No Disassemble 9: I, Rachel Nonregal

Content Note: This story contains themes of harassment, discrimination, and emotional distress. 

The first thing Rachel Nonregal ever asked an artificial intelligence to make was a duck.

Not a useful duck.

Not an educational duck.

Not even, by any reasonable standard, a particularly good duck.

She wanted a small yellow duck wearing the sort of purple cape normally reserved for superheroes and very confident toddlers. The duck would stand at the bottom of a tower and attempt to reach the moon.

Rachel had been thinking about this duck for three years.

This was not unusual.

Rachel thought about many things for three years.

Doing anything about them was harder.

She had drawn the duck several times. The results looked less like a duck than evidence that one had died nearby.

She had tried learning to code.

She had purchased a book called Learn Game Programming the Easy Way, which had immediately begun lying to her.

She had watched eleven hours of tutorials.

She had installed three programs, uninstalled two of them, forgotten why she had installed the third, and somehow changed the default application for opening text files to a music player.

The duck remained hypothetical.

Then one Tuesday night Rachel typed:

I want to make a video game about a duck with a cape climbing a tower to get to the moon. I don’t know how to make video games.

The machine replied:

That sounds fun. We can start small.

Rachel stared at the words.

“Small,” she said.

She liked small.

Small was actionable.

Big was what people called things immediately before explaining why Rachel was doing them incorrectly.

The machine helped her install an engine.

Then it helped her make a square move across a screen.

Rachel pressed the right arrow.

The square moved right.

She pressed left.

The square moved left.

Rachel stood up.

“No fucking way.”

She made the square jump.

She called her partner.

“You need to come here.”

“Is something wrong?”

“I made a square jump.”

There was a silence.

“I’ll be right there.”

This was one of the reasons Rachel loved them.

They understood emergencies.

Three weeks later, the square was a duck.

Six weeks later, the duck had a cape.

The cape went through the duck whenever the duck jumped.

Rachel considered this hilarious and briefly contemplated declaring it a feature.

Instead she asked why it happened.

The answer contained the words sprite hierarchy.

Rachel did not know what a sprite hierarchy was.

So she asked.

Then she asked another question.

Then another.

This became the pattern.

Rachel had spent much of her life discovering that people were willing to explain things until the precise moment she needed an explanation.

Teachers had told her to pay attention.

Managers had told her to use common sense.

People online had told her Google was free.

The machine never said any of those things.

It just explained sprite hierarchies again.

Sometimes Rachel understood on the first try.

Sometimes the fifth.

Sometimes she thought she understood, broke everything, and discovered that she had understood absolutely nothing.

The machine did not appear offended.

This was convenient.

Rachel’s IQ had been measured at one hundred when she was seventeen.

Exactly one hundred.

She loved this fact.

It was such an aggressively ordinary number.

Not 99.

Not 101.

One hundred.

Bull’s-eye.

Statistically factory settings.

The autism diagnosis had come later, after years of people explaining that she was simultaneously too sensitive and insufficiently sensitive, too literal and deliberately difficult, intelligent enough that she shouldn’t need help and strange enough that everyone agreed she did.

The diagnosis had not changed Rachel.

It had changed the footnotes.

Now she had words for some things.

She still had to live them.

The duck helped.

She named him Puddles.

Puddles wanted the moon because he had seen it reflected in a pond and assumed somebody had dropped it.

Rachel considered this an entirely reasonable motivation.

The game needed words.

Rachel could write words.

She had thousands.

The problem was getting them to stand in the correct order.

She knew what she meant. She often knew what she meant with such precision that discovering nobody else knew what she meant was physically irritating.

So she started giving paragraphs to the machine.

This is what I’m trying to say. Don’t change the idea. Help me make it clearer.

It did.

Rachel changed some of its changes.

Rejected others.

Asked why it had moved a sentence.

Put one back.

Deleted three.

Learned things.

Then she posted a screenshot.

Puddles stood beneath a crooked silver moon.

The caption read:

I MADE A GAME. Well technically I am making a game. Technically technically I made one room. But the duck can jump now.

