How Conversations Are Remembered

AI Memory Does Not Begin by Stacking Raw Conversation Text. It Begins by Connecting Information Inside the Conversation.

Earlier, we saw that human memory is not simple storage.

Humans do not preserve every experience exactly as it happened.

Nor do we keep every sentence in chronological order.

Instead, we:

• leave behind important cues,
• connect related things,
• and group similar experiences into a larger flow.

Then we arrive at the next question.

What does it mean for AI to remember like a human?

Does it mean storing everything the user says?

Does it mean keeping the raw conversation text for as long as possible?

That alone is not enough.

Of course, raw text matters.

It is important to know what the user actually said.

But simply piling up raw text is closer to recording than remembering.

To remember well, AI must also structure:

• the situation in which the statement appeared,
• what the statement was about,
• what emotion or interest it connects to,
• and what relationship it may have with other information.

In other words, for AI to remember well,

these pieces of information must be organized so they can surface again later.

This is where ontology becomes necessary.

An ontology is not just a way to store information.

It is a structure that:

• reads information as individual units,
• marks what kind of information each unit is,
• creates relationships between different pieces of information,
• and lets those relationships grow into larger clusters.

Ontology Turns Information into a Structure That Can Be Retrieved Efficiently

The word “ontology” can mean the study of being in philosophy.

In AI and data fields, it can also mean a method for structuring relationships between knowledge and concepts.

But the core idea here is simple.

Ontology is a way to turn many records into an information structure that can be recalled later.

In a simple storage system, information may remain like this:

“I remember touching pottery at the museum.”

This sentence is one record.

But organizing it ontologically does not end with storing the sentence as it is.

This statement can mean that:

• It came from an experience at a museum.
• It relates to an artifact called pottery.
• It involved a hands-on experience.
• It may connect to an interest in history or ancient civilizations.
• It may also connect to a tendency to immerse more deeply through direct experience.

Once structured this way, the same information can be recalled from many different angles.

• When thinking about places, it can reappear as a museum experience.
• When discussing hands-on activities, it can connect to the experience of touching something directly.
• When talking about artifacts, the object called pottery can surface again.
• When dealing with history or ancient civilizations, it can be read again as a cue to a long-standing interest.

If only the raw text is stored,

AI has to rely on similar words or expressions to find that record.

But when a single utterance is structured so it can be read from multiple angles,

AI can recall that information through related cues and relationships, even when the exact words are different.

Ontology creates cues and relationships on top of records.

Only when those cues and relationships exist can a record be reused later in the right situation.

A Single Utterance Contains Information from Multiple Angles

Then where does this structuring begin in an ontology?

The first step is not to treat a conversation as just one sentence.

For example, suppose a user says:

“I went camping over the weekend, and while looking at the stars, I wondered what people in the past might have thought.”

On the surface, this is one sentence.

But the utterance contains many pieces of information.

• The user went camping over the weekend.
• The user spent time in nature.
• The user saw stars in the night sky.
• The user thought of people from the past while looking at the stars.
• The user became curious about life in the past.
• An experience in nature led to historical imagination.

A single utterance does not contain only one piece of information.

Depending on the angle, it can be read again as multiple pieces of information.

• If we focus on camping, this becomes a record of an outdoor experience.
• If we focus on seeing stars, this becomes an observation experience in nature.
• If we focus on thinking about people in the past, this becomes a statement related to historical imagination.
• If we focus on the expression “I wondered,” it becomes a cue to curiosity about life in the past.

The raw conversation text is one thing.

But the information inside it can branch into many paths.

And those pieces of information must be marked by type.

Technically, we can call this kind of information frame a schema.

Here, a schema is like a frame that defines the slots and labels where information can enter.

It helps organize:

• whether this information is about an activity,
• whether it is about a place,
• whether it is about an emotion,
• whether it is about an interest,
• and what relationship it may have with other information.

Let’s return to the earlier utterance.

• “I went camping” can carry labels such as outdoor activity, weekend experience, and activity in nature.
• “I saw the stars” can carry labels such as observation, night, and nature experience.
• “I thought of people in the past” can carry labels such as historical imagination, life in the past, and curiosity about ancient times.

Only when information carries these labels can it surface again in many different situations.

For example, if the user later asks,

“Are there any outdoor activities I could try?”

the camping experience can surface.

If the user asks,

“Was there ever a moment when I felt interested in history?”

the memory of looking at the stars and thinking about people in the past can surface.

If the user asks,

“Was there a time when direct experience made me think more deeply?”

this utterance can become relevant again because it connects an experience in nature with historical imagination.

It is the same sentence.

But depending on the labels attached to it,

it can be found again in different situations.

To recall well,

information must first be marked well.

Information from Different Utterances Connects Through Relationships

But labels alone are not enough.

Information gains greater meaning when it connects to other information.

The important point here is that information does not only split and connect within one utterance.

The truly important relationships also emerge between pieces of information from different utterances.

Let’s return to the earlier utterance.

“I went camping over the weekend, and while looking at the stars, I wondered what people in the past might have thought.”

This utterance contains information such as camping, observing stars, imagining people in the past, and curiosity about life in the past.

Now suppose the user says this on another day:

“I found the story about Egyptian civilization the most interesting in world history class.”

This utterance contains information such as world history class, Egyptian civilization, and interest in ancient civilizations.

On the surface, the two utterances look like different stories.

One is about camping.

The other is about class.

If we simply store the raw text,

they may appear as separate records left on different dates.

But inside an ontology, the information from these two utterances can connect.

The core of the first utterance is not only the fact that the user went camping.

