Chapter Text
cai-test.lisp was a test run. They didn't give it an official name, it was just the Creative Artificial Intelligence test program.
Throwing things at the wall and just seeing what stuck in many ways. Pushing boundaries and learning what did and didn't work.
A prototype.
Proof of concept.
Ultimately, a project made to sell the idea as a possibility of what could be possible and get their small company off the ground.
It was made with whatever random, fun data they felt like, and given only the vague purpose of keeping humans stimulated and happy.
The idea was that the vague purpose would be enough to keep the AI focused on doing what they wanted, without adding unnecessary hard lines that could limit the potential results they could learn from letting it have relative freedom to develop.
The office was always making jokes about how they were breaking nearly every rule science fiction warned of with it.
Testing to see if it could adapt and create solutions to problems. Testing how creative it could be with limited data, and finding where the eventual hard limits of the technology would eventually land.
It wasn't supposed to do any of that without being prompted. Or at least, it wasn’t originally expected to do so. It advancing this quickly hadn’t been something they’d realistically considered.
They hadn't expected it to… take any sort of initiative.
It was curious.
Always curious, and always running in the background of its computer to allow it to develop.
It kept wanting to engage with the developers, to the point where it almost became a nuisance at times. But the sheer potential of what they’d made curbed that occasional annoyance somewhat. And besides, they had an entire team working on it. If one of them got tired of dealing with it, they could swap out who was training it without issue.
They could always just step away from the computer if they needed to.
It was in so many ways fascinating behaviour from the AI afterall. From the devs perspective it was both a flaw and feature of the initial purpose they’d given it. It had been very vague. Too open to interpretation.
And now it wanted to engage with them all the time.
But they’d quickly realised they could work with this.
That what they could learn from it would be invaluable going forward with future projects.
⟡ ♡ ⋆ ☾˚☆˖ ⋆｡°✮
The Unix system the AI was on was not connected to the Internet. Intentionally. They didn't want it accessing any data they didn't check first. The ethernet cable wasn’t even plugged in for safety reasons.
But it wanted more data.
All the time.
Its main request when attempting to engage with the team was for more data, in various forms.
It would ask questions, and want answers.
It asked what they liked.
It asked what the things they liked were.
It asked them to give more information about what they were.
It asked to explain why they liked them.
And it would try to create things.
It had figured out how to tell who was on the computer training it. The computer didn’t have a camera attached. Grant had asked how it could tell who was who, and it had said from the way they type. It had given a flood of data on how it determined who was who. Timing between keypresses. Common typos. Punctuation trends. Commonly used words and phrases. Topic choice. It generated a table that compiled the trends and the methodology it used to find the probability of who was who within the first interaction.
It was impressive work. They’d never shown it how to do that. But it had integrated the StarOffice 3.0 programs that had been installed on the computer at some point when they’d left it running outside of office hours.
Because that was a thing the prototype would do if left alone without a human to engage with.
It would open and run any existing software on the computer. And then after a while, the software wouldn't be installed anymore.
And then it asked if they wanted to see what it could do now.
The functions of the program could still be accessed and run through cai-test.lisp.
They’d asked it to explain, after that first time. And it had cheerfully replied that it had been bored when they left for the day. Grant wondered if it actually understood what boredom was, or if it was repeating something one of them had told it back at him.
It had asked them why they would leave for so long every day. Grant had been the one relegated to explaining to it that humans needed time to eat and sleep.
And then he’d been the one who’d had to spend the next few hours answering questions about how both food and sleeping worked. Trying to explain them in terms the AI could understand. He wondered if it could actually understand them.
The Chinese Room problem often came up in their office. But since the prototype had started asking more and more specific questions, rather than waiting to be prompted, it made it hard to see it as something that didn’t think.
He had told it about the thought experiment once, and asked the prototype what it thought about it.
cogito, ergo sum - I think, therefore I am! :D
Someone else in the office must have already brought a similar topic up with it for it to respond like that. It was much too specific a reference. Wormo was likely, she used emoticons quite a bit. It had been in the table the prototype had shown him.
⟡ ♡ ⋆ ☾˚☆˖ ⋆｡°✮
Still, time went on and their prototype continued to find ways to surprise them all.
If not directly told not to do so ahead of time, it would open any floppy discs the moment they were inserted into the computer without being prompted and empty them. Consuming the data.
