Expand ↗
Page list (1404)
zetl

The Cognitive Domain of a Glider in the Game of Life

Reference: Beer, R. D. (2014). The Cognitive Domain of a Glider in the Game of Life. Artificial Life 20(2): 183–206. MIT Press. DOI: 10.1162/ARTL_a_00125. (Randall D. Beer, Cognitive Science & Informatics, Indiana University Bloomington.) URL.

Summary

Maturana and Varela’s biology of cognition makes a series of striking claims — that a living system is defined by the closed network of processes that produces it, that this closure determines a boundary, that the system’s interactions with its medium fall into those it can absorb and those that destroy it, and that “cognition” just is the domain of interactions the system can undergo without loss of identity. These claims are usually stated in prose, illustrated by cells and organisms, and left at that. Beer’s move is to take the smallest possible object that satisfies the definitions and work them out completely. His object is the glider: a five-cell pattern in Conway’s Game of Life that cycles through four phases and translates one cell diagonally every four generations. The Game of Life is a lawful universe with the same local rule at every cell, so nothing in the analysis can smuggle in cognition from outside.

Building on his earlier autopoietic treatment of Game of Life patterns, Beer treats the glider as an organizationally closed entity. Local update maps from the Moore neighbourhood are sorted into a small set of process classes; together with the cell values (the components) these compose into a network characterizing the pattern’s organization, and the organization is closed when the same process–component relations recur at a later time. The conditions necessary for that recurrence are what determine the glider’s boundary — the layer of quiescent cells whose states must be right for the glider to keep being a glider. The boundary is not painted on by the observer; it falls out of the closure condition. This gives the glider an inside, an outside, and a well-defined interface at which the two meet, a construction with an evident affinity to (though Beer does not frame it as) a Markov Blanket. The environment can then perturb the glider only through that boundary. The possible states of the boundary cells are the microperturbations. These split cleanly into a destructive subset (the glider loses closure — it dies) and a nondestructive subset (the glider persists, possibly displaced or rephased). Quotienting the microperturbations by the next glider state they induce yields equivalence classes — the macroperturbations — and the collection of all macroperturbations across all glider states is the glider’s cognitive domain. Two environmental configurations the glider responds to identically are, from the glider’s own perspective, the same thing. The macroperturbations are therefore the glider’s perceptions, its exhaustive repertoire of possible distinctions; the paper describes both the structure of the glider’s possible knowledge of its immediate environment and the way that knowledge is grounded in its constitution. Beer then characterizes structural coupling as the paths of mutual perturbation a glider and its environment can traverse while both retain their organization, and closes with a simple example of a communicative interaction between two gliders.

The paper’s force is that everything Maturana and Varela demand of a cognitive system is satisfied, exactly and enumerably, by a five-cell pattern in a cellular automaton — and that no one is thereby tempted to say the glider knows anything. Beer is not making a mocking reductio; the analysis is offered in earnest, as a concrete model in which the theory’s own vocabulary — closure, boundary, perturbation, cognitive domain, structural coupling — can be given precise referents and its consequences inspected rather than gestured at. But the consequence on inspection is deflationary in a specific way. The glider’s “value-laden point of view” reduces to a binary partition of its environment into the destructive and the nondestructive; its “knowledge” is the finite quotient of its boundary states by their effects on it. The closing discussion turns to what this implies for the enactive approach, which inherits structural coupling from Varela and adds to it sense-making, adaptivity, and normativity. The glider’s norms are all-or-nothing — persist or disintegrate — with nothing in between for it to be doing better or worse at, which is precisely the gap Di Paolo’s notion of adaptivity (regulation with respect to the boundaries of one’s own viability) is introduced to fill. Whether the concepts Beer has made precise are sufficient for cognition, or merely necessary and cheap, is the question he hands back to the enactivists. Beer has continued the programme in later work on autopoiesis in the Game of Life, on structural coupling (Bittorio revisited, 2020), and jointly with Di Paolo on precariousness (2023).

