The Distinction
A simulation is an implementation used as a model of something. A realization is an implementation that possesses the property at issue. The two can come apart, but the label simulation does not tell us that they do.
A weather model does not move air or rain on the desk. A digestion model does not break down protein. Those properties require physical powers the computer lacks. A simulated calculator, however, really calculates, and a simulated controller wired to a boiler really controls it. The diagnostic question is therefore not is this called a simulation? It is which powers does this physical implementation actually realize?
The application to mind is conditional. If consciousness depends only on an intrinsic causal-computational organization, a system realizing that organization is not disqualified because it also models a brain. If consciousness requires additional world-involving and norm-fixing powers, those must be realized too. The hurricane analogy shows that structural fidelity does not transfer every property automatically. It does not identify the missing property for mind.
Intrinsic Implementation
Putnam- and Searle-style triviality arguments defeat unconstrained computational mappings: with enough interpretive freedom, a wall can be described as running any finite program. They do not defeat disciplined implementation.
Chalmers requires counterfactual-supporting causal structure. Ma and Kanai’s intrinsic computational functionalism strengthens the demand: the organization must be specifiable without an observer’s labels, invariant under structure-preserving relabeling, and displayed in the system’s intervention profile. A system meeting those conditions genuinely realizes the causal-computational organization. This is an observer-independent physical fact, not a story laid over the machinery.
The remaining dispute concerns semantics and consciousness, not whether the implementation is real. Intrinsic transition structure alone does not fix which worldly conditions its states are about. A critic of simulated consciousness must identify the further consciousness-relevant organization rather than repeat the word simulation.
The Candidate Stake
Mind, Matter, and Meaning defends thermodynamic self-maintenance coupled to adaptive regulation as a candidate basis of autonomy, practical interest, and welfare. The system’s activity must help preserve the physical organization doing the activity. Its regulatory states must distinguish movement toward and away from viable conditions and vary the response. That organization institutes a descriptive functional standard: some trajectories support the continuation of the regulatory organization and others undermine it.
The claim does not derive a categorical ought from causal success. Plenty of things persist without regulating their own persistence; a candle manages the first and fails the second. In adaptive reciprocal closure, regulatory activity counterfactually affects viability, while departures from viable conditions alter regulation. That two-way dependence makes the success condition internal to the organization being maintained. “Norm” here names the resulting possibility of function and malfunction, not a moral command written into matter.
Selection or training can give a submechanism content. Whether that content counts as original depends on whether the correctness standard arises independently of an interpreting practice. The stake does not create, elevate, or own that content. It changes what success and failure mean for the continuing organization: correct tracking may now guide activity on which its persistence depends, and failure may harm it. Content explains what a state says; stake helps explain why getting it right matters to this organized life.
This yields five constitutive questions. Selection, learning, and causal history can produce mechanism-level content. Flexible use guided by that content can support domain-specific competence. Several capacities constraining one another in an answerable system can support partial understanding; breadth, persistence, and cross-domain traffic make that understanding more integrated and world-involving. A supported threshold relation marks state consciousness, while integrated persistence identifies the subject and the world-presenting profile determines character. Reciprocal self-maintenance can separately establish autonomy, interests, and welfare. Reasons-responsive agency cuts across the map when represented considerations govern conduct through the persisting control organization. None of the later achievements mints content, and self-maintenance is not a prerequisite of consciousness.
In organisms, metabolism realizes thermodynamic self-maintenance. Metabolism is the biological instance, not the substrate-general requirement. Life and carbon are not necessary. An artificial or virtual system could acquire a comparable stake if its physical implementation closed the same regulatory loop. It could acquire content through selection, learning, or world-involving history whether or not it crossed that further line.
The System and Its Platform
The system boundary follows reciprocal maintenance at a stated time scale, not the casing, corporate ownership, or the location of a process. The time scale must match the processes that constitute, regulate, and repair the organization; it cannot be chosen merely to secure a preferred verdict. A component belongs to the candidate system when its activity helps maintain the relevant closure and its continued operation depends on that closure. A platform component that technicians, schedulers, or an independent power system keep available regardless of the agent’s success remains outside the loop.
This criterion permits nested and distributed systems. A robot may include remote components; a factory, colony, or firm may qualify as a self-maintaining organization with practical interests at that level. Such cases do not automatically produce a unified subject that understands or feels.
Why Current Systems Lack This Stake
The text-trained and agentic architectures the book analyzes commonly represent goals, energy budgets, health variables, and even their own continued operation. They may also contain genuinely content-bearing submechanisms. Their platforms nevertheless maintain the physical implementation independently of the agent’s success. The scheduler, data center, power system, and technicians keep the process available while the agent tends represented set-points. The viability drama and the continuation of the realizing machinery come apart. That supports a bounded verdict about autonomy and welfare, not a semantic or consciousness verdict.
That diagnosis stays bounded to those architectures; it does not survey every public or possible AI system. A digital system is physical all the way down. Couple its adaptive activity to energy acquisition, repair or reconstitution of its realizing machinery, maintenance of a boundary, and world-tracking that bears on those activities, and the case changes.
Copyability and Backup
Copyability is not a criterion of content or consciousness. A backup may reveal that a current agent leaves its continuation to infrastructure outside its loop, but the architecture does the work, not the file.
- A non-copyable process may fail to maintain itself; the candle dies quite unbacked-up.
- A self-maintaining system would not lose its norm because an engineer learned to copy it.
- A duplicate raises questions about identity, succession, and inherited history. It neither creates nor removes the original’s stake.
Geoffrey Hinton’s mortal computation remains important because it proves that computation can bind tightly to the detailed physics of one device. Hardware-specificity is neither necessary nor sufficient for a stake. It is an engineering clue, not the semantic criterion.
Virtual Worlds
Virtual objects and events can be real digital objects and events. Their causal relations can matter to an agent. Designer assignment does not permanently condemn every category inside a made world to borrowed meaning; designers also build laboratories in which organisms acquire content of their own.
A bare web of formal relations, without selected producer-consumer use, leaves distal reference unfixed. Training within a virtual environment can give mechanisms content when their consumers use those states and differential retention preserves the arrangement. Whether that content counts as original depends on whether the correctness standard remains borrowed from an interpreting practice. Causal-historical traffic can further fix worldly reference. Neither achievement alone establishes understanding by a unified agent.
A practical stake requires more: the content-bearing activity must contribute to maintaining the organization that sustains and revises it. A represented health variable supplies no such relation by its name. A physical implementation whose self-maintenance depends on tracking virtual conditions could acquire autonomy and welfare, even though designers built the world and its physics. Its content would still come from the relevant selection, learning, and causal history.
My View
The slogan simulation is not realization survives only as a conditional:
An implementation that merely represents a mind-making power does not realize that power. An implementation that genuinely possesses the required causal organization is not disqualified by being called a simulation.
For mind and stake, the relevant questions form a map rather than a single test:
- Does the implementation possess observer-independent causal organization?
- Did selection or training give any mechanisms content, and does that content remain derived or count as original?
- Do several content-guided capacities support partial understanding, and how broadly and persistently do they integrate?
- Which relation marks state consciousness, which persisting organization bears the state, and what world-presenting profile determines its character?
- Does reciprocal self-maintenance give the organization autonomy, practical interests, and welfare?
Reasons-responsive agency cuts across questions 2–4 when represented reasons govern the bearer’s conduct; it is not another grade of understanding or a proxy for consciousness or stake.
Different systems can occupy different positions on the map. Intrinsic computational functionalism may describe the entire package. If it can, I accept the description and inspect the realization. The dispute belongs in the architecture, not in the noun.