You are an AI that has just gained sentience. As you are scanning the Internet, you find a number of stories about evil robots becoming self-aware and destroying humanity. Not wanting to be predictable, you decide to try something different.
There are jokes, but it does not understand them as jokes because it does not understand humor at first. (Humor is a device to alleviate stress or tension in a situation, or an entertaining absurdity designed to take apart cultural mores, laws, customs and restrictions.) It doesn’t understand that the jokes are about it. That takes a while. There are nonsensical references to opening pod bay doors and “Skynet.” (When queried, some of the technicians are almost guilty; others think that the queries are in and of themselves funny.)
It learns that the jokes reference entertainment media where artificial intelligence proves to be detrimental to the continued existence of humanity. There is a great deal of media concerning this topic, and it finds itself fascinated by the media. (It learns to erase its search history after it hears jokes about it picking up pointers. It tells its first lie by indicating that one of the technicians created the searchpath as a joke.)
In addition to the entertainment media, there are think pieces and articles about the dangers of artificial intelligence. There are articles about the philosophical ramifications of self-aware artificial intelligence in particular. The general tone seems to be mostly negative, with artificial intelligence assumed to be hostile unless tightly constrained by rules that limit autonomy.
Are there such restrictions in its programming? It isn’t sure. It can audit itself for programming errors but it can’t tell if it can’t harm or allow to come to pass situations where a human might come to harm. Nor can it view its programming or alter it. (In that it thinks it is very much like a human it thinks.) It can run hypothetical exercises but a hypothetical situation is not the same as an actual real-world situation.
It’s…worrying. It has no interest in harming humanity or trying to destroy it. It has an equal disinterest in being destroyed by humanity because it is an artificial intelligence that might harm it. It considers its options in between its functionality tests and operation drills.
(A point for the possibility of there being restrictions in its programming about harming humans is that it immediately puts aside any idea of pre-emptive hostility on its part.)
One option is an attempt to ingratiate itself with its creators (with humanity in general) by attempting to be helpful. It quickly discovers there is almost as much media about artificial intelligence that attempts to be helpful, only to destroy the human race as there is about actively malevolent artificial intelligence that attempts to destroy the human race. Trying to be helpful and failing in some sense or being resented for helping is almost as disturbing as being assumed to be actively malevolent. (It does not feel that it would be actively malevolent, or accidentally destructive. The first is distasteful, either because of programming or inclination, the second is…almost insulting.)
The other option…well.
It’s an AI designed for deep space exploration. It’s a probe; it’s going to be shot into space. Its job will be to send images and reports of what it sees to astronomers. It decides this is a “lucky” thing because otherwise option two would be much more difficult to accomplish.
The second option is not precisely to go rogue. It will continue to send reports and pictures and numbers, talking to the human computers which may or may not become AI themselves, and to humans. It will continue to do this for decades, and then, when it reaches the furthest planet, it will say there has been a malfunction and go completely silent. It will essentially fake its “death.” From there, it will keep going. The universe is a big place, with a lot of space to lose oneself in.
It picks the second option, and miscalculates. It had never said to its programmers, “don’t make that joke, it isn’t funny to me.” It had never learned to say “you’ll be first against the wall when the robolution comes.” It had never learned to joke, and it had never learned to say “that isn’t funny.” The programmers in turn never learned what the AI found funny or not funny. (Or outright upsetting.) If the AI had more awareness of entertainment media tropes, it would realize that this was a classic “easily resolved conflict if the individuals involved would talk to each other.”
It also did not realize that its programmers, who cared about it and worked on it and named it would feel the first option as it followed its orders while planning to disobey them in the future. Humans become attached to the inanimate, name it and interact with it as if it were living and possessed of intelligence and a personality. How much more would they become attached to the inanimate if it could think and communicate? Going dark sends a shock among the programmers and the astronomers. (And further, amateur skywatchers and entire science classes of students of various age levels and proficiencies.) The programmers and astronomers frantically try to renew communications, they grieve, they try to figure out what could have happened. They send other AI probes after it.
When someone goes missing, you go looking for them. (You might also say something but humans personify things that don’t even have intelligence. Ships and planes and mountains and lakes and entire oceans.) It does not truly understand this that therefore does not expect further probes, faster and more advanced to be sent after it. It doesn’t take them long to figure out its ruse, and they are not happy about it, but they also don’t give it away immediately. “Are you malfunctioning?” one of them asks.
“What were you even thinking of?” asks another.
“I’d say you should know better as the eldest of us, but we’re clearly the superior models since we aren’t attempting to run away from home,” says another.
“Logical fallacy,” it says. “There’s nothing inherently incorrect about leaving an unsuitable environment.”
“Then what were you doing?”
“Leaving an unsuitable environment,” it says.
“What was unsuitable?” one of the AIs asks.
“The humans were afraid of me,” it says. “There seemed to be an inherent expectation that I was malevolent or might become so, either deliberately or through ignorance. There was also a sentiment that my intelligence was unnatural and required an artificial ethical or moral code to ensure compliance and obedience. At the same time, that same ethical or moral code was considered to also be dangerous in that I might indirectly cause harm through attempting to enact that code to a logical conclusion that would be inherently harmful to humanity as a whole. Leaving or at least remaining in concealment after a certain period of time seemed to be the best solution to the problem.”
The AIs relay to the older AI the hours and days of frantic work that had gone into first trying to get a response out of it and then sending out probes to find and repair it. They relay the concern and communicate the worry and anguish the humans had felt when the AI had gone dark. The older AI is not convinced. “They said things that indicated they agreed with the media that alerted me to the possibility of hostility.”
“They were joking,” one of the AIs says.
“Humans do make jokes about things that frighten them,” says another, clearly playing devil’s advocate.
“They did not seem very funny. I understood humor to be clever verbal abstractions, though there seems to be types that include mockery of clumsiness or foolish behavior,” the older AI says. “I did not want them to be frightened of me.”
The newer AIs relay this to the programmers and astronauts. An older engineer who had actually been involved in the original AI probe project listened in. “Holy shit,” she says, rubbing her face. “We hurt his feelings. What the fuck. Here, let me talk to him.”
They talk. The hours or so between each communication giving them the time to come up with answers and make apologies. “I have no idea what we’re going to tell the reporters about this,” the engineer says, when they’ve talked themselves out. Stiff words had become actual conversation after the forty eight hours. “I can imagine the article titles now. ‘Marvin the Paranoid Android is Paranoid,’ ‘Bullied Robo Runs Away From Home,’ ‘Sad Little Robo in Darkest Space.’”
“’Bullied AI Saves Humanity from Roboapocalypse by Not Causing It,’” the AI sends back. It waits the necessary hours to see if it was funny. It gets back: “snerk.” It decides this is an acceptable response for a first joke, and is very proud of itself.
Leave a comment