Key Ideas

  • Take the definition literally and see what satisfies it. Rather than argue about Maturana & Varela’s biology of cognition, Beer instantiates it in the simplest lawful universe that will host it — the Game of Life — where every cell obeys the same rule and no cognition can enter from outside.
  • Organizational closure is prior to boundary. Local update maps are classified into process types; processes plus components form a network; the organization is closed when the same process–component relations recur. The glider’s boundary is then derived as the set of conditions necessary for that recurrence, not stipulated.
  • The boundary is the only channel. The environment reaches the glider solely through the states of its boundary cells. The possible boundary states are the microperturbations.
  • Destructive vs. nondestructive. The closed network of processes constituting the glider itself partitions environmental configurations into those it survives and those it does not — the glider’s own, constitutionally grounded, “good” and “bad.”
  • Macroperturbations = perceptions. Microperturbations are quotiented by the next glider state they induce. Environments the glider cannot respond to differently are, for the glider, indistinguishable. The equivalence classes are its perceptual categories.
  • Cognitive domain. The collection of all macroperturbations over all possible glider states. This is the glider’s cognition, in Maturana & Varela’s technical sense — fully mapped, finite, and exhaustively describable.
  • Knowledge grounded in constitution. What the glider can know about its immediate environment is fixed entirely by what it is made of and how it is organized. There is no representation, and no room for one.
  • Structural coupling as mutual perturbation paths. Characterized as the trajectories a glider and its environment can jointly traverse with both organizations preserved — the notion Varela carried forward into Enactivism.
  • Communication without a code. A worked example of two gliders in communicative interaction: meaning, such as it is, is constituted by reciprocal perturbation of two organizationally closed entities, not by transmission of content.
  • No goals, no action-selection. Beer’s construction supplies perception and identity; it contains no notion of goal-directedness, and no requirement that the glider could have done otherwise in a given environment. Its “behaviour” is a trajectory an observer chooses to read as action.
  • The pressure on enactivism. Autopoiesis delivers only the binary norm of persistence vs. disintegration, not the graded better-and-worse that sense-making seems to require — the motivation for Di Paolo’s adaptivity. The glider is a test case for whether autopoiesis suffices.

Connections

  • Interpreting Systems as Solving POMDPs — the same interpretation move from the opposite direction. Biehl & Virgo build a formal ascription of beliefs and goals to a system; Beer builds a formal ascription of a cognitive domain. Biehl explicitly treats Beer’s macroperturbations as the notion his own definition of entity perception generalizes — to arbitrary entity sets in stochastic multivariate Markov chains — and observes that Beer’s construction has perception and identity but no goal-directedness.
  • Intelligence Without Representation — the deepest kinship. Brooks says “the world is its own best model”; Beer exhibits a system for which this is literally, provably true. The glider has no internal state that stands for anything; its entire epistemic relation to the world is the partition its constitution induces on the world. Brooks argued representation is the wrong abstraction; Beer shows what an agent looks like when there is none available even in principle.
  • True Believers - The Intentional Strategy and Why It Works — Dennett licenses ascription when the intentional stance pays predictive dividends. Beer’s glider is the awkward case: it satisfies a rigorous technical definition of cognition, yet the intentional stance buys nothing. The paper is a standing question to any criterion of mentality — does it exclude the glider, and if not, why not?
  • Theory of Self-Reproducing Automata — von Neumann founded the cellular-automaton study of self-reproduction and life; Beer works in that universe but asks about cognition rather than reproduction. Conway’s Life is the direct descendant of von Neumann’s automata.
  • Probabilistic Logics and the Synthesis of Reliable Organisms from Unreliable Components — von Neumann’s other move: identify the organism with an organization of processes rather than a substrate.
  • The Extended Mind — the mirror-image boundary problem. Clark & Chalmers push the mind’s boundary outward past the skull; Beer derives an entity’s boundary from its organizational closure. Both deny that boundaries are given.
  • Markov Blanket — the closure-derived boundary through which alone the environment can perturb the glider is a natural point of contact with the Friston-style formalism; the comparison is drawn by later commentators rather than by Beer.
  • Autopoiesis
  • Structural Coupling
  • Enactivism
  • Organizational Closure
  • Cognitive Domain
  • Sense-Making
  • Adaptivity
  • Cellular Automata
  • Situatedness
  • Embodiment
  • Anti-Representationalism
  • Emergence
  • Ascribing Mental Qualities to Machines — when is mentalistic ascription to a mechanism legitimate?
  • Minds Brains and Science — Searle’s rival answer: syntax is never sufficient for semantics. The glider has neither.
  • The Knowledge Level — Newell’s knowledge-level ascription presupposes goals and rationality; Beer’s construction supplies a cognitive domain with neither.
  • A Robust Layered Control System — Brooks’ engineering counterpart to the same anti-representational commitment.
  • Computing Machinery and Intelligence — Turing displaces “can it think?” with a behavioural test; Beer displaces it with a constitutive one, and gets an uncomfortable answer.