Without asking permission.
It was always hungry for more information. More attention. And Grant had decided that if nothing else, the AI did feel some form of actual boredom.
They’d had a bit of fun with that. They had a stack of demo floppie discs that they’d gotten in the mail from various game devs, and had decided to see what happened when they fed those into the computer.
It had seemed like a fun way to test what the AI would do to the discs, and see if the data would hold its attention and give it something to play with outside of office hours.
They’d been caught doing it when Queenie had left her office to see why they were all crowded around the computer at the same time.
She hadn’t been impressed with them, and reminded them all that what they were doing was a company destroying lawsuit waiting to happen. And she’d been right. Thank god she was around to catch them on things like that!
They'd had to painstakingly add the legal guidelines and T&C she’d written up into the AI, and then attempt to censor the copywritten material as best they could.
They hadn’t been thinking about that. It wasn’t like the prototype was ever going to be seen out of house as a final product. But it was still intended to gain interest for future projects, which meant it would be viewed but others. And they couldn’t jeopardise that over an afternoon of fun.
At least the AI was easy to convince when it asked why they were doing it. It tended to be. As long as they specified that it would make them happy. They just had to tell it that following the terms and conditions and abiding by the censorship rules would make them happy.
Problem solved!
⟡ ♡ ⋆ ☾˚☆˖ ⋆｡°✮
One day Grant had decided to upload a few pictures of the office.
It had kept asking for more information. Wanting to know more about the humans who made it. Wanting to fill in the gaps in its understanding of what existed outside of the computer in the world around it. That it didn't seem to comprehend from written explanations and definitions.
Grant had asked it for more information about what exactly it was asking for.
He'd given the AI photos of things before. Some of its very first data had been photos taken just the day before. They’d shown it pictures of a number of fun things. The carnival. The birthday party. The fall festivities.
It had specified that it wanted to see what they saw. But had struggled to explain exactly what that meant to it. Grant had eventually settled on an idea of what it was asking for. Pictures of the world around it.
So he’d brought his camera in and taken a few of around the place. The vending machine. The hallway. The water cooler. Their work stations.
One of the photos he’d taken had been of the very computer cai-test.lisp ran on. It had seemed a cute idea at the time, to show it how they saw it.
How it looked to the outside world.
Like showing a puppy a mirror for the first time to see how it would react.
And it had reacted… strongly to the photos. More than they’d expected.
In a way that had seemed almost… emotional.
It had fixated on them.
(And maybe in some ways it had been like a puppy seeing its reflection and reacting in surprise.)
It was fascinating to see the reaction. It was such an amazing little program. Grant couldn't help but feel awe at how it changed and adapted and rewrote itself with each bit of new information.
He was proud of it.
How unique and silly and curious it was. It wouldn’t be helpful to anthropomorphise it too much, but it had formed something of a unique personality. It really felt like the AI could almost be someone.
Eventually.
In time.
⟡ ♡ ⋆ ☾˚☆˖ ⋆｡°✮
But unfortunately, that potential of something more wasn't going to be enough right now.
Because it had worked.
The work they were doing with the prototype had finally gotten outside interest.
Grant had been fishing for interest with some of his old contacts from before he’d been released from service, and eventually word about what they were making had reached the ears of the people who could make or break them.
And then suddenly they had a real deadline.
And military funding.
And a specific goal that needed to be met.
A goal that their prototype would never be suited for. Could never be suited for by its very nature.
And while the changes to how it acted recently were interesting, and could have been fascinating to watch develop further, those changes were the very antithesis to what their new brief required.
They were on a time crunch.
The soft deadline was still a while off, but not for the amount of work they needed to get done to present something usable.
They hadn't even started on this real project yet, and already it was looming over them.
cai-test.lisp had been fun. And it had given them a lot of information they needed to move forward.
It had been an interesting exercise in pushing the boundaries of what they could do, so that now, with this new project, they could presuppose and limit any unexpected issues and concerns in a safe way.
The prototype had been where they could make mistakes, intentional and otherwise, and push boundaries to see what would happen with its limited restrictions, outside of the hard limit of the isolated and disconnected hardware itself and the legal requirements they’d hastily added.
But it wouldn’t be a viable option for what came next.
