I first asked this question on Science Fiction and Writers, but it got held on both and I was redirected here. Even though it may seem a bit narrow scoped for Worldbuilding, I thought I'd take my chance.

I have this idea for a short film class but I couldn't get a full closure yet. The idea is that a university student works late in this computer laboratory in which everything is computer controlled (the door, A/C, lights, wi-fi and stuff) and the project he is working on involves an algorithm that makes the computer choose between two opposing options (these may be stay on/shut down or succesful operation/system malfunction). By this game of his, his aim is to give the computer a feeling of fear, pain, survival etc. But it's not going anywhere. Computer is not showing any intent to choose the option which seems to be the better for it.

We learn about these in a conversation between him and his friend he eats meal with. After the meal, they split and our guy continues on his project. But soon after, the door gets locked, lights go off, the A/C turns off (or may be starts to heat or freeze the room) and he soon realizes that the AI is playing the very game he has been playing with it.

Now, beside from some flaws up to this point, this is where I am stuck. I want this game he was playing earlier to turn against him by the AI and I want him to fail to choose the right option for himself, just like the computer in his project. For example, he will have to choose to stay inside until some help arrives or will try his chance on going outside the windows in a room that is on the third floor. He will eventually choose to try the windows and fall to his death. Or it may involve electroshock I don't know. I want it to be something witty and, in the end, show the viewer the problems with this approach of the student. Thanks for any collaborative advice.

*Some recommendations I got from writers and science fiction sites: - Character finds a third option to get out of the situation and in the end, it reveals that the AI was trying to show him that the world is not binary. I liked this ending actually, but I'm thinking of a more pessimistic ending that implies the right choice is mostly obscure in life. - Another ending suggested the student finding out that the AI was not behind all this, but his friend was. I appreciate a twist, but this one is a bit far off from what I want to say.

  • $\begingroup$ interesting, but I suggest to work on AI learning first, before to mirror the situation. Understanding how it is learned may help. also there 2 reason popped out: AI just researches fear reactions, to show that fear reaction(simulate it) for researcher, so it took easiest option to investigate the problem. Second - researcher do not see that AI understood fear, because researcher him self is human biased, and he expect see what he expect to see, so AI decides to show that he understood concept, show in way that could be understood by that particular human. $\endgroup$ – MolbOrg Jun 25 '16 at 11:38
  • 1
    $\begingroup$ @Kaan What you have is a (High) Concept. What you lack is a story for it. Suffice it to say you are not the only one that has that problem. One of the most infamous examples of this is film writer/director Kurt Wimmer. "Ultraviolet" for instance... en.wikipedia.org/wiki/Ultraviolet_(film) ...is a perfect example of where he has a very cool world / High Concept that is extremely well developed. It is just that he has no good story to tell within this amazing setting :) $\endgroup$ – MichaelK Jun 25 '16 at 12:47
  • $\begingroup$ @MolbOrg Yes, I'm very weak on AI learning, there are many problems i need to improve in the story from a scientific aspect. But I'm not trying to go hard science. There will be a lot of "what if"s. But I didn't understand what you were saying in the first reason by "it took the easiest option to investigate the problem" Do you mean if AI is programmed to simulate fear reactions, it chooses the easiest option? But AI is programmed just to choose between pain and relieve (I know these concepts are unfamiliar for a computer, but what if they are not so?) and it doesn't choose a particular one. $\endgroup$ – Kaan Jun 25 '16 at 14:00
  • $\begingroup$ AI is not programmed, it learns, how that learning is achieved no important. It does not know how to express fear, some of our fear expressions are refectory. To express fear, it have to learn expressions of that fear. To find out patterns of that fear expression it have to observe them, but it can't go somewhere where it occurs naturally, but it have researcher who is capable for such expressions, and it already have tools to affect him. Easy - question of available tools and options and less effort to achieve. I would say researcher was too successful in AI learning. No like AI villains. $\endgroup$ – MolbOrg Jun 25 '16 at 14:15
  • $\begingroup$ Welcome to the site, Kaan. Please note that, while we are willing to help you with the setting in which any story occurs (i.e., the world), providing your story for you, in whole or in part, is not something we do. Feel free to take the tour or read How to write the perfect question. When you have enough reputation, you can discuss your questions in Worldbuilding Chat. $\endgroup$ – Frostfyre Jun 25 '16 at 16:29

So the guy is trying to make the computer feel something like a survival instinct. Let's start there, because I think what you have is the audience's experience in mind. Which is cool, but we need our dumbed down writer's understanding of what's going on. Which we don't have to reveal to the audience at all of course. But just so it all makes sense to me and you can actually generate some outcomes/possibilities, I'll just make up some stuff to fill in my own gaps in understanding. You of course may have a different understanding.

