Imagine, if you will, a multiplayer version of a survival/horror game which would randomly record the audio from you and the other players, and would substitute them for NPC/monster dialogue; perhaps even substituting for other players communications. Combine this with visual effects where your friends slowly begin looking more and more like whatever it is you are facing.
One of the players chimes in via voice saying "I found a clue", but the game intercepts it for you and substitutes an earlier bit of voice saying "Look out, shambling monster!". You run to where he is and see the shambling monster, but no sign of the other player. When you ask "That you, Bob?", and the player replies with "Yeah, that's me", it might come through, or it might get changed to another snippet of recorded sound, or the sound the shambling monster makes.
With the aforementioned shambling monsters also occassionally 'saying' some of the randomly recorded voices (including yours) instead of making the sound they normally make, you can run into instances where you are absolutely sure that the monster in front of you is real and attack; or is friendly and start yelling for them to stop attacking you, but it turns out it was the monster this time.
Throw in some creatures that will sometimes follow you instead of immediately attacking, and now you can run into a situation where there's 5 people in a room, only 4 players, and some of you are 'insane' to the point where you can't tell which one doesn't belong. Plus, if you can have the 'doppleganger' mirror the actions of the player to an extent, a few checks like asking the player to turn in a circle or hop to prove which is the real one won't always work.
