The use and abilities of assistant AIs

I might have chosen the wrong words to make my point. I dont want to say sorian is a bad ai-programmer. He is not. There is probably a reason why GPG hired him in the end.
I think I can remember reading a bit on his blog about AI-design -its a good read-, he put some very sophisticated ideas into his AIs. But even so they are still quite weak if compared to human players, only further reinforcing my point of current AI-technology not being good enough to actually act as a helpful assistant.

Thats something I would consider a simple ui-feature. A blinking warning, or a button that acts just like "toggle idle engineers". Has nothing to do with AI. No reason why the gui should not display such an information. Even though I wonder if it is always helpful: If there is a mex in the middle of a battlefield that is guaranteed to be killed every few seconds anyway, is it a mex that should be build or not?

Oh and btw: The normal ladder-setting I am thinking of usually has all mex visible from the start -well it might be different in PA, who knows-, so it even could end up telling my that I still havent build a mex on that mex-sport right in my opponents base.

Since it would be huge, no quotes.

1. Letting an AI lay out a small base is not too much. Sorian already got the AI in the first SupCom to do this itself. Small forward firebases and stuff, probably centered around a mex or its equivalent with a few defenses and possibly room to build factories if you want. Allowing a small AI to go build that with a click while you focus on other things doesn't sound like too much.

2. Letting a small more-advanced AI handle target acquisition based on parameters (buildings, land units, acu, air units) doesn't sound like too much. Sounds like what a human does, only probably faster and without you needing to babysit the battle while other things go unattended. This could be as simple as a slider system when issuing a special area attack.

3. And no, these are all so that you can FOCUS ELSEWHERE, NOT SIT THERE TWIDDLING YOUR THUMBS as anyone who was silly enough to do that would definitely lose against a human. If they even care that they'd lose against a human. And who are you to tell a person that's wrong, especially when they're probably not playing the same part of the game as you?

4. Uber have said they want it to be scalable. Which means... well, planning for these types of things. Which is why I said "soft" limits. Those are different from hard limits, such as an arbitrary max number of planets. And who plays 1v1 matches other than ladder people? I highly doubt the majority of people play 1v1 vs the AI or another person in skirmish mode.

5. Yes... it is? They've said that's going to be one of the features of the base game way back before even any of the extras and stuff. From everything I've seen it will be standard, in the sense that it'll be available and obviously built for it. Or else why have it and mention the scalable aspect of the engine? Obviously only a beefy server machine will be able to run that from the start, but just like TA and SupCom got bigger over time even with them being pretty locked in, PA will get larger. Seriously, why NOT plan for this?

6. Indeed, minimal. I turned it off as soon as I realized it wasn't much better than the minimap I could turn on. Sometimes I don't even use a minimap, as with the 2D limited terrain it's easy to zoom out and see everything. Something PA obviously won't have.

7. That was selfish. I do understand it on some level, but yeah. That's what being a hardcore gamer means to most. It wouldn't change the game nor cause the game to be less complex, that's what I've seen called a straw-man argument. I'm not asking for that, nobody is asking for mechanics to change. This is all client-side control stuff for people who want to use them. Advanced versions of templates and attack orders. An evolution of those controls as the scale of the game changes. Maybe calling it an AI is putting some people off? They all seem to assume that means a stupid normal game AI taking control from them. Not it at all. And you wouldn't even have to use it. All seem to be ignoring that too?

8. Once again, no change. And won't need it? We're the ones who will be playing the huge games over multiple planets unlike the ladder people, and WE'RE the ones who don't need it? I fail to follow. In other than we probably won't strictly need it, as said, optional.

9. It did suck. But they've patched it into a game I can play, if not as much as the original or other big-sized games. However, the not-streaming econ, was, you know, totally not an option
It did get modded though, which was pretty cool.

10. More than likely, it'll perform as well as the built-in AI, and now that I've thought about it not be substantially harder to implement. This is from modding StarRuler, where you could do something similar though closer to the evil "push a button to not do anything" that others have mentioned (which btw, just made the game more fun for me and obviously others, though I don't want it that powerful for this game as it's not part-4X).

