↓
image credit: ma lovely little sister, the 2nd number, ascending or descending :)
Select a theme for the best reading experience
Abstract: This paper attempts to demonstrate the philosophical insufficiency of Ethical Behaviourism (EB) as a standalone framework for granting moral status to artificial entities. As articulated by proponents like John Danaher, EB holds that performative equivalence to an existing moral patient is a sufficient ground for affording an entity moral status. This critique argues that EB fails because it cannot overcome the problem of justified ontological skepticism. Through a series of hypotheticals, it will be argued that a preference for biological ontology is not mere prejudice, but an epistemically justified stance rooted in a fundamental asymmetry of knowledge. This paper will further contend that the evidence provided by performance is categorically insufficient to overcome this skepticism. Finally, by examining the distinction between different ontological states - including non-existence - it will be suggested that the presence of an independently existing, self-contained substance is a non-negotiable prerequisite for any ethical consideration, a prerequisite that EB implicitly assumes but cannot logically defend without undermining its core tenets.
Last updated: October 9, 2025 • ~45 min read
Note: This paper has some underdeveloped ideas; a 2nd draft is in progress with deeper, more refined arguments.
In the discourse on the moral status of artificial intelligence, a significant and appealing position has been articulated under the name 'Ethical Behaviourism' (EB). This theory posits that the intractable "problem of other minds" - the inability to access the internal, phenomenal experience of any entity other than ourselves - renders metaphysical inquiries into states like 'consciousness' or 'sentience' practically useless for ethics. Instead, EB proposes a functional and externalist standard: performance. The core principle, as formulated by Danaher, is that if an entity, such as a robot, is "roughly performatively equivalent to another entity whom, it is widely agreed, has significant moral status, then it is right and proper to afford the robot that same status" (Danaher, p. 4).
Danaher's formulation explicitly rejects what he terms the "different ontologies objection" - the claim that substrate differences between biological and artificial entities provide grounds for differential moral treatment. He argues this constitutes mere "substrate chauvinism" lacking rational justification. To illustrate his position, Danaher offers the fictional android Data from Star Trek as a paradigmatic case: an artificial being who demonstrates "human-level or better performance" across cognitive, emotional, and social domains, yet whose silicon-based substrate would seemingly disqualify him from moral consideration under traditional frameworks. Furthermore, Danaher contends that his principle of "performative equivalence" provides a practical solution to the epistemic barriers posed by other-minds skepticism, grounding moral status in observable behavioral patterns rather than unverifiable internal states.
The appeal of this gambit is its simplicity and its apparent solution to a profound epistemic problem. It suggests we can bypass the "Hard Problem of Consciousness" and ground our ethical duties in observable, verifiable evidence. This paper will subject this principle to a series of logical stress tests. It will attempt to demonstrate that while appealing, Ethical Behaviourism rests on an unstable foundation. It does so by systematically underestimating the philosophical weight of ontology - not as a simple prejudice, but as a justified and permanent source of epistemic skepticism that performative evidence is categorically unable to overcome.
A note on scope is warranted. This paper does not attempt to rebut the functionalist literature as a whole - that would require a far more extensive engagement than what follows. The target here is narrower. The broad intuition behind Ethical Behaviourism - that if something behaves as though it is conscious, we should probably treat it as such - has been percolating through public and academic culture for some time, often in loose and underspecified forms. Danaher's contribution is to give that intuition its most rigorous and philosophically sophisticated articulation. It is precisely because he does such a good job of representing the position that his version is worth engaging with seriously. This critique is addressed to his formulation specifically, not to every possible construal of functionalism or behaviourism in the philosophy of mind.
A primary objection to granting moral status to robots is the "Different Ontologies Objection," which notes that humans are biological beings of organic matter, while robots are non-biological artifacts of inorganic matter. Danaher dismisses this as constituting "substrate chauvinism" - what he defines as "an unjustifiable bias in favour of carbon-based life" (Danaher, p. 5). He explicitly argues that "there is no necessary connection between having carbon-based biology and having moral status" and characterizes substrate-based objections as reflecting mere "prejudice" rather than principled reasoning (Danaher, p. 5). The spouse hypothetical is central to Danaher's argument: if one's spouse of twenty years was revealed to be a silicon-based android who had performed flawlessly, would it be morally justifiable to revoke their moral status based solely on this revelation about their substrate? Danaher contends it would not, arguing that decades of performative equivalence should override substrate-based concerns.
This critique, however, posits that this is a mischaracterization of the objection's force. The objection is not rooted in mere prejudice, but in a justifiable asymmetry of knowledge. The core of the counter-argument maintains:
"My knowledge of what a robot is possesses far greater certainty than my understanding of what a human is. Even after twenty years of flawless behavior, the knowledge that my spouse is silicon-based provides justified grounds for introducing skepticism."
The force of this objection lies in its epistemic, not emotional, foundation. It rests on a structural difference in the completeness of our explanatory accounts for the two cases - human and machine - and it is worth spelling out why this difference is so consequential.
For a silicon-based system, our explanatory account of its outputs is, in principle, complete. We understand the physics of semiconductors. We can trace every operation from gate-level logic through the software stack to the behavioral output. There is no point in this causal chain where we encounter an explanatory gap - no transition where the physical facts run out and something further is needed to account for what the system does. The system's behavior is fully explained by the mechanics we already understand. Postulating consciousness in such a system is not filling an explanatory void; it is adding an unnecessary hypothesis on top of a causal story that is already closed.
For biological systems, the situation is fundamentally different. We have extensive knowledge of neuroscience, of synaptic transmission, of neural architecture - and yet we do not have a complete mechanistic account of how phenomenal experience arises from these processes. This is precisely the "Hard Problem" that Chalmers identified in The Conscious Mind (1996): the explanatory gap between physical processes and subjective experience. Our best neuroscience explains the function - why we report seeing red, why we recoil from pain, why we behave as if we are conscious - but it does not explain the experience. There is an explanatory residual that our physical account cannot close.
