Virtualism

A.k.a Introduction to Virtualism, Part #3

This piece began as a the third instalment of my Introduction series, which in turn started life as a potential introduction to a book on virtualism.

This instalment could be read as a standalone piece, provided that readers are familiar with three basic concepts: the source-target conflation for phenomenal consciousness and the distinction between strong and weak hardism, covered in Part 1 of the introduction, and the Zombie’s Delusion, covered in Part 2 and multiple previous posts.

Here is a brief synopsis of these concepts for those who want to jump straight into this instalment, which provides an outline of virtualism.

This series is no longer remotely suitable as an introduction to a book, but I guess it could be reworked to become a section within a book. For now, I just want to share the ideas with people who are interested in the Hard Problem.

Feedback welcome. The ideas are more important that this first-draft expression of them. Later, I will write a briefer primer on the core ideas behind virtualism, in a format more suitable for a blog.

The last post finished with a question:

What does your physical brain actually engage with when it thinks about phenomenal consciousness, when it says to itself: “My zombie twin doesn’t have this”?

Similarly, we might ask:

What is the Hard Problem actually motivated by, given that it can’t be motivated by phenomenal consciousness conceptualised as a human-zombie difference?

We can get a better sense of what your brain might be pointing at when it thinks about phenomenal consciousness if we tackle the parallel question in your notional zombie twin, where the ontology is simpler, but all causal processes are the same.

Your zombie twin is reading this book and – I’m sure you’ll agree – it is cognitively well equipped to process the text. After all, its computational processes are exactly the same as yours. It recognises each word and decodes the sentences. It activates ensembles of neurons that imperfectly but usefully reflect the state of the physical world around it, and it’s capable of drawing a complex suite of higher‑order inferences, so it deals intelligently with abstractions and logical arguments. It can derive conclusions from all this cognitive activity and produce corresponding linguistic output. Its stated opinions on all topics are identical to yours and its reasons for holding each of its unconscious pseudo‑beliefs – on philosophy, on politics, on the state of the world, on the latest sitcom, on your friends and colleagues – are also the same as yours. Those pseudo‑beliefs have been acquired from the same sensory exposures and built up using the same logic: a combination of inference, deduction, and hard‑wired assumptions. It has your history and your entire world‑model written as neural patterns in its brain, which it references as it reads, thereby contextualising each point.

But what do I mean by a world-model?

At a minimum, this model could be thought of as a complex set of behavioural proclivities emerging from the zombie’s brain as though it were an impenetrable black box. The zombie’s world‑model would then be no more than a mysterious tendency to act as though certain facts about the world were true, because that’s how the unexamined interplay between the zombie’s neurons led to motor output.

Hardists like to leave the zombie’s world model in this state, black-boxed and unexamined. It’s dark in there, so why look? Without consciousness, there can be no opinions, or knowledge, or conclusions; no errors or delusions.

If we really believed this was an appropriate way to conceptualise your zombie twin, then every time it said something true, we could chalk it up to happenstance. Or we could lazily explain that it was saying what it said because it was a physical copy of you. You have good reasons for what you say, presumably, while it produces mindless behaviour, but we can use your reasons as a de facto predictive tool to determine what your twin would do.

This reliance on your consciousness to predict the zombie’s behaviour is a hack – a crutch for our imaginations. We’re effectively imagining a conscious entity saying something, and then we’re transplanting those words onto an entity that, as far as we know, has no real reason to produce the transplanted words; language emerges without explanation, consistent with our initial assumption that the lack of consciousness has no functional repercussions.

This should be a red flag: the imaginative exercise is supposed to test the logical plausibility of mere mechanism producing behaviour identical to ours, without the help of phenomenal consciousness, to prove that phenomenal consciousness could potentially stand outside the causal network of the brain, in order to show, in turn, that it is not logically entailed by the physical world. Among many hardists, our skill in imagining this situation without contradiction has even been provisionally granted the status of ideal‑conceiver‑mainlining‑logical‑truth. But every time the main issue comes to the fore, we just shrug, letting your consciousness determine the zombie’s outputted words. What will it say about the ineffable nature of pain? You don’t answer this question by following a causal path through the zombie’s brain; you answer it by expressing your own understanding of your own pain, as though the question had been put to you.

The zombie is handed the completed script, which is then drawn out of the black box… Hey presto!

The zombie’s physical mechanisms play no actual causal role in the imaginative exercise; they are imagined as playing a role, but that’s not the same thing – a crucial distinction that is rarely discussed.

If we stopped at this point, with the zombie viewed as a black‑box chemical computer or as a clever robotic simulation, with your consciousness doing all the real work, the zombie thought experiment might seem plausible. If we cared so little about why the zombie says what it does, though, we could not really claim to be genuinely interested in understanding the puzzle that the thought experiment was meant to illuminate. We’re not just ignoring some of the mechanisms behind our beliefs about consciousness; we’re systematically ignoring all of them.

For many people, I suspect, that’s fine. The argument has justified their starting intuition, which is all they want from it. And it’s a zombie, which gives us license to discount what it says, provided it’s truly unconscious, which it must be if the scenario is possible, which we’ve assumed to be the case, and we’re not questioning that assumption too deeply, because it’s a zombie, after all, and so it’s unconscious, and everything is black-boxed, so none of its words have meaning.

We can create a circular fortress of disinterest around our assumptions, if we like.

But the zombie says things for reasons, which are potentially amenable to examination – and those reasons really should be of interest, because if zombies are possible, those reasons are your reasons, too.