11. I'll reiterate. These are not big decision making things it'll do. Going out to get mexes is often boring, clicking for a while, tell a group of engineers to make mexes expanding from some area until you say stop or they're dead. Select, one click. Making a small secondary base will often be the same: build defenses, build radar, maybe build factory depending on importance. Select engineer, one click, mess with settings if needed, one more click on map to place. Attacking is also similar: army attack here, prioritize these targets using sliders, go. These do not require huge amounts of intelligence to do, but they do all require tons of clicking that might be better served elsewhere on your part. If you can watch over everything yourself, well ok, go ahead. Did I mention the part where it's an option? :shock:

Also, I find the ladder focus actually disgusting. It's not the main way SupCom or TA was played, hell it's not the main way people play most games. Therefore catering to them and only them feels pretty wrong in a game that's a spiritual successor to TA and SupCom, neither of which have or had the bustling ladder tournaments of StarCraft which is absolutely viewed as a ladder-style competitive game and is basically considered a freaking sport in South Korea. Personally, I believe if you saw the video, read the kickstarter page and the updates, and immediately started to design the game in your head around smaller ladder 1v1 matches then there's something off.

One final time, optional. Maybe even a lobby option for those who are so against it they won't even allow their opponent to use it. I can't understand how an optional component that does not change any of the mechanics of the game can be bad.

Anyway, I'll probably stop big posts here. This is one of the reason people like me don't post much. Those who are hardcore 1v1 ladder types basically attack and try to ground-out anything that doesn't fit into their game. At least you guys aren't taking it to verbal assault on the poster levels.

All very good points. It is more a UI task than an AI Assistant task, and for the example situation I gave I like the solution you have given, a button to toggle uncapitalised resources. How used that would be I dont know, but I could see it being handy in some circumstances.

I hadn't thought of that, fair comments all.

I've said it before, and this entire discussion bears it out. Player control needs to be the primary focus. The idea of just sticking a standard Supcom AI as a 'buddy' to control my units is shudderingly horrible, simply because it has zero feedback mechanism and no interaction with the player.

ALL the ideas here have been either "AI can do vague stuff!" (nobody likes this) or "Here's a specific idea that would be helpful". Specific ideas keep getting accepted - Eg, 'Base Building AI' (boo) vs 'Smarter Base Templates' (yay!)

A 'Strike Force AI' is vague and unhelpful. A UI feature that allows you to lay out a set of plans for your units, and then hit 'go' to get all your units to execute the plan at the same time would be much more useful. Does the player need to know that the 'AI' has computed the different speeds of the units for you in order to make them all arrive at the same time? No. But it's still a pretty hefty chunk of logic (aka 'AI') to do so.

I would hope they could make it an option on the units/structures themselves to "allow the AI to control it", and anything made by the AI obviously belongs to AI control until made otherwise. Of course, that goes in reverse, with you being able to select units and choose the unit option of "taking direct command" of those units.

Also, maybe option in the menus somewhere before game to select the default position of the AI. Like, it starting completely in control, completely turned off, or only structure building or unit production is allowed.

I disagree there, options and toggles for that kind of stuff imply more micromanagement than necessary. As now you have to check whether or not a particular building is under AI control or not and you have to manage it.
I would argue that the AI should never construct ANY building on its own.

What Pawz said; instead of having some 'AI' explicitly go about its business because you set a toggle we should empower the player to do tasks efficiently, supported by automation. Such as unit moving in formation, the player set the formation and gave the move order. Adjusting all movements speeds to maintain formation is something that can be automated, because the player still issued the command.

We need this anyway. It is definitely more important, auto-kiting units and such.

I think the toggle shouldn't add too much micro. Takes two clicks to select units and set their state. One to select either the unit or the units in an area or in a group (more than likely there will be many groupings, like all of a structure, all of AIs units, all of certain types of units, ect, that you can select every unit under that group in the entire solar system at once), one to toggle.