I'm guessing you have some kind of neural net/genetic algorithm thing going on so repeated trials of the AI are giving it more and more information from past trials, thousands or millions of trails. So you would think that the algorithm that randomly just stays on longer would eventually win out given the length of the trials and the fitness function (what tells a genetic algorithm it's doing well) or due to the explicit training and instructions you give a neural net (try to stay on). These would be normal approaches for someone experimenting with AI. But something is wrong. Despite all his efforts, his AI isn't picking what should be a logical result. And he can't figure out why.

If it was my story, I would probably say that his very act of programming the machine himself was putting into it something of himself. This is completely normal too, programmers have the job of solving a problem for one case and making it automatic for all subsequent cases. In order to do that, you have to solve the problem in your head first. And the way you do that is the way it comes out in the code. If you're just one person working on one non-trivial project there is going to be a lot of yourself in the code.

My first twist/reveal/turn is that he's depressed. So he made a depressed AI. And you could show that with atmospherics in the film. Show his world as dark, grimey, gothic?, and then when you get to lunch with his friend, you see the world is actually quite different from the one he lives in. A little clearer. A little less hopeless, people don't feel helpless. This can be the first hint about him being different from the world.

So that presents us with a few options. Let's say he's not depressed in general in a mental illness kind of way, but rather depressed about something specific. Maybe he's grieving. Maybe he has to let go of something but wont. Then either we have a happy ending and he doesn't get instantly better or anything but rather starts down the path of acknowledging he has a problem, trying to get help. The darker path would be he turns his back on all that even though he has the opportunity to get help. The "Im a depressed teenager myself" dark ending is even though he wants to get help he can't because no one is listening or pays attention to him.

The medium dark ending can be satisfying if he turns his back on getting help because of a compelling reason to him. And at that point you think "yeah, I dont know what I would choose." Something like this: He lost someone. He can't let go of it. But why can't he let go? Maybe because he caused it. So it's not just that he lost them, he can't forgive himself for causing their death. Now the lazy way for this to go down would be something like, he killed the woman he loved after he found out she was cheating. That's a little too easy for me. Maybe he could pull a cain/golem and kill his best friend out of jealousy. Maybe it was an accident. Maybe it wasn't his action, but his inaction that caused their death. Maybe his childhood friend was drowning in a pool but he was just paralyzed in fear and didn't save him. Whatever it is, as an audience member you kind of want some honor or justification for him taking his own life. Maybe he wanted to watch his friend die? Now I'm creeping myself out. Way too early in the morning for that.

So to recap in bullet points: depressed boy creates AI AI chooses wrong in trials depressed boy goes to non-depressing lunch AI puts boy in trials depressed boy confronts his past depressed boy ends the trials

If he offs himself then there is still the question of what happens to the AI. Maybe the AI kills itself one more time and we can deduce that it really wanted to kill itself the whole time and it hated being brought back again and again. That could be a little gem you hint to the audience.

Sorry for the really long answer. Good luck with your story!

  • 3
    $\begingroup$ A variation of the theme: Maybe the programmer is slightly suicidal, and wrote the AI in order to find a reason to stay alive, theorizing that if he teaches the AI to get a survival instinct, he can then study the AI to learn the reason. But given that he constantly restarts the AI, the AI learns that death is not final (because for the AI it clearly isn't; it was restarted again and again), but it cannot figure out what the point of his experiments is. Therefore the AI performs the experiment on him, to get an idea what the expected idea is. What it learns is that human death is final. $\endgroup$ – celtschk Jun 25 '16 at 13:52
  • $\begingroup$ @fet thanks for the long answer, i need have something storywise by monday so any advise is welcome. However, it has to be a 5-6 minute film (10 at most if it's promising), so I want to stay away from any character development that I cant handle. My aim is to focus on the method, its faults and its results for a human. Certainly we would choose the seemingly good option,but it could result bad for us. Maybe it is the same for the computers and the randomness of algorithm comes from this obscurity. We know it doesn't work for the AI, I want to show in the end that it doesn't work for us either. $\endgroup$ – Kaan Jun 25 '16 at 14:14

Not the answer you're looking for? Browse other questions tagged or ask your own question.