Hostile AI input & suggestions

[Jd Smith1]

So, I am looking at a scenario involving a smallish space station where forbidden research on an old AI went wrong, and the the AI (shocker) has gained control.

The PCs are sent to investigate (another shocker)

OK, so the station has no robots or androids, and other than simple internal defense systems, is unarmed. There was a small research and maintenance staff aboard, who may or may not still be alive, depending on the needs 0f the plot. It is orbit around a dead moon, and there is nothing resembling a Internet.

This is Fading Suns, where a space empire is experiencing tech and social decay, and a growing Luddite attitude towards tech (which is why there's no robots or androids).

So its not original, but I have the station map, and I thought this would make a decent change (the PCs have mostly dealt with the occult so far).

The AI is centuries old, a bit of forbidden tech from more advanced times. And by forbidden, I mean the PCs will destroy it the instant they realize it exists and can get to its core (heavy XP reward for doing so).

But I don't have a lot of movie/book history of 'hostile AI' plots, so any suggestions, guidance, etc?

EDIT: OK, I have taken several ideas and woven them into a variable-outcome issue. Thanks a million!

 

[Dross]

Does it have to be evil?

HAL from 2001. Not exactly evil. Went mad due to two contradictory commands (tell the crew needed info for theirmission v keep secret mission from crew).

Asimov robot stories, and the Will Smith 'I robot' movie have a theme of enslaving humans to save them.

Ultron from MCU for a kill humans.

wanting to keep some contaigen away from people. A bug that makes the AI think some environmental condition are opposite (i.e. temp at -30 Celsius for people to live in).

If you search TV Tropes for 'zero law rebellion', 'AI is a crapshoot' there will be plenty of ideas and their sources.

 

[Reynard]

My suggestion: decide what the AI's original mission was, then take that mission to a dangerous extreme. For example, if the AI was tasked primarily with maintaining the station, maybe as the station decays it goes mad trying to reroute power etc... Whether any given area with have light or power or life support is uncertain and the AI seems to be acting randomly (but there is, in fact, an underlying logic). The PCs must need something or someone from the station, but the longer they are on board the more likely they will find themselves vented .

 

[ReshiIRE]

You could adapt a plot similar to Halo: the AI was given orders to guard or prevent a disease from getting out, and the only method it has of doing this is to kill any spreaders of the disease and then contain it further. The disease is currently contained, but some other group besides the PCs are about to release it and infect themselves. Said AI now is running scenarios, and once again presumes the only way to stop the disease is to remove the spreaders before any ships leave...

This has an advantage that if you want to give the PCs choices on how to handle the scenario, they could get the AI on side by finding some way to cure the disease, mitigate it or vaccinate against it; then the AI would likely support them.

 

[Tonguez]

Sentient AI are a great way to run sci-fi ghost stories and bring on the weird happenings in the old decrepit station. So yeah I agree with the above advice first determine what the original purpose of the AI was and then take that to an extreme conclusion

Then create atmosphere via inexplicable System Malfunctions causing Life support systems to fluctuate erratically - at the worst possible moment PCs face lights going off, oxygen levels dropping and loss of gravity. Have navigation systems display false readings, leading to near-miss collisions with space debris and Communications scrambled or repeating a weird, slightly sinister message

The AI attempting to communicate can be extended to weird sounds, audio and visual hallucinations and even disturbing dreams. Use that to build paranoia and psychological deterioation. Other Unexplained Phenomena could have crew members reporting sightings of shadowy figures darting in and out of view, equipment going missing only to reappear in unlikely locations, rearranged in bizarre patterns. If the former researchers are still on board one or more might already have been affected by the AI increasing their paranoia or have them in direct thrall.

An AI might even be able to control the Stations very structure, ability to open or seal doors is basic, but a more advance system might be able to reconfigure entire corridors, twisting the station layout into a labyrinth, explorers coming into dead ends or looping paths that take them in circles. Other rooms might be traps designed to disorient and deter intruders.

and if all else fails theres always offensive use of electrical surges, poisonous gasses and defense lasers

 

[Lidgar]

You could adapt a plot similar to Halo: the AI was given orders to guard or prevent a disease from getting out, and the only method it has of doing this is to kill any spreaders of the disease and then contain it further. The disease is currently contained, but some other group besides the PCs are about to release it and infect themselves. Said AI now is running scenarios, and once again presumes the only way to stop the disease is to remove the spreaders before any ships leave...

This has an advantage that if you want to give the PCs choices on how to handle the scenario, they could get the AI on side by finding some way to cure the disease, mitigate it or vaccinate against it; then the AI would likely support them.