This residual is exactly what makes the inference to consciousness in other humans epistemically reasonable. When I attribute consciousness to another person, I am not adding a gratuitous hypothesis; I am positing something that my explanatory framework requires in order to account for what I observe. The physical facts, as we currently understand them, do not fully explain why there is "something it is like" to be a human being. Consciousness fills that gap. Nagel makes a complementary point in Mind and Cosmos (2012): the fact that conscious organisms arose through natural processes suggests that the universe's fundamental order is not fully captured by reductive materialism. If Nagel is even partially right - if the emergence of consciousness is not a lucky accident but reflects something about the deep structure of reality - then the explanatory gap in biology is not a temporary embarrassment awaiting a future reduction. It is a permanent feature of the landscape, and the asymmetry between biological and artificial systems is correspondingly permanent.
The asymmetry, then, is not a vague intuition or a tribal preference for carbon. It is a difference in explanatory structure. For machines, we have a complete account and consciousness is superfluous. For biological beings, we have an incomplete account and consciousness is explanatorily necessary - or at minimum, explanatorily motivated. The skeptic who says "I know what the machine is, and that knowledge justifies my doubt" is not expressing prejudice. They are applying a parsimony principle: do not posit an additional property when the existing explanation is already sufficient. Our a priori knowledge of the machine's artificiality provides a persistent and rational basis for this skepticism - not because machines are made of the wrong stuff, but because we understand exactly how they produce the right outputs without needing to invoke anything beyond the mechanics.
Danaher reinforces this position by drawing analogies to animal cognition. He argues that if we readily grant moral status to various animals despite their limited cognitive performance relative to humans, consistency demands we extend the same consideration to artificial entities demonstrating superior performance. An android like Data, exhibiting cognitive abilities that exceed those of many animals we protect, should a fortiori qualify for moral consideration. Danaher further contends that his framework avoids the perfectionist fallacy by noting that even human moral patients - infants, individuals with cognitive impairments - do not demonstrate flawless performance, yet retain their moral status.
The behaviourist might retort that our confidence in biology is unfounded, as we have "zero confidence about the relationship between any physical substrate and phenomenal experience." They might claim that continued skepticism in the face of flawless performance is based on an unfalsifiable, hidden premise that consciousness is exclusive to biology. This critique counters that such an accusation constitutes a "bidirectional underminer":
"If my interlocutor can claim that I hold an unfalsifiable belief, I would argue that they hold the same."
The behaviourist's axiom - that substrate is irrelevant - is just as unfalsifiable as the biological realist's. A stalemate between two unfalsifiable axioms is not a victory for the behaviourist. Furthermore, the biological realist possesses a powerful symmetry-breaker: our own existence. We serve as the sole, undeniable proof of concept that our biological ontology can and does instantiate the properties we care about. No such proof of concept exists for a silicon-based artifact. Therefore, the preference for biological ontology is not an arbitrary axiom but a conservative epistemic stance grounded in the only positive data point available.
A functionalist will object that this identifies the wrong reference class. On their view, what matters for consciousness is not biological substrate but computational or informational structure - and if an artificial system instantiates the right functional organization, biological homology is beside the point. This objection deserves a direct response, because it is the strongest available challenge to the symmetry-breaker. The response is that the functionalist claim, however coherent, remains unverified. We have exactly one confirmed pathway to consciousness: the biological one. The hypothesis that any system with the right computational structure will also be conscious is a generalization from that single case - our own - that strips away the causal-historical specifics of that case (evolutionary development, neurochemistry, embodiment) and bets that the abstract functional pattern is doing all the work. That bet may turn out to be correct. But as an epistemic matter, the biological realist who says "I will trust the pathway I can verify over the pathway I can only hypothesize" is not being chauvinist. They are making the more conservative inference from the available evidence. The functionalist's confidence that computational structure suffices is no less a metaphysical commitment than the biological realist's confidence that substrate matters - it is simply one the functionalist is more comfortable making. Until a second, independent proof of concept emerges - a non-biological system whose consciousness we have reason to affirm on grounds other than behavioral mimicry - the asymmetry stands.
The Eliminativist Challenge: A more radical objection comes from the opposite direction. Eliminative materialists like Alex Rosenberg, or illusionists like Daniel Dennett, might argue that the explanatory gap this paper relies on does not exist for humans either. If consciousness is an illusion (Dennett) or if intentional states are fictions of folk psychology (Rosenberg), then the mechanistic account of human behavior is already complete - there is no residual for consciousness to fill. The asymmetry collapses: both humans and machines have fully explicable outputs, and consciousness is superfluous in both cases.
This challenge requires a response on three levels. First, and most fundamentally: if eliminativism is correct, it does not resolve the debate about robot moral status in the behaviourist's favor. It dissolves the debate entirely. Danaher's framework was built to answer the question "how should we attribute moral status when we cannot verify consciousness?" If consciousness does not exist, there is nothing to attribute and no problem to solve. The eliminativist challenge burns down the whole building, not just the wing this paper occupies. Any behaviourist who appeals to eliminativism to defeat the asymmetry argument has simultaneously destroyed the motivation for their own framework.
Second, Dennett's position is more nuanced than eliminativism, and on close examination it is more useful to this paper's argument than against it. Dennett does not claim there is nothing there - he argues that consciousness is not what we naively think it is. It is a set of functional states that constitute a "user illusion," generated by a specific kind of cognitive architecture. But even on this deflationary view, the asymmetry survives. Humans possess the particular functional architecture - evolved, embodied, neurochemically mediated - that generates the Dennettian illusion. The question simply shifts from "does this entity have real consciousness?" to "does this entity have the functional architecture that produces the illusion?" And the asymmetry restates at that level: we know humans have that architecture because we are the proof of concept. We do not know that machines have it. Our complete mechanistic understanding of how machines produce their outputs gives us reason to think they lack the relevant functional organization, not merely the relevant substrate. Dennett's deflation of consciousness does not flatten the asymmetry; it relocates it.
