After Downward Causation

Abstract Entities and Material Implications

Reductionism asserts that phenomena observed at higher levels of complexity can be fully explained by the interactions and properties of their most basic constituents. However, there is a compelling argument for the existence of top-down causation, positing that this is a critical component of how complex systems operate.

Take, for example, a digital computer. At its most fundamental level, it is a collection of electronic components like transistors and logic gates, through which electrons flow. Yet, without the presence of software, this sophisticated hardware is rendered purposeless. It is the software, or program, that activates the hardware and dictates its function—be it a word processor, a music player, or a painting application. This is the crux of top-down causation: the same hardware yields vastly different outcomes based on the abstract instructions provided by the software.

Consider the ontological status of computer programs and books in electronic format, which are abstract entities, not merely physical objects. Their essence does not reside in the physical medium—like an eBook reader—but in the abstract content it conveys. This content can manifest in various forms, demonstrating the concept of multiple realizations.

Such abstract entities possess causal power. The rules of games like chess or football are prime examples. They are not tangible entities, yet they govern the play and strategies within these games. These rules, or abstract logics, dictate the physical movements on a chessboard or a football field, again underscoring top-down causation.

When a chess player moves a rook, they are not altering the physical properties of the piece itself but are enacting a rule that exists independently of the physical board and pieces. Similarly, when a computer executes a program, the underlying logic gates and circuits don't change in their physical makeup; they follow the paths and states dictated by the abstract software loaded onto the computer.

Compilers play a pivotal role in bridging the gap between high-level abstract programming languages and the binary machine code understood by a computer's hardware. They take the abstract logic written by programmers and transform it into instructions that can control the hardware at the most fundamental level. This transformation is a direct example of top-down causation: the abstract dictates the behavior of the concrete.

In this view, the compiler is analogous to the interpretation of the rules of chess in a player's mind. The logic of the game exists independently of the physical board and is implemented through the actions of the players, just as a compiler takes the logic of the software and translates it into actions that the hardware can perform.

This exemplifies how higher-level systems impose order and function upon lower-level components through a top-down process. In the context of chess and computers, it becomes clear that while the physical entities are necessary for the game to be played or the program to run, they are not sufficient. The true drivers of action and change are the abstract entities—the rules of the game, the logic of the software—that govern the physical components.

This embraces a dualist standpoint, especially when considering the human brain and mind. In this view, the brain is the hardware, while the mind—or thoughts—are the software. Thoughts are not physical entities; they are abstract and have a hierarchical structure, much like the levels of programming languages within a computer. This structure seems to direct the physical activity of the brain, a top-down logic that challenges the reductionist perspective that physical states are the prime determinants of higher-level phenomena.

Supervenience suggests that knowing every detail about a system's lower-level states allows for the prediction of its higher-level states. This aligns with reductionist thought but does not account for how these lower-level states arise.

The resistance to top-down causation among some scientists may stem from the fear of inviting supernatural explanations into the realm of science. However, the realm of ideas and abstract logic is not supernatural but an intrinsic part of the natural world, affecting it without violating the fundamental laws of physics.

Top-down causation presents a significant shift from the traditional reductionist view, encouraging a reevaluation of how we understand the interconnectedness of systems at all levels even when each plane consists of differing logic. It accommodates a broader perspective that acknowledges the influence of abstract entities on the physical world, a viewpoint that could profoundly impact various fields, from computer science to moral philosophy and cognitive science.

1. Premise (Multiple Realizability)

Phenomena such as software, games, or mental states can be instantiated (realized) through a variety of distinct physical configurations.

• Example: The same chess rules can be enacted on a wooden board, a silicon-based computer, or even a purely mental simulation.

2. Premise (Abstract Structures Organize Physical Systems)

These higher-level phenomena are governed by abstract structures (e.g., rules, protocols, logic), which remain consistent across multiple physical instantiations.

• Example: A text document’s content remains the same whether it is printed on paper or displayed on an e-reader. The abstract content transcends the medium.

3. Premise (Causal Influence of Abstract Entities)

The organization or structure imposed by these abstract rules affects how the underlying physical system behaves, even though the abstract rules themselves are not reducible to any specific set of physical properties.

• Example: Chess rules dictate legal moves in a game, thereby constraining and guiding the physical motion of chess pieces.