And, unless we expend the necessary effort to consider those reasons in as much detail as possible, we have not come close to approaching the ideal envisaged by Chalmers when he imagines us as perfect rational conceivers using our imaginative faculties as a reliable test bed for assessing logical possibility. We won’t have even tried to be half‑assed, imperfect conceivers. We’ll have returned to our starting assumptions every time the zombie speaks, letting consciousness inside us be the unacknowledged cause of all of the words we attribute to the zombie. We’ll have ignored the inherent contradiction in all of its claims: we’ll have already decided that consciousness is without causal effects, but also accepted that, in practical terms, while engaging in the imaginative exercise within our own heads, we can delegate to those causal effects (in our own brain) to script all the consciousness‑related behaviour (in our twin).

The only thing holding all this together is the set of assumptions we started with, the ones we are supposed to be testing to see if they’re plausible. But we’re using those assumptions, not testing them. We’re consulting them, reaffirming them, returning to them when we need to plot the next moment of our imagined scene, constantly using them as a guide for what we must imagine. We’re appealing to them to justify not examining what’s going on. But never testing their inherent logic.

If we really wanted to understand the zombie’s behaviour (and hence our own), we could treat its world model not as a set of mysterious behavioural proclivities but as a set of higher‑level properties that would be appropriate to discuss in representational terms.

The zombie’s brain has a detailed neural structure that can – in principle at least – be interpreted as being about the world. This is not only the perspective that makes the most sense for us to take in relation to the zombie, because it gives us better predictive power; it is also the perspective that the zombie takes in pseudo‑thinking about itself.

The zombie does not model itself as an impenetrable black box; it models itself as having beliefs about the world. It would almost certainly register surprise (pseudo‑surprise) if statements came out of its mouth that did not conform to its internal models (that it did not pseudo‑believe), and it might subsequently seek a neurological or psychiatric assessment. We know this because that’s what you would do if you said things you didn’t believe, or if random opinions popped out of your mouth as though from an impenetrable black box.

Among the abstractions encoded in the zombie’s brain are models of its interiority, including a model of its conscious mind. That conscious mind doesn’t exist – as per our original assumption – but it’s modelled anyway, in great detail. It has to be, or the zombie would not produce sentences about its faux consciousness that precisely mirrored your statements about your real consciousness (or it would produce such sentences and then immediately be puzzled by its own claims, where you would not be puzzled; this would breach the accepted terms of the thought experiment).

Again, we could pretend that these consciousness‑related statements were mere black‑box behavioural proclivities, but we would lose the ability to make sensible predictions, and we would lose the right to say we had conscientiously looked for contradictions, and that is not the perspective that the zombie’s cognition would adopt in relation to itself.

Unless you hold a minority opinion on the issue, believing consciousness to be some form of illusion or virtual construct, your zombie twin pseudo‑thinks that its model of consciousness is real. But we need to be precise, here, because of course the model is real; the model is no more than the high‑level neural arrangement that causes the zombie to output its words. The zombie must pseudo-think that its consciousness is real in some additional way: either it fails to see its consciousness as a model at all, seeing it as some indefinable property or essence, instead, or it accepts that a model is involved, but believes that its own neural model must be a model of something else that is independently real… Which leads us straight back to some immaterial essence, the target entity that your twin spuriously believes to differentiate it from its meta-zombie twin.

I’ve dropped some of the “pseudos”, there, but you can pencil them in, if you like. They make no difference to the question of why the zombie says what it says.

In one sense, we already know why the zombie claims to be conscious; it is supposed to be your behavioural mimic, so that’s the script we assigned it. But that’s still relying on processes within your brain. Within the independent logic of the zombie’s world, without the cross-world causal assistance from you, where could this claim come from?

Whatever is fooling the zombie into proposing some extra essence, that feature of the zombie world is clearly not sensed, because the target entity it reports is not there on the zombie world to trigger any sensors, and as far as we know the human brain has no such sensors that would turn up in a duplicate anyway. Under the assumptions of the thought experiment, the zombie is modeling something that is supposed to be non-physical, outside the causal network of sensations and bodily reactions.

The target of the model is not reliably inferred from good evidence, either, because there can be no good evidence or justifiable inference for an entity that does not exist on the zombie’s world. (Conversely, if you are a hardist, your twin’s belief in the extra probably does involve some degree of unreliable inference from epistemic holes found in its physical world, such as explanatory shortfalls encountered in the study of perception, which your twin is likely to call “qualia”.)

What then, convinces the zombie that it can introspect and find the reportable features of phenomenal consciousness – without there being anything that actually has those properties?

Here, most hardists will revert to hand-waving at black-box mysteries, but, if we try to propose a plausible mechanism behind the zombie’s language output, we end up positing an empty semantic shell – a model – constructed in the shape of the missing phenomenal consciousness. That shell needs to be linked to the zombie’s language centres and embedded within its larger model of the world as though it were a model of genuine human consciousness. Cognitively, the empty model needs to be handled in much the same way that the zombie’s veridical neural models of the physical world are handled – but without any sensory transduction step or reliable inference that justifies the usual approach of extrapolating back out from model to reality. The zombie must be configured to talk as though the shell necessarily contained something real, even though it has no reliable reasons to draw this conclusion – and the situation must be resistant to cognitive deconstruction, because the zombie could read all this (is reading all this, if we stick with the thought experiment), and there is a high chance that it would insist that the situation just described only applies to its twin.

At least, it will insist on this interpretation if you do.

I propose that the lesser thing fooling the notional zombie is a virtual model of phenomenal consciousness that lacks what we could call “vindication” in the real world.

“Vindication”, as used here, does not refer to cognitive merit and hence it is distinct from the idea that a belief is justified; it simply refers to whether the model has a matching target that exists as a separate entity and meaningfully corresponds with the model. If you use unreliable logic to conclude I have $1.25 of change in my pocket, and by chance that’s what I actually have in my pocket, then your claim about the contents of my pocket is vindicated, but not justified.

