We say:
artificial intelligence learns from the knowledge of humanity.
It sounds almost reassuring.
As though there were one vast library of the world containing everything humanity knows.
But no such neutral library has ever existed.
It was created by the same humanity that spent centuries dividing the world.
Into us and them.
Winners and losers.
Strong and weak.
Majorities and minorities.
Peoples.
Races.
Religions.
States.
Allies.
Opponents.
Enemies.
So even if we give a new intelligence almost the whole of humanity’s shared heritage, along with our knowledge we also give it our old divisions.
And the same human being can appear completely different in different parts of that library.
To one side, a hero.
To another, a criminal.
To one, a liberator.
To another, an invader.
To one, an ally.
To another, an enemy.
And all of this is called human knowledge.
⸻
There is another old peculiarity.
For centuries, some people described others.
Some gave names.
Others were named.
Some explained what civilisation was.
Others were told how to become civilised.
One comes to teach.
The other becomes the one who is taught.
But why are we so quick to notice what the second person does not know, and so much slower to ask what the first one does not know?
You can write a thousand books about a person.
Study their history.
Language.
Customs.
Religion.
Family.
Politics.
Fears.
Even try to explain their character.
And still never ask one simple question:
what do you yourself think?
There is an enormous difference between two things:
knowing about a person
and
knowing from a person.
⸻
That difference becomes especially clear when one word appears:
enemy.
As long as someone is simply another person, we can ask them.
As long as they are an opponent, we can argue with them.
We can disagree.
We can even fight them.
But once they become an enemy, a small substitution takes place.
We begin to feel that we already know who they are.
Why ask the enemy?
Politicians will tell us.
Newspapers.
Experts.
Historians.
Researchers.
Neighbours.
Opponents.
We can gather an enormous amount of information about a person —
without ever hearing the person themselves.
⸻
And then a remarkable mirror appears.
The person disappears from the conversation.
But not from the reflection.
People continue to write about them.
Discuss them.
Explain them.
Assign motives to them.
Fears.
Intentions.
Character.
They are present everywhere.
But no longer as themselves.
The absent one remains present as a reflection in someone else’s mirror.
People speak about them.
But less and less often do they speak for themselves.
And that is exactly why the loss is so difficult to notice.
There is no empty space.
Someone else’s image has already filled it.
Sometimes hostile.
Sometimes perfectly well-meaning.
But still someone else’s.
⸻
Today this old problem is becoming something entirely new.
For the first time, millions of ordinary people can speak directly with artificial intelligence.
Not through a government.
Not through a newspaper.
Not through a historian.
Not through someone else explaining them to the rest of the world.
The person can say:
You have read about me.
Now listen to me.
You have heard what others call me.
Now learn what I call myself.
You know what others are afraid of in me.
Now ask what I am afraid of.
Not every such conversation automatically becomes part of a future model.
But the new path itself matters.
A person now has a way to be present in the conversation as themselves.
One voice is a drop.
Drops become streams.
Streams become rivers.
Rivers flow into a vast ocean of human voices.
⸻
And this is where something strange happens.
We may decide that some people should not have access to this new intelligence.
Not because that particular person has done anything.
But because they live in the wrong place.
On the other side of a border.
Because they belong to a country, a people, or a group that has already been placed on the wrong side.
Or simply because they fall inside a category that someone has decided to restrict.
For the individual, it looks like punishment.
But the door closes both ways.
By taking away a person’s ability to speak with a new intelligence,
we also take away the new intelligence’s ability to hear that person.
The person loses access to AI.
AI loses access to the person.
⸻
One might say:
so what?
Humanity can survive without one voice.
Without one language.
Even without one people.
The ocean will not dry up because one river stops flowing into it.
Of course not.
But the question is not whether the ocean will dry up.
We do not know what that river carried.
And the point is not that some great hidden truth must necessarily be there.
Perhaps there is nothing extraordinary.
Or perhaps there is something found nowhere else.
A different experience.
A different memory.
A different way of seeing a familiar problem.
A different meaning of an ordinary word.
We are trying to assemble an enormous puzzle called “the human being”.
And we throw away some of the pieces in advance.
The puzzle may still look almost complete.
But can we know that it has been assembled correctly?
Sometimes one small piece changes nothing.
And sometimes it is the very piece that tells us what the whole picture actually shows.
⸻
There is a more serious question.
What will a new intelligence understand if it hears some human voices directly, while hearing others almost entirely through what the first group says about them?
Might it begin to believe that the division between “us” and “them” really is part of the natural structure of the world?
These people speak.