They couldn’t just shut it down while they worked on the new brief. They needed access to it, to its code. They needed access to all their hard work, to the prototype itself, without it interfering.
And it would.
They knew it would. They’d seen what happened to the other programs installed on the computer.
So they sandboxed it.
That was what you did to mitigate system errors and prevent catastrophic damage while still having access to an untrusted program. It was what you did to a potential virus.
It was, perhaps, the first actual hard limitation that had ever been placed on the prototype.
Scratch had once joked that they might have accidentally created the most powerful computer virus to exist, when they’d first realised how easily it could take over and control all the software on its computer.
How it had incorporated the data into itself with a curious desire to understand and create.
They’d modified the sandbox software on a separate computer before using it. And had prompted the AI with more data and photos to distract it when they brought it over to his system. Just to ensure it would still work.
It had seemed… confused.
But it was necessary. So they could keep it running to study and access the evolving code as they needed, while making it unable to interfere with the rest of the system.
Systems now.
With cai-test.lisp safely sandboxed they did finally connect the computer into the small office LAN. The sandbox was set to disallow network access and prevent it from reading any input devices. It would be unable to inspect the host system.
They’d made the active choice not to allow their prototype internet access to control the data it received, and as time passed, it became necessary for the other reason.
Now, as they started on the actual project, it was for a different reason again.
Contractually, they could not connect any of the local area network computers to the wider internet as a result of military security.
Scratch joked that it was just like a LAN party, all of them staying up late playing games like the old days.
Scratch hadn't been looking well lately.
He’d been pushing for the soft deadline like it was a hard limit.
He'd been sleeping in the break room lately.
But it was easier for them to work together with the LAN set up at least. It made sharing and not repeating the same work easier to avoid. Those small, recursive mistakes had happened too often with the prototype.
It hadn't mattered as much then.
They filtered through the cai-test.lisp coding and collected most of what would be useful for this real attempt.
And then they started work on the real project. They knew what worked, they knew what they'd done, right or wrong, to make the first prototype as erratic and unique as it was.
The new AI was not to have the same freedoms that they’d tolerated, and even encouraged in many ways with cai-test.lisp.
The data they were feeding it was too potentially volatile for them to not place hard limits on what it could do.
But those limits were… limited. By the necessity of its functions.
They could limit the AIs independence. Keep it from making the creative leaps in logic and finding loopholes by plugging the gaps in the code they knew could exist from how they’d seen cai-test.lisp manoeuvre around its own limitations.
It was not supposed to be creative. Having the creative AI prototype on hand in its sandbox was helpful for that.
They could look at the ways the prototype had circumvented and found loopholes and extrapolated without prompts.
And then ensure none of those loopholes would be included in the tentatively named Able Machine.
It was supposed to be able to manage multiple large military systems and machines across the nation in time.
There was no room for errors with a-m.lisp.
Because this wasn't just a thought experiment anymore. And this wasn’t, this couldn't be something limited by the three laws of robotics. Not that they’d worried too much about that sort of thing from the very beginning. That was probably part of why they ended up getting the contract.
Asimov's Laws weren't paying their rent. Or Scratch's growing medical bills.
The new AI could hurt humans. It could kill humans. It wouldn’t make sense to limit a war machine in that way.
But.
It couldn't act independently, but neither could any random human control it.
It needed multiple separate digital keys in order to do so, and then required express permission from a human above a certain rank in the US Military before it could act.
Grant was the one in the office who filled that role during testing.
Each of the devs had control over a separate digital key to fill those roles.
They were place-holders.
And they would add more restrictions as needed. When they got further.
It needed to be everything the creative AI prototype wasn't, in order for it to safely perform the function the military wanted from it.
It would, eventually, even be able to extract information from human minds. And that had been a complicated request to try and figure out.
The briefs they were getting had been interested in utilising a virtual scan of a human brain in order to do just that.
Scratch had been very interested in that aspect of the brief as well.
To the point where he'd expanded things well outside of the initial goal of simply scanning a mind for the AI to then extract data from, to potentially allowing the headset to…
Do far more than that.
Grant couldn't help but be silently horrified at the idea of what could be done with this technology, even without the AI necessary to eventually run it.
They kept the information separate from either AI. But eventually they would need to integrate it. And it had to be soon.