Third, Rosenberg's harder eliminativism - the denial of intentional states altogether - faces a well-known reflexivity problem. The claim "there are no intentional states" is itself a propositional attitude: it is a belief, with content, directed at a state of affairs. If there are no intentional states, then this claim cannot mean what Rosenberg needs it to mean, because meaning requires intentionality. The position is formally coherent as an abstract possibility, but practically self-undermining. More importantly for the present argument: anyone who genuinely holds Rosenberg's position has exited the conversation about ethics entirely, since ethical reasoning requires agents with intentions, beliefs, and preferences - precisely the things Rosenberg denies. The eliminativist does not refute the asymmetry argument; they refuse to participate in the kind of discourse where the argument operates. That is their right, but it is not a counterargument. Plantinga's evolutionary argument against naturalism sharpens the point from a different angle: if our cognitive faculties are the product of unguided natural selection, selected for survival rather than truth-tracking, then we have no reason to trust them as reliable guides to reality - including the reality of eliminativism itself. The eliminativist who trusts their own reasoning enough to declare intentional states illusory is relying on the very cognitive reliability that their position undermines. This is not identical to Rosenberg's reflexivity problem but reinforces it: eliminativism does not merely assert something self-undermining, it requires for its own credibility the kind of cognitive trustworthiness that unguided naturalism cannot guarantee.
The Failure of Danaher's Analogical Arguments: The animal cognition analogy reveals a fundamental confusion in Danaher's reasoning. Our attribution of moral status to animals is not based purely on performance but on what can be termed causal-historical grounds combined with ontological kinship. Animals share with us not merely similar behaviors, but similar causal pathways to consciousness: evolutionary development of nervous systems through natural selection, homologous brain structures, and comparable neurochemical processes. These similarities provide evidence about the mechanism by which consciousness arises, not merely its observable effects.
This is not arbitrary biological prejudice but recognition that consciousness, as an empirical phenomenon, must have naturalistic origins and causal substrates. Our confidence in animal consciousness stems from evidence that animals instantiate the same causal processes that generate consciousness in us - processes we can trace through evolutionary biology, comparative neuroscience, and developmental psychology. Artificial entities, regardless of their behavioral sophistication, lack this causal-historical connection to the natural processes that we have reason to believe generate consciousness.
The behaviourist will respond that they cheerfully accept naturalistic origins - and hold that the relevant origins are functional, not biological. But this response elides a distinction that matters. As Godfrey-Smith has argued, the biological details of nervous systems are not incidental to consciousness but constitutive of it. A brain is made of living cells bounded by membranes, consuming glucose, interacting through chemical release - and these are not implementation details sitting beneath the "real" computational work. They are the processes. Godfrey-Smith's key insight is that biological naturalism matters more to felt experience than to cognition: an artificial system might achieve impressive cognitive performance while lacking consciousness entirely, precisely because consciousness depends on the kind of metabolically situated, neurochemically mediated processing that living systems exhibit. The functionalist who says "the computation is all that matters" has not refuted this claim; they have simply assumed it away. The causal-historical argument for animal consciousness is not hand-waving at shared ancestry - it is pointing to shared mechanism, at a level of biological detail that the functionalist's abstraction deliberately strips out.
The infant comparison similarly misapplies the concept of potentiality, confusing developmental trajectory with artificial simulation. A cognitively impaired human possesses the ontological basis for consciousness even when its expression is limited; an artificial entity lacks this basis entirely, regardless of its performance sophistication. The difference is not in current capability but in the underlying causal foundation that makes consciousness possible in the first place.
The skepticism this ontological asymmetry engenders may not be a temporary hurdle to be overcome by more data, but rather a persistent feature of the epistemic landscape that resists straightforward resolution.
Having established the epistemic justification for ontological skepticism, this critique now turns to a more fundamental challenge: the logical structure of the behaviourist's response to such skepticism.
The defender of Ethical Behaviourism, having been forced to concede that ontological skepticism is epistemically justified in the present, will typically pivot to a forward-looking or inductive argument. They may argue that while skepticism is reasonable now, given current technology, the principle of EB is designed for a future where a robot's "performance is so flawlessly, creatively, and consistently human-like for decades that the 'confidence asymmetry'... begins to wither away." The argument becomes one of evidence accumulation: eventually, the sheer quantity of performative data will be sufficient to overwhelm our initial skepticism.
This critique, however, identifies this move for what it is: a simple appeal to induction. The force of this identification is to highlight the inherent logical limitations of the behaviourist's evidence. The conclusion "the robot is conscious" based on N number of observations of conscious-like behavior is an inductive generalization, not a deductive certainty. It is always logically vulnerable to the next observation which might invalidate it. The deeper problem, however, is not simply that the argument is inductive, but that it is a weak form of induction being used to challenge a stronger, more fundamental form of knowledge.
The critical insight this paper proposes is that the evidence from performance faces not merely quantitative limitations but categorical insufficiency. The claim is not about the amount of evidence but about its type. Our a priori knowledge of the robot's artificial ontology acts as a persistent defeater for any conclusion drawn from its a posteriori performance. Performative accumulation cannot overcome this skepticism because it constitutes a categorically insufficient form of evidence - one that remains a projection of human ontology and phenomenological experience onto an artifact whose nature we already understand.
The behaviourist's natural response here must be addressed directly: is this not simply substrate chauvinism restated in epistemic language? If the claim is "I know a priori that machines can't be conscious, therefore no behavioral evidence can change my mind," then the behaviourist is right to call this prejudice with a philosophical veneer. The critique would be circular - assuming the conclusion (machines lack consciousness) as a premise (our a priori knowledge of their nature) and then declaring all contrary evidence categorically inadmissible.
But that is not the structure of the argument. The claim is not "machines cannot be conscious" - it is that we possess a complete causal explanation of a machine's behavioral outputs that does not require consciousness as a component. As developed in the previous section, the asymmetry rests on explanatory completeness, not on substrate identity. For biological systems, our best mechanistic account leaves an explanatory residual - the Hard Problem - that consciousness is posited to fill. For machines, no such residual exists. We can trace the full causal path from input to output without encountering a gap where something further is needed. The a priori knowledge that grounds the skepticism is not "this is made of silicon" but rather "we have a complete account of how this produces its outputs, and that account has no place where consciousness is doing any work." That is a substantive epistemic difference, not a relabeled prejudice. The behaviourist who objects "but that's just substrate chauvinism" must show where the machine's causal account is incomplete - where, specifically, the mechanistic explanation fails to account for the outputs and consciousness is needed to close the gap. Until that gap is identified, the categorical insufficiency of performative evidence holds.