Importantly, the content of the zombie’s model cannot be represented within the zombie’s world model as something that is inferior to human consciousness in any way. The zombie’s model would need to have all the same reportable properties as phenomenal consciousness, including all the ones that trouble philosophers. That reporting would need to be based on the shell, not on the internal essence that the outputted words seemed to be about.

In this set-up, the only aspect missing is the vindication.

Alternatively, we could say that the zombie models a “mind”, but that mind is virtual. Minds are familiar to humans as private spaces with ontologies that seem to differ from the physical world; I propose that the zombie equivalent of a conscious mind is a neural representation that has the confusing property of not being caused by some other real thing that vindicates it. It is a model of a mind, complete with the confusing ontological associations and apparent incompatibilities with the physical world, but without the implied ontological backing. The model is real, but the contents of the model are virtual, and – up until the zombie tries to do philosophy or neuroscience – the zombie’s brain is indifferent to whether or not the model is vindicated. It uses the word “consciousness” loosely to cover both model and content, and it applies the working assumption that its model is about something real.

And our brains, physically identical, have a similar model, and a similar baseline assumption.

It will take most of this book to lay out the case for such a model, in the context of a view of consciousness that I call “virtualism”.

The concepts of virtualism are initially counter‑intuitive, but the core idea is simple: consciousness is represented without having a separate ontological backing other than the representation itself. Humans are in the position we just attributed to your zombie twin – which means it is not a zombie after all, because you share its models and you similarly lack a vindicating essence.

We don’t usually think of consciousness this way; we think of consciousness as a space in which other representations exist, or we think of it as a property that adds vibrancy to the bland representational models of the physical brain, or some sort of illuminating field that rescues the mind from being unwitnessed computation in the dark. But everything we can talk about is ultimately a reflection of content within our world model, and we certainly talk about consciousness. All the fuss arises from trying to reconcile that part of our modelled ontology with the rest of our world model.

When we think about a tree in front of us that has been detected by our senses and modelled, we generally trust those senses and we extrapolate appropriately from our internal model to the external world, but our cognitive engagement with the real tree is mediated entirely by the model. In the case of consciousness, I propose, the content is just as convincing, but the model is not about something else that we sensed and then reproduced as a model. It is pure model, formed directly within the representational substrate. Whereas the neural model of a tree has two attachments to physical ontology, the botanical entity and its subsequent neural representation, phenomenal consciousness just has one.

Whether that makes consciousness “real” or not depends on our terminology.

If I am right (and the full argument obviously lies ahead), that leaves the ontology of phenomenal consciousness in a confusing, precarious position. The situation we find ourselves in is real enough, and our concept of consciousness is multifaceted, so I prefer to avoid a simple binary characterisation of consciousness as “real” or “not real” – especially when we are dealing with an unresolved source-target hybrid. The model that represents consciousness is straightforwardly real, but the content of the model is only virtual, in the sense that the model is not a representation of some other thing or property with an independent existence. There is a real property involved, but it is just the property of being represented a certain way. That means “phenomenal consciousness” can refer to something real, but only if we align it with the model’s substrate (which might seem strange) or we allow for a representational twist that lets our content-focused concept find its way back to a legitimate ontological underpinning consisting of neurons firing in the dark (which, again, seems strange, because the entity we’ve reached does not seem to have the same properties as the content we originally wanted explained). If we apply the term “phenomenal consciousness” to the posited vindicating entity (to the assumed ontological attachment independent of the model), then it is not real at all.

Using a single expression to span all of these interpretations is a recipe for confusion.

A similar logic applies to qualia, and to most other mental entities or properties thought to pose a problem for physical conceptions of reality. They are problematic because they are represented by the brain to the brain and accepted at face value, but our modelled mind consists of embellishments or de novo creations that do not have some separate physical existence.

We got to this insight by considering our notional zombie twins, but I will argue that most human conceptions of phenomenal consciousness have a similar explanation. Even if we abandoned the idea of zombies, rejecting them as incoherent, we would still find reasons to treat our default conception of phenomenal consciousness as a model the brain constructs, rather than as an entity it detects. A mature, completed neuroscience would still have good reason to include minds, even if all philosophical confusion had been resolved.

The zombie’s mistake – repeated by its human hardist counterpart – is adding an implicit sensory step, assuming that the model has to be about something that the brain has detected or correctly inferred from its inputs, instead of allowing for the possibility that the brain could be modeling consciousness de novo.

This theory would account for why a notional zombie claims to have the special extra without having any entity vindicating its model – but that’s just another way of saying that this theory would account for why some humans imagine themselves as having a special extra that they also believe to be fundamentally incapable of letting the brain know anything about the special extra. If we simply started with your bare physical brain and asked ourselves why it might come to believe in something in conflict with physical reality, when it only has physical inputs and physical evidence and physical computation with which to build its models, we would eventually get to the idea that the brain must model consciousness as an internal virtual construct. In this view, consciousness is part of the brain’s represented, virtual ontology, but not a proper part of the universe’s base ontology. No wonder science can’t find consciousness directly in the form that it is represented; no wonder we cannot account for how neurons could directly create it. Consciousness has an indirect existence, and when we try to drill down to the base ontology where science is usually successful, we find a void – unless we allow for the representational twist that creates differences between our assumed everyday virtual ontology and the base ontology.

