Why do single-player games have trouble offering significant replay value? I’ve been thinking about this question quite a bit lately, being in the process of building primarily single-player games.
I just came across an old Jason Rohrer article that attempts to answer the question:
Can you make an AI-free, randomness-free, physical-challenge-free, single-player game with gameplay depth akin to that of Go?
Is there any hope for the single player art game that seeks to provide that kind of depth at the gameplay level?
I now firmly believe that the answer is “no.” The proof comes from considering how one might go about winning, or doing well at, such a game. If there is a single, optimal path to victory, then systematically finding that path is the main task in the game. Once the path has been discovered and documented for future use, the game’s depth is exhausted. If there are multiple possible paths to victory, finding the rest after you’ve found one is an optional act of completionism, an exploration of mechanical depth.
Rohrer went on to make Sleep is Death, which requires the scenario creator to be live and in-game. He simply isn’t confident in the authentic depth of single-player games.
Rohrer gave his argument a lot of restrictions, though: AI-free and randomness free? No physical challenges? Multiplayer is his only solution: each player is matched with other humans. But even multiplayer has (at least) two big problems:
- It relies on the availability of another person with a similar expertise level. If I were to play a game of Starcraft against an expert, it’s likely they’d consider me less enjoyable and interesting than an advanced AI opponent.
- It presupposes that the player will enjoy either the idea or actuality of having a human opponent.
Rohrer’s examples of Chess and Go as the perfect multiplayer games, ones that can keep us playing forever, reinforces these points in my mind. For 1), I’ve rarely met anyone who provides an enjoyable challenge in such games — the gulf between beginners and advanced practitioners mean they are either easily defeated or wipe the floor with me. Even matchmaking systems are generally flawed. For 2), quite a few people (including me) just don’t always enjoy multiplayer. Humans, in many ways, are just as imperfect as game mechanics.
One more assertion Rohrer makes is that AI has mostly disappeared from modern games. I suppose that’s true in a sense, but I’m also interested in the possibility that we’ve just gotten better at using simple mechanics to produce interest.
This comment below Rohrer’s article caught my eye:
“ I remember years ago playing Quake with bots. The bots were exceptionally simple AIs: they were able to navigate the terrain collecting health, and they had perfect aim. The only thing that made them semi-enjoyable opponents, was their persistence on using the shotgun and ONLY the shotgun. They were hard to fight. At close range they were lethal, and even at long range they were punishing. By an “intelligent” standard they were far from perfect players, but that is what made them “enjoyable”. Their imperfect understanding of the mechanics allowed for a win condition.”
I remember those Quake bots too, and agree that they were enjoyable — as much so as a “smart” AI that knows how to flank me but has imperfect aim.
True, their appeal runs out quickly. But why not update the old Quake formula? Indie game design is currently preoccupied with random generation of static content (rooms, enemy placement, etc) but it's conceivable that AI could be coded in the same way.
For example, drawing from some aspects of Spore:
- Create an alien rig that allows for variations in size and movement speed
- Create a set of weapons with variations in damage, area, range and speed
- Create sets of simple AI behaviors: guard, flank, attack, retreat, etc
Mix and match all three categories in each new “encounter”; preferably, the aliens will be stupid and somewhat crippled, but appear in overwhelming force until the player figures out the right exploit. For bonus points, keep statistics on player matches and save enemy types that are in the right “challenge” range.
Would it work? I’m not sure, but it sounds fun as an idea.
The idea of keeping players unbalanced and uncertain of what’s coming next reminds me of an old question about who would win between a samurai swordsman and fencer. The answer was that both would die, given that they started unaware of each others’ style. The reason: both types of swordfighting are reliant on a small set of moves that are too fast and deadly to counter, without the ability to predict your enemy.
Swordfighting, although a “multiplayer” game, is just as much about the rules and properties as the intelligence of the opponent, and until you’ve seen many similar scenarios, a simple set of trained behaviors would be just as good as a human opponent.