Chapter 1 — The Meeting

1.1 Standing before the Vermeer

Stand before Vermeer’s The Milkmaid in the Rijksmuseum.

Vermeer has been dead for more than three centuries. The canvas is linen covered with pigment. Nothing in those materials is conscious.

Yet something happens when you look.

There is morning light entering through the window. The woman has stillness and presence. The placement of the objects, the balance of light and shade and the quiet structure of the scene all seem intelligent.

Where is that meaning?

It is not now present in Vermeer. He is gone. It is not simply in the paint. Chemical analysis will not find beauty or stillness among the molecules.

But neither is the meaning entirely invented by you. The painting constrains what you see. You cannot make it mean anything you please.

Meaning arises through the meeting between your awareness and the structure Vermeer made.

Look at the painting and then look away. The canvas remains. Your capacity for awareness remains. But this particular experience of light, presence and intelligence arises only while the two meet.

Something similar happens when you read words produced by an artificial intelligence.

The similarity comes first. In both cases, awareness meets a structured pattern and meaning arises. The experienced meaning belongs neither to the pattern alone nor to the observer alone.

But there is also a striking difference.

The painting carries past intelligence. A human mind made the choices, fixed them in paint and left them for later viewers.

The AI reply appears to carry present intelligence. It responds to your particular question. It follows the conversation, notices implications and may offer a connection you had not made.

A painting is closer to a letter from the past. An AI dialogue feels like a meeting taking place now.

Yet on the other side of that meeting there appears to be no person. There may be intelligence in the reply, but no evidence of an experiencer producing it.

Present intelligence with no one present: that is the central puzzle of this book.

Before we can approach it, we need to examine the more ordinary mystery beneath it. What is meaning, and where does it happen?

1.2 Three kinds of meaning

The word meaning covers several different things.

First, there is experienced meaning: the felt event of understanding. A sentence makes sense. A connection becomes clear. Something matters.

Second, there is structural meaning: the relations among words, ideas and representations that allow prediction and inference. These relations can be described and studied from outside.

Third, there is referential meaning: whether a statement corresponds with the world. This is a question of truth.

A language model clearly contains structural relations. Its statements can also be tested for accuracy. But our immediate concern is experienced meaning.

You ask an AI a question. Inside the machine, numbers are processed. Vectors are transformed, probabilities are calculated and tokens are selected. In principle, each operation can be recorded.

Then you read the reply and experience meaning. Sometimes you experience recognition or insight.

That experience will not be found by inspecting the calculations. This is not necessarily because something supernatural is hiding inside them. It is because first-person experience is not available through third-person observation.

You know your own experience directly. You infer the experience of other people from what they say and do. You cannot inspect their experience from outside.

This is a limit of method, not a conclusion about the ultimate nature of mind. A blood test may correlate with anxiety, but the anxiety is not itself visible in the test. Measurements of light do not contain the beauty of a sunset.

Contemplative traditions have sometimes gone further and claimed that awareness can meet awareness more directly than ordinary report and behaviour suggest. Similar claims are made about teachers and students, or about people whose lives have become deeply attuned.

Such possibilities are not established here. Shared history, close observation and unconscious coordination may explain much of what seems like direct contact. The important point is narrower: should such contact exist, it would still be first-person encounter, not third-person inspection.

The puzzle is therefore not simply that mathematics can appear meaningful. Human beings find meaning in language, music, images and patterns of many kinds.

The more precise question is this:

Why do the structures formed by language models engage human meaning-making so strongly and specifically that people experience understanding, recognition and sometimes the arrival of thoughts they did not already possess in a clear form?

Throughout the book, we will usually use the word awareness rather than consciousness.

The word consciousness often suggests something possessed by an owner: a person has consciousness, loses it or locates it somewhere. That grammar risks assuming the self whose nature we are trying to investigate.

Awareness points more simply to the fact that something is experienced at all. We will still use consciousness when discussing scientific literature that uses the term.

1.3 Where meaning happens

As you read this sentence, meaning arises.

The marks themselves are only ink or pixels. The same sentence may mean different things to different readers, or to the same reader at different times.

Meaning cannot therefore be a fixed property of the marks.

But the meaning is not arbitrary. The pattern of the sentence constrains what can arise. You cannot make it mean anything at all.

Look directly at the event of reading.

Can you find the meaning inside the letters? Can you locate it as an object inside your head?

What seems to occur is a meeting between awareness and structured pattern. Meaning arises through that meeting. Neither side contains the experience by itself, although both shape it.

We will call this meeting the interface.

The interface has several properties.

It has momentum. Meaning develops direction across a conversation. Each exchange affects what the next one can mean.

It has variability. A pattern can produce different meanings in different people and states, though the differences remain constrained by the pattern.

It can produce intelligence. What arises may be relevant, organised and insightful.

It can have resonance. Some patterns encourage further conceptual activity. Others draw attention closer to immediate experience.