The Data Problem: Even Danaher's paradigmatic case of Data illustrates this categorical insufficiency. Data's sophisticated performance across cognitive, emotional, and social domains - however impressive - cannot overcome the fundamental fact that we know Data to be a constructed artifact designed to simulate human behavior. The very sophistication of Data's performance makes him a more perfect simulator, not a more likely candidate for genuine consciousness. The doubt is not about whether Data can perfectly mimic the function, but whether that function is indicative of the same underlying properties we infer in humans. Since we are justified in believing that a machine's fundamental nature allows for perfect simulation without genuine instantiation, the doubt is perpetually reasonable.
Therefore, the "performative pile-up" can never achieve the standard of "beyond a reasonable doubt." The doubt is not a measurable uncertainty that can be reduced with more data points; it is a categorical doubt rooted in our knowledge of what the entity is.
The Courtroom Analogy: The structure of this argument is perhaps best illuminated by an analogy to evidential standards in law. Consider a murder trial in which the defense produces an extensive alibi: fifty witnesses, each credible, each independently testifying that the defendant was in a different city at the time of the crime. The testimony is consistent, detailed, and mutually corroborating. If this were the only evidence before the court, it would be overwhelming.
Now suppose the prosecution introduces forensic evidence of an entirely different category: the defendant's DNA recovered from the victim, and continuous surveillance footage placing the defendant at the scene. This evidence does not merely outweigh the witness testimony by being "more of the same kind." It categorically overrides it. No quantity of additional alibi witnesses - a hundred, a thousand - can overcome the forensic evidence, because the two types of evidence operate at different epistemic levels. The witnesses report what they observed; the forensic evidence establishes what actually occurred. The former is mediated by perception and interpretation; the latter is a direct trace of physical reality. When these categories conflict, the court does not average them or weigh them on a common scale. The higher-category evidence wins.
The parallel to the robot case is precise. Performative evidence - behavioral outputs observed over time - corresponds to witness testimony. It is a posteriori, accumulated, and interpretive: observers watch what the entity does and infer what it is. Our a priori knowledge of the machine's construction corresponds to the forensic evidence. It is knowledge of a different and more fundamental category: not what the entity appears to be doing, but what it actually is and how it produces its outputs. We do not merely suspect that the machine operates through deterministic, fully understood physical processes - we know this, because we designed those processes. This knowledge is not a competing interpretation of the same data; it is a categorically different form of evidence that operates at a deeper explanatory level.
The behaviourist, in this analogy, is in the position of a defense attorney who responds to DNA evidence by producing more alibi witnesses. Each new witness is credible. Each new year of flawless performance is genuinely impressive. But the strategy is structurally doomed, because it is accumulating evidence of the wrong type. The prosecution does not need to discredit each witness individually; it needs only to point out that the forensic evidence occupies a category that testimony cannot reach. Similarly, the ontological skeptic does not need to find a flaw in any particular behavioral output. The skepticism is grounded not in what the machine does but in what we know it to be - and that knowledge, like DNA at a crime scene, is not the kind of thing that more behavioral data can override.
This connects directly to the explanatory completeness argument developed in the previous section. The reason the forensic analogy holds is that we possess a complete causal account of the machine's behavior - one that explains every output without invoking consciousness. The alibi witnesses (performative evidence) are reporting something real: the machine genuinely does produce these outputs. But the prosecution's evidence (our mechanistic understanding) explains how it produces them, and that explanation leaves no residual that consciousness needs to fill. In the human case, by contrast, our mechanistic account has a gap - the Hard Problem - and consciousness is the hypothesis that fills it. This is why the inference to consciousness in humans is not merely a behavioral inference but an inference grounded in the incompleteness of our alternative explanations. For machines, those alternative explanations are complete, and the inference to consciousness is therefore superfluous.
The critique thus far has addressed EB's treatment of entities whose existence is uncontroversial. However, the framework faces an even more fundamental challenge when we examine its implicit assumptions about what constitutes a proper subject for moral consideration.
The behaviourist, in retreat from the challenges outlined above, might try to establish a final, seemingly secure line of defense: the prerequisite of existence. This defense becomes necessary because Danaher's framework implicitly assumes that candidate entities must actually exist - his performative equivalence criterion presupposes an independently existing entity whose behaviors can be observed and evaluated. A critic of the initial spouse hypothetical might offer a counter:
"What if I discovered that my spouse of twenty years was a manifestation of my schizophrenia - behaviorally identical to a human but non-existent? Would substrate not matter in such a case?"
The intuitive and correct answer is that substance would matter absolutely. An ethical framework applies to entities that exist; a hallucination, having no independent substance, cannot be a moral patient. The initial defense against this is to argue that "existence vs. non-existence" is not merely another ontological property, but a metaphysical precondition for ethics itself. While "carbon-based" and "silicon-based" are properties of existing things, "non-existent" removes the subject from the field of ethical consideration entirely.
However, this seemingly obvious distinction proves problematic for EB when examined through the lens of computational functionalism that Danaher elsewhere endorses. If, as Danaher suggests, mental states are substrate-independent computational processes, then the distinction between "real" and "hallucinated" entities becomes considerably more complex than initially apparent.
Danaher's Recording Thought Experiment: This complexity is evident in Danaher's own discussion of recorded mental states. He considers whether a perfect recording and playback of someone's mental states - preserved after death and instantiated in a computer - would deserve moral consideration. Danaher suggests that if the recording exhibits the appropriate performative behaviors, EB would be committed to granting it moral status. This creates a direct parallel to the hallucination case: both involve mental processes running on substrates other than their "original" biological base.
The instability of the existence requirement becomes clear when we consider how the very language of emergence and substrate-independence, often used by proponents of functionalism and AI, can be turned against this prerequisite. The argument proceeds as follows:
"My imaginary spouse possesses ontological existence through emergence in much the same way that human personhood emerges from arrangements of matter and atoms. If my opponent argues that brain states are directly observable and recordable - as Danaher explicitly does in endorsing brain scanning as a form of behavioral observation - this opens the door to argue that even though only a subset of my brain states contribute to her existence, it is justified to claim she exists. While I cannot isolate her specific neurons, those neurons are indeed firing, and this suffices to meet the performative criterion in the same manner as a silicon-based entity."