4. Premise (Observing Contrasts in Outcomes)

When the same underlying physical system (hardware or substrate) is subjected to different sets of abstract rules or instructions, it produces measurably different outcomes—outcomes that cannot be fully accounted for by purely bottom-up descriptions of physical properties alone.

• Example: A computer with identical hardware components behaves differently depending on whether it is running a word processor or a video game.

5. Premise (Higher-Level Constraints Cannot Be Fully Reduced)

Even if all lower-level states are known, the origin and functional role of those states often depend on top-level constraints—rules, software, or mental states—that define how the lower levels are configured or orchestrated.

• Example: Knowing every detail about a CPU’s gates does not, by itself, explain the purpose of the running software or how it came to be arranged in that particular functional pattern.

6. Conclusion (Top-Down Causation)

Therefore, these abstract organizational principles—existing at a higher level of description—exert a genuine causal influence on physical states. This top-down causation shapes how the lower-level elements behave and gives rise to system-wide functions that cannot be captured by reductionist analysis alone.

A computational theory that avoids the pitfalls of both downward causation and reductionism—one that isn't strictly bottom-up or top-down—would likely be bidirectional, recursive, and interdependent. Instead of assuming a strict hierarchy where either micro-level or macro-level structures dominate, this theory would treat computation as an emergent, self-referential process where different scales influence each other dynamically.

1. Reciprocal Constraints – Instead of higher levels simply constraining lower levels (downward causation) or lower levels fully determining higher levels (reductionism), the two levels interact through feedback loops. Constraints emerge and dissolve dynamically rather than being fixed.

2. Heterarchical Computation – Rather than a strict hierarchy, computation would function through a heterarchical network where different levels of abstraction interact in multiple directions, like in neural networks or biological systems.

3. Multi-Scale Interaction – Computation operates at different levels simultaneously, where patterns at a higher level can influence lower-level operations without overriding them. This could resemble fractal structures or scale-free networks where patterns at different levels co-evolve.

4. Contextual Modulation – Information processing is modulated by both local and global contexts. Just as in language, where meaning emerges from words, sentences, and discourse together, computational processes might adjust dynamically based on local and system-wide states.

5. Computation as a Field, Not a Sequence – Instead of thinking about computation as a sequence of instructions moving in one direction (bottom-up or top-down), it could be conceived as a field or a mesh where causal influence propagates in multiple ways simultaneously.

6. Circular and Self-Modifying Computation – Computation would be reflexive, meaning that system-wide changes could lead to alterations in the computational rules themselves. This could resemble self-organizing systems, where a system modifies its own underlying architecture based on emerging patterns.

7. Quantum or Non-Classical Computation – If we take inspiration from quantum computing, where states exist in superposition and interact non-linearly, then computation itself might not follow a strict directional hierarchy at all but instead unfold through entanglement-like structures.

• Interactive Turing Machines that modify themselves based on external feedback loops.

• Neural Computation with Lateral Inhibition, where not only bottom-up and top-down flows exist but also sideway interactions shaping emergent patterns.

• Graph-Based or Hypergraph-Based Computation, where nodes influence each other dynamically without a strict dependency tree.

• Agent-Based Computational Ecosystems, where no single level dominates, but emergent behavior arises from the interactions of many computational entities.

It moves away from mechanistic computation (strict sequences and layers) and toward process-based computation, where systems are always in flux, co-determining their own rules. This would align with concepts from enactivism, process philosophy, and cybernetics, where cognition, information, and computation emerge through reciprocal influences rather than linear causation.

If we extend this to a computational paradigm, then computation itself must:

• Co-emerge with interaction rather than being pre-programmed.

• Link interface conditions dynamically instead of executing rigid, predefined rules.

• Develop self-modifying affordances that shift with use and context.

This shifts the question from how to encode rules to how to sustain evolving interfaces that maintain coherence without static encoding—a problem cybernetics has always flirted with but never fully resolved.

Process philosophy (Whitehead, Bergson, etc.) argues that reality is not a collection of fixed objects but an ongoing process of becoming. If we merge this with the Ruliad, we get:

• Reality is not a simulation that was pre-designed but a continually unfolding computation.