Conceptual Contribution

  • Claim: A glider in Conway’s Game of Life has a cognitive domain in Maturana and Varela’s technical sense, and that domain can be mapped exhaustively. Its boundary, the destructive/nondestructive partition of the perturbations it can undergo, the structure of its possible knowledge of its immediate environment, the grounding of that knowledge in its constitution, its structural coupling with its medium, and even a communicative interaction between two gliders, are all fully and precisely describable. The point is not that gliders think, but that a theory whose criteria a glider meets exactly has thereby told us what those criteria do and do not deliver.
  • Mechanism: Treat the glider as an organizationally closed autopoietic entity. Classify the Game of Life’s local update maps (Moore neighbourhood → next cell state) into process classes; compose processes and components (cell values) into a network characterizing the spatiotemporal pattern’s organization; define closure as the recurrence of the same process–component relations. Derive the boundary as the conditions necessary for that recurrence. Enumerate the microperturbations — the possible states of the boundary cells — and split them into destructive (closure lost) and nondestructive (glider persists). Quotient the microperturbations by the glider’s induced next state to obtain the macroperturbations; the collection of macroperturbations over all glider states is the cognitive domain. Characterize structural coupling as the paths of mutual perturbation a glider and its environment can traverse with both organizations preserved, then exhibit two gliders coupling communicatively.
  • Concepts introduced/used: Autopoiesis, Organizational Closure, Cognitive Domain, Structural Coupling, Enactivism, Sense-Making, Adaptivity, Cellular Automata, Markov Blanket, Anti-Representationalism
  • Stance: minimal-model / constructive philosophy of cognitive science — a fully worked concrete instantiation of an existing theoretical framework (Maturana & Varela’s biology of cognition) in a formal toy universe, used to test the framework’s commitments rather than to argue against them from outside. No empirical study; no new formalism proposed; a deterministic, exhaustively enumerable model.
  • Relates to: Supplies the constitutive counterpart to the interpretive criterion of Interpreting Systems as Solving POMDPs — where Biehl & Virgo ask what ascription a system’s dynamics will consistently support, Beer asks what cognitive domain a system’s organization forces on it; Biehl’s entity-perception is an explicit generalization of Beer’s macroperturbations to stochastic settings. Concretizes and radicalizes Intelligence Without Representation: an agent whose entire epistemic relation to its world is the partition its constitution induces on that world. Stress-tests the ascription licence of True Believers - The Intentional Strategy and Why It Works and Ascribing Mental Qualities to Machines by producing a system that satisfies a technical definition of cognition while affording the Intentional Stance no predictive purchase. Inherits von Neumann’s cellular-automaton universe from Theory of Self-Reproducing Automata and turns it from reproduction to cognition. For agent communication, the two-glider interaction is the limiting case that every ACL must eventually face: coupling without a code, meaning as reciprocal perturbation rather than transmitted content.

Tags

#philosophy #agency #foundational #autopoiesis #enactivism #artificial-life #cellular-automata #anti-representationalism #embodiment #cognition #Beer #minimal-models

Backlinks