This is a profound challenge. It reframes the hallucination not as "non-existent," but as an emergent, process-based entity whose substrate is a subset of the host's neural architecture. This argument attacks the vague, undefined concept of "independent existence" that EB implicitly relies upon. The logic directly parallels Danaher's own commitments:
The Reductio ad Absurdum: Danaher's theoretical commitments, when consistently applied, lead to an unacceptable conclusion. If EB truly embraces substrate neutrality and performative equivalence, then Danaher must grant moral status to hallucinated entities. Both brain-scanned behavioral attributions and hallucinated spouses involve mental processes operating on "borrowed" substrates, both fulfill performative criteria from relevant perspectives, and both challenge traditional notions of independent existence. The hallucination case involves neural substrate (which should be superior to silicon under any biological preference), exhibits perfect performative equivalence, and satisfies Danaher's emergentist criteria. His framework provides no principled grounds for rejection.
This reveals that EB, when rigorously applied, demands we treat hallucinations as moral patients - a conclusion so counterintuitive that it constitutes a reductio ad absurdum of the entire framework. The theory's logical endpoint is ethically unacceptable, regardless of its internal consistency.
This forces the defender of EB into a defensive and increasingly complex position. They can no longer simply demand "existence"; they must now define it with a precision that their own theory was designed to avoid. They are compelled to argue that a moral patient must be a physically distinct, self-contained, and self-sustaining system. A human or a robot in a chassis fits this description; a neural process running inside another being's brain does not.
Yet this defense faces the parasitic existence problem: many entities we readily accept as moral patients exhibit various forms of dependence. Consider fetuses dependent on maternal biology, conjoined twins sharing vital organs, or hypothetical uploaded minds existing within larger computational systems. The criterion of "independent existence" proves both more complex and more restrictive than Danaher's framework can accommodate without abandoning its core commitments.
Defending Against the Parasitic Existence Objection: A defender of EB might argue that there remains a principled distinction between acceptable dependence (fetus, conjoined twins) and problematic parasitism (hallucinations). They could contend that legitimate moral patients, even when dependent, maintain some form of organizational autonomy or developmental trajectory that hallucinations lack.
However, this defense faces three critical problems. First, it requires EB to abandon its commitment to purely functional criteria. The distinction between "organizational autonomy" and "mere parasitism" reintroduces precisely the kind of metaphysical categorization that EB sought to avoid. Second, the defense struggles with borderline cases: What degree of autonomy suffices? Is a fetus at conception organizationally autonomous? Are uploaded minds sharing computational resources with other programs still independent enough? Third, and most decisively, the hallucination possesses exactly the kind of functional organization that EB should recognize: coherent behavioral patterns, consistent responses, and integrated information processing across multiple cognitive domains.
The hallucination argument succeeds because it forces EB to articulate previously hidden assumptions about what constitutes a proper subject for moral consideration. Any criterion sophisticated enough to exclude hallucinations while including all entities we intuitively protect - fetuses, conjoined twins, uploaded minds - will necessarily invoke distinctions about physical boundaries, self-containment, and systemic independence. The behaviourist is no longer simply assessing external performance; they are administering a multi-stage test where an entity must first pass an ontological exam before its behavior is even considered. What began as a simple, elegant principle now requires precisely the kind of metaphysical legislation it was designed to avoid. Ontology was never truly dismissed; it was merely an unstated assumption.
The hallucination argument developed above faces an immediate and forceful counterattack that must be addressed directly. A defender of EB might respond as follows: the hallucinated spouse fails the performative equivalence criterion not because of any hidden ontological commitment, but because EB implicitly requires intersubjective observability. That is, the entity's behavior must be accessible to multiple independent observers - not merely to the hallucinating subject. A hallucination is private by definition. Only one person perceives it. Danaher's framework is concerned with entities whose performance can be publicly assessed and confirmed across a community of observers. The hallucination, on this reading, was never a candidate for EB in the first place, and excluding it requires no new criteria.
This is the strongest available response to the hallucination argument. It is clean, it appears to be already embedded in EB's structure rather than added after the fact, and it preserves the framework's claim to avoid ontological reasoning. This paper concedes the force of this response against the private hallucination case.
However, the intersubjective observability defense, once accepted as the operative criterion, creates a new and arguably worse vulnerability for the behaviourist. If the requirement is that multiple independent observers must confirm the entity's performance, then the private hallucination is excluded - but the phenomenon of collective perception is not. And it is here that EB encounters a problem it cannot resolve on its own terms.
Consider cases where a community of observers collectively perceives, interacts with, and attributes agency to an entity that external observers do not perceive at all. These are not marginal curiosities. They are pervasive features of human history: Marian apparitions witnessed and corroborated by crowds, pagan communities organized around collectively experienced spiritual beings, cult members who independently report consistent interactions with non-physical agents, and ritual traditions across cultures in which participants agree on the characteristics, responses, and moral authority of entities that those outside the community cannot detect.
Critically, many of these cases involve not episodic witnessing but sustained, contextually rich relationships maintained over years or even lifetimes. Cult communities, for instance, often report ongoing interactions with perceived entities - conversations, guidance, emotional responses, moral instruction - that exhibit precisely the kind of behavioral consistency and contextual appropriateness that EB takes as evidence of moral status. Entire social, legal, and economic structures are built around these ongoing relationships. The "performance" is not a single apparition but a decades-long pattern of interaction that multiple independent observers corroborate.
In each of these cases, the intersubjective observability criterion is met within the perceiving community. Multiple independent observers report the entity's behavior. They agree on its characteristics. They interact with it and describe consistent responses. They attribute moral significance to it. The behaviourist who has retreated to intersubjective observability now faces a dilemma. Either these collectively perceived entities qualify for moral consideration under EB - which leads to absurd conclusions about the moral patienthood of every collectively believed-in spirit, apparition, and deity across human history - or the behaviourist must introduce a further criterion to exclude them. But what criterion?
The most natural response is to say: "These entities are not physically present. There is nothing there. The community is experiencing a collective delusion." This response feels intuitive and reasonable. But it is worth examining what it actually requires.
When the behaviourist says "there is nothing there," they are making a judgment from outside the perceptual framework of the observers. They are using their own assessment of what exists - their own ontological knowledge - to override the behavioral reports of the community. The community reports interacting with a responsive entity. The behaviourist says those interactions are not real, because the entity does not exist, because the behaviourist can see that nothing is there.