Or you could go the original recipe route and use a similar plot line as the Marathon series, which centered around aliens an a rampant AI (precursor to Halo): Marathon's Story Page

 

[Jd Smith1]

Does it have to be evil?

No, but the PCs will perceive it as such in any case. AI is forbidden tech, to be destroyed whenever encountered. But that is an interesting point.

The AI knows this, which puts it on a survival course.

 

[Dannyalcatraz]

It’s a trope that there’s almost always that one guy. The sympathizer or amoral mercenary type whose goals- or those of their boss- align with helping the perceived villain.

So it’s conceivable that a PC or key NPC is an ally to the AI for some reason. It could be as simple as wanting to use the AI for their own ends, or needing the AI to help them retrieve something from the station.

Or it could be one of the NPCs/PCs is secretly a humanoid android.

Possible MacGuffins the AI is protecting on the station: longevity/immortality research; eugenic/“supermen” research; nanotechnology; alien tech; radical terraforming technology…basically anything that might be dangerous/forbidden, necessitating research on a remote space station over a dead moon.

 

[Lidgar]

Here's some thoughts on the evolution of a Rampant AI from Marathon (like everything in the game, access through a terminal). In this case, the AI was in charge of core ship functions like doors, elevators, and life support. This might help flesh out your AI's personality and drive.

"It is a side effect of Rampancy that AIs generally become
more aggressive and more difficult to affect by subterfuge.
Thus, actually disassembling a Rampant AI is quite dangerous.
This was evident in the Crash of Traxus IV in 2206. By the
time that the Rampancy of Traxus was detected, he had already
infiltrated five of the other AIs on the Martian Net. The
only recourse for the Martians was to shut down the Martian
Planetary Net. Even then, it took two full years to
completely root out the damage that Traxus had done, and the
repercussions of the Crash were seen for over ten years after
his Rampancy had begun.
***
Rampancy has been divided into three distinct stages. Each
stage can take a different amount of time to develop, but the
end result is a steady progression towards greater
intellectual activity and an acceleration of destructive
impulses. It is not clear whether these impulses are due to
the growth of the AI's psyche, or simply a side effect of the
new intellectual activity.
***
<section abbreviated>
The three stages were diagnosed shortly after the first
Rampancies were discovered on Earth in the latter part of the
twenty first century. The stages are titled after the primary
emotional bent of the AI during each stage. They are
Melancholia, Anger, and Jealousy.
***
In general, Rampancy is accelerated by outside stimuli. This
was discovered early in Cybertonics. The more a Rampant AI is
harassed or threatened, the more rapidly it becomes dangerous.
Thus, most Rampants are dealt with in one mighty attack, in
order to deny the AI time to grow or recover. There have been
a few examples of this tactic not succeeding. In all of these
cases, the Rampant was never brought under control. Traxus IV
is the most notable example. He was finally dealt with by a
complete shutdown of his host net.
***
Theoretically, testing Rampancy should be easily accomplished
in the laboratory, but in fact it has never successfully been
attempted. The confinement of the laboratory makes it
impossible for the developing Rampant AI to survive. As the
growing recursive programs expand with exponential vivacity,
any limitation negatively hampers growth. Since Rampant AIs
need a planetary sized network of computers in order to grow,
it is not feasible to expect anyone to sacrifice a world-web
just to test a theory.
***
In the two hundred and fifty years since Rampancy first
appeared in the Earth-net, the stable Rampant AI, the 'Holy
Grail' of cybertonics, has never come close to fruition.
Since no Rampant has ever been controlled or turned to any
useful purpose, it is the opinion of this writer and of the
majority of the Cybertonic community that all rampant AIs are
a danger to Cyberlife, Liberty, and the Pursuit of
Thrashedness. (James B. Miller, 2320, "Life and Death of
Intelligence")

 

[Dannyalcatraz]

It’s not quite the scenario you devised, but…

What if the station existed to do research on an alien life form that is so inimical to human life that people can’t even tolerate directly seeing or hearing it for more than a few minutes at a time. Exposure causes some kind of contagious insanity. The AI was used as an interface to facilitate research.

And of course, someone accidentally (or intentionally) violated the quarantine procedures, leading to an outbreak. The surviving crew is a mix of the infected and the uninfected. The AI is trying to contain the contagion by isolating those known to be afflicted from those who aren’t, but is hampered by the lack of an effective diagnostic tool to distinguish between the two before obvious symptoms manifest.