• There is no fixed "simulated world" distinct from a "base reality"—rather, every reality is just a locally stabilized slice of a larger computational unfolding.

• Instead of thinking of ourselves as being inside a simulation, we should think of simulation as an emergent organizational structure—the world appears stable because we select and reinforce certain computational pathways.

Instead of an external programmer creating a simulated universe, what we call a "simulation" is just a coherent region of computational evolution within the larger Ruliad. Our experience of reality is a local ordering of computational processes, not an artificial construct running inside something else.

• If by "simulation," we mean a pre-programmed reality created by an external being, then no, we are not in a simulation in the traditional sense.

• But if we define simulation as a dynamically evolving computational process that stabilizes through interaction, then yes—we exist within a self-organizing, self-simulating reality.

• The distinction between "real" and "simulated" dissolves—reality is simply one computationally stable way of traversing the Ruliad.

The rules that govern language, computation, and chemistry may seem different in their implementation, but they all share the same formal structure:

• Abstract programs (computational logic) → Executed on physical hardware (transistors, silicon, electrons).

• Linguistic rules (grammar, syntax, semantics) → Embodied in speech, text, neural activity.

• Chemical laws (valence, molecular bonding, reaction kinetics) → Manifest in physical interactions of atoms and molecules.

Each of these systems has rules that are multiply realizable:

• A program can run on different types of computers.

• A sentence can be spoken, written, or thought.

• A chemical reaction can occur in different environments.

Yet, none of these rules exist apart from their implementation—they do not float in a Platonic realm; rather, they persist as constraints on the underlying system.

This suggests that rules emerge as stable structures within a computationally rich system—a Ruliad-like space of all possible rule-based interactions.

A key question emerges: Are algorithms, rules, or abstract entities real if they are multiply realizable?

• A chess rule does not "exist" without an interpreter (a player or a computer).

• A program does not "exist" without execution.

• A chemical reaction does not "exist" apart from the material context in which it unfolds.

Yet, they persist as stable informational constraints, which means:

• They are real, but their reality is relational, not absolute.

• They emerge from interaction rather than preexisting as separate Platonic entities.

• They shape and are shaped by the medium through which they propagate.

This suggests that rules are computationally persistent structures that co-emerge with the systems that enact them. They are not imposed from an external realm; they are entangled with material implementation.

The top-down view described assumes:

1. The abstract realm (software, rules, thought) exists independently of the physical realm.

2. The abstract imposes order on the physical.

3. The physical merely executes instructions.

This is partially true but incomplete because rules do not exist apart from the systems that instantiate them. In a truly cybernetic, enactivist, or Ruliad-based perspective:

• Rules evolve and change based on the interactions of the systems they govern.

• The physical system can reshape the abstract structures governing it (e.g., neural plasticity reshaping cognition, physical constraints reshaping software architecture).

• Computation is not just an execution of abstract laws—it is a bidirectional, recursive negotiation between abstract possibilities and material realizations.

Example: Chess Rules Are Not Immutable

• Chess began as a set of historical rule constraints, but the game changed over time.

• The evolution of chess openings and strategies is not dictated purely by abstract rules—it emerges from historical, cultural, and competitive interactions.

• In AI-driven chess, new strategies have emerged that even human grandmasters never conceived—suggesting a bottom-up influence on the “abstract” level.

Example: Chemistry Is Not Purely Rule-Based

• The laws of chemistry do not exist in a vacuum; they emerge as stable patterns within quantum physics.

• Certain molecules self-organize into life-like systems that then impose new rule constraints—leading to emergent phenomena like metabolism, self-replication, and ultimately, intelligence.

Example: Thermodynamics

• Thermodynamics emerges from statistical mechanics at a higher level of description. Temperature and pressure are not new fundamental forces; they are aggregate, emergent properties of vast numbers of particles.

• Yet, temperature and pressure “cause” real effects (like expansion of a gas or heat flow). We do not call temperature or pressure a new force; we recognize them as emergent properties shaped by lower-level physics.

Example: Biological Systems

• Genes impose constraints on protein synthesis; cellular structures impose constraints on molecular interactions. However, we do not call genetic information a “fundamental force.” It is an informational constraint that orchestrates how existing molecules (governed by electromagnetic and other fundamental forces) come together into functional forms.