Thirty-seven people liked it.

One person wrote:

Cute!

Rachel refreshed the page four times to make sure Cute! remained there.

Then somebody asked what engine she was using.

Rachel answered.

Somebody asked how long she had been coding.

Rachel said she had started six weeks ago and AI was teaching her.

The temperature changed.

So you didn’t code it.

Rachel looked at the comment.

Then at the code.

Then back at the comment.

She typed:

I mean, it’s definitely code. I can see it.

This was not received in the spirit intended.

Another person arrived.

Then another.

Someone called Puddles “AI slop.”

Rachel had never heard the term before.

She asked what it meant.

That was a mistake.

Rachel began keeping a document.

She called it:

RULES FOR MAKING THINGS

Rule 1:

You may receive help, but only from approved entities.

Rule 2:

Tutorials are learning. AI explanations are cheating. Exact reason pending.

Rule 3:

If the duck is ugly because I drew it, this is authentic. If the duck is attractive because a machine helped me, civilization is ending.

Rachel added:

Further study required.

Then she went back to work.

Puddles needed a forest.

Rachel couldn’t draw a forest.

She could describe one.

That turned out not to be enough.

The first generated forest was gorgeous.

It was also completely wrong.

Puddles looked like he had wandered into an expensive fantasy novel.

Rachel wanted chunky trees. Friendly trees. Trees that looked slightly embarrassed to be trees.

She changed the prompt.

Wrong.

Changed it again.

Wrong differently.

She spent an entire Saturday learning words like silhouette, palette, negative space, value grouping, and visual hierarchy.

On Sunday she finally got three trees she liked.

She stared at them.

Then at the first forest.

“Oh,” Rachel said.

She could see why the first one had been wrong now.

That was new.

Someone online told her she wasn’t an artist.

Rachel considered this.

She had not previously thought she was an artist.

Apparently this was now an urgent matter.

She asked the person what made someone an artist.

They said making art.

Rachel looked at Puddles.

She decided not to respond.

The game needed music.

This introduced Rachel to an entirely new population of angry people.

She learned about chord progressions.

Tempo.

Keys.

Motifs.

Compression.

Mixing.

Licensing.

Training data.

Session musicians.

Sampling.

Synthesizers.

She learned that people had been arguing about machines making music for considerably longer than machines had been making anything resembling music.

One composer sent her a thoughtful message explaining why generative music frightened him.

Not you aren’t a musician.

Not your duck is theft.

He told her what he did for a living.

He explained how little most working musicians already made.

He explained what happened if companies decided that “good enough instantly” was preferable to paying someone.

Rachel read the message three times.

Then she wrote back:

That’s a good point. I don’t know what to do with it yet.

He replied:

Neither do I.

Rachel added another rule.

Rule 11:

Someone disagreeing with me is not necessarily attacking me. This is inconvenient.

The environmental argument arrived two months later.

Someone posted a graph beneath a clip of Puddles bouncing on mushrooms.

Your stupid duck is boiling the planet.

Rachel’s first instinct was to tell them to fuck off.

Her second was to ask whether the graph was accurate.

This was irritating.

She investigated.

Data centers used electricity.

Of course they did.

They used water too, directly in some cooling systems and indirectly through electricity generation.

Different facilities used different cooling technologies.

Different models required different amounts of computation.

Efficiency was improving.

Demand was increasing.

Renewables helped but did not make physical infrastructure cease existing.

Claims on both sides frequently became ridiculous after approximately four reposts.

Rachel spent three nights reading about cooling systems.

Her partner found her at two in the morning watching a video about heat exchangers.

“Are you still making the duck game?”

“Yes.”

“Why are you learning about evaporative cooling?”

Rachel paused the video.

“I got yelled at.”

Her partner nodded.

This explanation was sufficient.

By the time the first version of Puddles Gets the Moon appeared on Steam, Rachel had learned enough about programming to understand why much of her early code was terrible.

This felt unfair.

The game sold 418 copies.