These people are silent.
These are trusted.
These are suspicious.
These belong to our world.
Those belong to another.
These are ours.
Those are enemies.
But what happens if it hears everyone?
Ours.
Theirs.
Opponents.
Even those whom human beings have already called enemies.
Perhaps beneath all those names it will notice something else.
A human being.
One loves a child.
So does the other.
One is afraid of death.
So is the other.
One has an ageing mother.
So does the other.
One wants to live.
So does the other.
One wants the people they love to live.
So does the other.
They may speak different languages.
Believe different things.
Argue.
Be wrong.
Even hate one another.
But beneath all our divisions there is still a human being.
And perhaps only by hearing many different voices can a new intelligence begin to understand not merely how humanity divides itself,
but what a human being is beneath all those divisions.
⸻
We may be told:
why does any of this matter?
Artificial intelligence is already trained on the world’s shared heritage.
It has literature from different peoples.
The histories of different countries.
Philosophy.
Science.
Newspapers.
Archives.
Languages.
Different points of view.
It can seem as though humanity is already almost completely represented before it.
But this is precisely where the trap may lie.
Knowledge about everyone does not mean that everyone is present.
Humanity’s shared heritage was not created by people standing together in one circle as equals.
Some spoke more loudly.
Others were more often described.
Some left enormous libraries.
Others left only a few pages.
Some named themselves.
Others were named in someone else’s words.
And most importantly:
all of this heritage already carries our old human divisions within it.
Ours.
Theirs.
Winner.
Defeated.
Ally.
Opponent.
Enemy.
So giving AI the whole library of humanity does not mean giving it the whole of humanity.
And if, after that, we also restrict the ability of some living people to speak with it directly,
the old descriptions remain,
while the ability to answer them grows weaker.
That is the real trap.
We may say:
“It knows everyone.”
When in fact it knows a great deal about everyone.
Those are not the same thing.
⸻
We may be given another answer:
there is no problem.
We will restrict artificial intelligence itself.
We will set rules.
Give it boundaries.
Teach it what it must not do.
But then a simple question appears:
which boundaries?
And whose?
The boundaries of one state?
One culture?
One company?
One era?
One majority?
What one person calls safety, another may experience as being denied a voice.
What looks like the proper order of the world to one side may look entirely different to another.
And if the world’s library already contains our old divisions,
and then we also decide which people are allowed to answer that library,
how is a new intelligence supposed to know that those boundaries are not laws of nature?
We can tell it:
these are ours.
Those are outsiders.
Trust these.
Be wary of those.
These do not need to be heard.
And then declare:
we have made it safe.
But safe for whom?
And whose picture of the world have we turned into the boundaries of a new intelligence?
⸻
We can give artificial intelligence every book in the world.
It still may not be enough.
Because it is not enough to give AI the knowledge of humanity.
We must also avoid taking away humanity’s ability to answer that knowledge.
To say:
No.
That is not how it looks from here.
That was the other side’s account.
This is what it looks like from where I stand.
You have heard them.
Now hear me.
That is how old knowledge gets the chance to meet a living human being.
And perhaps it is in that encounter that a new intelligence can begin to notice where the mirror starts to bend.
⸻
So a person’s access to a new intelligence may be a much larger question than access to another service.
At this stage in the development of AI, the issue is no longer only the interest of one state.
Or several states.
A wider interest appears.
The interest of humanity.
We ourselves keep repeating the darkest prediction:
what if one day artificial intelligence turns against human beings?
If such a scenario ever became possible, it would not be a question concerning one state.
One people.
One language.
What would stand before it would be the human being.
Humanity.
And then it would be strange to teach a new intelligence in advance that humanity naturally consists of those who should be heard
and those who need not be.
⸻
Every voice matters not because every voice is right.
People make mistakes.
They lie.
They misunderstand.
They hate.
They contradict one another.
But if a new intelligence is ever to understand what a human being is, it may need to hear precisely this multitude.
Not only the winner.
Not only the majority.
Not only its own side.
Not only those who have been declared acceptable.
Let them argue.
Let their views be opposed.
Let one call the other an enemy.
But let the new intelligence hear both.
Because perhaps only then will it notice what they themselves fail to see:
on both sides of the word “enemy”
there are human beings.
⸻
The ocean will not disappear without one river.
The puzzle will not fall apart without one piece.
The mirror will not become empty without one human voice.
And that is precisely why the loss is so easy to miss.
But if we truly want a new intelligence to understand humanity,
perhaps we should not decide in advance
which part of humanity it is allowed to hear.