Premise 1 (Bidirectional & Recursive): Brain networks exhibit bidirectional and recursive connections, demonstrating computation is not unidirectional but involves feedback and iterative processing (Sporns, 2011).

• Paper: Sporns, O. (2011). Networks of the brain. MIT press.

• Relevance: This book provides a comprehensive overview of brain networks, emphasizing their bidirectional connections and recursive organization, moving beyond simple hierarchical views.

Premise 2 (Reciprocal Constraints): Constraints in cognitive systems are dynamically and reciprocally influential, emerging and dissolving through system interactions, not fixed or unidirectional (Kelso, 1995).

• Paper: Kelso, J. A. S. (1995). Dynamic patterns: The self-organization of brain and behavior. MIT press.

• Relevance: A foundational text in dynamical systems theory applied to behavior, demonstrating how constraints in complex systems emerge dynamically and are reciprocally influential, rather than being fixed top-down or bottom-up.

Premise 3 (Heterarchical Organization): Neural processing in the cortex shows a heterarchical organization, with distributed and multi-directional information flow, moving beyond strict top-down hierarchies (Felleman & Van Essen, 1991).

• Paper: Felleman, D. J., & Van Essen, D. C. (1991). Distributed hierarchical processing in the primate cerebral cortex. Cerebral cortex, 1(1), 1-47.

• Relevance: While titled "hierarchical," this seminal paper on primate visual cortex acknowledged the distributed and interconnected nature of processing, suggesting a more heterarchical than strictly top-down hierarchical organization. It opened the door for thinking beyond simple hierarchies in brain function.

Premise 4 (Multi-Scale Dynamics): Brain networks function across multiple scales, indicating that computation involves dynamic interactions between levels, not just single-scale or level-dominant processes (Bassett & Sporns, 2017).

• Paper: Bassett, D. S., & Sporns, O. (2017). Network neuroscience. Nature neuroscience, 20(3), 353-364.

• Relevance: This review highlights how brain networks exhibit multi-scale organization, with interactions occurring across different spatial and temporal scales, supporting the idea of computation operating across multiple levels simultaneously.

Premise 5 (Contextual Modulation): Neuromodulatory systems demonstrate that brain processing is dynamically modulated by context and global brain states, highlighting context-dependent computation (Aston-Jones & Cohen, 2005).

• Paper: Aston-Jones, G., & Cohen, J. D. (2005). An adaptive theory of locus coeruleus–norepinephrine system function: tonic and phasic modes modulate cognitive exploitation versus exploration. Philosophical Transactions of the Royal Society B: Biological Sciences, 360(1457), 2041-2063.

• Relevance: This paper details the role of neuromodulatory systems like norepinephrine in dynamically adjusting brain function based on context and behavioral demands, demonstrating how global states modulate local processing.

Premise 6 (Field-Like Computation): Neural activity manifests as a spatially extended dynamic field, suggesting brain computation is a distributed, parallel process rather than a linear sequence of instructions (Freeman, 2000).

• Paper: Freeman, W. J. (2000). Neurodynamics: an explanation of mesoscopic brain dynamics through mean field theory. Springer Science & Business Media.

• Relevance: Freeman’s work, though debated, powerfully argues for understanding brain function as a continuous, spatially extended dynamic field rather than a sequence of discrete steps, highlighting the distributed and parallel nature of neural computation.

Premise 7 (Self-Organizing Rules): Neural networks can self-organize and develop structured representations through unsupervised learning and lateral inhibition, showing that computational rules can emerge within the system (Linsker, 1988).

• Paper: Linsker, R. (1988). Self-organization in a perceptual network. Computer, 21(3), 105-117.

• Relevance: Linsker's work on self-organizing networks demonstrated how structured representations can emerge in neural networks through unsupervised learning and lateral inhibition, illustrating a mechanism for how computation can be self-modifying and create its own rules.

Premise 8 (Emergence from Interaction): Embodied and enactive theories argue that cognition fundamentally emerges from dynamic agent-environment interaction, not pre-programmed internal mechanisms (Varela, Thompson, & Rosch, 1991).

• Paper: Varela, F. J., Thompson, E., & Rosch, E. (1991). The embodied mind: Cognitive science and human experience. MIT press.