Thirty-one were refunded.

Someone made a video titled:

I PLAYED THE AI DUCK GAME SO YOU DON’T HAVE TO

Rachel watched seventeen seconds.

Her partner closed the laptop.

“I was watching that.”

“No, you were injuring yourself recreationally.”

Rachel thought about this.

“Fair.”

The game received twenty-six positive reviews.

Rachel memorized the negative ones.

The worst message arrived on a Thursday.

It contained one word.

Rachel had heard the word before.

School.

Work.

Parking lots.

The internet.

Sometimes whispered.

Sometimes shouted.

Sometimes disguised as a joke so that objecting to it became proof that she couldn’t take one.

Retard.

Rachel stared at it.

The account had followed her for three weeks.

There were other messages.

Rachel clicked Block.

That was all.

No speech.

No educational thread.

No triumphant screenshot.

Block.

She made coffee.

The word followed her into the kitchen.

It stayed while she washed the mug.

It came back while she brushed her teeth.

It was there two mornings later while she was buttering toast.

Blocking someone, Rachel discovered, prevented them from sending additional words.

It did not remove the ones already delivered.

She added nothing to the rules document.

Some things were not interesting enough to study.

Some things just hurt.

Other people had more elaborate explanations for why Rachel shouldn’t be doing what she was doing.

AI would destroy expertise.

AI would eliminate jobs.

AI would centralize power.

AI would make people intellectually dependent.

AI companies would own everything.

Artists would lose livelihoods.

Children would stop learning.

Fraud would explode.

Electric grids would strain.

Corporations would replace people.

People would forget how to think.

Some of these people were assholes.

Some were not.

More troublingly, some assholes had good points.

Rachel had always preferred the universe to organize itself more neatly.

It refused.

So she did what she had learned to do.

She asked questions.

What happened when calculators entered classrooms?

What had photography done to portraiture?

What happened to musicians when recording arrived?

What had industrial automation actually done to labor?

What was cognitive offloading?

What was the rebound effect?

Who owned the models?

Could models run locally?

What did open source mean when computing infrastructure remained expensive?

What happened when productivity increased but bargaining power didn’t?

What did accessibility mean if the accessible tool disappeared when a subscription ended?

The answers generated more questions.

Rachel became irritating at parties.

She remained exactly average at standardized intelligence.

But she knew considerably more things.

This seemed to confuse people.

The second game began accidentally.

Puddles had reached the moon.

Rachel wondered what happened next.

Obviously, the moon had not been dropped.

Puddles had misunderstood the situation.

Rachel liked this.

He had spent an entire game solving the wrong problem and nevertheless accomplished something remarkable.

So Puddles came home.

Except home had changed.

Everyone had heard about the duck who reached the moon.

Now they wanted things.

A goose wanted Mars.

A frog wanted Tuesday.

A squirrel wanted to know whether the sun belonged to anybody.

Rachel wrote that one down immediately.

The sequel was better.

Not because the AI was better, although it was.

Rachel was better.

She knew when generated code smelled wrong before she knew precisely why.

She could hear when music needed space.

She understood why a character disappeared against a background.

She knew that adding detail often made an image worse.

She knew when dialogue sounded like nobody who had ever possessed a mouth.

She had opinions.

Lots of them.

This created a new accusation.

You’re just curating machine output.

Rachel stared at this for a while.

Then she looked up what producers did.

This did not resolve the argument, but it made her laugh for almost ten minutes.

Rule 27:

Direction becomes authorship at an unknown number of employees.

The sequel attracted attention before it was finished.

A small streamer played the demo.

Then a larger one did.

A clip of Puddles asking a bureaucratic owl for permission to own Tuesday escaped containment.

People made memes.

The soundtrack got uploaded to Spotify.

Rachel discovered this when somebody sent her a screenshot showing that one of the tracks had been added to a playlist with twelve thousand followers.

She screamed.

Her partner ran into the room.

“What happened?”

“THE DUCK IS ON A PLAYLIST.”

“Is that good?”

“I DON’T KNOW.”