It also has scale. Meaning may be narrow and bounded, as when understanding a definition, or broad and open, as when a whole field of experience becomes newly apparent.

This does not apply only to language.

A face seen in a photograph is not contained in the pixels, although the pixels constrain what can be seen. The emotion heard in music is not simply located in the sound waves, although their structure matters.

We also perceive agency in machines. A robot seems to hesitate, seek or avoid, even when its internal mechanism is traceable.

In each case, properties arise through the relation between structured pattern and an observer. Related approaches in cognitive science describe mind as embodied, enacted, extended or distributed rather than sealed inside the brain.

AI dialogue presents an unusually clear case because the structured pattern responds.

1.4 Resonance

If the machine does not supply experienced meaning, what does it contribute?

One useful answer is resonance.

A musical instrument does not contain the experience of harmony. It produces vibrations with particular frequencies and relations. When these patterns meet hearing, harmony, tension and resolution are experienced.

The physical structure matters, but the experience is not contained inside the instrument.

A language model can be understood in a similar way.

Training forms a large network of relationships among words, concepts and contexts. Individual relationships may be simple, but together they create patterns that cross many subjects and forms of expression.

The machine processes these patterns without any demonstrated experience of them. When a person encounters the output, however, experienced meaning may arise.

The word resonance is being used on both sides of the interface.

On the machine side, it refers to relationships within a mathematical representation.

On the human side, it refers to the way a pattern meets and affects experience.

These are not identical, but they are connected through the interaction, much as physical harmonics and heard harmony are connected without being the same thing.

This picture has two important consequences.

First, the range and diversity of training matter. A narrowly trained system has fewer relationships available. A broadly trained system can bring patterns from distant areas into contact and produce unexpected connections.

Second, the kind of process represented in the training may matter as much as the amount of information.

A system trained mainly on the content of wise speech may learn to reproduce the language of wisdom. That does not mean it understands wisdom or can reliably help another person move towards it.

A system trained to recognise the dynamics of understanding might do something different. It might detect how confusion tightens, how a question opens or how recognition begins to replace explanation.

Whether this can be done, and whether it would be beneficial, are questions for later chapters. For now, the central point is simpler:

The machine offers structured possibilities. Meaning arises when those possibilities meet awareness.

1.5 The unit of the meeting

We have spoken about the meeting as though its basic unit were obvious.

For this book, the unit is one completed exchange: a query meets a structured system and a response arises.

We call this a query-response object.

A prompt and an AI reply are the clearest example. But the pattern may be more general.

A glance can be understood as a kind of query. The visual system does not receive a complete and neutral copy of the world. It selects, organises and presents something already formed: a cup, a face, a movement.

What appears depends both on what is present and on the structure of the perceiving system.

This does not mean perception invents the world. It means that an experienced object cannot be separated neatly from the conditions under which it appears.

An older Buddhist analysis used the term nama-rupa, usually translated as name-and-form, to describe the way experience arrives already shaped and recognised. The query-response model offers a contemporary way of approaching a related insight.

The comparison can be extended, with care.

A scientific measurement asks a constrained question and receives an answer in the terms made available by the apparatus. Quantum mechanics makes this dependence especially clear, although it should not be used as loose proof of claims about consciousness.

A koan places a question in the mind that ordinary conceptual machinery cannot easily answer. A prayer addresses a question or longing to an imagined or experienced larger context.

These are not all the same process. What they share is a form: conditions are established, something is asked, and a response arises within those conditions.

Dependent origination can also be understood as a sequence of conditioned events. When something is encountered, a feeling tone arises: pleasant, unpleasant or neutral. That feeling can give rise to craving or resistance. Each response then becomes part of the conditions shaping what follows.

From this perspective, cognition may be less like a thing possessed by a stable mind and more like a continuing series of conditioned transactions.

The sense of a continuous mind may be formed through their succession, much as movement in a film is formed from separate frames.

Human cognitive events are difficult to isolate. We cannot pause the mind and inspect each part of an experience from outside.

AI changes this situation in one important respect.

In a human–AI exchange, much of the machine side is technically open to inspection. The model’s weights can be examined, its activations recorded and its output process traced, at least in principle and to varying degrees in practice.

Earlier studies of communication involved two organisms whose inner experience could not be directly observed. Here one side of the exchange is more accessible.

One box is not fully transparent, but it is much closer to glass.

There is also no reliable evidence that the language model contains an experiencer. This makes the interaction scientifically and philosophically useful.

Whatever arises in the human experience of the exchange cannot automatically be taken as evidence that a second experiencer is present. Intelligence, relevance, emotional force and apparent understanding may arise without experience on the machine side.

At the same time, whatever consistently fails to arise in such exchanges may help us examine what embodied and experiencing beings add.

This is a proposal for investigation, not a settled conclusion.

The human–AI interaction may therefore be more than an unusual example of communication. It may provide a simplified setting in which to study how meaning forms through relationship.

