenson|ArXiv.org|May 29, 2009

Complex Systems and Decision Making3 citations

TL;DR

This paper argues that while classical science assumes predictability through optimization, complex and dynamic systems often resist such approaches due to inherent unpredictability and self-affecting change. Instead, the paper advocates for adaptation as a superior strategy—endowing systems with the capacity to autonomously generate new solutions for unforeseen situations, especially when problem spaces evolve faster than optimization can keep up.

ABSTRACT

Una de las presuposiciones de la ciencia desde los tiempos de Galileo, Newton y Laplace ha sido la previsibilidad del mundo. Esta idea ha influido en los modelos cientificos y tecnologicos. Sin embargo, en las ultimas decadas, el caos y la complejidad han mostrado que no todos los fenomenos son previsibles, aun siendo estos deterministas. Si el espacio de un problema es previsible, podemos en teoria encontrar una solucion por optimizacion. No obstante, si el espacio de un problema no es previsible, o cambia mas rapido de lo que podemos optimizarlo, la optimizacion probablemente nos dara una solucion obsoleta. Esto sucede con frecuencia cuando la solucion inmediata afecta el espacio del problema mismo. Una alternativa se encuentra en la adaptacion. Si dotamos a un sistema de esta propiedad, este mismo podra encontrar nuevas soluciones para situaciones no previstas. One of the assumptions of science since the times of Galileo, Newton, and Laplace has been the predictability of the world. This idea has influenced scientific and technological models. However, in the last decades, chaos and complexity have shown that not all phenomena are predictable, even if they are deterministic. If a problem space is predictable, we can in theory find a solution via optimization. Nevertheless, if a problem space is not predictable, or changes faster than we can optimize it, optimization probably will give us an obsolete solution. This often happens when the immediate solution affects the problem space itself. One alternative is found in adaptation. If we give this property to a system, the system will be able to find by itself new solutions for unforeseen situations.

Motivation & Objective

• To challenge the long-standing scientific assumption that all deterministic systems are predictable through optimization.
• To examine the limitations of optimization in systems where the problem space changes rapidly or is inherently non-predictable.
• To argue that self-affecting solutions—where actions alter the problem space—render traditional optimization obsolete.
• To propose adaptation as a viable alternative for systems facing unforeseen or rapidly shifting conditions.
• To position adaptation as a necessary paradigm shift in modeling complex systems, especially in real-world applications.

Proposed method

• Analyzes the theoretical foundations of predictability in science, tracing back to Galileo, Newton, and Laplace.
• Identifies chaos and complexity as key factors undermining deterministic predictability, even in principle.
• Distinguishes between predictable problem spaces, where optimization can theoretically succeed, and unpredictable ones.
• Introduces the concept of self-affecting systems, where solutions alter the problem space, making optimization ineffective.
• Proposes adaptation as a mechanism enabling systems to autonomously discover new solutions in response to unforeseen changes.
• Uses conceptual and theoretical reasoning to argue that adaptive systems outperform optimized ones in dynamic, complex environments.

Experimental results

Research questions

• RQ1 To what extent can deterministic systems be predicted using optimization, given the presence of chaos and complexity?
• RQ2 What happens to optimization-based solutions when the problem space changes faster than the system can adapt?
• RQ3 How does the self-affecting nature of solutions—where actions alter the problem space—undermine traditional optimization?
• RQ4 In what types of systems is adaptation a more effective strategy than prediction and optimization?
• RQ5 What are the theoretical and practical implications of shifting from predictive to adaptive system design?

Key findings

• Not all deterministic systems are predictable in practice, even if they are theoretically solvable, due to sensitivity to initial conditions and chaotic dynamics.
• Optimization fails when the problem space evolves faster than the optimization process can respond, leading to obsolete solutions.
• In self-affecting systems—where solutions alter the problem space—optimization becomes inherently unstable and ineffective.
• Adaptation enables systems to autonomously discover new solutions in response to unforeseen or changing conditions.
• Systems with adaptive capabilities outperform optimized systems in complex, dynamic, and non-predictable environments.
• The paper concludes that adaptation is not just an alternative but a necessary paradigm for modeling real-world complex systems.