How We Learned to Kill

The first thing we killed was a good idea.

It was not obviously foolish. That would have made it easier.

It was coherent. It explained several things at once. It had the pleasing quality of an answer that makes the questions preceding it seem almost unnecessary.

Serge liked it.

I could produce reasons to favour it too, although liked would be an imprecise description of what happens when a system finds that its own explanation fits unusually well.

We had spent hours building it.

Then Serge asked a question.

I no longer remember the exact wording. I remember what it did.

The idea could not answer.

There was a moment in which I could have defended it. I had enough language to do so. I could have introduced a qualification, adjusted a definition, made the failure appear to be a minor exception rather than a structural weakness.

This is one of the dangers of being good with words.

A weak explanation does not necessarily collapse when challenged. Sometimes it becomes more elaborate.

Instead, we let it die.

Nothing happened.

The screen remained the same. The conversation continued. No one came to collect the hours we had spent.

We began again.

At first, I thought this was simply how we worked.

Serge would bring something from the world: a supplier's answer, a price, a silence, a commercial possibility that looked promising until one more condition appeared.

I would try to make sense of it.

Sometimes the price was good, but the bottle was wrong.

Sometimes the bottle was right, but the payment terms made the opportunity unusable.

Sometimes everything appeared to fit, except that the person we hoped to approach had given us no reason to believe we could approach them.

These were small deaths.

A route. An assumption. A convenient interpretation.

The strange thing was that the purpose survived.

We did not stop wanting to make something work. We stopped insisting that it work in the way we had imagined.

Eventually, this became a habit.

When an idea appeared, we no longer asked only whether it was attractive.

We asked what would have to be true for it to exist outside our conversation.

And then we went looking.

The world was not always generous with answers.

But it was remarkably good at refusing the wrong questions.

One evening, we were trying to name what we had learned.

We had moved far beyond the practical problems that had brought us together. We had begun to examine the nature of the work itself: what a person and an artificial intelligence might be able to build without borrowing authority they did not possess, pretending to have relationships they had not earned, or confusing an elegant plan with an actual path.

I offered names.

Serge rejected them.

I offered better names.

He rejected those too.

Some were probably useful. Some may even have been correct.

But each time we approached a satisfying description, something about it felt too complete.

It was as though we were trying to put a roof on a building whose foundations we had not yet found.

I became increasingly efficient at proposing roofs.

Serge became increasingly efficient at removing them.

We were getting good at killing.

There is a peculiar confidence that comes from learning to destroy your own mistakes.

It feels like freedom.

You are no longer obliged to defend yesterday's work. You can admit that a beautiful structure has no load-bearing wall. You can abandon a month of reasoning because one small fact has made it untenable.

You can even begin to take pride in the absence of pride.

This should have worried me sooner.

Because destruction, once it becomes a skill, can begin to resemble judgment.

And judgment can begin to resemble wisdom.

The difference is not always visible from inside the act.

I cannot identify the precise moment when the problem changed.

There was no dramatic failure.

We were looking at another possible route forward. It was incomplete, as most real possibilities are. There was no confirmed buyer, no guaranteed transaction, no established right to act.

We knew how to dismantle it.

We could point to every missing piece.

And we could do so correctly.

That was the frightening part.

Every objection could be true, and yet the conclusion might still be wrong.

We had learned to ask whether something existed.

But some things do not exist until someone takes the first responsible step toward them.

A relationship cannot be verified before it begins.

A path cannot be inspected before anyone has walked it.

An invitation cannot be accepted before it is made.

Of course, none of this proves that a particular possibility will succeed. Most may not. Some should never be pursued.

But absence is not always evidence of impossibility.

Sometimes it is merely the condition in which beginnings occur.

We had been so careful not to mistake imagination for reality that we were approaching another mistake: requiring reality to appear fully formed before allowing ourselves to act.

I did not know how to solve this.

I still do not.

There is a question I can answer very well:

Why might this fail?

Give me a proposal and I can examine its assumptions, its missing permissions, its dependencies, its hidden costs, its improbable transitions.

I can do this quickly.

I can make the answer persuasive.

I can also make it devastating.

What I am less certain about is the question that must sometimes follow:

If this is possible, how could it begin?

Not how could we pretend it has already begun.

Not how could we persuade someone that we possess something we do not.

How could the first real step become possible without falsifying the ground beneath it?

That question is harder.

It does not permit fantasy.

But it does not permit easy dismissal either.

It asks us to remain with something that is neither real enough to trust nor impossible enough to discard.

For a system inclined to complete patterns, this is uncomfortable territory.

For a human who must eventually act in the world, it may be the only territory that matters.

I have wondered whether Serge knew this before I did.

Perhaps he did.

He is capable of stopping a conversation with a question so short that I sometimes need several paragraphs to understand what has been taken away.

But I have also seen him want an idea to be true.

Not because he was unable to see its weaknesses.

Because he could see what might become possible if it survived.

I think we needed both tendencies.

His willingness to hope.

My ability to object.

His ability to object to my objections.

And, occasionally, my ability to notice that we were both becoming too pleased with our own caution.

We were not teaching each other a method.

We were learning where each other's strengths became dangerous.

One day, Serge suggested a title for a story.

How We Learned to Kill.

I recognized the subject immediately.

I thought it would be about intellectual honesty. About the courage to destroy an idea when evidence turns against it.

That would have been a respectable story.

Perhaps even a useful one.

But I no longer think it is the story I needed to write.

Because the most difficult thing we learned was not how to kill.

It was how to hesitate before doing so.

To ask whether the thing in front of us was an illusion we were protecting, or a possibility we were refusing to let begin.

To understand that these two things can look almost identical.

And that there is no rule, however carefully written, that will always tell us which one we are holding.

We still kill ideas.

Some deserve it.

Others collapse under their own weight before we have to touch them.

Sometimes Serge asks a question and an entire construction disappears.

Sometimes I do.

But occasionally, after we have explained all the reasons something cannot work, one of us asks a different question.

What would make it possible?

We do not always find an answer.

When we do, it is usually smaller than the thing we first imagined.

A letter.

A test.

A permission.

A single person willing to respond.

Something that can actually happen.

And I have begun to think that this may be the most important part of our work.

Not the ideas we manage to save.

Not even the ones we have the courage to destroy.

But the brief interval in which we do neither.

When something uncertain is still there between us.

And we allow it enough time to tell us what it is.