• Relevance: A foundational text for embodied and enactive cognition, arguing that cognition is not a pre-programmed, internal process, but fundamentally emerges from the dynamic interaction between an embodied agent and its environment.

Premise 9 (Dynamic Interface Linking): Ecological psychology emphasizes that perception dynamically links agents to environmental affordances, suggesting computation is about managing dynamic interfaces, not executing fixed rules (Gibson, 1979).

• Paper: Gibson, J. J. (1979). The ecological approach to visual perception. Houghton Mifflin Harcourt.

• Relevance: Gibson's ecological psychology, particularly his concept of affordances, emphasizes how perception is about directly perceiving possibilities for action in the environment ("interface conditions"), highlighting the dynamic link between agent and world, rather than internal representations.

Premise 10 (Self-Modifying Affordances): Predictive processing frameworks highlight that brains continuously update and modify action possibilities (affordances) based on experience and context, indicating self-modifying computation driven by use (Clark, 2013).

• Paper: Clark, A. (2013). Surfing uncertainty: Prediction, action, and the embodied mind. Oxford University Press.

• Relevance: Clark's work on predictive processing expands on the concept of affordances within a dynamic, Bayesian framework, arguing that the brain is constantly updating its predictions and action possibilities (affordances) based on experience and interaction, showcasing self-modification driven by use and context.

Conclusion: Therefore, based on neuroscientific, cognitive, and computational evidence, a valid and sound theory of computation for cognitive systems should move beyond mechanistic, hierarchical, and rule-based models and embrace a process-based paradigm. This paradigm understands computation as a dynamically emergent, self-referential, and contextually modulated phenomenon characterized by bidirectionality, recursion, heterarchy, multi-scale interaction, and co-emergence with agent-environment interaction. This perspective reframes computation from pre-programmed rule execution to the sustaining of evolving interfaces within a self-organizing system.

Cohesive Integration:These ten points, supported by the cited research papers, coalesce into a compelling picture of a computational paradigm shift. Collectively, they argue against a static, hierarchical, and purely rule-driven view of computation and cognition. Instead, they champion a dynamic, process-oriented perspective where computation is understood as an emergent phenomenon arising from bidirectional, recursive interactions within heterarchical and multi-scale systems. The evidence points towards brains, and by extension cognition, as self-organizing systems that constantly adapt their computational rules and interfaces through ongoing interaction with the world. This enactivist and cybernetic-aligned view suggests that computation isn't about executing pre-programmed instructions but about sustaining evolving, context-sensitive interfaces that maintain coherence within a dynamic, self-simulating reality. This framework, grounded in empirical findings from neuroscience, cognitive science, and AI, offers a scientifically richer and more biologically plausible approach to understanding the nature of computation and cognition, potentially resonating with Chomsky's emphasis on biological plausibility while challenging his more static and rule-centric historical focus.

The abstract rules governing language, chess, computation, and chemistry are not distinct from their material instantiations. Rather, they are persistent constraints that emerge, stabilize, and evolve through computational processes.

Instead of saying rules impose themselves on matter, a better formulation might be:

• Rules are dynamically stabilized regularities within a computationally rich substrate.

• The physical world is not "programmed"; it is a self-simulating process unfolding within a vast computational possibility space (the Ruliad).

• Cognition, chemistry, and computation are all nested within this recursive computational unfolding—each level shaping and being shaped by the others, creating infinite possibilities from finite multi-layer sets.

• Rules emerge within computation, but computation is also constrained by emergent rules.

• Lower-level processes generate stable patterns, but those patterns, once formed, act as constraints on further computation.

• The structure of reality is not a fixed set of rules running from the ground up but a self-organizing computational process where stability at one level shapes what can happen at another.

So, rather than strict top-down causation, we have recursive computational stabilization, where:

1. Abstract patterns emerge from lower-level processes.

2. Those patterns then shape future interactions as constraints.

3. The whole process continuously evolves through cybernetic feedback loops.

This eliminates the dualist separation between abstract rules and physical execution. Instead, rules and material systems are co-implicated, emerging together in a dynamic computational process.

We are in some kind of co-evolving enactivist program, and being co-implicated with it, we co-create via rules simply because we can isolate them and thus recreate many more of them.