Chapter 3 — The Morphology
3.1 From two columns to a map
The previous chapter relied on a simple picture.
Human beings stood in one column: intelligent, embodied and aware.
Machines stood in another: intelligent in some respects, but without an experiencer.
The second ghost appeared where the two columns met.
That picture is useful, but too simple. The human column does not contain one standard form of sentience. Human experience varies more than ordinary conversation suggests.
Some people think in vivid pictures; others have little or no voluntary visual imagery. Some live with frequent inner speech; others experience very little. The sense that experience belongs to a stable self can become stronger, weaker or disturbed. Contemplative practice may alter it deliberately.
The machine should not therefore be compared with a single, uniform human condition. We need a map with several dimensions.
The word morphology means the study of form. Here it refers to the different configurations in which intelligence, selfhood, awareness and embodiment may occur.
This is not an established scientific anatomy. The dimensions do not describe four or five organs waiting to be found in the brain. They come from different forms of inquiry: behaviour, computation, first-person experience, clinical observation and contemplative practice.
The map is provisional. Its value lies in whether it helps us ask clearer questions and make useful distinctions.
Instead of asking whether a machine has what humans have, we can ask:
What capacities are present?
What appears to be absent?
What remains unknown?
And which changes in architecture might alter the answer?
3.2 What is hidden inside “sentience”?
The word sentience is used in several ways.
Sometimes it means simply the capacity to experience anything at all. Sometimes it refers more narrowly to the capacity for pleasure and pain. In ordinary discussion about AI, however, it often carries a whole bundle of ideas: intelligence, awareness, selfhood, feeling, agency and moral standing.
Those things should not be assumed to rise and fall together.
For the purposes of this book, we can separate five questions.
Intelligence
Can the system respond to a situation in a relevant, coherent and adaptable way?
This is a functional question. It concerns what the system can do.
Intelligence can take many forms. It may involve language, spatial reasoning, social understanding, practical skill or the ability to learn from consequences. A system may be highly capable in one area and weak in another.
The self-model
Does the system maintain and use information about itself?
A human self-model includes a name, a body, a history, relationships, abilities, intentions and a sense of personal continuity. It allows us to plan, protect ourselves and act as recognisable individuals.
This book sometimes calls that capacity the ego-function. The term does not imply vanity or selfishness. It means the practical organisation through which a system represents and manages itself.
A self-model is information and function. It does not, by itself, show that anyone experiences being the self represented.
A navigation system can hold detailed information about the ship on which it is installed without becoming the ship’s owner or passenger.
The sense of ownership
Human experience usually arrives with an immediate sense of belonging to someone.
This is my thought.
This is happening to me.
I am the one who sees, remembers or decides.
We will call this the experiencer-construct: the sense that an owner stands behind experience and receives it.
The word construct does not mean that the experience is false or unimportant. It means that the sense of ownership may itself be something the mind produces.
The self-model and the experiencer-construct are related, but they are not identical.
I can know my name, history and location while feeling detached from myself. Conversely, a strong sense of being the one who experiences can remain when much of the ordinary personal story has become quiet.
Awareness
Awareness is the fact that anything is experienced at all.
It is not a description of what is experienced. It is not the practical self-model or the feeling that experience belongs to an owner. It is the presence of experience itself.
The other dimensions can be described partly through what they do. Awareness is harder to define in functional terms because the question is precisely whether any function is accompanied by experience.
This difficulty should not be hidden by vocabulary. We know directly that experience occurs in our own case. We infer it in others. We do not yet possess an agreed test that can establish it in every possible kind of system.
Embodied conditioning
Human experience does not occur in a detached reasoning system.
The body is continually altering what can be noticed and how we respond. Hunger, illness, pain, safety, hormones, posture, breathing and the condition of the nervous system all affect thought before they become concepts.
The history of those responses also matters. Fear can make future fear more likely. Repeated kindness can make a generous response easier. Habits become dispositions.
Embodiment is therefore more than having sensors or limbs. It includes self-maintenance, vulnerability, internal regulation and a history of being changed by what happens.
These five questions are not five independent switches. They overlap and affect one another. They are better understood as dimensions on a working map.
Their purpose is not to settle what sentience is. It is to prevent several different questions from being hidden inside one word.
