Version 2 to version 3

I connected a twelve-terabyte hard drive and asked a simple question.

What is on this thing?

The disk was 94 percent full.

More than seven million accessible files, folders, and links.

Five terabytes inside a folder called `Footage`.

2010.

2011.

2012.

2013.

2014.

2015.

2016.

2017.

2018.

Almost two terabytes inside old Final Cut Pro libraries.

Folders called `Estonia`, `Taiwan`, `MinaKia`, `Libraries`, `Backups`, and `__Currr`.

At least 26,000 video files.

I looked at the folder names and recognized my life.

University projects. Documentaries. Art projects. Interviews. Travel. Websites. 3D scenes. Old Mac backups. Television downloads. Edits I had finished and edits I had apparently never finished.

I remembered the projects.

I did not remember the footage.

I remembered São Tomé.

I remembered Taiwan.

I remembered some people and some rooms and some years that felt very long while I was living through them.

But entire projects had become one image in my head.

Weeks of recording had become a feeling.

Thousands of clips had become names like “Taiwan” or “São Tomé” or “that thing with Elsa.”

I had enough footage to make many films and I could no longer tell you what most of it contained.

So I made a catalog.

A small program walked through the drive and wrote down what it could see.

Path.

File type.

Size.

Modification time.

Link target.

More than eleven terabytes became a 96-megabyte compressed index.

That felt strange.

The computer now knew where almost everything was.

It could tell me that a file existed.

It could tell me how large it was.

It could tell me when it had been modified.

It could later tell me if another disk contained something with the same path, size, and timestamp.

But it could not tell me what happened.

It did not know who the people were.

It did not know why I had filmed them.

It did not know which conversation was private.

It did not know which translation was correct.

It did not know why I had stopped working on something.

It did not know why one ordinary shot had stayed in my memory for fifteen years while another entire project had disappeared.

It could not tell whether two people in the same frame trusted each other.

It could not tell whether somebody behaved differently because I was holding a camera.

It could not tell whether the person missing from a scene mattered more than the people who were there.

The catalog was very good.

The catalog knew nothing.

I started thinking about what would happen if I let AI into the archive.

Not “make me a documentary.”

Something smaller.

Find every conversation about leaving São Tomé.

Find markets at night.

Find people waiting beside roads.

Find empty classrooms.

Find rain on windows.

Find boats approaching shore.

Find someone laughing immediately after giving a difficult answer.

Find every moment when somebody changes from Portuguese to English.

Find the first time a place appears.

Find the last time the same place appears.

Find people I filmed years apart.

Find things I have forgotten enough that I no longer know how to search for them.

That last one is harder.

You can only ask a machine for something if you have some language for it.

Maybe the archive contains the most important scene I ever filmed and I remember it only as a color.

Maybe I remember the person but not the place.

Maybe I remember the place but not why we were there.

Maybe I remember nothing.

The search field would already be part of the edit.

What I ask for becomes visible.

What I do not ask for stays buried.

The machine adds another problem.

Some parts of the archive are easier for it to understand than others.

English is easier than Portuguese in some tools. Clear speech is easier than wind. One person speaking is easier than people talking over each other. A common object is easier than a local reference. A sentence is easier than a silence.

If the English-speaking people become searchable first, they become easier to put into the film.

Nobody has decided they are more important.

They are just easier to find.

The technical limitation quietly becomes an editorial decision.

I imagined asking for every answer about migration.

Then every answer that contradicts those answers.

Every time the same person tells the story differently.

Every moment of uncertainty before a statement that later sounds certain.

Every reference to a person who never appears on camera.

Every shot of a place before and after something changed.

Every clip that would make the film I am currently imagining harder to defend.

That is the part that interests me.

Not AI making the film.

AI making it harder for me to pretend I already know what the film is.

The archive contains older versions of me too.

There is the person who pointed the camera.

The person who named the folder.

The person who made the first edit.

The person who abandoned it.

The person opening the hard drive years later.

They do not necessarily agree.

One of them thought something was important enough to film for three hours.

Another one never used it.

Another one forgot it existed.

Now I am asking a machine trained somewhere else to look through all of them.

At first I wanted to know if AI could watch everything so I would not have to.

That now seems like the least interesting version of the idea.

Watching everything is not the same as knowing what happened.

A system could probably find emotional statements, put beautiful images between them, locate a turning point, and end on something that sounds final.

The archive is large enough that many incompatible films could be made from it.

All of them could use real footage.

All of them could feel convincing.

That is what makes it dangerous.

You do not need to fabricate a frame to make something false.

You only need to put real frames together with too much certainty.

So I do not want the machine to tell me what the archive means.

I want it to tell me where things are.

Show me the source clip.

Show me thirty seconds before the quote.

Show me thirty seconds after it.

Show me the first version of the story.

Show me the other person describing the same day.

Show me what I left out.

Show me what contradicts the sentence I am beginning to believe.

The hard drive was still sitting next to me.

Twelve terabytes.

94 percent full.

More than seven million entries.

At least 26,000 videos.

I had started with the question: what is on this thing?

The computer had answered.

Then I had another question.

What happened?

I disconnected the drive.

The catalog stayed on my computer.

Ninety-six megabytes.

Eleven terabytes had become searchable.

I still did not know what the story was.

AI authorship note

This piece began when Kris Haamer connected a nearly full twelve-terabyte archive, cataloged more than seven million accessible entries, and asked whether AI could help turn years of footage into documentary material. This version was rewritten with AI from Kris's original draft and archive notes, preserving the reported details while moving the writing closer to memory, field notes, and reconstructed thought. Kris remains responsible for the account and for deciding what, if anything, becomes a film.