What sorts of game design mechanics can one use to minimize reliance on AI in a strategy game? You've probably heard of game projects which crashed and burned due to AI issues and played your share of games with AI that simply fails to measure up. What game rules / mechanics can you use to eliminate the need for much beyond rudimentary AI? It strikes me that what you really need to consider are the choices the player will want to make surrounding a given situation. If you lack a reasoning opponent, you may have to substitute randomness, arbitrary constraints and a host of conditional rules (which ironically may actually be more brittle that AI programming!) More specifically, I've been thinking about the game design hole I've dug for myself in wanting to allow players to take on a wide variety of roles, from mobster to captain of industry. One way I saw that this might be possible was to have a map abstract enough so that movement and specific positioning (such as lines of supply or kill zones) was mostly non-existent or vaguely represented. The game would need to rely on possible permutations of rules to achieve any depth. If there was no "move troops behind lines to take HQ" then there'd have to be rules such as "HQs cannot be attacked unless combat strength was 2x the defender." This is boring by itself, so whatever process getting combat strength up was would have to be interesting and it would be crucial that this not be the ONLY rule for victory. Taking this approach, you'd lose flexibility but sidestep really tough issues like trying to detect patterns in player behavior or involving positions on the board (especially future ones like imminent envelopment). You might also depict a wider variety of strategic situations. Bank Heist / Prisoner Rescue / Surprise Bombing RaidHow many threats do I face? Where are they concentrated? What stops them from all massing against me? What tactics can I use? (Stealth + Surprise, Hit and Run, Siege) What does tactic X, Y or Z gain me? How might they counter tactic X, Y or Z? What do these tactics cost, both short term and long term? How can this situation morph into another (e.g., calling for backup / reinforcements)? Do you think it would provide a decent amount of strategic complexity if these elements were presented to you in the form of 1) Overall plans and 2) Potential interrupts once a plan was in progress? For instance, a bank heist, prisoner rescue or surprise raid might involve identifying the number of defenders. Without specifying individual position, we could portray this abstractly, such as "concentrated," "dug in" or "scattered." Some resource (like time/turns) could then identify how long it takes for the threat we face to morph into a different kind of threat-- for instance representing how long it takes for every defender to mobilize/mass against us. Tactics then could help control this, and defender states might be represented as a series of category values. If we had attack and defense stats, maybe attacker and defender lose their defense stats when we employ "Blitz" whereas maybe a long range attack exposes us to less damage but costs far more time. It might be possible to work out the abstract basics for a wide variety of situations then layer and balance from there. Might it even be possible to do with a strategic map all together? Thoughts?

Consider this:

Start with Master of Orion. Only, instead of competing with other AI empires, the focus is shifted. You don't have other AI empires. You have vague, somewhat simple-minded, enemies out there waiting to wreck your empire.

And, instead of focusing on the warfare/conquering aspect, instead focus on the... internal affairs and infrastructural aspects. Make it more like SimCity in space, if you will.

Still requires AI in a sense but a much more rule-driven one.

I've made a strategy game, Imperii, that only uses a very basic AI since it's a defense game. You build a defense around a base and the enemies try to walk through it. Even if the AI is a simply path-finding algorithm the game is very strategic. The difficulty is to build and upgrade the right turrets in the right time and at the right place. The whole layout of the defense is also very important.

The game:
http://www.gamerendering.com/Imperii

/

Consider a simple cellular atomata aproach.
Set each location on your map (grid, planet, whatever) as having a defense and value.

*Value brings traders/industry.
*Traders/industry bring pirates.
*Pirates drop security.
*Low security + Traders brings military.
*Military increases security to owner, but lowers value to owner.
**Lost trader ships in an area causes a temporary value loss to the owner.
**Enemy Military == Pirates to the enemy AI's evaluation of security.

Now, the player will be able to place anything he wants far more strategically than any AI would. But the AI would seem to be acting intelligently. It would attempt to counter your military to protect traders, traders would move around to capture high value areas, etc. Much like the fish/sharks/algae atomata, you'd see everything swimming about the map, acting in its own best interest.
To make things more "strategic" on the side of the computer, limit the military and trader sizes by some economic value for the empire, so the resources of the empire have to move around.


If you can distill the experience down to one or two features, it is easy to make an AI that SEEMS like it is doing good. A lot of RTS games deal with this by coding in what units are hard/soft counters to what other units. Causing any AI that sees you build Air to produce AntiAir to counter your exact move. The AI might not do anything else "smart", but it tends to have the right unit combination to counter your moves, making anything it does do seem all that more effective. Or again, some RTS games manage to always place units at the right spots (chokepoints, near resourcers, etc.) making it much harder to kill/counter at any encounter, because of good positioning.

