Another way to look at AI?

Reluctantly, I am makeing a topic about AI, as it has being done LOTS before, but I feel like its a very important topic to address. Ok, so many people are saying how the AI sould be able to do menial tasks, like replacing lost buildings, and some think it should be able to do basic base management, all the way up to attacking for you. Well, how about have all that fancy, shiny AI, just a basic AI, to "housekeep" planets when your not veiwing them. In this game, a human that is Microing should always be able to beat a AI (unless the human is bad at RTS (like me)). So, why not have it so, when the play veiws a planet, the AI shuts off, and lets the player control the planet, while it is left to "housekeep" plants not being veiwed by the player activly. That way, the player is always looking at the most imprtant thing, and not having to worry about rebuilding on the 7 other planets he has.

How is this in any way different to what was being suggested in the assist AI thread?

This, exactly.

Regardless (thanks for pointing that out, Sorian!) of everything, I think many people who try and defend things on the basis that it "worked fine in TA/SupCom/FA" doesn't work in many ways when talking about Planetary Annihilation; remember, we are having to deal with multiple planets at once. It'll be hard to completely micromanage all that.

On another note, however, we don't know how well this would work in practice. In most RTS games, once you attack a base you aim to destroy it, only sending the little raiding parties to minor bases which can easily be taken. We could find that having auto-build just wouldn't work because there would never be enough minor damage for it to matter to rebuild it. Obviously you won't want auto-rebuilding engineers when the enemy is trying to blow them all up, so it could never be used.

I'd rather stick with things like area-reclaims and area-bombing runs, that type of stuff, just because I think auto-building engineers wouldn't be useful.

This is just a suggestion, but what about having the player CREATE the A.I. that his bases, engineers and commanders will used while he's not looking. This could be a very simple yet powerful interface, that could even be done out of game. You could set engineers to build things only if certain conditions are met. This introduces real intelligence in thinking of how it will work, because realistically, a commander of an intergalactic war really should be telling people how to do things and when, not ordering them around. You could also specifically set A.I. patterns to specific bases, planets or systems.

Here's an example.

Engineer:

Build Factoryx2 if Base Health ≦ 60%

You could even add more things, like time, place, etc. But you get the point.

Blue is for actions, green is for objects, red is for values, cyan is for operations.

This is not a simple solution. Reasons:

Build where?
Build with which engineer?
Build with how many engineers?
What if there is a shortage of mass/energy?
What if we have a million factories already?
What type of factory?
What is a base?
What is the extent of a base, what is included in the base, what isnt?
What does health mean? Average health of all buildings? Based on some arbitrary template?
How do I prioritise this with every other order you have given me?

Here is another way to look at it see how we go with out it on alpha then decide if every one is having a hard time trying to manage every thing.
Then see what we can do when it is and if it is a problem for all we know the ui might be that awesome that we don't need ai to help.

The example I provided there was just a very basic draft that I had thought up on the spot. As I mentioned, you could add a lot more conditions and parameters such as the area in which to be built, which engineers to use, what sort of factory, etc. I didn't mean to exactly define everything that is definitely in the game, which is why I used such a simple example. What I was going at was not what it was to look like, but simply the basic idea. I'm sure if the devs used something like this, they'd sort all those questions out.

My point is that there has to either be a few thousand lines of code hidden from the user which answers all those questions and simultaneously limits a lot of control for the user, or every "simple" script will either be 100+ lines long or rely on a lot of supporting scripts.

Either way this is not simple and is basically assist ai's/outright modding.

I think what would be better is improvements just to unit AI, and the ability to add more advanced commands/restrictions.

One that REALLY irks me in FA, is that my patrolling engineers, who I want to reclaim trees, will stop reclaiming if my mass income goes positive, and attempt to assist in building tasks instead.

The number of times I've caught the little critters upgrading a mex instead of eating trees. It would be good to give my engineers instructions to reclaim only. Likewise, I should be able to set them to 'repair' only, or 'factory assist', etc. Or a mixture of each behaviour.

Another thing is patrolling combat units that decide to stop and shoot. With land units especially, this negates most of the reason why they are on patrol in the first place - to avoid taking fire.

Perhaps a thread discussing unit behaviours that we would like to see would be more in order than what essentially is an AI Co-Commander.

I know we're into 'beating a dead horse' territory now. But I do have one thing to add:

I have to agree in that a singular 'patrol' command is too broad of a function. It's just something you have to hope your engineers are skilled enough to do and that no shiny objects that need nanolathing lay between what you want and them.

I've recently started playing Zero-K some so I could get a feel for what some people on this forum were talking about and I pretty much think it has the best system of just splitting up patrol for mexes, patrol for repair, and patrol for reclaim into three separate commands that allow me to lasso areas on the map to specify location. Really I can't think of a way to improve that, so since sorian may be reading anyway, that might be a good thing to consider when programming unit AI. Or to anyone who is programming the UI in general.

I guess what I'm trying to say is: I want you to take some of those great UI features from Zero-K. Yes, it's true some of the ZK developers read this forum, and that's good, they made one of the best ways to command engineers on a large scale and I can hardly picture playing with anything less now after maybe four days of lightly experimenting with it.

So you basically say that you want the player to be able to beat the AI by bugusing? Because thats what microing is all about, abusing flawed game mechanics to achieve unintended advantages (like virtually invulnerable units by abusing the range system or dodging projectiles) which have not been considered in the initial unit balancing. You can't solve this problem by simplifying the game mechanics, you would have to simplify it far to much to prevent bugusing at that level. The only solution is the opposite approach, make those bugs a part of the "equation" by making the units intelligent enough to use those behaviors on their own and balance the game based on their real efficiency.

By the way: I hate the fact that many people on this forum apparently can't differentiate between the AI which is responsible for the unit behavior and the large scale AI which is responsible for making strategical decisions like sending units to the front line or even another part of the AI which is responsible for constructing a base. If you say "AI", please make clear what aspect of the AI you are talking about. Especially don't mix unit behaviors in battle situations with base construction or even strategic decisions. Those have nothing in common, there are completely different systems behind those tasks and they shouldn't even be announced in the same thread, especially not when talking about "balancing" or even mixing in advanced controls.

This thread is a great example, even the OP started mixing individual unit behaviors (micro management and attacking) with base construction and even managed to bring strategical decisions in ("housekeeping" the planet).

Oh, and we already have several threads about better unit commands, both about being able to give "area commands", and being able to simply "toggle" parts of the individual unit behaviors, like turning off assist for engineers or turning off the chase option for units without affecting the other options. (TA had something alike, SC and FA didn't!) And it's simple too, if the most appropriate action (as rated by the unit AI) is a disabled one, then it simply defaults to the most appropriate, still allowed option, as long as it makes any sense. (Thats not a new idea, just repeating what was discussed about a month ago!)

I second this. There is a huge difference between low-level AI and high-level AI. In fact, low level AI is generally solvable on a discrete scale, whereas high-level AI has a certain level of uncertainty behind it. Therefore, problems that are easily solved by machine should be solved by machine. That's what computers were created for in the first place. It leaves the more complex, high-level problems for humans to solve, unencumbered by low level issues.