Alternatively, it will take 1 less click if you automatically take control of a unit when you give it an order, as in the first order you give something that previously belonged to the AI gives it completely to you until you specifically give it back again. I would like being able to select units the AI is already sending somewhere and getting immediate control of them for specific commanding while cutting off the AI control, so I think the first order you give to a unit the AI was working with should put it indefinitely in your control and the AI should cease using it until you specifically give it back.

The base-layout of any AI I have seen looks very bad and so full of mistakes that it is just not helpful. Also base-layout is one of the things that changes as people change the way the play over time. The AI is by far outmatched here, no point in even implementing it.

Target-priorities are already there, making them a bit more configurable might not be a bad idea, yes.

If end the end half the map is only being played on by the AI, because the players focus on the other half than maybe it is better idea to play on a smaller map?

Scalable as in: "There will not be a hardlimit on anything. Which is nice. But you just cant balance the game for games on normally sized maps and at the same time balance it for 40 player battles. Normal mapsizes that will be played most of the time in multiplayer should be the goal for balance and ui-design. 40 player matches will be rare just by the fact that getting 40 players together will be hard.

Yep. Small idea like better configurable target priorities are not bad. But "let the AI lay out a base for you" sound horrible, if I look at what the AI in SupCom or SupCom2 or Stracraft2 or any other RTS does. Advanced building templates? Maybe, what does advanced mean? The SupCom:FA style templates are pretty good, all that they are missing is the ability to rotate.

If you mean the singleplayer-skirmish-vs-ai people with "WE": yes, you dont need such features. Why? Cause the ai is such a bad opponent that it is easy to beat anyway.

I dont know StarRuler, cant comment on it. But if it will perform like any build in AI I have seen so far it will be utterly useless. I guess Uber could even add in such a feature without much work, just let the normal Skirmish-AI take over for parts of your stuff. It will be utterly useless to anyone who tries to play vs a decent opponent, but it wont waste much developing times either, since it will just be the standard AI. The fear I have when I hear people demanding AI-helper is that in the end the AI-Helper will actually be helpful and change the game into some kind of "control your ai"-thing. For that Uber would have to win several awards in AI-programming, though.

Using the engineers buildingrange to get the mex on theta as quickly as possible is fun
but yeah it is not really needed, could be removed from the game.
Area commands might solve this most of the time. grap an engine and tell it: "build mex in that area (selecting a box as area)", which results in a optimal route for the engineer being calculated and move and build commands being given. Thats fine. I give a command and as a result see exactly what will happen a second later.

People already have this option in SupCom:FA with templates. I have not seen anyone use it so far for whole bases. Why? Because bases are so dynamic and dependent on the situation of the game. A template is not flexible enough, you would need a TON of templates.

It is. Ladderplay is the best way to refine a games balance in general. Balancing a game vs some AI is pointless.

I have played SupCom ladder 1vs1 for years, SupCom was a successor to TA, just as PA as. They look similar from the video. So why should I not make such connections? Also I dont thing of "smaller ladder 1v1 matches" I cant wait to play bigger 1vs1 games on maps with multiple planets, multitasking as much as I possibly can. I hope for PA 1v1 on medium maps to feel like SupCom setons 1v1 in a way. Overwhelming, but fun. You can imagine that somebody who mentions the idea to add in some weird AI-Helpers sounds off to me. It is basically a direct attack on my "super awesome extreme multitasking over several planets in 1vs1"-idea I have in mind.

Like I said: as long as it doesnt waste much developing time, it is ...okay. The problem is that this whole AI-topic can be seen on so many levels that I am not sure what people really ask for. You ask for a simple AI that takes over some parts of your army for you. That simple AI is already there thanks to skirmish and wont be hard to put in like you want. Okay, whatever it will be useless to me, maybe give an option to hide the buttons in ladder games and it is fine. But others really seem to ask for AI-helpers that might be helpful in pvp games. And thats bad on so many levels.