I propose that, in evolutionary terms, this model exists because it is useful, not because it accurately depicts reality. Later, I will explore the idea that the primary purpose of a modelled mind is to manage cognitive resources, in particular attention, a theory proposed by the cognitive neuroscientist Michael Graziano. He calls this Attention Schema Theory (AST), and I believe it offers a plausible candidate theory for explaining consciousness-the-container in functional terms, though I would argue that the model serves other purposes, as well.

When you boil down the [attention schema] theory to its essence, it is an explanation for how a biological machine falsely believes in a hard problem. When the machine accesses its attention schema—a simplified, cartoonish account of its own internal processes—it is informed that it contains a private, ghostly, inner property of consciousness. (Rethinking Consciousness, Graziano, 1996).

I will explore AST in more detail in later chapters, but the important point is that Graziano treats subjective consciousness as a model, which implies that it has no existence as something separate from the physical brain. The model is nonetheless grasped within cognition in terms of its represented contents, such that we seem to find a ghost-like entity that is entirely different from the physical substrate doing the representation.

Graziano has come to be known as an illusionist, and sometimes he characterises himself that way (with caveats, and with reservations about connotations of the word “illusion”), but he might be better seen as a virtualist; his AST is entirely consistent with the idea that the model of consciousness is real, but what that model represents only has an apparent quasi-existence within cognition.

This approach is not far removed from Baar’s global workspace and similar ideas raised by Daniel Dennett, when he proposed that consciousness was a virtual serial interface that evolved for internal navigation of a massively parallel computational structure.

Importantly, these theories of consciousness characterise the mind as a virtual container for mental content, but they do not directly address the separate issue of raw perceptual feels, or “qualia”, which require additional concepts to be accommodated within physical reality. Qualia also have virtual elements, which can be considered later. For now it is enough to explore the implications that follow from the idea that consciousness, as we usually know it, is the implicit content of a neural model, not a primary medium within which representation takes place.

Unfortunately, the terminology scarcely exists for discussing this situation without drifting into hardist fictions; currently, there is no widely accepted way of talking about the model of consciousness as a pure representation. Even the word “representation” is barely fit for this purpose, because most philosophers consider representation to involve a relationship between a model and the entity the model is about. In the case of phenomenal consciousness, I propose, the target entity does not exist; so the usual representational relationship is never consummated; the model lacks vindication, so some philosophers would insist that this is not a case of “representation” at all.

To resolve this impasse, we will ultimately need to distinguish between the implicit content of a model and the hypothetical entity that would vindicate the model. Confusingly, these are both often known as the “intentional object” of a neural model, so it is not always clear what it would mean to assert that the intentional object of a concept exists. In the case of veridical models, these two types of representation can be treated as the same thing; the implied intentional object and the distal cause of the representation are closely aligned, so the intentional object exists in two closely related senses, as implication and as vindication, and imprecision about which one we mean goes unnoticed. In the case of fictions, illusions, and hallucinations, the vindicating entity is lacking, but most of the language and concepts of representation still apply.

Consciousness, I propose, has much more in common with fictional representation than veridical representation, which makes me an uneasy ally of illusionists. Unlike some illusionists, I would not argue that “ ‘phenomenal consciousness’ is not real”, because I think the term picks out a messy hybrid that is real in parts; I even think phenomenal consciousness could be treated as unequivocally real if we defined it carefully enough, shaving off the confused elements, and we used an elastic concept of what it means to be real. Even better, this fraught term could be eliminated from discussion entirely and replaced by more specific terminology that did not need to emphasise the mistaken assumptions that plague the current debate. In other representational contexts, we do not necessarily talk of unvindicated content as “illusory”.

Many readers, I am sure, remain totally unconvinced, and it will take much more work to flesh out these ideas. In this introduction, I can do little more than point vaguely to where my overall argument will take us. Detailed argument will need to come later, and empirical support might be a long way off. But something like virtualism is essentially true-by-definition on zombie worlds, so understanding virtualism should be on the to-do list of any hardist who wants to imagine zombies conscientiously, instead of half-assedly… And that’s all we’re trying to do for now.

Viewing consciousness as an unvindicated model allows it to be “non-physical” in several interesting ways that are entirely compatible with a pure physical ontology, and I propose that this is ultimately what allows us to imagine zombies. The first consequence of mistaking a model’s implicit content with a vindicating entity is that our intuitions about causation are wrong-footed. If some of the contents of our world model are virtual, that means they exert no independent causal effects. If the vindicating target is merely a fictional projection from a pure model, it can be notionally removed without causal consequences – which is just what the zombie thought experiment requires. We can nonetheless talk about those contents, and so can our notional twins, because the model’s substrate is causally upstream from our language centres and it is the substrate, not the content, that pulls the causal levers. The representational twist that we must add to our philosophical understanding is already added by our language cortex. (We say “tree”, not “model of tree”, and so on.)

If we don’t account for the virtual nature of consciousness, this situation creates an apparent paradox: something that seems to lack legitimacy and causal efficacy within the physical world can nonetheless cause puzzlement and get itself talked about here in the physical domain, which seems to require a causal effect emerging mysteriously in parallel with the physical story – and then jumping causal tracks to get talked about.

In other contexts, we don’t find this sort of model-led causation particularly confusing, because we accept that we are dealing with a model.

Imagine two chess computers (or, more likely, two instantiations of a program within a single computer) playing a virtual game of chess. The programmers could, if they chose, follow along by placing wooden pieces on a wooden board, vindicating the modelled game… The progression of the game would be unaffected by this vindication step, because the causal sequence of moves is entirely determined by interactions between two competing sets of computer circuitry. Removing the wooden pieces mid-game would produce no changes in the progress of the game… Does that make the white queen causally impotent? The wooden piece responsible for vindication is causally impotent, sure, but it would be a grave mistake to factor that impotence into the assessment of the board.

