5

As a kid hanging out in my local arcade, me and my friends used to swear up and down that our computerized sf2 opponents would learn our strategies.

I wonder if anyone here might know whether that was the case. It feels as if it couldn't have been as simple as a finite state machine, because the opponent's tactics were definitely different depending on how early in the session you'd encounter them. And as each fight went on, it was as if old tactics would stop working. Brilliant games :)

anyways, I know it's a relatively open ended question. I had so many great memories playing that game, and it would be interesting to know whether the AI was that sophisticated, or otherwise what techniques it may have used so I can potentially use similar techniques in my own game designs.

Tetrad
  • 30,124
  • 12
  • 94
  • 143
Joel Martinez
  • 1,258
  • 7
  • 14
  • 2
    I wouldn't think SF2 had true 'learning' but rather very adaptive behavior. The game may start at a low baseline difficulty and as you play better, it ramps up the challenge. I've never played SF, but I've seen other games find patterns (combos) then create counter-patterns, based on a pre-made algorithm. Or else the game would end once you found a combo that worked over and over. – Adrian Seeley Aug 23 '11 at 12:02
  • 3
    You might get a better opinion/speculation discussion on this topic on the gaming SE site instead of GameDev :) – James Aug 23 '11 at 17:43
  • Linked: http://gamedev.stackexchange.com/questions/17447/games-containing-machine-learning-ai – e100 Apr 24 '12 at 14:10

2 Answers2

10

As someone who, more than once, beat the (SNES version of the) game on the hardest difficulty level with one repeated move (M. Bison's Psycho Crusher), I feel I can safely say it did not learn from strategies.

Though, this question does remind me of something I read in Ernest Adams' Fundamentals of Game Design:

Most current video games do not, in fact, contain much real AI. The point of video games is to entertain, not to simulate intelligence in depth, so they usually contain just enough AI to make the player feel as if the software is reasonably smart. The players—who are already immersed in the make-believe world anyway—are often happy to give the game credit for intelligence that it doesn't really possess.

Source

chaosTechnician
  • 1,735
  • 1
  • 11
  • 15
  • 2
    and sometimes bugs in the AI make it feel smarter/more realistic (source: http://blogs.msdn.com/b/shawnhar/archive/2009/12/29/bug-or-feature.aspx) –  Aug 23 '11 at 15:59
  • 2
    Very true. Pinky (one of the ghosts in PacMan) exhibits unplanned behavior when PacMan is facing "up" because of an overflow bug: http://donhodges.com/pacman_pinky_explanation.htm – chaosTechnician Aug 23 '11 at 16:13
  • 1
    Well the arcade version was totally different from the SNES version. For one the arcade is a dedicated machine and stays on all the time, so its memory doesn't get cleared. It was a lot more expensive than a simple cartridge and its conceivable they stuck in some pretty crazy programming. – bobobobo Aug 23 '11 at 16:20
  • 3
    Not to mention that arcade machines are designed to make you pay them more money to continue playing the game. I believe at the end of the day there is no adaptive learning however, it is infinitely more likely that the system simply knows everything it Can do at any given time (proper time for counters, blocks, etc) and just based on the difficulty it is supposed to be at, either does it or doesn't within some level of timing to avoid the ever present 'There is no WAY he could have done that!' chimed in with asking for another quarter :) – James Aug 23 '11 at 17:42
1

I played the game a lot on the Snes, and a little in the Arcade, and I know that era of videogames well cause I was a kid and I loved em... And I'm pretty sure the AI was absolutely unadaptative. I, myself, didn't see any kind of learning at all, but maybe someone from Capcom could prove me wrong...?

Sirs
  • 121
  • 2