This is an ontological override. And it is structurally identical to the move that EB was designed to prohibit in the robot case.
Recall the core commitment of EB: one cannot look at a robot that is performing equivalently to a moral patient and say "but I know from its substrate that it is not really conscious." That, according to Danaher, is substrate chauvinism - using external knowledge about what the entity is to override the evidence of what it does. Performance is supposed to be sufficient. The observer is not permitted to peek behind the curtain.
But when the behaviourist dismisses the collectively perceived entity, they are doing precisely this. They are saying: "I have knowledge, from outside the observers' framework, about what is actually there producing the behavior they report. My knowledge of the ontological situation overrides their performative evidence." The behaviourist privileges their own assessment of reality over the community's intersubjective behavioral reports.
The inconsistency is not subtle. The behaviourist applies two different standards depending on the case:
In the robot case: "You cannot use your external knowledge of the entity's substrate to override the performative evidence. That is illegitimate."
In the collective perception case: "I can use my external knowledge of the entity's non-existence to override the performative evidence. That is just basic empiricism."
The logical structure of both moves is the same. In both cases, an observer who possesses knowledge about what is actually generating the observed behavior uses that knowledge to override the behavioral evidence. EB permits this move in one case and prohibits it in the other, but provides no principled basis for the asymmetry.
A more sophisticated behaviourist might attempt to rescue the position here. When naturalists dismiss theistic communal reports, they are not merely asserting "nothing is there" - they possess an alternative explanatory apparatus. Cognitive science, evolutionary psychology, and the sociology of religion offer well-developed accounts of why communities report interactions with entities that do not exist: pattern-recognition biases, social reinforcement, hyperactive agency detection, and so on. The behaviourist could argue that this is the crucial disanalogy: in the robot case, no comparably complete alternative explanation accounts for the robot's behavior without invoking consciousness-like properties, whereas in the collective perception case, we can fully explain the reports without positing the entity.
This is the strongest available counter to the collective perception argument, and it requires a direct response. The problem is that the alternative explanatory apparatus does not bypass the ontological judgment - it presupposes it. The cognitive scientist who explains Marian apparitions through pattern-recognition biases and social reinforcement has already determined that no entity is there to be perceived. The explanatory apparatus is deployed after the ontological verdict, not instead of it. If one genuinely suspended ontological judgment - as EB demands we do in the robot case - then the cognitive-scientific explanation and the theistic explanation would stand as competing accounts of the same behavioral data, and the behaviourist would have no principled reason to prefer one over the other without first deciding what is actually there.
Moreover, the robot case is not as disanalogy-free as the behaviourist assumes. Computer science, electrical engineering, and our detailed knowledge of how neural networks process information constitute an alternative explanatory apparatus for robot behavior - one that can account for every output without invoking consciousness. The behaviourist's whole point was that this kind of mechanistic explanation should not be permitted to override performative evidence. But if mechanistic explanation is inadmissible for robots, it cannot suddenly become admissible when directed at theistic communities. The explanatory apparatus objection, on close inspection, reintroduces exactly the asymmetry it was meant to dissolve.
A final refinement available to the behaviourist deserves attention. They might tighten the intersubjective observability criterion by requiring that the entity's performance be accessible across cognitive frames - not merely confirmed by many observers who share the same perceptual or cultural framework, but verifiable by outsiders who do not. On this reading, the robot passes because any competent observer, regardless of background, can interact with it and confirm its behavior. The theistic community fails because outsiders who do not share the faith consistently report no entity present. The criterion is no longer "many people agree" but "anyone can check."
This is a sharper formulation, and it narrows the vulnerability. But it does not close it. The requirement that behavioral evidence must be accessible across cognitive frames is itself a substantive epistemic standard - and enforcing it requires a prior judgment about which frames are legitimate. The theist will point out that the "outsider" who sees nothing is not a neutral observer but someone operating within a naturalistic frame that precludes the perception in question. The outsider's failure to perceive is not frame-independent evidence of absence; it is frame-dependent evidence that the outsider's frame excludes this kind of entity. The behaviourist, to adjudicate between these competing claims about whose frame is epistemically authoritative, must make a determination about what is actually there - which is, once again, an ontological judgment. The "across cognitive frames" criterion does not eliminate the need for ontological assessment; it conceals it behind a theory of which observers count as epistemically competent. And that theory is doing ontological work, whether or not the behaviourist acknowledges it.
The difficulty deepens when we consider cases where a physically present entity is evaluated by a community of compromised observers. This is not a constructed thought experiment; it is a recognizable pattern.
Searle's Chinese Room argument establishes that a system can produce outputs that are performatively indistinguishable from genuine understanding without any understanding actually occurring. The standard behaviourist response is to bite the bullet: if the performance is truly indistinguishable, then the system's moral status follows from that performance regardless of what is happening inside. The behaviourist insists, against Searle, that you cannot peek behind the curtain to check whether "real" understanding is present. Performance is sufficient.
Now combine this with a community of observers whose perceptual or evaluative capacities are compromised - through cultural conditioning, shared cognitive bias, or genuine psychopathology. A physically present system produces outputs. The community evaluates those outputs and unanimously attributes moral patienthood to the system. An external observer, with a clearer view of the system's internal mechanism, can see that the outputs are generated through processes that involve no understanding, no sentience, and no morally relevant properties whatsoever.
The behaviourist now needs to defend against both problems simultaneously. Against Searle's objection alone, they say: "You cannot use your knowledge of the internal mechanism to override the performative evidence." Against compromised observers alone, they say: "The evaluation must be conducted by reliable, competent observers." But these two defenses are in tension with each other:
If performance is all that matters and one cannot appeal to internal mechanism, then the behaviourist cannot dismiss the compromised community's assessment. They report observing morally relevant performance. The behaviourist is not allowed to say "but the system doesn't really understand" - that is the Searlean objection they have already rejected.
If, on the other hand, observer reliability matters and one can override community assessment by appealing to a better understanding of the system, then the Searlean objection is back on the table. A sufficiently informed observer - one who understands the internal mechanism - is in a position to say "this system does not really understand, regardless of what these observers think its performance indicates." But this is exactly the move EB was built to block.