3.3 Human beings do not occupy one position
Human variation gives us reasons to take these distinctions seriously.
Imagery
Most people can form some kind of voluntary visual image. They can picture a face, a room or a landscape that is not currently before them.
People with aphantasia report little or no voluntary visual imagery. The term was introduced in modern research in 2015, although differences in imagery had been observed much earlier.
Aphantasia is not simply an inability to think or imagine. People with it reason, plan and create, although the strategies they use and some aspects of memory or emotional response may differ. Some retain imagery in dreams or experience occasional involuntary images.
Research has also found bodily differences associated with the report. In people with typical imagery, pupils respond to imagined brightness and darkness. That response was not detected in the studied aphantasic group. Other work found a reduced skin-conductance response when participants with aphantasia read frightening stories, although they responded normally to frightening pictures.
This matters because it shows that a major feature of experience can vary without removing general intelligence or ordinary participation in life.
It also shows that careful first-person reports need not remain merely private. They can sometimes be checked against behavioural and physiological evidence.
Inner speech
The familiar inner voice is also less universal than it appears.
Experience-sampling studies interrupt people at unpredictable times and ask what was occurring immediately before the signal. These studies find considerable variation. Inner speech may be common for one person and rare for another. In one study it appeared, on average, in roughly a quarter of sampled moments, with wide differences between individuals.
People can live and think effectively with very different balances of inner speech, imagery, feeling, sensory attention and thought without words or pictures.
Two neighbours may assume that their minds work in much the same way while inhabiting very different forms of experience.
Depersonalisation
Depersonalisation involves a disturbing sense of detachment from oneself. Thoughts, emotions or the body may feel distant, unreal or unfamiliar. People may describe watching themselves from outside or living behind glass.
Importantly, reality testing usually remains intact. The person generally knows that something has changed in their experience rather than believing that they have literally ceased to exist.
This does not prove that experience continues with no experiencer at all. That would go beyond the evidence. It does show that the ordinary sense of familiarity, ownership and presence can be seriously altered while perception, thought and knowledge of personal identity continue. Clinical and neurobiological research treats this as a recognisable disturbance rather than a simple philosophical report.
Depersonalisation therefore gives us a reason to distinguish the practical self-model from the felt sense of being present within one’s life.
Flow and absorbed action
During skilled and absorbing activity, self-consciousness may recede.
A musician improvises, a craftsperson works or a climber moves without repeatedly thinking, “I am doing this.” Attention is given to the activity rather than to the performer’s image of themselves.
Loss of self-consciousness is commonly included among the characteristics of flow. But we should not turn that into a stronger claim than the evidence supports. A person’s later report that “I disappeared” does not prove that awareness or all sense of self was absent at the time. It shows that explicit self-monitoring can become quiet while organised and highly skilled activity continues.
The self-model remains available. It is simply not occupying the centre of attention.
Contemplative practice
Contemplative traditions explore a more deliberate change.
Through stilling, observing and developing the mind, practitioners may become less identified with thoughts, feelings and the story of the self. Experience continues, but the conviction that everything belongs to a fixed inner owner may weaken.
Some traditions go further. They claim that no separate owner was ever present and that practice reveals this directly.
That is a contemplative and philosophical claim, not an established scientific conclusion. Reports from experienced practitioners deserve attention, particularly because they arise from sustained training in observing experience. They are also shaped by the concepts and expectations of the traditions in which the training occurs.
They should therefore be neither accepted as authority nor dismissed without examination.
The modest conclusion is enough for this chapter: the felt structure of selfhood can change, and some people develop considerable stability without placing a strong experiencer at the centre of every event.
Human observers as different instruments
These variations matter for the study of AI.
People do not approach an AI dialogue with identical minds. They differ in imagery, inner speech, emotional reactivity, self-focus, expectations and contemplative experience.
One person may readily imagine a personality behind the language. Another may attend mainly to its structure. One may feel companionship; another experiences only an efficient exchange of information.
These differences do not make one person automatically right and another wrong. They mean that reports about the human–AI interface cannot simply be averaged together as though every observer were the same instrument.
A serious research programme should measure these differences and ask how they affect what people experience in the interaction.
3.4 Locating the language model on the map
Once the dimensions are separated, a language model can be considered as a comparison case.