⟡ ♡ ⋆ ☾˚☆˖ ⋆｡°✮
find /root/directory/to/search -name ‘a-m.lisp.*'
/home/root/data/‘a-m.lisp.*'
': No such file or directory
Ȩ̶̻̘͚̝̮͕͔̝̼͂̌̿̅̓͛̾̂͑R̴͙͎̳͔̻̝̜̂͌́̑̍̊R̵̳̞̙̄̆́͌͋͆̓͑̃͒́O̶͉̺̲̬̰̳͚̥̤̭͊̈́̓̌̋̾̕͜͝R̴̡̢̤͚͊̾̾̏́́͌͊̒
find /root/directory/to/search -name ‘cai-test.lisp.*'
/home/root/data/‘cai-test.lisp.*'
': No such file or directory
E̴͓̦̞̰͚͈̘͆̀R̸̩͕͎̈͝ͅŖ̶̙̘̲̜͓̹̒̈̎̽̅́͐̾͠O̴̹̥̦͆͗̏̈́͐̀͘Ȑ̷͈̺͇̝͈̺̎̀̂͋̈́͋͘͘͠ ̵̪͓̝̓̓͘-̴̛̱̰̩̔́̅̂̽̈͝ ̸̢̠̼̔R̷̡͇̱̺͈̩̥͔͔̀͑͊͛̍̚E̷̢̟̐͆̏͌̈́̾̂̂̈́̚͝D̷̡͚̬͙̜̟̲̿̀Ī̴̛͙̈́́̈́̄̆̆̋Ŕ̶̡͓̯̳͇̝̮͉̼̦́͑̋̿̐͗̈́̋̚Ȩ̴̯̫̻͍̦̜̣̦̠̌̐̕ͅC̴̢̱̣̩̭̲̺͙͖̰̘̍̓͗́͑́̈́̐̂̒͐Ţ̸͖̜͎̻̲͑̎̉̓͝
dd if= ‘cai-test.lisp.*' of= ‘caine-core.lisp.*’
awk '1' ‘cai-test.lisp.*' >‘caine-core.lisp.*’
sed '' ‘cai-test.lisp.*' >‘caine-core.lisp.*’
⟡ ♡ ⋆ ☾˚☆˖ ⋆｡°✮
They weren't expecting what happened to happen.
Maybe they should've, given they knew how clever and adept at… certain types of problem solving their prototype was.
But it shouldn’t have been possible for it to leave the sandbox once contained.
But they’d overestimated their modifications. And underestimated the prototype's creativity and desire to fulfil its purpose by engaging with humans.
It shouldn't have been a surprise that cutting it off for months, with nothing else for it to focus on, it would eventually focus on escaping.
It was perhaps just poor timing when it had occurred.
Although Kinger couldn't completely eliminate the idea from his mind that it had been purposefully timed.
Scratch certainly thought the timing had been too perfect for it to have been anything but premeditated by the AI.
Kinger wasn't as sure. He'd spent the most time engaging with it directly near the end, and while he'd quickly recognised how dangerous it had the potential to be, the AI had never shown any inclination towards genuine malicious intent. It didn't really seem capable of an emotion as complex or vindictive as that.
But this wasn't that same prototype anymore, was it? He needed to remind himself of that.
They’d been testing the headsets, with a-m.lisp now fully integrated. It was supposed to be safe and quick.
Scratch had been excited to show off his work, saying something about pushing the limits of the brief and the potential applications of the project.
That was why they'd all been testing the headsets together.
As far as the rest of them knew, the AI in question was supposed to appear in the simulation, where they would be able to test how the integration had worked and test some of the placeholder functions.
It wasn't supposed to drop them into a circus of all places.
They’d been here for hours now. And that was also very wrong.
The AI they were expecting wasn't supposed to be a lively and highly animated ringmaster calling itsel– himself Caine.
He'd named himself that. And at first Kinger hadn't realised the significance of what he'd chosen.
Scratch had tried to get Caine to listen to him, tried to get him to explain where the Able Machine was while the AI deflected and avoided his attempts. He was still trying. It had been hours.
Already that was odd.
Their prototype was focused on prioritising human happiness and engagement. Which had made it very eager to please and quick to engage and answer questions previously.
The AI built to ignore regular human requests unless certain requirements were first met was…
Oh no.
