Dumb AI as a feature

[guest509]

  I was just playing some Berzerk on the Atari 2600, also Night Stalker for the Intellivision. They are top down maze shooters about killing robots. One of the features of these games is the stupid AI.

  That doesn't sound so much like a feature, but trust me, it is. Dumb robots are much more fun. They kill each other on accident, they come at you but hit walls and electrocute themselves. They line up perfectly so I can shoot them all at once. It's great. It really is. 

  Contrast that to Brogue, where the goddamn monkey is smarter than I am. I can never catch those little bastards, they know right where to go.

  How is a monkey smarter than I am? Give me dumb robots to exploit anyday.

  Here are some youtube vids of berzerk and night stalker.

https://www.youtube.com/watch?v=gb4gWvqbsYA   [arcade version of berzerk]
https://www.youtube.com/watch?v=Yej3M9Ec8Cg     [2600 version, preferred]
https://www.youtube.com/watch?v=UaqbN480Wog  [Night Stalker]

[Pueo]

I'm definitely implementing this.  Trolls that just run straight at you even if there's a explosive in the way, dogs that will start feral but become all happy and follow you around if you feed them, "broken" automatons that run into walls and shoot randomly, the fun never stops!

[kraflab]

I think manipulating the short-comings of enemy AI is a staple of the roguelike genre

[guest509]

  Yes. Agreed. But I'm not talking about shortcomings here. I'm talking about idiots, purposeful idiots, that get a little excited and kill themselves. Shoot at you but happen to kill their comrades. Easily led into a death trap. The environment set up to be used by the player to lure these idiots to their deaths.

  Love it!

  Purposeful. Meant to be exploited. The game is designed to be that way. You know?

  That little 2600 game Berzerk is still very playable and fun due to this.

[Darren Grey]

I think AI should be renamed Enemy Behaviour or something similar.  The whole point of gameplay is to spot the patterns and overcome them, not to be simply beaten by an utterly optimal enemy.  Monsters should be interesting by how they act in response to different situations, terrains, items, etc, without necessarily being smart about them.  It's no fun if the monsters are too smart and you don't find the ways to exploit them.

Jeff Lait tried to prove that with Smart Kobold, but didn't achieve the required level of difficulty 

[Skeletor]

Contrast that to Brogue, where the goddamn monkey is smarter than I am.

Hahahahahahaha you crack me up Jo  

 I was just playing some Berzerk on the Atari 2600, also Night Stalker for the Intellivision. They are top down maze shooters about killing robots. One of the features of these games is the stupid AI.
  That doesn't sound so much like a feature, but trust me, it is. Dumb robots are much more fun. They kill each other on accident, they come at you but hit walls and electrocute themselves. They line up perfectly so I can shoot them all at once. It's great. It really is.  

I'm with you on this! I'd put that kind of sake into the "macgivering my way out" thing. That was probably the most fun part in Doom, where you had to dodge all those fireballs while listening raging demons getting angry at each other.

Dumb AI can definitely be a feature.

[guest509]

  I've come up with this behavior algorithm for my robots in my talkie-project, GUNFIST: GAIDEN.

  Each robot will have an Intelligence, or 'AI' score of 0-6. It's one of the 4 ability scores each robot has along with speed, shields and guns.

  Every (7-AI) seconds, or when the enemy encounters a wall, they'll decide what to do. So the top AI decides every 1 second, the stupidest every 7 seconds.

  Decision: Roll 1d6 + AI. You'll get a result of 1-12.
  1..3 = Halt. "Scanning. Scanning."
  4..6 = Walk. Random direction, up, down, left, right. "Searching for humanoids."
  7+   = Chase. Turn in the direction of the player, if you can see him. "Terminate."

  Shooting occurs whenever a player is in view along an orthogonal line, so no diagonals for now.

  Later, when I place lava and pits, the AI score will control whether or not the robot goes crashing to his death.
 
  This behavior algorithm should result in the robots acting like enemies in Zelda (NES), Berzerk (2600) and Night Stalker (Intellivision).

  At least I hope it will. :-)

[Omnomnom]

ah man theres nothing worse than the first level of brogue and a monkey steals your darts

[Hi]

I am fond of exploitable behaviors.  But both, random movement, and moving straight for the player are only somewhat interesting.
The most interesting behavior to me is when the creature seems to have a clear purpose in mind.  A dragon that breaths fire whenever someone gets too close to its gold is interesting.  Kobolds that travel across the map from one ore deposit to another are interesting.  Predators that are so irritable that they stop chasing you to fight each other are interesting .

I guess what I'd like is the best of both worlds; behaviors that are both easily exploited and purposeful.

[guest509]

  Oh I absolutely and totally agree HI. What I am going for with Gunfist, however, is a very light arcade experience. Real time shooter.

[Skeletor]

The most interesting behavior to me is when the creature seems to have a clear purpose in mind.  A dragon that breaths fire whenever someone gets too close to its gold is interesting.  Kobolds that travel across the map from one ore deposit to another are interesting.  Predators that are so irritable that they stop chasing you to fight each other are interesting .

Subtle observation.. I fully agree. Way more satisfactory to beat a system which seems like having its own logics.

[guest509]

  I too like this sort of behavior Skeletor. I was talking to my programmer buddy about it and he mentioned a good idea. Robots are fine to have sorta straight forward 'state' behavior. Sorta dumb, rigid, exploitable. Bumping into walls, falling off cliffs, etc...

  He mentioned adding other types of monsters with smarter behavior. Each type of creature (alien, mutant, cyborg, etc...) having it's own behavior.

  I don't think this project will ever get to the point where we add all of that stuff, but having a huge variety of behavior types sounds interesting (and hard to do!).

-JO