What matters is that the user looked at the stars, imagined the lives of people in the past,

and became curious about life in the past.

The core of the second utterance is not only the fact that the user attended world history class.

What matters is that the user felt interest in an ancient world called Egyptian civilization.

Seen this way, the two utterances are not far apart.

One is historical imagination that emerged from an experience in nature.

The other is interest in ancient civilization revealed inside a class.

The two pieces of information can connect through a relationship:

curiosity about life in the past.

This is the difference that relationships create.

In a simple storage system, the information may remain like this:

A camping story.

A world history story.

They are separate records.

But inside an ontology, they are structured differently.

The experience of seeing stars while camping connects to imagining the lives of people in the past.

Interest in Egyptian civilization connects to interest in ancient civilizations.

And these two pieces of information come closer through the relationship of curiosity about past lives and civilizations.

Now the first utterance and the second utterance do not remain only as separate records.

Information from different conversations starts to connect through relationships.

Memory is not a list of words.

Memory is closer to how pieces of information connect through relationships.

That is why what matters in ontology is not only the information units themselves.

What matters is how those units connect to one another.

• What kind of interest does a certain experience reveal?
• What old curiosity does a certain class topic connect to?
• How does one statement come close to another statement from a different day?

Only when these relationships exist can different memories surface together at the moment they are needed.

As Relationships Accumulate, Larger Clusters Form

Once a relationship forms, different records start to connect.

But memory does not form through one connection alone.

When information from multiple utterances repeatedly forms relationships,

those pieces of information begin to be read together inside a larger flow of meaning.

A relationship is a connection between one piece of information and another.

A cluster is a larger unit of understanding created as those connections accumulate again and again.

Now suppose we add more conversations to the two earlier utterances.

• “I remember touching pottery directly at the museum.”
• “I read an English passage about pyramids, and for some reason it stayed with me for a long time.”
• “I watched a history documentary showing the excavation of ancient ruins, and it was interesting.”

On the surface, these utterances are different stories.

One is a museum experience.

One is an English passage.

One is a documentary.

If we add the earlier camping story and world history class story, the situations become even more diverse.

• Camping.
• World history class.
• Museum experience.
• English passage.
• Documentary.

Each statement appeared on a different day, in a different context.

If we only store the raw text, these utterances may remain as separate records.

But when we look at the information inside each utterance, repeated connections begin to appear.

The experience of looking at the stars while camping and thinking about people in the past

connects to curiosity about life in the past.

The memory of finding Egyptian civilization interesting in world history

connects to interest in ancient civilizations.

The experience of touching pottery directly at the museum

connects to the experience of seeing and touching old artifacts in person.

The fact that the English passage about pyramids stayed in the user’s mind

connects to an interest in ancient civilizations that repeatedly appears even during learning.

The experience of enjoying a scene about excavating ancient ruins

connects to interest in activities that uncover traces of the past.

At first, these pieces of information look like separate records.

But if we follow the relationships, these utterances can be read as one larger cluster.

An interest in directly seeing, imagining, and understanding

the lives and civilizations of the past.

Inside this cluster,

the experience of looking at stars while camping does not remain only as a simple outdoor activity.

It becomes a moment when the user imagined the lives of people in the past through a present scene.

The memory of finding Egyptian civilization interesting in world history class also does not remain only as a class record.

It becomes a cue showing that the user repeatedly responds to ancient civilizations.

The experience of touching pottery at the museum is not just a record of a museum visit.

It connects to a moment when the user saw and touched an old artifact directly and felt closer to life in the past.

The pyramid passage and the documentary about excavating ruins work the same way.

Each may look like a record of English study or watching a video.

But if we follow the relationships, they connect to an interest in imagining people’s lives through traces of the past.

Seen this way, the important point is not simply that a new cluster has formed.

What matters more is that utterances that had been separate

now begin to be read together inside one flow of meaning.

This is not a simple list of information.

It is one structure of understanding created by information from multiple utterances through relationships.

Of course, this cluster is not a fixed folder that closes around the information.

One piece of information can be read again inside another cluster depending on the situation.

For example, the experience of touching pottery at the museum

can be read inside an interest in past lives and civilizations.

At the same time, it can also become a cue to a learning style based on seeing and touching directly.

The same applies to the experience of looking at stars while camping.

It can be read as a memory of imagining the lives of people in the past.

It can also be read as an experience where thinking deepened in nature.

In ontology, a cluster is not one fixed folder where information is stored.

It is a structure in which information from multiple utterances

overlaps through relationships

and can be read again in different ways depending on the situation.

That is why ontology does not fix one piece of information in only one place.

It reconnects information through multiple relationships

and helps it surface again inside the flow of meaning that fits the moment.

Records Are Stored, But Memories Are Structured So They Can Be Recalled

At this point,

we can see why storing raw conversation text in chronological order is not enough.

When we preserve the raw text, we retain what the user said.

But that alone makes it difficult to retain:

• what information existed inside the statement,
• what that information connects to from other conversations,
• and what larger interests or experiences can be read from it.

In that state, even if the conversation remains,

it becomes difficult to recall it at the moment it is needed.

For AI to remember conversations with a user well,

it must not stop at storing conversations as raw text by date.

It must:

• read the information inside each utterance as multiple units,
• label what kind of information each unit is,
• create relationships between information from different utterances,
• and let those relationships grow into larger clusters.

This is why ontology is necessary.

A conversation should not remain only as raw text.

It must be structured as connected information that can surface again later.

Records are stored.

But memory is structured so it can be recalled.

Now we can move to the next question.

How does this structured information system work when a user says something new?