242 | David Krakauer on Complexity, Agency, and Information

Episode Summary

Executive Summary: Sean Carroll and David Krakauer argue that complexity science is a coherent field centered on teleonomic matter: systems that encode their environment, adapt, and build internal models. They trace its roots to 19th-century ideas about control, entropy, evolution, and computation, distinguish emergence from mere self-organization, and explore implications for individuality, noise, life, intelligence, AI, and social systems.

Main Topics: What counts as complexity (Priority: 5/5): Krakauer defends complexity science as a real field, but defines its core object narrowly: systems that carry internal information about the world and act purposefully. He excludes purely physical self-organizing phenomena like hurricanes or vortices as insufficient on their own. Historical foundations of complexity (Priority: 5/5): The discussion traces complexity to the industrial revolution and the convergence of thermodynamics, statistical mechanics, evolution, control theory, logic, and computation, highlighting Weaver, Simon, Maxwell, Darwin, Boole, Babbage, Wiener, Shannon, and Turing. Adaptation, agency, and teleonomy (Priority: 5/5): Krakauer argues that adaptation is not just an action principle or optimization problem; it is the acquisition and storage of information from the environment. Agency arises when systems follow policies and maintain internal models of future states. Emergence and symmetry breaking (Priority: 4/5): Emergence is framed as weak emergence: broken symmetries allow effective theories at macro scales that are dynamically sufficient without new physics. Krakauer rejects strong emergence and downward causation as physics, preferring pluralistic higher-level theories. Individuals, demons, and natural selection (Priority: 4/5): The conversation explores how individuals are emergent causal units discovered by information-theoretic criteria. Natural selection is recast as a distributed 'demon' that inspects and preserves information-bearing bits across generations, with complex environments acting as selectors. Noise, exploration, and persistence (Priority: 4/5): Noise is defended as constructive for complex systems because it enables exploration, mutation, stochastic resonance, and equilibrium selection. Complex systems often balance exploration with exploitation through controlled stochasticity. Intelligence, AI, and cognitive artifacts (Priority: 5/5): Krakauer treats intelligence as a universal problem-solving phenomenon and argues that large language models represent a shift from complementary tools to cognitively opaque artifacts that may exclude humans from deliberation. Complexity science and the social world (Priority: 4/5): SFI’s interdisciplinary model is positioned as especially suited for climate, pandemics, inequality, and other problems where biological, economic, political, and informational factors are inseparable.

Key Arguments: Complex systems should be defined by teleonomy: they encode the world internally and act with purpose, unlike passive self-organizing physical systems. The field of complexity has common intellectual roots in 19th-century developments in entropy, control, evolution, logic, and computation, making it a coherent paradigm rather than a loose collection of metaphors. Adaptation differs from action principles in physics: an adaptive system is not merely following a trajectory, it is building a map of its environment and using stored information to guide behavior. Self-organization alone is insufficient for complexity; a hurricane may be ordered, but it does not represent or model its environment. Emergence is best understood as symmetry breaking plus effective theories: higher-level descriptions are valid when they are dynamically sufficient and more useful than microscopic descriptions. Individuals are not always the smallest units; the right causal unit may be a genome, organism, colony, or social collective depending on which unit propagates adaptive information forward in time. Noise is not merely a nuisance; it enables exploration, mutation, stochastic resonance, and the discovery of long-lived adaptive states. Intelligence is broader than human cognition and should be studied as a general algorithmic capacity for problem solving, though modern AI raises issues of opacity and human exclusion. Large language models differ from traditional tools because their mechanisms are not internalizable in the way that maps, numbers, or abacuses can be internalized. Complexity science is especially valuable for real-world problems like pandemics, climate change, and inequality because these require integrated biological, social, and policy analysis.

Data Points: Santa Fe Institute founding horizon: 40 years - Carroll notes that complexity science and SFI have matured over roughly four decades. Historical consolidation window: 1840–1870 - Krakauer says the core ideas of complexity emerged in this period through Boole, Babbage, Maxwell, Wallace, Darwin, and others. Weaver paper year: 1948 - Krakauer identifies Warren Weaver’s 'Complexity in Science' as a foundational paper. Simon paper year: 1962 - Herbert Simon’s 'The Architecture of Complexity' is cited as a key systems-oriented contribution. Kolmogorov algorithmic complexity year: 1968 - Krakauer references Kolmogorov’s work on description length/incompressibility. Maxwell’s regulator paper year: 1868 - Krakauer cites Maxwell’s paper on governors and regulators as an origin of control theory. Large language model emergence: Recent months/years - Discussed as a contemporary turning point in cognitive artifacts and intelligence.

Pivotal Quotes: "We study teleonomic matter. We study matter with purpose." — David Krakauer: His concise definition of what makes complexity a distinct field. "It’s a map that’s being drawn of a landscape." — David Krakauer: Used to explain adaptation as internal encoding and learning, not a ball rolling downhill. "The parts reading off the states of the whole." — Sean Carroll (referencing Jessica Flack): Used in the discussion of downward causation and emergent organization.

Implications: Complexity science is presented as a genuine interdisciplinary framework for adaptive systems, not a loose buzzword. It may be crucial for understanding life, intelligence, AI, and societal crises where information, history, and purpose matter more than reduction alone.