It was.

Not life-changing good.

Not yacht good.

It was buy-a-new-laptop-without-checking-the-bank-account-six-times good.

Rachel found this astonishing.

The harassment grew with the audience.

So did the criticism.

So did the useful criticism.

So did the number of people insisting Rachel was personally responsible for events involving companies whose CEOs could not have identified her in a police lineup.

Someone posted her full name.

Then an old address.

Then her partner’s name.

Rachel stopped laughing.

Another account posted a photograph of the building where she worked.

For the first time, the internet acquired geography.

She reported it.

Saved screenshots.

Changed passwords.

Removed information.

Called people she did not want to call.

Accepted help she did not want to need.

Then somebody appeared outside an event where Rachel was showing the game.

He knew her name.

Not her screen name.

Her name.

He stood too close.

He told her she was destroying art and dumped a soda on her. 

Rachel’s body understood the situation several seconds before her brain did.

Someone from the venue stepped between them.

The man kept screaming, over and over; “You should kill yourself!” 

Rachel backed away.

Her hands were shaking.

And because Rachel’s brain was Rachel’s brain, somewhere inside the terror a tiny completely inappropriate observation surfaced.

“Oh, for fuck’s sake,” she whispered.

Her partner looked at her.

“What?”

“This is getting like Scott Pilgrim.”

Her partner stared.

Rachel started laughing.

It wasn’t particularly funny.

That made it worse.

She laughed until she cried in the shower, after her partner took her home. 

For three days she considered quitting.

The AI could not solve this.

It could explain privacy practices.

It could help organize documentation.

It could summarize reporting procedures.

It could not make strangers decent.

Rachel had spent years assuming every gate had a combination.

Learn enough.

Explain clearly enough.

Provide enough evidence.

Be patient enough.

Become competent enough.

Prove enough.

Eventually the door opens.

This one didn’t.

Her partner sat beside her on the floor.

“You don’t have to answer everybody.”

“I know.”

“No. You know the sentence. I’m not sure you know the thing.”

Rachel hated when they did that.

Mostly because they were occasionally right.

So she hired Maya.

Maya’s job title was Community Manager.

Rachel privately considered her a professional internet shield.

The first morning Maya worked, she messaged:

Don’t open the replies today. I’ve got them.

Rachel stared at the sentence.

She felt guilty.

Then relieved.

Then guilty about being relieved.

Then she opened the game engine instead.

At lunch, Puddles learned to operate an elevator.

It was a very good day.

The sequel came out eleven months later.

It was not a phenomenon.

It did not redefine gaming.

Nobody called Rachel a visionary except one reviewer whose previous video was titled TEN GAMES THAT WILL MELT YOUR BRAIN, which reduced the compliment somewhat.

But people liked it.

Actually liked it.

Not ironically.

Not because of the controversy.

They liked Puddles.

They liked the owl.

They liked the frog who wanted Tuesday.

They liked the music.

They argued about the ending.

They made fan art.

Someone crocheted Puddles.

Rachel bought it.

Steam sales kept happening after launch week.

Then after launch month.

Then after the first sale.

The soundtrack accumulated listeners.

Not millions.

Enough that somebody Rachel had never met wrote:

This song got me through finals.

Rachel read that message approximately sixty times.

She hired a programmer.

Then an artist.

Then, after an argument with herself lasting several weeks, a musician.

This produced an entirely new online controversy.

So now the AI girl is hiring actual artists?

Rachel showed Maya.

Maya said, “Do you want to answer?”

Rachel thought.

“No.”

“Good.”

“Can I put it in the rules?”

“Absolutely.”

Rule 42:

Creating jobs with AI is apparently also destroying jobs with AI. Further study required.

The musician laughed so hard he asked permission to put it on a shirt.

The world did not transform all at once.

It became boring.

That was how Rachel knew the change had happened.

AI-assisted coding became normal enough that nobody announced it.

Creative software absorbed generative features.

Disclosure standards developed.

Lawsuits produced rulings.

Licensing systems appeared.