So, anything you do ontop of the simple AI system is just flavor, the AI picked units and positions that were good. Now you get to pick how the dice roll in combat, ie.
he picks randomly and hidden from you 'turtle'.
You pick 'stealth'.
He gets a low attack(can't maneuver, reducing his options), high defense and scout. 3/7/7
You get a bonus to getting past scouts, very high attack (surprise), low defense. 7/2/4
So, he sees you, does moderate damage, but you still catch him somewhat off guard, doing moderate damage.
A diplomacy option would end up with both sides not losing anything, but all the enemy units being removed from the tile, and yours placed there instead. etc.

If you can represent all decisions as slider values you can create a system of equations that is then solved numerically to get an optimal solution. Computers are good at that. The AI can also approximate the probable values the player will use and expand the equations into several steps, thus looking into the future. Some of these equations can include map coordinates, morale, firepower, formation, area knowledge, ammunition and whatever else you can think of.
Then let the GPU solve that matrix.

EDIT: Note that map coordinates can also be considered slider values. And you might put in equations to prevent them from being to close to each other.

Have a stupid but overwhelming ennemy : ZOMBIES !
I never played DOTA but apparently this is the same motivation.

Tower defense games involve a predictable, stupid but overwhelmingly powerful enemy. It takes skill and strategy to defeat those.

Alternatively, you could have a game where a natural process is the "enemy" : manage an ever growing population, manage resources in a sea-rising world, build a civilization/city/base with a countdown timer that unleash a vague of destruction you have to resist for the longest time, etc...

Considering how many fps games where the AI is so horrendously stupid it isn't fun to play I've been trying to figure out how to make them seem more intelligent in generally easy ways and the approach I've come up with would be using elements from rts games.

If you use layers of AI so to speak I think you could create a rather intelligent but simple enemy for different types of games.

At the lowest level we have the traditional things most units need to have, path finding, aiming and maybe some way of finding cover and similar.

By adding another layer we can create squad behavior, a commander who analyses a tactical situation. From here order like flanking maneuvers or similar could be given to the units.

We can keep adding layers with command over lover levels until we tailor it in such a way that we get the effect that we want. Here more strategic decisions are taken, that area needs to be secured, defend this area.

As layers are added as needed we can get to the point where the main goals are defined for the game. Is it to rob a bank, conquer the world, defend something or whatever you need.

Maybe this doesn't really fit what you are looking for but by separating single units from the main enemy you can tailor it to fit different roles.

Different levels in this hierarchy could then take care of a variety of tasks, the tactical threat could be assessed by a commanding layer for that squad or unit. A higher commander could asses where the unit would be best deployed or most likely to encounter a more dangerous enemy, spotting weaknesses and order deployment, you name it.

A few points:

Don't think so much about *smart* enemies. Think about interesting enemies.
Don't think about *human-like* enemies. Think about challenging enemies.
Don't think about *perfect* AI. Think about fun AI.

Here's some ideas.

Good AI is meaningless without giving the enemy things to do. If you have perfect AI but the enemy is only capable of running and shooting, it's not very interesting. On the other hand, if you give the enemy interesting actions and some simple rules about when to do these, you can have simpler AI that's more fun to play with.

Quote:

Original post by Psilobe ... If you use layers of AI so to speak I think you could create a rather intelligent but simple enemy for different types of games. ...

This is the direction I've been looking at going with my game. It seems like this would help ensure a single AI doesn't completely loose track of itself and keeps the overall stratigy going the right direction. Also each higher level of AI could be run less often than the lower levels which would save on processing.

I definitely think this style of gameplay has enough depth to sustain a game if done properly. Looking at a simple example (which I now realise is fairly similar to KulSeran's):

The game world is made up of a number of territories.
Each territory has revenue and defense values and belongs to a faction.
If the player's faction owns a specific territory he can choose the level of extra defense to give it.
If the player's faction does not own a specific territory he can choose the level of attack to give it.
Attack and defence cost revenue.
If the ratio of attack to defence is high enough the territory is captured.
There are attack and defence bonuses for territories based on borders/distance from enemy/depth within friendly territory.

As I'm sure you can imagine the strategic possibilities are huge. There are also three main benefits from an AI point of view:

1) Computers are very good at making reasonable decisions in this type of game, its essentially an optimisation problem.
2) Most of the show-stopping AI issues are null and void. There is no need for pathfinding and combat AI.
3) Humans tend to ascribe intelligence where none exists in this kind of situation. It will feel like the computer is making complicated plans when its just trying to maximise its territory.

So yes, I do think gameplay this abstract can work. How to create a working mechanic for your examples is a harder problem. If it was to follow the game described above, which I am not suggesting but just using as an example, the territories could represent the captain of industry's facilities or the mob boss's front businesses. There could be more than one revenue type associated with each territory and some territories could need to use one type of resource to create another.