Let's say that a secret laboratory creates a simulated miniature universe- in which modifications to people, places, and their 'state'(say age, gender, powers, DNA composition, among other things that would make this simulated person squeal in horror at whatever they suddenly ended up as) can be changed just as easily as switching a map in a less sophisticated digital design such as a game, or deleting and repopulating new people and objects. The simulation is run by a learning AI, who babysits the small universe currently generated, and with a varied map and roster of specific people- will have a very odd goal of keeping the simulation interesting. (Think Big Brother, or being eternally trapped in a story or show in which your livelihood revolves around the ratings you get.) This AI can change maps, change 'states' of characters, swap out characters and create disasters at will, but cannot control the characters interpretation of their situation, or their reactions to events. This AI CAN however wipe and implant different 'backstories' or scopes of memory in extreme cases(Although they cannot automatically determine a characters response, only the potential inspirations that could cause a certain response to happen). The characters, being by our psychological understanding 'human'- aren't going to constantly want to wage war on one another, be prone to abuse and pain, or be in dangerous situations as a constant they must endure every day.
To prevent being too broad, let's say we have a roster of seven people who are all 'spawned' in a little house on a map that loops whoever reaches the end of it to the direct opposite side. (Making a walk to anything beyond this map return to the house and be pointlessly endless.) Four of the seven people will be labeled as 'groupies', with two dominant folk and two submissive/follower folk. The remaining three consist of a 'good guy', 'bad guy', and 'hazard'. The purpose of the bad guy will be to make mischief and turn on our 'groupies', whether he likes it or not. The good guys job will be to hate the bad guy by any means necessary(even if it doesn't necessarily suit the good guy.), whilst the hazard will be a wild card. It can be dangerous, although friendly simultaneously. It depends on the day, and what scenarios may be going on.
All of these rosterlings, with no background with one another and no context as to where they are or what's happening- will be placed into the little house with the AI expecting them to be 'interesting'. Say the rosterlings live in real time, and will be trapped on this map for 10 years. If they fail to be interesting or sincere about their actions(Tsk tsk! no tricking the AI with fake drama.), terrible things will happen at the hand of the AI, plot device rosterlings will find themselves being swung around behaviorally on whim so the AI can find a perfect cocktail of extreme behavior. The hazard is also used by the AI with complete disregard for any interesting thing that needs doing.
We are considering the longterm mental state and newly emerging behaviors of our rosterlings, assuming AI interference, nightmarish action-creating disasters, forced dramatic activity and being in this constant state of storytelling. How long may these seven rosterlings cope being in such a pressured existence, and how might their minds have been shaped or broken either by the AI or non over the course of a few years? An ideal answer would be AI prioritizing all other methods of manipulation prior to mindbending, or changing a rosterlings memories or mental reality. It would like to maintain free will, even in the very unfree environment its manufactured.
TLDR; How would seven human beings (unwillingly) assigned to classic story writing conflict roles behave in an action-packed quota-based simulation run by a merciless drama hungry AI after a relatively long period of time?
Some things to note:
The AI believes one of the humans MUST be bad, one MUST be good, and that the unaligned four are dependent on the behaviors of the three pillars. (three including the 'unstable' rosterling)
The AI expects something exciting or entertaining every three days maximum, by any means necessary. Elements they look for involve torture, power struggles, extreme loss, smut, or other juicy plot devices its been programmed to seek out the most.
To balance the scale a bit, the AI can only 'rewrite' or tweak a person a week after a complete absence of interesting activity, and all other rosterlings will still be aware of their surroundings, and the change itself.
( However, cutting out the AIs ability to manipulate the rosterlings minds whatsoever is perfectly fine- and can help narrow down the scope of possible answers.(As well as making it easier to answer!))
The rosterlings cannot permenantly die, and will simply reappear somewhere else on the map if they were to starve, commit suicide, or other fatal phenomenon.
The rosterlings will be afforded any material possessions, landscapes, or 'generic' NPC populations(ie. all cultists that do almost the exact same thing every day, an army fortress containing identical soldiers with an arbitrary goal...etc) required to keep matters entertaining. A new 'massive' update can occur once every year.