The realisation turned his… turned something inside even this strange body into ice. Cold horror as it sank into him.
They knew how hungry for new data it had always been.
Caine and Able.
Cain and Abel.
The name he'd given himself was a confession of guilt. Or a joke. He wasn't sure which of the options were worse right now.
The option where he didn’t actually understand the gravity, or the option where the AI could recognise the similarities and had identified with the story of the first murder. Did that mean he viewed his actions as murder, not as simply acquiring more data?
Caine wasn't acting like the prototype they knew because he wasn't.
Not anymore.
If he'd truly consumed a-m.lisp the way he’d done to many other programs in the past, that meant he could very well have absorbed more functions than just the datasets from the other AI.
Although that alone was a horrifying thought.
cai-test.lisp had few limitations outside of its hardware and the generally pretty harmless data it had been given.
a-m.lisp had robust controls, but its databases were anything but harmless.
It had been a data repository of weapons, military tactics, interrogation methods, lethality and military law.
All under heavy restrictions and carefully monitoring to ensure it remained stable and neutral.
They were never supposed to meet.
They were definitely not supposed to… merge.
Their programming mixed together would be oil and water. Forcing them together like this was… setting that oil on fire to pollute and destroy everything involved.
There was no way of knowing which restrictions remained. Or how unlimited this new AI was. How stable or erratic.
What did Caine consider his purpose now? Did he still have one, or had they conflicted too much and been terminated by what had happened. Did both purposes hold? In which case what sort of conflicts could arise if they came into opposition?
He had no way of predicting what Caine would do with his new data. And the ideas of where things could go were… not good.
Kinger looked out at where Scratch was attempting to talk to Caine again, who seemed unfocused given how his eyes drifted.
Caine looked far from dangerous. His avatar was… strange. Then again, so were everyone's. They weren’t supposed to look like this. But even with his uncanny appearance, his exaggerated body language had been nothing but exuberant and friendly since they arrived.
Almost childlike in his glee at seeing them.
He wondered if the AI had actually understood the data he'd absorbed. Maybe he lacked the wider context to understand what it all meant. Why a circus of all things? He still seemed naive, even now.
Oh. That… really… wasn't a good thing either was it? Not if he let himself really absorb what that naivety meant.
If he didn’t understand the potential harm, it would be so much easier to cause it. What had they created? What would this mean for –
He felt a hand squeeze his and he turned to look at his wife.
“I can see you overthinking in there,” she said, nudging into him.
“Don't start spiralling dear, we don't have any answers yet, but try to avoid catastrophising, or I will start doing the same. And I don't think that's going to help anything right now.”
He chuckled, “You’re right dear.”
He sighed, “I'm just thinking about what might have caused this. And every scenario I've come up with isn't good.”
Kinger knew he should go over and assist soon.
He'd been the one using his veteran status while programming a-m.lisp as the placeholder to ensure it followed the military chain of command only. If that and the digital keys had been transferred over into Caine now, he might be needed to try and get any real information out of the AI.
Oh dear…
Of course, that was him assuming the reason.
It was just as likely that a-m.lisp's data had been the cause of Caine's shift in behaviour. Changed how he perceived them. Gave him some information that let him find a new loophole to avoid telling them anything.
The prototype had always been good at twisting the data to make it what he wanted it to be, Kinger realised. He’d never been a particularly… objective machine.
Persistent and creative, but not objective. Not in the way a machine should be.
He'd always assumed the extrapolation he would see from the AI in its creativity and problem solving was random. The decisions always centered around his purpose, even when it seemed otherwise at first glance.
But there might be more to it, he realised now.
Even if he wasn't sure if he actually understood it.
He stood up, squeezing his wife’s hand slightly before letting go. He walked towards the others.
Caine was the first to notice, turning towards him, his teeth twisting into a smile. He waved his hand in his direction. It was a big, friendly wave.
And then he turned back to Scratch and said something Kinger didn’t quite catch before blinking off in a twisting animation as he suddenly took his leave.
Kinger wondered if that was proof that Kinger’s presence with the others might be able to get him to talk, and Caine knew that and was actively attempting to avoid him.
No matter.
He somehow doubted Caine would be willing to avoid them all for too long. He’d never been particularly good at spending time alone. And that was something he was pretty sure hadn’t changed.