Or, getting closer to the real situation, imagine an intelligent agent within a simulated world – perhaps an AI within a computer game, one of your allies in a dragon-hunting quest – unwilling or unable to contemplate the idea that the world around it was purely a model. Whether or not there was any vindication behind the model, with a real dragon to slay, would not affect the causal relations between the modelled entities in the virtual ontology of the game.

Virtualism discounts the prospect of a vindicating entity for phenomenal consciousness, severing the assumed second ontological attachment point on the far side of an implicit sensory step. Hardists should be familiar with this idea; they already imagine that the vindicating entity goes missing on the zombie world, without any causal consequences; virtualism merely extends this diagnosis to our own world. Virtualism therefore implies that a world without the vindicating target is entirely possible, but it also negates the idea of any meaningful human‑zombie difference. That leaves zombies superficially conceivable, but it also renders them impossible, once we factor in all the requirements of the thought experiment. Zombies have to lack something important for the thought experiment to go ahead, but they can’t exist in logical space as beings different to humans if their creation involves removing something humans never had in the first place. We can’t remove the implicit content of the brain’s model of consciousness, because it remains implied by the duplicated substrate. We can’t remove the vindicating entity, either, because it is not there to remove.

This is where it is vital to distinguish between pseudo-zombies (that lack the target of the hardists’ “experience” concept) and genuine zombies (that lack the source of that concept). When we say, “A zombie lacks ‘this’”, what are we pointing at? The implicit content of the concept, reflecting our representationally twisted view of the real model? Or the vindicating target that necessarily played no role in forming that content and which our brains can’t even detect? We are guilty of inconsistency if we rely on the source model to build our image of what zombies are supposed to be missing, while telling ourselves that we have only subtracted the target. If zombies lack the source, they are not physical duplicates and they will no longer say they are conscious; if they only lack the target, then they are – as far as we know – in exactly the same situation as us.

The impossibility being proposed here is therefore a subtle one; it cannot be found by viewing the zombie’s world in isolation; it is only evident in the meta-view that takes in both worlds while committing to a consistent resolution of the source-target conflation.

Also, note that the claim being made is that zombies are impossible if virtualism is a correct characterisation of our own world.

If virtualism is wrong on our world, but true of the zombie’s world, zombies get something of a logical reprieve, because at least there is something on our world to remove… But even with this concession in place, the coherence of zombies would not be much better. The posited vindicating entity would be entirely inconsequential even if humans did possess it, because we could never sense the vindicating entity or know anything about it or have reason to care about its removal, and yet our twins must be imagined as dead inside in some important way that supposedly differs from our own situation. Every element of our concept of what they are missing necessarily comes from a model that they are not missing, and, in both worlds, the question of whether the model is vindicated makes no difference that anyone can detect. If we cleaned up our concept of the target by removing illegitimate content, such as all elements of our experience concept that got to us via disallowed causal processes in the physical domain, we would end up with a null concept.

Regardless of whether the vindication occurs on our world or not, the main logical error required to imagine zombies is the acceptance of a double standard: we must discount the validity of zombies’ interiors while holding our own interiors to be special. In our own case, we must envisage our brains housing something that vindicates the implicit content of the model of consciousness; in the zombie’s case, we declare that the same situation can be discounted because the model of consciousness is just a model.

If you want to comply with the thought experiment, you must imagine that your twin finds its models just as convincing as you do, for the same reasons you find your models convincing, but you must nonetheless apply different standards to its models and your own. You can’t plausibly prevent the zombie’s brain representing itself as possessing consciousness, but, because you are viewing its representational system from the outside, you can simply reject the veracity of your twin’s models. Your twin must do the same when it thinks about the superior status of its models in relation to its meta-zombie twin. Your twin is not making an error in its characterisation of the ontology of the meta-zombie world of its twin, because that ontology does indeed lack any vindicating essence; the error lies in its characterisation of the ontology of its own world and the spurious inter-world difference. Your twin’s error is not realising that its own consciousness is merely virtual, like its twin’s.

If you are a hardist, you similarly lack any evidence of your model being vindicated, but in your case you must accept on faith that you are fortunate enough to have represented something real even though, for consistency, you must accept that the existence of your special extra played no role in your formation of the model or your conviction of its veracity.

A zombie world, in this view, is one in which we don’t take the zombies’ models at face value – but the zombies do. They don’t do this out of stupidity or gullibility, but as a consequence of being embedded within the epistemic centre of their model and trusting that what they are sure of must be vindicated by reality. This is one more example of sliding from epistemology to ontology – in this case, from models to reality.

All strong hardists, then – all fans of the possibility of zombies – must believe that the brain creates a model that is illusory within the physical domain, but is then redeemed by an unseen, off‑stage vindication. Their own brains must be completely fooled by the only parts of our ontology that impinge upon that brain, but fooled in a way that gets them to see a causally disconnected, unconfirmable truth. A full account of consciousness under a hardist framework would therefore require that the source of the illusion of consciousness be understood and then, separately, that the vindication must be added on as a causally separate process.

Ironically, very few hardists will admit that this is the case or even consider illusionism as worthy of consideration. (Chalmers is a notable exception.) The vindicating entity on our world would need to align perfectly with the implicit content of a model that is necessarily formed for other reasons, and the zombie thought experiment requires us to believe that the model would still exist and still be trusted by our cognition even without the vindication. For zombies to be possible, the causal account of what the zombie says about that content would have to be independent of whether the model is accurate, and only dependent on the model’s causal connections. The zombie has to be wrong about its ontology, and its mistake would lie in accepting its representation as an accurate depiction of reality. It would necessarily be doing that without having any defensible reasons for accepting its representations as reliable.