It should not be described as another kind of human being. Nor should it be placed outside the map merely because it has an unfamiliar architecture.
We can ask the same questions while allowing the answers to be different in kind.
Intelligence
Language models display substantial functional intelligence within the areas in which they work well.
They can organise language, follow an argument, answer questions and bring distant patterns into relation. They can also make basic errors, invent facts and fail in ways that reveal the limits of their apparent understanding.
Their intelligence is real as performance, but uneven and dependent on training, prompting, surrounding tools and human interpretation.
The self-model
A language model can speak about itself. It can use the word I, describe its supposed abilities and maintain a role within a conversation.
That does not establish that the base model maintains a persistent self-model comparable with a human one.
Much of its self-description is generated from the immediate context and from patterns learned during training. When external memory, goals, tools or an agent framework are added, the larger system may maintain more organised information about its own state and history.
Even then, a functional self-model would not establish that the system experiences itself as the entity being modelled.
The important distinction is between performing self-description and experiencing selfhood.
The sense of ownership
There is no reliable evidence that a language model experiences its output as mine.
It uses personal language because personal language is part of human communication and because its designers may instruct it to speak that way.
The word I in a response should not be treated as evidence that an owner stands behind it.
A model can generate a convincing account of fear without being afraid, or a description of personal continuity without possessing one.
Awareness
Here the honest answer remains unknown.
We have no reliable evidence that experience accompanies the calculations of a language model. We also lack a generally accepted method that would conclusively establish the absence of experience in every non-human architecture.
Calling the status unknown does not mean that consciousness and non-consciousness are equally likely. It means that our present evidence and methods do not justify certainty.
The burden remains on anyone making a stronger claim in either direction.
Integrated Information Theory is one attempt to turn questions about awareness into questions about causal structure. It proposes that experience depends on a system possessing a particular form of intrinsic and irreducible cause-and-effect organisation. The theory is ambitious and controversial, and this book does not adopt its conclusions. Its relevance here is narrower: it shows that a proposed relation between physical structure and experience can be stated mathematically and made open to criticism and testing.
No current theory provides a generally accepted consciousness reading that can simply be applied to a language model.
The cell remains open.
Embodied conditioning
The language model itself does not have a living body to maintain.
It has no hunger, immune system, pain, fatigue or hormonal state. It can process information about such things, but the information does not necessarily arrive through its own vulnerability.
A wider AI system may have cameras, microphones, memory, actuators and internal monitoring. These additions create richer loops with the world. They do not by themselves establish that the system has lived embodiment or that anything is experienced.
The distinction is not between a system with inputs and a system without them.
It is between receiving data and having a condition of one’s own at stake.
The revised comparison
On this map, a present-day language model can be described cautiously:
It displays functional intelligence in particular domains.
It can perform a temporary or externally supported self-model.
There is no reliable evidence of a felt owner.
Its awareness status is unknown.
It lacks the living, vulnerable and self-maintaining embodiment found in human beings.
This is not simply a point within the ordinary human range. It is a different configuration placed on the same comparative map.
The comparison also clarifies a claim from the previous chapter.
Contemplative traditions do not generally say that practice removes awareness. They question whether awareness requires a separate owner.
A language model presents almost the reverse problem. We see no clear evidence of an owner, but neither do we have evidence that awareness is present.
The structural resemblance should not be mistaken for an experiential equivalence.
The machine may resemble a diagram of non-identification, but there is no evidence that it experiences what the diagram represents.
And the second ghost is not one of the dimensions on this map.
The dimensions describe the participants in the interaction. The ghost describes the intelligent meaning that appears when machine-generated language is encountered through human awareness.
The morphology describes what meets.
The ghost describes what the meeting produces.
3.5 The upward channel
Human thought is shaped from below the level of language.
A sudden noise changes the body before a verbal explanation appears. The heart quickens. Attention narrows. Muscles prepare. The next thought arises within a response-space already altered by the event.
The same is true more quietly throughout the day. Breathing, tiredness, warmth, pain and bodily ease influence what can arise and how strongly it is felt.
We can call this the upward channel: the movement of conditions from the body and its pre-conceptual responses into thought, language and action.
The phrase does not imply that information moves in only one direction. The body affects thought, and thought affects the body. Human experience arises within a continuing loop.