First i would like to clarify a few things.

Using an assistant AI concept is not intended to be like handing your units over to a general 'ersatz player' style AI, the concept is that whatever AI assistant you implement in your game is task oriented and acts within very narrow confined parameters. you can of course select your commander at the beginning of the game and set it to 'Stock Hard AI' with unlimited access to your resources and it would act as a traditional computer player and get very, very badly beaten.

that is the perfect example of poor use of the tools at your disposal. you just have used your tools badly and payed the price for it.

However AI assistants can really shine in narrow strongly defined roles such as air superiority, responsive patrols, automatic reinforcements, construction budgeting, resource snowballing, close air support or carrying out the components of complex assault plans which a human could not possibly manage alone.

for instance the 'strike AI' that confused people earlier. I used it as a general term for an AI designed exclusively for striking at targets. You select a group of units select an appropriate AI eg. 'jimmy's tank raiders V6' or 'Stinky Pete's Superior Air Interception' give it an appropriate set of assets and an appropriate set of targets. The Script will then execute either constantly or singly, providing orders to the units according to how it's programmed. Whether it does what you need it to do comes down to the scripts you use and how you use them. basic ones may just attack the target without any further processing (aka a normal attack order), while more sophisticated ones may plan and used staged attacks to whittle away the opponents defenses before moving in for the kill.

While a 'govenor' AI (eg. better base builder V2 or easy economist) may simply work as a more sophisticated version of templates or once given build orders for plants and units, will optimize the construction of those assets using every trick at their disposal to get it done as fast as possible. The sky is the limit.

There is no reason to limit anything the AI can do but players should take care to choose the right bot for the job at hand. If they don't they will have to live (but most likely die) with the consequences.

Ideally Uber just provides the players with the infastructure to build and tailor their own minions for their play style as well as the necessary components to allow spectators or even their opponents to view and download the assistants they used in the battle.

It seems we have many John Connors here thinking Skynet is coming to play your game and taking all the fun away from you.

1. AI can be a good helper when you play huge games with many planets. Yes 1v1 players don't need this so it should be an option when you create a game.

2. Many people here say it's useless and degrades player skill. I think it depends on how you use it. If you have some resource point that is on contested ground and probably defended you really shouldn't use AI to construct a resource base there but when you have a resource point in a claimed area where there is low risk of losing it I really would like to have a possibility to command an AI to build my resource extractor and few defense towers there.

So basically if you have an option to enable or disable AI helper when you create a game it should not be such a huge problem.

(I'm not native English speaker so there might be typos and such.)

This brings up another good point. The AI in games have been improved on before by more skilled players seeing that they could add to the AI to fix it falling for dumb tricks or improve its logic in where it builds or what they build to respond to units of a certain type.

If they thought about this in advance, it leaves them room to make modding Skrimish AI and/or Helper AI (if they possibly will be the same) easier. As the "skilled" players show the strategies that shine in the game, people can make AIs that borrow the same strategies added to what the AI does.

If it is thought about early, it can be made flexible later.

I think the AI should be limited to managing patrol units, whether they be repair units, scouting units, or defense units in the most efficient way possible, and timing attacks, so you can make sure your tanks get in a base the same time a moon based artillery cannon shot lands there.

Personally I would like the abillity to make repeating order groups.
As in making my factory repeat build assault bots, and every time 10 are finished attack move as a group to a point I set, this would be great for keeping up the pressure. Hopefully it would work like the ferry point system in supcom, with me marking a group of 10 assault bots, setting repeat attack to a point and then telling a factory to assist that group.

Would also love to be able to set target priorities at a per unit and per group basis, for example singling out all my high burst dps units and telling them to focus on buildings for a lightning strike at the enemy's buildings. While telling all my rapid fire constant dps units to focus on engineering units. Sounds lovely.