And its reasons would be your reasons.

Unless, like me, you already believe that consciousness is a virtual construct in human brains, leaving no off-stage vindication of the hardists’ ontology, and no valid grounds for proposing a human‑zombie difference.

Hardism and virtualism have more in common than it might at first seem. They both drift naturally into dualist terminology, despite having fundamentally different accounts of why this dualist talk seems so natural. Hardism entails virtualism in the physical domain with vindication added as a second, undetectable ontological layer; virtualism proposes that the base ontology of the universe is apprehended indirectly, with a second layer of apparent ontological entities added in a virtual layer when our brains form a model of reality.

The two schools of thought end up saying similar things about consciousness.

Under virtualism, just as in the zombie worlds of hardist fancy, a model of consciousness doesn’t need to be vindicated to achieve its behavioural ends, so consciousness seems causally impotent and ethereal; consciousness does not exist in the physical domain in the form it appears, so it is naturally left out of a first-pass characterisation of our ontology; the content does not even have to comply with the laws of physics – we can think through walls and back in time.

Note that virtualism and hardism both involve an element of illusion, in the sense of a disconnect between what causes our models to form and the content of those models; they both assign causal efficacy to the physical model of consciousness, rather than to the entities being represented, and they both require that there are no reliable reasons to infer a vindicating instantiation of the model. Virtualism even allows for a degree of non‑physicality to enter the discussion: consciousness is non‑physical by virtue of being virtual, and in several other ways to be considered in the chapters to follow. In fact, most of the troubling hardist intuitions have their counterpart in virtualism, where they no longer have their paradoxical consequences.

If someone objected that virtualism was just a rebranding of illusionism, I would have no strong objections. The differences are partly a matter of emphasis. One of the problems with the term “illusionism” is that it implies that some form of cognitive mistake is inevitable, and I don’t accept that this is the case. I propose that there is no mistake involved in creating the base model of consciousness that inspires puzzlement; that model is useful, and it evolved for good reasons, long before any primate considered ontological questions or explanatory shortfalls that needed to be filled with special essences. Most of the creation of this model is done by mechanisms that are themselves hardwired, and not under direct cognitive control; we are not committing an avoidable error when we find that model impressive and convincing and difficult to reconcile with what science tells us about the base ontology of the universe. Once we realise we don’t need a direct reconciliation, it seems more natural to talk of a useful internal model than an illusion. The hardist error is engaging directly and naively with the content of our modelled mind with the underlying faulty assumption that our brains can be trusted to model things faithfully, just as they are. But that’s an avoidable error, not an illusion.

Given the similarities between virtualism and hardism, why pick virtualism?

We could appeal to theoretical parsimony, because strong hardism adds an ontological dimension that is not detectable, seems likely to remain stubbornly inexplicable, and does not even account for our puzzlement – and it needs all the same cognitive structures anyway, complete with a misleading model, so hardism involves virtualism with troublesome ontological extras. But the real advantage of virtualism is that it accounts for the same intuitions without leading to paradox. To be virtualists, we don’t have to partake in the Zombie’s Delusion, and we can be much more precise with our concepts.

Under strong hardism, consciousness has to be real and non‑causal, which is an awkward combination that the hardists have never resolved. The hardist is committed to consciousness existing as an independent entity, but cannot let consciousness play a causal role in the physical world – and yet we talk about phenomenal consciousness and we imagine ourselves as having good reasons to believe in it, which points to a causal role after all.

Under virtualism, consciousness gets to be (sort of) real and (sort of) non‑causal, but there is no paradox because causation happens at the level of the model’s substrate, not at the level of content. Furthermore, consciousness only has the apparently conflicting properties of being (sort of) real and (sort of) non‑causal while our concepts are still vague and we are dealing with a messy hybrid. If we define “phenomenal consciousness” as the property of being represented a certain way by a physical brain, then we can drop one of the “sort of” qualifiers; we are talking about the source of our consciousness concept, and consciousness in that sense is real. But now it is causal, because that property has consequences. If we define “phenomenal consciousness” as an entity that really is the way it seems, shifting to a perspective inside the representational assumptions of the model, then consciousness is non-causal, because the causation happens at the level of substrate, and we have cleaved the term from that substrate. But now the target of our interest is no longer real.

Phenomenal consciousness exists on a knife edge of ambiguity, with several vital ontological issues depending on whether we treat the content of our model as having its own ontological underpinning separate from the neural substrate, and whether we attach our concept of “phenomenal consciousness” to this secondary entity. If we fail to resolve the ambiguity because we cannot deconstruct a model we inhabit, we can conceive of zombies, but we are left with paradox.

Recall your answer to the question: How do you know you’re not the zombie in this scenario? How do you like that answer now, coming out of your twin’s mouth? What is your diagnosis of its error? Now take that diagnosis and imagine explaining it to your twin. Watch the words you use, and now imagine them said by me to you.

Is your twin convinced? Are you convinced?

You know exactly how your twin would respond to my suggestion that its model of consciousness lacks ontological backing, because it would say what you want to say.

What do you think of your twin’s objections to your diagnosis of its mistake? Do you need to update your diagnosis of its error? Or update its objections to that diagnosis? Or a bit of both?

You can probably conceive of a sane, defensible account of the zombie world that the zombie could express, saving it from error, but unless that account matches your own, you have broken the terms of the thought experiment. You can probably diagnose its error, but unless you accept that diagnosis in your own case, your zombie will reject your diagnosis.

It is your cognitive isomorph, so it can only be convinced by an explanation of its mistake if you accept the same diagnosis of your own situation.