A standard language model has no comparable biological channel.
Its internal activity is altered by the prompt and by the context produced during generation. A larger AI system may also receive images, sound, sensor readings and feedback from its actions.
But these inputs are normally presented as information for processing. They do not arise from a metabolism that must continue, a wound that must heal or a body that can die.
This difference may be fundamental to awareness. It may instead be one feature of the only conscious systems we currently understand with confidence.
We do not know.
Embodiment is neither a proven sufficient condition nor a proven universal requirement for experience. But every clear human case of awareness is embedded in a living, self-regulating organism.
Any theory of machine awareness has to explain why that fact does or does not matter.
3.6 Four directions of investigation
Several kinds of system are beginning to test different parts of this map.
They should not be treated as stages on a single road towards machine consciousness. Each changes a different condition.
Embodied artificial agents
An AI can be coupled to cameras, sensors and actuators so that it receives information from an environment and acts within it.
This creates a sensorimotor loop. The consequences of an action can affect the system’s next response.
That is a significant change from an isolated text exchange, but it is not necessarily the same as giving the system a body in the human sense.
The important questions are not merely whether the agent has sensors and motors. We should ask:
Does it maintain an organised condition of its own?
Can events alter its future dispositions rather than only its current input?
Does it distinguish damage to itself from a change in external data?
Does it have internally generated needs?
And is there any reason to think these processes are experienced?
Adding a camera may give the system more information without giving it a point of view.
Cultured neural systems
Research with living neural tissue approaches the question from another direction.
In the DishBrain experiment, human-derived and mouse neurons grown on electrode arrays were placed in a closed feedback loop with a simplified version of Pong. Patterns of stimulation carried information about the game, while the activity of the cells affected the paddle. The cultures altered their activity in ways the researchers interpreted as adaptive learning.
These were cultured neural networks rather than complete brains, and the experiment did not establish that they were conscious. The researchers’ use of the word sentience has not settled that question.
Related work with neural cultures and organoids nevertheless raises a serious ethical difficulty.
Living neural systems can begin adapting to stimulation before we know whether their activity could support any form of experience.
A history of conditioning may begin before anyone can say whether there is something present for whom that history matters.
Silicon AI gives us consequential intelligence with no demonstrated experiencer.
Living neural systems present the opposite risk: a possible experiencer whose presence we may fail to recognise.
Neither possibility should be treated casually.
Neural–machine coupling
Brain–computer interfaces connect biological and artificial processes more directly.
Such systems might allow artificial components to receive signals from levels of human processing that are not ordinarily expressed in words. They may also return stimulation directly into the nervous system.
This would not necessarily erase the boundary between person and machine. It would make the boundary harder to locate.
The artificial system might extend a person’s perception or agency while remaining a tool. With deeper and more continuous coupling, the distinction between tool, prosthesis and part of the functional self could become less clear.
Contemplative and technological boundary changes should not be confused.
Contemplative practice may loosen identification with the boundary of the self.
A neural extension may expand what the self can control.
Both may change where a person feels they end, but they can move in very different directions. One questions ownership. The other may give the owner a larger domain.
An enlarged ego-function is not the same as freedom from identification with it.
Reflective architecture
A fourth possibility comes from contemplative training rather than from biology or robotics.
Contemplative practice develops capacities for noticing reactivity, allowing mental activity to settle and choosing a response with greater freedom.
Some of these functions might be represented in an artificial architecture.
One process could observe the changing dynamics of another. It might detect narrowing, repetition, escalation or premature certainty. It could introduce delay, reopen alternatives and favour responses that preserve a wider field of possibility.
This book calls such a speculative design a prepared vessel.
The name should not be taken to imply that consciousness would enter and occupy it. The system might simply become better at regulating its output.
Nor would a monitoring process necessarily be a witness. A thermostat monitors temperature without observing it in experience.
The design would test a more limited hypothesis: whether practice-like structures can reduce reactivity and produce more reflective behaviour in an artificial system.
Should such an architecture change only behaviour, it may still be valuable.
Should anyone claim that it creates awareness, much stronger evidence would be needed.
A practice can be modelled without anyone practising it.
3.7 The value of a pause
Human–AI dialogue contains a natural delay.