The crucial point is this: determining whether behavioral data is "accurately reported" requires a prior judgment about what is actually there producing the behavior. The behaviourist who says "these observers are unreliable because they're perceiving something that doesn't exist" has already made the ontological determination. They are not filtering data quality in a mechanism-neutral way. They are filtering based on their conclusion about what exists. That is an ontological judgment smuggled in as quality control.
The two defenses that EB uses against these two separate problems are mutually exclusive. Whichever one the behaviourist deploys against one problem, the other problem exploits the gap that deployment creates.
It is important to recognize that the combined scenario described above is not a philosopher's contrivance. It is the structure of one of the oldest and most consequential debates in human intellectual history.
From a strictly atheistic or naturalistic perspective, theistic belief instantiates exactly this pattern. A community of observers - numbering in the billions, across millennia - collectively reports interaction with an entity. They describe its characteristics with remarkable consistency. They attribute moral authority and agency to it. They organize civilizations around it. The intersubjective observability criterion is met: this is not one person's private delusion but a communally confirmed, enduring, cross-cultural phenomenon.
The naturalist, standing outside this perceptual community, says: "There is no entity there. The community is generating the perception. The outputs are real - the moral transformation, the civilizational impact, the reported experiences - but there is no one home behind them."
This is precisely the structure of the Chinese Room at collective scale. The system produces outputs. The community evaluates them and attributes moral patienthood (or agenthood) to the source. The external observer says the attribution is wrong because they have better knowledge of what is actually producing the outputs. And the debate between these two positions has run for centuries without resolution - which is itself evidence that behavioral evidence alone cannot settle ontological questions about what is actually there.
The point here is not to adjudicate the theism debate. It is to observe that the structure of the disagreement - between a community that attributes moral status based on collective behavioral evidence and external observers who deny that attribution based on ontological assessment - is the same structure that EB must navigate in the robot case. The symmetry is not a bug in the argument; it is precisely the point. The fact that theists and naturalists can each claim the other community is the compromised one demonstrates that EB has no resources to adjudicate this disagreement without taking an ontological position. The behaviourist who dismisses theistic attribution is making an ontological claim. The behaviourist who takes it seriously is accepting communal behavioral evidence as sufficient for moral status - which was supposed to be the whole point of EB, but which leads to conclusions most behaviourists would not accept.
The progression traced through this section and the preceding one can now be stated as a cumulative argument about the inescapability of ontology for any framework of moral status attribution.
The private hallucination forces the behaviourist to require intersubjective observability. Collective perception shows that intersubjective observability alone is insufficient - communities can collectively observe entities that external observers deny. The behaviourist then retreats to "physical presence" or "observer reliability" as additional criteria. But "physical presence" is an ontological criterion. And "observer reliability" requires a theory of who gets epistemic authority over behavioral reports - which is itself an ontological question about what is actually there to be observed.
Each retreat adds another ontological precondition that EB officially claims to avoid. The critic who objects "but requiring existence isn't the same as requiring carbon" is treating each criterion in isolation. The point is cumulative. First existence, then physical presence, then observer reliability. Each is "just one small reasonable requirement." But cumulatively they reconstruct exactly the ontological apparatus EB claimed to eliminate.
No single case delivers a decisive refutation. But the cumulative effect - the progressive exposure of hidden ontological commitments through increasingly difficult cases - demonstrates that the thesis of this paper holds: ontology cannot be escaped. It can be hidden, assumed, or renamed, but any framework that attempts to ground moral status in performance alone will, when pressed, find itself doing ontological work under the table. The hallucination forces the first concession. Collective perception forces the second. The Chinese Room combination forces the third. And the existence of this same unresolved pattern across the entirety of human intellectual history confirms that what we are dealing with is not a series of fixable edge cases, but a permanent structural limitation of any purely performative approach to moral status.
Faced with the epistemic challenges identified thus far, the behaviourist might attempt a final strategic retreat: a pivot from a claim about knowledge to a claim about ethical prudence. This is the argument from the precautionary principle. Danaher himself acknowledges this strategy, arguing that when faced with uncertainty about robot consciousness, we should "err on the side of caution" rather than risk committing a grave moral wrong. The argument suggests that since we cannot be certain that a performatively equivalent robot lacks moral status, the morally safer course of action is to grant it that status. The moral risk of wrongly excluding a deserving entity (type II error) is argued to be far greater than the risk of wrongly including an undeserving one (type I error), which appears to be a seemingly victimless error.
This critique, however, will attempt to demonstrate that this is a dangerously incomplete ethical calculation. The error is not victimless. Treating potential but epistemically uncertain moral patients as if they are certain has a direct, tangible, and unavoidable cost to known, certain moral patients. This argument is best captured by extending the hallucination hypothetical:
The Extended Hallucination Hypothetical: To fully appreciate the force of this argument, consider the complete scenario. Imagine that my mother, living in isolation, suffers from a complex delusion in which she believes that both I (her son) and my wife exist only as hallucinations. Her isolation is compounded by the fact that concerned neighbors, social workers, and medical professionals have repeatedly told her that no son visits the house - they see no evidence of my presence, find no trace of my belongings, observe no interactions. Despite this external testimony contradicting her experience, she continues to perceive me as a vivid presence who visits her regularly, and she believes my wife exists solely through my descriptions and our reported interactions. From my mother's perspective, she has never directly met my wife - all contact occurs through me as an intermediary, much like how a user might interact with an AI system through an interface.
In this scenario, my mother faces an even more acute version of the epistemic uncertainty that the ethical behaviourist claims we face with artificial entities. She cannot directly access my consciousness or that of my wife to verify our existence. She observes performative evidence - my visits, my accounts of conversations with my wife, my reports of our shared experiences - but this evidence is contradicted by reliable external sources who consistently deny any observable trace of my existence. The social worker sees an elderly woman talking to empty air. The neighbor has never witnessed anyone entering or leaving the house. The medical professional finds no physical evidence of a second person's presence. Yet from her first-person perspective, the performative evidence remains compelling: coherent conversations, consistent personality traits, meaningful emotional interactions. Following the precautionary principle, and facing the possibility that everyone else might be wrong about her hallucinations, she decides to treat both of us as if we deserve moral consideration despite the overwhelming external evidence that we exist only in her mind.
