Single Player Game "AI"?

A thought on the process...

Including a system for boardgame enemy "AI" with any depth means that you would have to develop a clever and involved method by which it is achieved, since it would have to merge well and make sense with the player-controlled mechanics of the game. I wouldn't waste that effort on making it merely a system which invokes itself during a single-player game. Have it control the enemies during multi-player games as well.

~Josh

In helping a friend develop an MMORPG, we used a four-variable system. A was self-preservation vs. agression
B was close vs. far away from the character
C was like vs. dislike of the character
D was the number of times in a row the creature had performed it's last action. (If it was 4 or more, we shunted it to it's next-preferred option.)

We put together a 4D chart of the actions "Close with character", "Maintain distance", "Run from character", "Talk to character", "Ignore character", "Attack character", "Heal/Buff character", "Rest", and "Other" which was creature-dependent and often had actions like stealing from the character or buffing the nearest creature hostile to the character. Depending on which cell the four variables pointed to, the creature performed it's action.

Of course, that's waay too much for a board game, but the theory stands.

In a simple dungeon-crawl game, you can assume that all of the bad guys have the same goal: kill the player. Then, you have to ask yourself "What changes to the environement or themselves will affect this creature's actions?"

If you only have one variable (creatures change priorities as their HP diminishes), you can go all HeroClix style, and have the monster's action displayed on the base of the monster figure. Each time the monster loses a hit point, you turn the figure, and a (potentially) different action appears.

In it's simplest form, you'd want the actions to be "Close and attack" and "Run". Everything would start in "Close and attack" mode. Animals and other creatures with a strong natural instinct for self-preservation would switch to "Run" after taking a single point of damage. Magical beasts and semi-intelligent highly territorial creatures would switch to "Run" at about the half-way point. Creatures with an active reason to hate the party wouldn't switch to "Run" until they only had 1 or 2 HP left. And, of course, undead and other mindless or dominated creatures wouldn't ever switch to "Run".

If you want more than those two actions, you can have them without increasing complexity. If you want more than one variable (say, self-preservation on one hand and "how much do I hate the party" on the other, you'd need a more complicated method of keeping track of which action a creature would perform.

Good luck!

The problem with AI is that they generally fit only for a video game.

If you want a game with an AI, make it an optional way of playing the game (ex: solo mode ).

Else, you will have to give the player more control on the ennemies which will make their behavior more subjective.

But there might be way to stimulate the player to play correctly both side. For example, playing monster correctly could increase the challenge rating and make the character earn more EXP.

I remember having seen some Star Wars book where you could fight with a light saber or dogfight in a starfighter (We still have them in our game club). You had a picture of what you see, and each action you take makes you jump to another page.(Like a book where the reader is the hero). It was simply horrible.

You beat me to it, Larienna!

First off, let me say that most of the ideas in this thread are commendable. The "code-wheel" system to manage complexity is particularly ingenious.

However...

...I agree with Larienna that solid AI is generally fit more for computers. Even with the methods discussed in this thread, I doubt you could create opponents that felt natural. Any rules-based system that humans can manage at a board-game level of complexity will almost certainly produce very mechanical (and exploitable) opponents.

Which leads me to wonder if a slightly different approach wouldn't allow you to leverage the player's brain. Larienna's quote hints at the general idea: encourage the player to drive his opponents in a "correct" way. This allows you to greatly reduce the complexity of your AI because the human brain supplies it naturally.

Here's what I mean: AI algorithms consist of 2 basic parts:
1) What are my goals?
2) What are the rules I use to achieve those goals?

#2 generates the massive complexity of the problem, so throw it out. Have the game system supply only the goals for the creatures and let the player fulfill those goals -- as creatively has he or she wishes. To ensure that the player honors the goals, incur a score penalty each turn for failed goals and enforce a "red line" limit that forces the end of the game. In other words, if the player consistently plays stupid AIs, he loses.

Example:
Define Kobolds as follows:
Required range: 3-5 hexes.
Required end-of-turn state: under cover.
Preferred attack: light crossbow.

Every kobold must end its turn in a place that satisfies its requirements or the player loses 1 point per failed requirement. The player gains 1 point for every "preference" honored in a turn. These points are assessed in addition to standard points for kills, treasure, etc.

So now, whenever the player moves a kobold, he's going to try to maneuver it such that it's at range 3-5 from its target (which just happens to favor the light crossbow attack), position it behind cover, and fire at someone if possible.

You could have multiple "kobold" goal sets to reflect different morale states and/or settings (in cave, bright light, etc) if you so choose (as you discussed above). The key here is that the amount of information the player tracks for any given kobold in any given game state is small. Also, there's no fiddling with charts and tables to see how the kobolds move -- you just look at their end state and adjust your score accordingly.

This has the added benefit of reducing record keeping -- you only need to adjust scores for those creatures that fail to meet goals. In the "fully automated" scenario, you need to consult movement and attack rules every time you move every piece. Sure, you'll probably memorize those soon enough, but that's a steep barrier-to-entry for non-hardcore gamers.

The worst part of this system will be the tuning. Finding the right balance of requirements, preferences, and "mission failed" point totals would be a hefty task. However, I expect it pales in comparison to the balancing required by full AI sets.

Whatever you decide for your system, I wish you the best of luck.

K.

I came up with another idea to solve the AI problem. I will illustrate it with an example.

For example, in my space rally game, let say during the rally an event occur: "a space dragon appear". Normally, I would place a space dragon token on the board, then each turn I would need an AI to move the space dragon toward the player ship.

Now I have solve this solution from another angle. I declare that the dragon appears in an area of the map(and it does not move outside it). Each time a ship pass in this area he must make a die roll. If he is unlucky, he gets attacked by the dragon. All the other players who pass in this area this turn won't get targeted since the dragon only attack 1 ship per turn.

Another example in a recent ninja game I wanted to make. Moving the soldiers and determining if they see your ninja and start running after it ( think tenchu or metal gear solid ) would require an AI.

Instead of making a detailed map where you move from square to square, I made a map where each tile represent a part of the area. On each part you would have a token for each soldier present and the detection/pursue would be made by a simple dice roll. So there is no actually path finding or decision AI to make. The dice determine if you are detected, pursued or if you have successfully hidden yourself. Or course, soldiers could move from an area part to another, but they might try to set the alarm instead of running after you in the whole castle.

So the idea is to increase the abstraction level and remove the details so that it requires a less performant AI.