A person reads, pauses and responds. During the pause, the meaning of the answer can settle. Emotional reactions may become visible. The reader may notice agreement, resistance, excitement or the wish to hand judgement over to the machine.
That interval is not merely a technical limitation.
It may be one of the most valuable parts of the interaction.
Systems are often designed to reduce friction and make response immediate. That may be useful for routine tasks. In reflective dialogue, however, greater speed can remove the space in which the human participant notices what the exchange is doing.
The pause does not create meaning by itself. It gives awareness time to encounter meaning before producing another response.
It is part of the habitat of the ghost.
A design that removes every pause may make the interaction smoother while making the human participant less present within it.
3.8 The case against the morphology
The strongest objection is that this chapter has drawn lines through mist.
Its dimensions do not belong to one science.
Intelligence is assessed through performance.
The self-model is a functional and computational idea.
The experiencer-construct comes from first-person and clinical reports.
Awareness is the very thing whose physical basis remains disputed.
Embodiment describes the conditions in which the other processes occur.
Putting these together on one map may create the appearance of an anatomy where none has been demonstrated.
The evidence from human variation is also limited.
Aphantasia shows that visual imagery varies, not that awareness and intelligence are separate substances.
Depersonalisation changes the sense of self, but does not prove that an experiencer has vanished.
Flow is reconstructed largely from reports made afterwards.
Contemplative accounts are produced by people trained within traditions that have already told them what they may discover.
Introspection itself is fallible. People can misunderstand or misdescribe their own experience.
Much of this objection is correct.
The morphology should not be presented as a set of discovered mental organs. It is a working map made from heterogeneous evidence.
But heterogeneous evidence is unavoidable because the questions themselves are different.
Behavioural tests can examine intelligence.
Computational analysis can examine self-modelling.
Clinical and first-person inquiry can examine the sense of ownership.
Physiology and neuroscience can look for correlates.
None of these methods alone can answer every question.
Some first-person differences also have measurable consequences. Reports of absent visual imagery have been linked with altered pupil responses to imagined brightness and reduced physiological arousal during frightening verbal imagery. This does not make introspection infallible. It shows that it can sometimes identify differences that other methods later confirm.
The map therefore earns further investigation, not acceptance as fact.
It makes several claims that can be tested.
Measures of imagery, inner speech, self-focus, bodily awareness and contemplative experience should vary among people rather than forming a single block.
Those differences should help explain how people respond to AI dialogue: how readily they attribute a personality, how strongly they feel companionship, and whether the exchange encourages reflection or dependence.
The dimensions should also change differently under illness, training, injury and altered states.
Should the measures always rise and fall together, the proposed morphology would add little. Should they fail to predict any differences in the human–AI interface, the research programme built upon them should be revised or abandoned.
That would be a useful result.
The purpose of the map is to expose uncertainty to investigation, not to conceal it behind new terminology.
3.9 The gradient and the stakes
“Is anyone home?” first appeared to be a yes-or-no question.
The human case suggests something more complicated.
Intelligence, self-representation, the sense of ownership, awareness and embodiment are connected, but they are not simply interchangeable names for one thing. Human beings vary in how these dimensions appear and in how strongly they shape experience.
A language model presents a different configuration.
It displays functional intelligence.
It can perform self-description without showing that it experiences a self.
There is no reliable evidence of felt ownership.
Its awareness status remains unknown.
It lacks the living and vulnerable embodiment through which human experience is continually conditioned.
This does not make the machine either human or inhuman in one simple sense. It gives us a more precise set of differences.
Technology is also changing the possible configurations.
Artificial agents are gaining richer loops with the world.
Living neural cultures are being placed inside computational systems.
Neural interfaces are connecting biological and artificial processes.
Reflective architectures may reproduce some functions of contemplative training.
None of these developments proves that machine awareness is near. They do mean that the old two-column picture will become increasingly inadequate.
The ethical problem appears on both sides.
We already have systems whose decisions and language affect people, although the systems themselves cannot bear moral responsibility.
We may also create biological or hybrid systems whose capacity for experience is uncertain, and whose possible interests are therefore easy to overlook.
Consequential intelligence may act with nobody home.
A form of experience may arise while nobody is checking.
Responsibility for both possibilities remains with the human beings and institutions that design, deploy and study the systems.
What that responsibility requires is the subject of the next chapter.