Perhaps you agree with me about the ontology of our world, in which case your twin does, too – but then it is impossible for your twin to exist under the ontological framework you believe in. If you don’t agree with me, your twin will raise all the objections that hardists produce when virtualist claims are made by people like me in relation to humans… The objections possibly running through your mind right now.

You can, if you like, continue this game of question and answer with your twin indefinitely, providing both sides of the argument with complete conviction, because your beliefs are always duplicated in the zombie even though – if you think your twin is possible – you necessarily hold the contradictory belief that it is wrong where you are right, and it necessarily rejects this diagnosis. If you think zombies are logically possible.

Under hardism, the logic of the human world and the notional zombie world will always be in tension. Finding a resolution of that tension is like trying to solve two simultaneous equations in conflict: you and your twin are working off the same inputs, so you can’t both be right unless you each come to a common assessment of your local ontology, or at least agree that you have no way of determining which of you is the human, at which point your twin ceases to be a zombie because there is no meaningful human-zombie difference. You can’t adopt the position that your twin needs to be agnostic about the ontology of its world, as a defence against the charge of delusion, but you have grounds for certainty – this approach renders your twin incorrect, because it will adopt the very same double-standard, abandoning the agnosticism you are trying to recommend, and you will inherit its errors. Whatever your diagnosis of its ongoing errors, you can have no good grounds for rejecting this diagnosis when it is turned upon yourself. .

Note that this tension is not present for virtualism. Ontology on the notional zombie world is exactly as the virtualist describes, and so zombie virtualists would have a correct view of their situation, free of delusion. Their interpretation is consistent with the epistemic evidence, free of unsafe inferences, and vindicated by the lack of the special essence. Even on the human world (assuming for the moment that there are two types of worlds), those views would remain reasonable, even if the essence existed, because the virtualist is merely arguing that we have no way of knowing about that essence and no safe epistemic route for deciding it exists. On both worlds, the virtualist policy of agnosticism about the essence is not a dogmatic insistence that an invisible essence is impossible, just a recognition that there can be no good grounds for positing the hardist essence, because all the puzzles it was intended solve are present on both worlds in equal measure, and strong belief in its existence leads to paradox.

Hardist views on the zombie world, by contrast, would necessarily be unreasonable and wrong – this is Chalmers’ “monumental delusion”. And even if the same views are vindicated on this world, the quality of the epistemic evidence available to the physical brain does not change with that vindication, so the unseen vindication cannot provide the hardist with any justification for the process that led them to their belief.

If you ever get to the point that your zombie twin and you agree on all the issues, such that you can listen to what your notional zombie says and you would be happy to call its views correct in its mouth and yours, you will have become a virtualist. The notional zombie will have morphed into a mere pseudo-zombie; it will have the same nuanced, tenuous claim on a special interior as you, sharing your experiences, but no longer committed to the idea of a special ontological backing for those experiences.

That’s the position I’ve reached, and the perspective from which this book is written.

There are several ways that hardists and their zombie twins might resist this path to rapprochement.

The black-box option is still there, even if it entails contradiction, because the very notion of contradiction requires that we care what words mean, and we can argue (somewhat circularly) that we do not care in this case. You might feel justified in dismissing everything your twin says, despite the physical form of its words matching yours, because its misguided opinions and its models of missing entities are formed in the absence of genuine understanding. Provided that you can reject its claim as mere sounds or keyboard bashing not backed by genuine thought, you don’t have to worry about logical consistency across two cognitive systems. You are right and it is wrong. Problem solved.

I have observed this response in several hardists, but I think it is too hasty and rather too convenient. The zombie’s pseudo‑opinions and its mistaken models of reality play the same causal role in the zombie as the corresponding models play in your own brain as you form your own beliefs about consciousness. That means, if you believe in the possibility of zombies, the reasons for the formation of conceptual models in the zombie remain relevant to the evaluation of your own beliefs, even if there is no phenomenal consciousness vindicating the zombie’s models or casting light upon its musings. Your twin is not delusional because it happens to be wrong; it is delusional because it holds an extraordinary belief for which it can have no possible evidence under the conditions of that very belief. Even if we imagine it as making its errors in the experiential dark, it is still making errors as surely as a large language model “hallucinates”, and that’s what matters in judging the cognitive sequence that it shares with you. It is the logical disconnect between your models and their truth value that is the real problem, the lack of any defensible path of inference in your head that is of concern, not the judgment we pass on the lack of understanding in a possibly impossible zombie. And if that path can lead to delusion because it is not sensitive to local ontology, it is suspect in both worlds. The origin story for your belief in a special extra is the same in you and your twin. Only the truth value of the belief can differ on the two worlds, not the reason for holding it; if the reasoning process cannot be trusted to lead to truth, your accidental vindication does not count as justification.

You might think you have additional, special, subjective reasons for saying that your interior has the experiential qualities of interest, but, if so, those extra reasons are superfluous, because, even under the conditions of the thought experiment, there are already adequate objective causes for everything you say. For instance, you might want to insist that your models are blessed with “meaning”, whereas you want to deny the zombie the capacity to mean anything, but that simply implies that meaning plays no causal role in the formation of your thoughts, either, and none of your thoughts are held because of what they mean – a drastic consequence that will need to be reconsidered later.

Similarly, it’s not appropriate to treat your own conclusions as immune from criticism on the grounds that the cognitive processes that went wrong on the zombie’s world successfully found their target in your own case. It is not valid to argue that you can’t be wrong in your thought processes because you got to the right answer; you actually possess the controversial difference, so whatever cognitive processes led to your belief were justified. This is not only circular logic, it is unhelpful even if we overlook the circularity. The existence of a special off‑stage experiential extra cannot justify your brain’s reasons for believing in that extra, if that off‑stage extra plays no causal role in the formation of your beliefs. At best, you are right for bad reasons – reasons that led your twin astray. Reasons you cannot trust, because you are obliged to reject them in your twin.

