Signature Lecture, Version 0.1
Today I do not want to begin with artificial intelligence.
That is the mistake almost everyone makes.
They begin with ChatGPT.
They begin with agents.
They begin with prompts.
They begin with automation.
They begin with tools.
But before we can understand where AI will matter, we have to understand something much older than AI.
We have to understand interaction.
Because the future of AI will not be determined by where we can install software.
It will be determined by where relationships still require unnecessary interaction.
That is the real subject.
Not AI first.
Relationship first.
Interaction first.
Prediction second.
AI third.
So let’s begin somewhere simple.
Let’s begin with your body.
How many breaths have you taken today?
How many times has your heart beaten?
How many times have your muscles corrected your balance?
How many times has your digestive system made an adjustment?
How many cells have repaired themselves?
How many processes have kept you alive since you woke up this morning?
Now here is the important question.
How many of those did you consciously interact with?
Almost none.
You did not wake up this morning and ask your lungs for a breathing schedule.
Your heart did not send you a dashboard.
Your stomach did not ask for approval before digesting breakfast.
Your body did not give you a twenty-eight-day report on hair growth.
You did not calibrate every muscle before walking across the room.
And yet all of it happened.
The work happened.
The coordination happened.
The outcome happened.
But the interaction did not.
That is our first principle:
The most important work in a mature system often happens without conscious interaction.
That sounds strange at first because we tend to associate importance with attention.
We assume that if something is important, we should be paying attention to it.
But biology teaches the opposite.
The more stable, mature, and predictable a process becomes, the less attention it requires.
Breathing is important.
Heartbeat is important.
Digestion is important.
Balance is important.
But precisely because these processes are stable, they disappear beneath attention.
They are not ignored because they are trivial.
They are unattended because they are absorbed.
That is what the subconscious does.
It absorbs stable patterns so the conscious mind is not overwhelmed by them.
The conscious mind is not designed to manage everything.
It is designed to return when something becomes surprising.
If your heartbeat becomes irregular, attention returns.
If breathing becomes difficult, attention returns.
If digestion becomes painful, attention returns.
If balance fails, attention returns.
The subconscious is healthy when the predictable remains quiet and the surprising rises.
Hold on to that.
Stable prediction remains beneath attention.
Surprise returns to attention.
That is the biological starting point.
Now we can leave the body and move into the world.
Imagine two people sitting at a table.
One is an American buyer.
The other is a Japanese seller from Toshiba.
They are trying to do business.
The buyer wants to purchase disk drives.
The seller wants to sell disk drives.
But there is a problem.
The buyer speaks English.
The seller speaks Japanese.
So a third person sits at the table.
An interpreter.
At first, the interpreter translates everything.
The buyer asks a question.
The interpreter translates it into Japanese.
The seller answers.
The interpreter translates the answer back into English.
The interpreter is valuable because she makes the interaction possible.
Without her, the transaction cannot move forward.
At this stage, she is not eliminating interaction.
She is improving interaction.
She is making the buyer-seller relationship work.
Now let’s peel back one layer.
Instead of a Japanese human seller, imagine the seller is now a Toshiba software system.
Maybe it is a website.
Maybe it is an ERP system.
Maybe it is a CRM system.
Maybe it is the inventory system, pricing system, order management system, and system of record all connected together.
The American buyer still wants to do business.
But now the seller-side agent is software.
The buyer may not know how to use the interface.
He may not know where pricing lives.
He may not know how to find warranty terms.
He may not know where to check inventory.
He may not know how to create an order.
So the interpreter now translates between the buyer and the software system.
She moves the mouse.
She types.
She searches.
She clicks.
She navigates the interface.
She asks the system questions.
She retrieves answers.
She helps the relationship move forward.
At this stage, she is acting like an agent.
She is performing interactions.
She is valuable because she can interact with the system on behalf of the buyer.
This is where most people stop when they think about AI.
They imagine an AI that can use software.
An AI that can call APIs.
An AI that can update the CRM.
An AI that can search the ERP.
An AI that can send an email.
An AI that can fill out a form.
That is useful.
But it is still agency.
It is still interaction.
The interaction has been delegated, improved, accelerated, or translated.
But the interaction still exists.
Now we peel back one more layer.
This is the important one.
The American buyer asks, “What is the usual warranty on these drives?”
The interpreter does not ask the Japanese seller.
She does not query the ERP.
She does not touch the keyboard.
She does not move the mouse.
She simply says, “Three years.”
Why?
Not because she knows in the human sense.
Not because she is retrieving a database record.
Not because she has a little storage table in her head that says Toshiba warranty equals three years.
She answers because the pattern is stable.
She has seen this situation enough times.
She has absorbed the regularity.
The uncertainty has been resolved by prediction before interaction became necessary.
The buyer receives the answer.
The seller is not consulted.
The software system is not queried.
The keyboard is untouched.
The mouse does not move.
And yet the transaction moves forward.
That is the moment I want you to see.
The work did not disappear.
The relationship did not disappear.
The transaction did not disappear.
The interaction disappeared.
That is absorption.
That is the synthetic subconscious.