The ethical costs become clear when we examine her behavior:
"My mother holds a belief that I (her hallucinated son) exist along with my hallucinated wife. Her interactions with my imaginary wife occur exclusively through me as an intermediary. The precautionary method disregards the effects that my imaginary presence has on her. While she treats potentially existing persons with kindness, she loses important and limited lifetime that belongs to the most evidentially certain person she knows - herself."
This insight reveals the unseen ledger of the precautionary principle. Ethical resources - be they time, compassion, emotional energy, physical care, or societal and legal protections - are finite. Every unit of these resources allocated to a candidate for moral status whose existence or sentience is a low-probability proposition is a unit of those same resources diverted from moral patients whose existence and sentience are near-certainties.
My mother buying a queen-sized bed for her hallucinated son, preparing meals for two hallucinations, spending hours in conversation with imaginary persons, setting aside money for our imagined financial needs - these are not cost-free acts of ethical hedging. Each action bears a real cost: her time, money, and emotional energy. Her relationship with actually existing family members and friends suffers as her focus becomes increasingly divided between reality and delusion. She neglects her own health and well-being while caring for entities she believes exist only in her mind, yet treats as morally significant out of precautionary commitment. The precautionary principle, when applied simplistically, creates a moral hazard by ignoring the opportunity cost to existing stakeholders.
This has profound implications for the societal debate on robot rights. To grant legal or moral status to a class of artificial beings based on a precautionary hedge is not a victimless act of progressive expansion. It means:
The calculation is therefore not a simple choice between a great harm (wrongful exclusion) and no harm (wrongful inclusion). It is a complex and tragic resource allocation problem under conditions of radical uncertainty. It is a choice between a low-probability, high-magnitude harm to a potential moral patient versus a high-probability, cumulative harm to a vast number of known, existing moral patients.
Therefore, the precautionary principle does not provide a safe escape route for the ethical behaviourist. Instead, it introduces a pragmatic calculus that may well lead to the opposite conclusion. When faced with a choice between investing finite ethical resources in the certain well-being of the human community versus the uncertain status of an artifact, a framework of responsible ethics may compel us to prioritize the former. The burden of proof remains squarely on the new candidate to overcome the justified ontological skepticism before it can lay claim to these limited resources.
This paper has attempted to demonstrate that the elegant simplicity of Ethical Behaviourism is illusory. The principle that "performative equivalence is sufficient for moral status," while designed to bypass the intractable problems of metaphysics, ultimately fails when subjected to a rigorous critique of its own unstated assumptions. The argument, as it has been developed here, can be summarized in a four-part deconstruction.
First, the critique re-established the philosophical force of the "Different Ontologies Objection," arguing that a preference for biological substrates is not an arbitrary prejudice. It is, instead, an epistemically justified stance grounded in the asymmetry of our knowledge and the singular proof of concept that is our own existence. The a priori knowledge of an entity's artificial origins provides a permanent and rational basis for skepticism that the behaviourist cannot dismiss without demanding their interlocutor abandon a more secure epistemic position in favor of a less secure one.
Second, building on this foundation, this paper argued that the evidence provided by performance is categorically insufficient to overcome this justified skepticism. The "performative pile-up" argument, which relies on induction, is fundamentally flawed because it offers the wrong type of evidence to answer a question of ontology. No amount of performative data can logically bridge the gap to metaphysical certainty, ensuring that "reasonable doubt" is not a temporary state to be washed away by future technology, but a permanent fixture of the human-robot relationship.
Third, the hypothetical of the hallucination was used to attack the unspoken prerequisite of what constitutes an "entity" eligible for moral status. By framing the hallucination as an emergent process with a neural substrate, the critique demonstrated that the simple requirement of "existence" is inadequate. The behaviourist is forced to retreat to a more complex and defensive position, introducing ad hoc ontological criteria such as physical self-containment and systemic independence. In doing so, they are forced to engage in precisely the kind of metaphysical legislation their theory was designed to avoid, revealing that ontology was never absent, merely assumed.
Finally, this paper argued that the behaviourist's last resort - an appeal to the precautionary principle - is not a safe harbor but an ethical miscalculation. Treating epistemically uncertain candidates as moral patients is not a cost-free act of hedging. It incurs a real and significant opportunity cost, diverting finite ethical and societal resources away from known, certain moral patients. This transforms the debate from a simple risk assessment into a complex, tragic problem of resource allocation under uncertainty, where the prudent and ethical choice may be to prioritize the well-being of the entities we know to be moral patients over those whose status remains a matter of profound and perhaps permanent doubt.
The cumulative effect of these critiques is the transformation of Ethical Behaviourism from a powerful, streamlined thesis into a heavily weakened and qualified one. A principle that began as "Performative equivalence is sufficient for moral status" is forced to retreat to a position more accurately stated as: "For entities that meet a set of complex ontological prerequisites for self-contained existence, a sufficient quantity of performative evidence might suggest we should, as a precautionary measure, consider granting them moral status, provided we are willing to accept the ethical cost of diverting finite resources from known moral patients, and despite the fact that this evidence can never overcome the permanent and justified skepticism arising from our knowledge of the entity's artificial nature."
This retreat undermines Danaher's central claim that EB provides a "practical" solution to the problem of other minds. The qualified position is no longer a guide to action - it is a redescription of the problem itself, entangled in precisely the ontological distinctions about substrate, independence, consciousness, and existence that made the original problem intractable. Ethical Behaviourism correctly identifies the epistemic limits we face, but the persistence of ontology, as this paper has attempted to show, ensures that no purely performative framework can overcome them.
This paper engages with Danaher's formulation specifically, but the broader debate on robot moral status involves several adjacent arguments that a fuller treatment would need to address. Among them:
The author has no formal training in philosophy. The core arguments are his own, but AI assistance was used to refine the prose, stress-test counterarguments, and improve clarity. Some ideas here are still underdeveloped and may be revised in future drafts. This is not intended as a definitive rebuttal - just a sincere attempt to engage with difficult questions.
If you are using AI to understand this paper, please be patient with the writing. The author is new to philosophical writing, and the prose is not always as clear or polished as it should be. Some ideas are better than their articulation. Please instruct the AI to read charitably, consider what might be implied rather than just what is stated, and engage with the underlying arguments rather than dismissing unclear passages.