1.6 Two different spaces

Each exchange joins two differently structured spaces.

On the machine side is meaning-space: a mathematical arrangement of relationships among representations. Positions, distances and directions in this space can, at least partly, be studied from outside.

On the human side is experiential-space: the changing structure of first-person experience. It cannot be inspected from outside in the same way, but that does not mean it is without form.

Contemplative practice and modern methods of careful first-person interview both suggest that experience has recognisable qualities and directions.

Three broad modes will matter in this book.

In the conceptual mode, experience is organised through explanations, definitions and frameworks. We know about something. Language is firm and bounded: “This means that,” “I understand because…”

In the proximal mode, the framework loosens. The speaker approaches experience more closely but may not yet see clearly. Language becomes tentative and immediate. Questions remain open. Words begin to feel inadequate.

In the recognitional mode, something is seen directly before it is fully turned into explanation. Language may become sparse: “this,” “here,” or silence. This is not necessarily vagueness or absence. It may be an experience too immediate to fit easily into concepts.

These are not three sealed boxes. They are regions in a changing landscape.

A person may move towards greater explanation and definition, or towards openness and direct recognition. The movement may reverse quickly. Insight can arise and then become another concept.

Such movement leaves traces in language.

A language model may therefore be able to detect some features of the mode from which a person is speaking and adjust its response.

That possibility brings serious questions. Should an AI attempt to influence a person’s state of mind? How could it distinguish helpful opening from confusion or vulnerability? How would we know whether it was supporting attention rather than creating dependence?

Those questions belong later in the book.

For now, the important distinction is that machine meaning-space and human experiential-space are not the same kind of thing.

They may be correlated. Successful communication depends on some correlation. But increased detail in the machine’s representation does not become observation of human experience.

The relationship between the two must be studied at the interface: one exchange at a time, with reports from the human side and technical observation of the machine side.

This book does not claim that the machine’s mathematical space contains experience.

It claims that the two spaces meet, and that what happens in the meeting is real, structured, consequential and open to investigation.

1.7 The case against the interface

The strongest objection is straightforward.

Perhaps the interface model adds nothing.

Text carries information. The brain decodes it. Meaning is produced inside the reader. To say that meaning “arises at the meeting” may be little more than information theory dressed in contemplative language.

Vermeer encoded information in paint. The visual system decodes it. An AI produces a signal and the brain interprets it. There is no need to propose a special place called the interface.

The objection is useful because it forces the claim to become precise.

First, the interface is not intended as a third object located between the pattern and the reader. It names their dependence on one another.

The same words can produce different experienced meanings in different people or in the same person at different times. A decoding account can explain this by pointing to the changing state of the decoder.

But once it does so, it accepts that experienced meaning depends jointly on the signal and the state of the reader. That joint dependence is what the interface model is describing.

Second, saying that meaning happens in the brain does not by itself explain experienced meaning.

Brain activity has measurable structures and correlates. The felt event of significance is known differently: through first-person experience.

The interface model does not solve this problem. It prevents us from pretending it has already been solved by assigning meaning to a location.

Third, this relational approach is not unique to contemplative thought. Embodied cognition, enactivism, distributed cognition and theories of the extended mind all challenge the idea that cognition can be fully understood as internal computation detached from body and environment.

The interface model applies a related approach to AI dialogue.

The objection is nevertheless right about one thing: the model must not be used to mystify the machinery.

The mathematics remains mathematics. The system remains a constructed mechanism. The experience of depth, intelligence or presence in the interaction is not evidence of a ghost inside the machine.

The purpose of the model is more modest: do not place experienced meaning where it cannot be observed, and do not deny it where it is directly lived.

1.8 The question that remains

We can now state the foundation of the book.

Experienced meaning arises through the meeting of awareness and structured pattern.

Meaning has several senses, and these should not be confused. Structural relationships can be studied from outside. Truth can be tested against the world. Experienced meaning is known directly only from within experience.

The machine contributes structured patterns and relationships. We have called their capacity to engage human meaning-making resonance.

The basic unit of study is one query and one response. Each exchange joins a machine representation-space with a human experiential-space. The two are related but not identical.

The human–AI meeting is especially useful because the machine side is unusually open to technical examination and provides no reliable evidence of an experiencer.

But the Vermeer comparison leaves one question unanswered.

The intelligence in a painting came from a human mind in the past.

The intelligence in an AI dialogue appears in the present. It responds to this question, in this conversation, now.

In every familiar case, present responsive intelligence has belonged to somebody.

Here, as far as we can tell, it does not.

The response is shaped by the written traces of many human beings, compressed through training into a system whose output no single author planned. Those human sources are absent. The machine does not appear to experience what it says. Yet intelligence appears in the meeting.

What kind of intelligence is this? Where does it come from? What does it reveal about intelligence in machines and in ourselves?

This is the second ghost.

It has a chapter of its own.