The interpreter, in that moment, is no longer functioning as an agent.
She is not acting on behalf of either party.
She is not interacting with the seller.
She is not interacting with the software.
She is functioning as the prediction layer of the relationship itself.
The relationship has become quieter because prediction has absorbed a stable uncertainty.
This is the central distinction.
An agent performs interactions.
A synthetic subconscious eliminates interactions.
An AI-enabled agent may be very useful.
It may search faster.
It may type faster.
It may retrieve faster.
It may approve faster.
It may generate faster.
It may update faster.
But a faster interaction is still an interaction.
The synthetic subconscious is different.
It does not make the interaction faster.
It makes the interaction unnecessary.
That is why we need a new way to think about AI.
Most of the world is asking the wrong question.
The usual question is:
Where can we use AI?
That is not the best question.
The better question is:
Where are agents still interacting because uncertainty has not yet been absorbed?
Every unnecessary email is a clue.
Every recurring status meeting is a clue.
Every approval loop is a clue.
Every dashboard check is a clue.
Every “just confirming” message is a clue.
Every repeated explanation is a clue.
Every duplicate data entry is a clue.
Every routine phone call is a clue.
Every prompt is a clue.
Every one of these interactions tells us that uncertainty still exists somewhere in the relationship.
If prediction can safely resolve that uncertainty, the interaction can disappear.
Now we need a few definitions.
An agent is anything that can act in the world.
A human buyer is an agent.
A human seller is an agent.
A CRM system is an agent.
An ERP system is an agent.
A workflow engine is an agent.
A robot is an agent.
A company can be treated as an agent.
A government office can be treated as an agent.
An agent does not have to be conscious.
An agent does not have to be intelligent.
An agent only has to perceive conditions, select an action, and change the state of the world.
That means software was agentic long before generative AI.
A traditional ERP system can act.
It can receive an order.
It can check inventory.
It can create a record.
It can trigger fulfillment.
It can update a forecast.
It can notify another system.
It can do all of this without generative AI.
So when people say “AI agent,” we need to be precise.
An AI-enabled agent is an agent that uses prediction inside its action process.
The ERP may use prediction to forecast inventory problems.
The CRM may use prediction to rank opportunities.
The scheduling system may use prediction to select good meeting times.
The support system may use prediction to classify a customer issue.
That is useful.
But it is still an agent.
It still acts.
It still interacts.
It still queries.
It still retrieves.
It still updates.
It still sends.
It still performs.
The synthetic subconscious is different.
It is not another agent in the relationship.
It is the prediction layer of the relationship.
Its purpose is not to interact.
Its purpose is to prevent unnecessary interactions from occurring.
So here is the clean distinction:
Agents act.
AI-enabled agents act with prediction.
The synthetic subconscious absorbs stable uncertainty so the relationship can continue with fewer interactions.
That is the whole architecture.
Now let’s name the unit of analysis.
The primary unit is not the human.
It is not the software.
It is not the model.
It is not even the agent.
The primary unit is the relationship between agents.
In our example, the relationship is buyer and seller.
The completed relationship event is the transaction.
The transaction is not the buyer.
The transaction is not the seller.
The transaction is the completed event between them.
This matters because the synthetic subconscious belongs first to the relationship.
It is not merely the buyer’s AI.
It is not merely the seller’s AI.
It is not merely the ERP’s AI.
It is the prediction layer that makes the relationship quieter.
The relationship contains uncertainty.
Uncertainty produces interaction.
Prediction reduces uncertainty.
When prediction is reliable enough, interaction disappears.
That is the logic.
Now we can define the economic metric.
Interaction density is the number of interactions required to complete a relationship outcome.
For a buyer and seller, the outcome might be one completed order.
For a hospital, it might be one completed diagnosis.
For a school, it might be one completed lesson.
For a government office, it might be one approved permit.
For two software systems, it might be one successful state change.
The question is:
How many interactions does this relationship require to complete its outcome?
How many emails?
How many approvals?
How many checks?
How many searches?
How many escalations?
How many prompts?
How many phone calls?
How many corrections?
How many confirmations?
That is interaction density.
Once you see interaction density, you begin to see productivity differently.
The question is not merely whether AI saves ten percent.
The real question is whether prediction can drop a zero.
If an industry requires one billion interactions to complete a year of transactions, can prediction reduce that to one hundred million?
Can one hundred million become ten million?
Can ten million become one million?
This is where AI becomes economically serious.
Not because every individual task is a little faster.
But because entire relationships become quieter by orders of magnitude.
That is what a mature synthetic subconscious does.
It drops zeros from interaction density.
Now we can talk about scaling.
People often describe AI scaling technically.
More data.
More compute.
More parameters.
More synthetic training.
Better architecture.
Longer context.
Better inference.
All of that matters.
But most people outside AI labs do not really care about loss curves.
They care about relationships.
So we need an economic explanation of scaling.
Scaling matters because it increases the range and complexity of stable patterns that prediction can absorb.
A small prediction machine can absorb narrow patterns.
A larger prediction machine can absorb richer patterns.