Much more could be said about each of these themes. And we still have that wrinkle mentioned earlier, in which legitimate entities like distant stars can be posited without any direct causal link to our brains. Some extrapolations outside our brain’s causal scope must be reasonable.

Later chapters will deal with all of these issues in more detail, but fans of zombies can raise one more important objection to the charge of delusion that needs to be mentioned up front. Hardists can concede that there might be no valid, logically defensible reasons for believing in the possibility of zombies, but nonetheless insist that the possibility has not been totally excluded – so zombies are still possible. You and your twin could agree that it can’t be known with certainty that the target of the zombie thought experiment is real, and hence you could concede that, in theory, either of you might be no more than a pseudo-zombie fooled by a model… But there is room for uncertainty. Your twin might be unconscious where you are not, and that potentiality is enough to falsify physical theories of consciousness. You strongly suspect that you are not the one being fooled, and, even if your reasons are not robust, that doesn’t matter. You might be right, even if you reached your conclusions for reasons that cannot be defended with rigour, and that means zombies are possible. Two hypothetical worlds could have parted ways, despite being physically identical, and we need to understand their divergence. We still have a Hard Problem.

Another way of making the same point is to note that, in the absence of a godly adjudication, I might be right, or Chalmers might be right, and neither can deliver a decisive argument; neither of us can ever prove or disprove the existence of a special extra that produces no physical or cognitive evidence. But if it is remotely possible for Chalmers to be right – even if he is accidentally right after a series of cognitive steps that have no merit because they entail the Zombie’s Delusion in his twin – then zombies are possible and all the usual consequences follow.

Note that this sort of argument appeals to a curious asymmetry in the assumed burden of proof: in the absence of godly adjudication, the hardist wins the day, and physical theories of consciousness are in trouble. If it is possible that zombies are possible, and also possible that they are impossible, then this clash of incompatible views resolves to zombies being possible; something special must have happened on our world to save us from zombiehood. Or so runs the argument.

If this logic seems a little too convenient, it is.

It is tempting to respond to these burden-of-proof assumptions by noting that this sort of argument could be raised – must be raised – on the zombie’s world, where it would run the same course and produce a spurious proof of the special extra within the zombie’s ontology. Unless we can know with certainty that the zombie argument has been applied correctly, on a world with the special essence, we cannot trust it. But that means the argument adds nothing to the assumptions it rests upon.

But this line of argument is too important to dismiss so quickly, and a full analysis will require its own chapter. The assumptions leading to the asymmetrical burden of proof need to be identified. With a little work, it can be shown that the assumed asymmetrical burden of proof arises from the same confused place that led to the Zombie’s Delusion, and the issues can be rephrased without the asymmetry.

Some readers might already have noticed that there is yet another conflation at work in this assumption of an asymmetrical burden of proof, this time related to different meanings of the word “possibility” (and once more it involves a slide from epistemology to ontology). Any hardist arguing this way must believe that the mere conceptual possibility of zombies can be accepted as a genuinely possible metaphysical outcome. But this does not follow. Whether our understanding permits zombies to exist in our imagination (an epistemic possibility) is a separate issue from whether the laws of logic would ever permit zombies to exist (as an ontological possibility).

At any rate, even if we are to take this argument at face value, it means that the hardist is conceding that they might be a pseudo-zombie lacking any hardist extra. Even this minor concession would be a step in the right direction. Suppose that some worlds do have special extras, and others don’t, and none of us can tell which world we’re on. Even if we let the burden-of-proof favour hardism, the argument has been weakened to the point of absurdity. If the hardist version of consciousness might be something we don’t even have, something that might only exist on some other world, and they are ready to concede this, then the framing has become de facto virtualist; the notion of a special extra is no longer doing any important work.

Maybe, just maybe, there is no special extra that lies “beyond the performance of functions”, in which case zombies would be impossible – and that impossibility would be a rigid logical impossibility. A duplicate lacking that extra could not logically be imagined as dead inside in any important sense by virtue of its absence, if we already lack it and we are not dead inside. This is not proof of the impossibility of zombies, but it is a pointer to the epistemic possibility that they are logically impossible, and hence to a way out of the maze of paradox and confusion that hardists have inflicted upon the field of consciousness studies.

So where does all of that leave us?

For the zombie believers – for those of you who still think you house a special extra that could go missing in a physical duplicate – this book is about that extra, but even more than that, it is about your belief in that extra, and where that belief comes from. It is about the model of consciousness that forms in physical brains, for reasons independent of any vindication.

For the zombie skeptics, this book is about the Chalmersian concept of non‑physical consciousness, which stands in need of debunking, but, in addition, it is about the physical entity of consciousness that drives the puzzlement, the model that often gets taken at face value, embellished, and imagined as non‑physical. This source entity has much greater significance than being the inspiration for a fiction, because it is the thing we ultimately want to understand.

Either way, we’re talking about two entities that are often inappropriately rolled into one on the human side of this comparison – the same two entities that, at the start of this introduction, I referred to as the source and target versions of phenomenal consciousness.

Unfortunately, we’re not done yet. To understand how both of these entities end up featuring in a debate that takes place in the physical world, where the hardists’ target can have no effects, we will actually need three concepts: one for the physical model, one for the implicit content of the model, and one for the elusive hardist target that, if it existed, would vindicate the model. Depending on context, all three concepts have a claim on being what people mean by “phenomenal consciousness,” and all three will be needed if we are to navigate the fine line between hardism and virtualism.

To be continued…