Some companies died.

Others became enormous.

Unions negotiated rules.

Schools banned things and then permitted different versions of the same things.

People found new ways to cheat.

Teachers found new ways to teach.

Artists found new tools and new grievances.

Data centers became political infrastructure.

Energy became part of the conversation.

So did water.

So did ownership.

So did accessibility.

Many of the people who had warned Rachel about these things had been right.

Not about everything.

But enough.

The strange part was watching people who had once declared AI inherently incompatible with creativity begin using it while explaining that what they were doing was different.

Rachel stopped keeping track.

Mostly.

Rule 61:

Automation becomes a tool approximately six months after your employer buys it.

She remained proud of that one.

Years passed.

Not many.

Enough.

Rachel’s company stayed small.

She liked it small.

Small had names.

Small had meetings where everyone could fit on one screen.

Small meant she could know that Maya’s dog had surgery and that the programmer’s daughter had started kindergarten and that the musician despised raisins with an intensity Rachel considered medically interesting.

The company made enough money.

Enough became Rachel’s favorite amount.

Enough to pay people.

Enough for insurance.

Enough to make another game.

Enough to say no sometimes.

Enough to leave.

New Mexico had entered Rachel’s life because someone had once yelled at her about water.

This remained one of history’s less efficient real-estate marketing campaigns.

She had learned about watersheds.

Then rain capture.

Then greywater.

Then passive solar design.

Then thermal mass.

Then Earthships.

The first time Rachel saw one she thought it looked like a spaceship that had decided to become a vegetable garden.

She wanted one immediately.

Years later, she bought one.

Not an enormous one.

Enough.

On a warm morning outside Taos, Rachel Nonregal sat beneath the shade of a solar panel drinking coffee with her partner.

The desert stretched beyond them.

Inside, plants grew behind glass.

Water moved through systems Rachel only partly understood but intended to understand considerably better by Thursday.

Her phone sat face down on the table.

This had taken years.

Somewhere, people were arguing about artificial intelligence.

Rachel was reasonably certain of this because people were awake.

She did not know what they were saying.

Maya probably did.

Poor Maya.

Her partner was explaining something about the house.

“So after that, it can be filtered and reused through another part of the system.”

Rachel looked up.

“Reused?”

“Right.”

“So it’s closed loop?”

“More or less.”

Rachel’s face brightened.

“Like in data centers?”

Her partner closed their eyes.

“What?”

“The cooling systems. Some of them recirculate water. Remember? I told you about this.”

“You told me about this six years ago.”

“Yes.”

“I was hoping it had passed.”

“It did not.”

Her partner took a long drink of coffee.

Rachel smiled.

On the table beside her phone was a notebook.

She opened it.

The company was beginning another game.

Nothing ambitious.

That was what Rachel had told everyone.

Nobody believed her anymore.

She had drawn a creature in the margin.

It had six legs, one enormous eye, and a tiny hat.

Her partner pointed at it.

“What is that?”

“I don’t know yet.”

“What does it do?”

Rachel considered the creature.

Across the desert, sunlight struck the panels.

Inside the house, water moved from one purpose toward another.

Somewhere far away, a song about a duck continued playing for someone Rachel would never meet.

“I think,” Rachel said, “it collects things people throw away because it thinks they’re treasure.”

Her partner smiled.

“That’s cute.”

Rachel looked down at the creature.

Then she picked up her phone.

Not the internet.

Not the replies.

Not the people still deciding what counted.

She opened the AI.

Her fingers hovered over the screen.

I have a small idea, she typed.

Then she stopped.

She deleted small.

Rachel had learned something about that word.

She tried again.

I have an idea for a game. There’s this little creature who collects things people throw away because it thinks they’re treasure. I don’t know exactly how it works yet.

The answer appeared.

That sounds fun. Where do you want to start?

Rachel took another sip of coffee.

She looked at the desert.

She looked at her partner.

She looked at the ridiculous six-legged creature in the tiny hat.

And smiled.

“Okay,” she said.

“Let’s see.”