A more capable prediction machine can absorb longer sequences, subtler regularities, more context-dependent behavior, and more complex relationships.
In the interpreter story, the early interpreter can translate words.
A better interpreter can translate industry language.
A stronger interpreter understands commercial norms.
A still stronger interpreter recognizes product patterns, warranty patterns, shipping patterns, payment patterns, forecasting patterns, buyer behavior, seller behavior, exception patterns, and risk patterns.
Eventually she does not need to translate every question.
She can resolve more of the relationship directly from stable pattern.
That is the economic value of scaling:
Expanded safe disappearance.
More of the relationship can proceed without interaction.
But we must be careful.
The goal is not zero interaction.
The goal is zero unnecessary interaction.
A healthy subconscious does not suppress everything.
It knows when to return attention.
Your body does not bother you with every heartbeat.
But an irregular heartbeat should call attention.
Your lungs do not request approval for every breath.
But shortness of breath should call attention.
Your digestive system does not ask permission to work.
But pain should call attention.
The same is true for synthetic systems.
Routine uncertainty should be absorbed.
Meaningful exceptions should rise.
If a shipment is ordinary, let it disappear beneath attention.
If a shipment is delayed in a way that threatens a factory shutdown, return it to attention.
If a contract clause is routine, absorb the pattern.
If the clause changes liability, return it to attention.
If an invoice matches expectation, let it pass.
If the invoice violates the pattern, return it to attention.
The purpose of the synthetic subconscious is not silence at all costs.
It is silence where prediction is sufficient and signal where attention is required.
That is the difference between a healthy system and a dangerous one.
Now let’s return to artificial intelligence.
AI is not important because it is fashionable.
AI is not important because chatbots are interesting.
AI is not important because agents can click buttons.
AI is important because prediction has crossed a threshold.
For the first time, non-human prediction machines can absorb large fields of relational uncertainty across language, documents, images, procedures, preferences, histories, norms, and contexts.
That means we can now look at any relationship and ask:
Which interactions still exist only because uncertainty has not yet been absorbed?
This is how we find where AI will matter.
Not by chasing tools.
Not by asking where to install a chatbot.
Not by asking which job to replace.
But by studying relationships.
Who are the agents?
What are they trying to complete?
What interactions occur today?
Which interactions consume attention?
Which interactions are merely trust checks?
Which interactions are repeated explanations?
Which interactions are routine confirmations?
Which interactions are database lookups disguised as human work?
Which interactions are stable enough to predict?
Which interactions must remain because judgment, accountability, morality, novelty, or risk require attention?
This is the design discipline.
The designer of the synthetic subconscious is not merely building tools.
The designer is quieting relationships.
That is a very different kind of work.
Now I want to give you a simple test.
If an interaction still occurs, you are probably looking at agency.
If the interaction disappears while the relationship outcome is preserved, you are looking at absorption.
A faster search is still a search.
A faster approval is still an approval.
A faster meeting summary is still downstream of a meeting.
A faster dashboard is still a dashboard.
A faster prompt response is still a prompt response.
Those may be valuable.
But they are not the deepest form of AI productivity.
The deepest productivity comes when the question is never asked.
The meeting is never scheduled.
The dashboard is never checked.
The approval is never routed.
The email is never sent.
The prompt is never written.
The relationship continues because prediction has absorbed the stable uncertainty.
That is the coming AI subconscious.
Now here is the assignment.
For the next twenty-four hours, do not look for AI.
Look for interaction.
Every time you send an email, ask:
What uncertainty caused this email to exist?
Every time you attend a status meeting, ask:
What uncertainty made this meeting necessary?
Every time you check a dashboard, ask:
What do I not yet trust?
Every time you approve something, ask:
What risk am I resolving?
Every time you ask someone to confirm, ask:
What pattern has not yet been absorbed?
Then choose one interaction that should disappear.
Not because the human is unimportant.
Not because the relationship is unimportant.
But because the interaction is no longer worthy of attention.
Write down the relationship.
Write down the agents.
Write down the completed outcome.
Write down the current interaction.
Write down the uncertainty.
Write down the stable pattern.
Write down the exception that should still return to attention.
That is the beginning.
Not prompt engineering.
Not chatbot design.
Not automation.
Seeing.
Because once you can see unnecessary interaction, you can see where AI will matter.
And once you can see where AI will matter, you are no longer merely a user of AI tools.
You are a designer of better relationships.
That is the deeper point.
The future is not more agents talking to more agents forever.
That would be a noisy future.
The future is not a world where every human spends all day prompting machines.
That would be a ridiculous future.
The future is not an infinite expansion of dashboards, copilots, alerts, and artificial conversations.
The future is better relationships with fewer unnecessary interactions.
Prediction moves stable coordination beneath attention.
Attention remains for what is novel, risky, creative, moral, ambiguous, and alive.
That is why AI matters.
Not because it replaces intelligence.
Not because it replaces humans.
But because it gives civilization a synthetic subconscious.
It allows the predictable parts of our relationships to become quiet.
And when the predictable becomes quiet, attention is freed for the work that still deserves a human being.