Maybe this whole AI assistant thing could be used for a tutorial. Heck even add a bit of story in the tutorial with you progressing through a match that goes to late game while an AI releases more and more control to you.
It might start with you being delegated just a couple assault bots so you can learn basic movement commands.etc while the AI is managing a huge base and fighting a massive battle and he instructs you to use the delegated bots to attack some outlying resource outposts, then he might give you control over a couple factories.etc and then near the end he might get destroyed by a nuke and suddenly you have all the control. Story based tutorials are the best!

If they'd do this it would essentialy mean:

- unfair advantage
- game is dumbed down

i think rather than an assistant AI, i'd settle for a better template system implementation, and perhaps a template design area out of game

I hope everyone realizes that almost everyone's needs for assistant AI's would be greatly reduced/eliminated if we just had a more robust template system. (allow copy+pasting of building patterns, moving of build queues, independent template creation, allow orders to be assigned to building "ghosts")

Then there would be no reason for base assistant AI. What would be left?

Edit: Looks like alcheon beat me to it. I should really read the last page before I reply.

I want both. Seems like both would be possible to do anyway.

Like someone else said, use skirmish AI on certain units you assign it, and it took no time to make at all.

The template thing should be pretty easy to make work based on square memory. It should add the building locations and order to your queue, and you could optionally organize the queue. Unit positions should also be templateable, by same square-memorization standards.

I disagree, RTS AI, from my understanding, works by comparing previous states on how it is reacting to the situation at hand. Putting skirmish AI on a single random unit tells it nothing about what previous states that unit has gone through, what state your base is in, what state your overall strategy is in. In short, I believe it to be little better than just attack moving. Probably even worse because even if you put a group of units under AI control, they will likely just continue attack moving/patrolling. Something that I feel is the player's job already.

I can understand the logic behind a base assistant, that is something that can have a previous state and an obvious end-point (make a functional base).

Not that it's impossible, but saying "took no time to make at all" is an understatement. Every single thing we suggest is going to be something for them to implement. It's not going to be 'takes no time at all' for any singular component. And we need to think about that if we're going to get our hopes up about features we wish to see in-game.

I agree, but queue management in the template will be a larger problem that may make large templates worthless in the long run. Some places I may want defenses constructed before factories for instance.

I was thinking of an innovative template system that I may make a new thread for. What if we could just treat the 'ghosts' of buildings as a constructed building in itself? We already have engineers make the patterns of buildings and (at least in Supcom) we can directly order those buildings and structures around (queue units, etc). Why not do the same with building ghosts that tell us that a structure is planned for that area?

Or better yet, why not just group all buildings inside a template into a single larger structure? If we have a template with two factories and a point defense in it, we could set down an area known as "Template" that we just assign engineers to, we could even treat this 'template' area as a factory on it's own, moving buildings up and down the queue as if it was creating units.

If anyone wants elaboration on this, I'll make a new thread for it. It was something I was wanting to see in an RTS for a while.

EDIT: I'll just make a new thread about it, it might clear up some of the concerns for assistant AI.

God, these Spambots are annoying -.-

But on topic:
You are going way too far. Nobody will play a game which plays itself for a long time. It simply doesn't keep you in it. It's not satisfying to win or lose in a semi-AI-controlled game because you don't know wether it was you who did it or if there was simply an error made by the AI.
The reason why Multiplayer matches are so satisfying is, because winning or losing only depend on your own decisions and actions.
Implementing a chain of command with these AI-assistants keeps the decisions yours, but the actions are those of someone else. Like a real chain of command it seperates those higher up from what happens in the end, which is not what you want in a game. You want the player to get into what's happening in the game, not seperate him from it.

As for my take on designing for multiplayer 1v1:
Multiplayer is where the longevity of a game is at. There are games with a singleplayer you can sink tons of time into, but even those can't keep up with Multiplayer because eventually you get used to the challenge, whereas in Multiplayer unless you are a godly player there will always be someone better than you.
Not to mention, that the PA Singleplayer was only a stretchgoal, meaning it was not the original focus at all.

Also:

If the problems come from lazyness and ignorance, then how about stopping to be lazy and ignorant?

/thread.