The use and abilities of assistant AIs

reading all this i see a basic division into three different camps:

1/. the "I don't want AI Sub-commanders and i don't want you to have them either" camp

2/. the "I would Like AI Sub-commanders to be able to be ordered to construct bases of different types at specific locations but have no control over my unit production queue's or army"

3/. and finally the "i want AI Sub-commanders to control any game aspect that i don't want to do myself" group

personally i favor the second of these options, i don't want an AI to have control over my unit production queues or major army movements but i do want the ability to order an AI Sub-commander type unit to travel to a location and construct the basic form of "X" type of base at that location after which i can come in and tweak the base to better suit my needs.

eg. if i want a basic resource gathering operation on a moon or asteroid i can simple give the command to the sub-commander, saying i want him to go there and build up a basic mining operation.

if i require a basic military production facility closer to the front lines again , i can give the command to the sub-commander, saying i want him to go there and build up a basic military production base then i can expand the base or make improvements myself if i need the base to be permanent


its a trade off, you cant be everywhere at once so you trade the ability to specifically plan and design every aspect of your base so that its absolutely perfect in every way, for a more one-size fits all approach and go for a prefab construction method, ie. build me the bare bones of an operation because i need an imperfect base now, not a perfect base 10 minutes from now. once ordered it would free you to direct your attentions to other areas that perhaps require a more significant portion of your attention, maybe you have an area that really needs a base designed specifically for it, giving a basic prefab construction order to a sub-commander would allow you to get creative on the base that needs it while trusting that the base that just needs to be built will get built and be effective if not perfect.

Excerpt from PC gamer Interview with Mavor:

My expectation for 60 moons and 40 AI opponents is unrealistic, I admit. The core of my argument remains, SupCom scale tools won't be enough to manage large PA games, be them massive or merely large.

I encourage ÜBER to explore ways for smooth gameplay at small and large scales: AI assist, area commands, co-op single army, gameplay pauses every 30m for players to qeue list of commands, broad high-level commands, commands stand apart from units and units get assigned to them, etc...

Some of these ideas might suck, they might fail miserably, but trying and failing is how we learn, it's how progress is made. I hope ÜBER keeps experimentation at heart.

Seems the AI assistant debate has developed dramatically since I first saw it. I was originally quite enamoured with the idea of an AI assistant for all of the possibilities that it suggested, however in the light of arguments against them I have to adjust my views.

What Thefirstfish has suggested in the quoted text is what I would like to see. Adding complexity to your orders, or contextual sensitivity to your orders. So I guess I come down on the side of UI power in this debate?

Another thing I would like to see is an evolution of the template system seen in FA. It allowed for the minimising of repetitive aspects of gameplay within a match without taking away from the player.

Finally there was something mentioned by Neutrino (where I first came across the idea) that I think paints an AI assistant in a different better light. http://forums.uberent.com/forums/viewtopic.php?f=61&t=34299&start=20 - It has been mentioned that it may be possible for more than one person to control a single army (as opposed to two armies allied) - I for one like this idea. I'd be able to play games with my less RTS fluent friends and lend logistics support while they get better. Besides the idea of co-op commanding a single army is pretty damn cool.

Now; if we assume it is that kind of game, where both sides are co opting command of armies and someone doesnt have a person to co opt with, I see no issue with having an AI fill that space than having an AI ally in a 2v2 match with only 3 players.

say that pending further discussion this is where I'd stand on the issue. UI power, templates and the option for co op with AI for those lonely nights without a friend. The later of course being an option for those kind of games where such a system is appropriate.

This whole game seems to be quite the experiment, as is the level of community involvement in the discussions on PA. I think some of the debates get very heated and at times seem to be trading insults as often as ideas.

Despite this overall I think it is great to be able to have a critical debate on features and try hash things out and have the developers watching to impliment developed ideas with promise. I dont think you have much to worry about concerning their openness to experimentation

there seems to be some confusion about what i meant when i proposed the topic and a lot of people are getting quite worked up.

The concept is that you would temporarily cede control of your units to a script with a defined set of inputs and behaviors and it would follow its default behavior based on its inputs until it was either deactivated or its assets were all destroyed.

Units in TA and SC/FA already do this with patrol commands, assist commands and their normal idle behaviors.

the Crux of this concept is that the UI allows for more sophisticated scripts to be used in addition to the already defined scripts which have served us all well in the previous games.
These more sophisticated scripts are the assistant AIs. they can issue any command the player can through their interface with the game and if they are appropriately programmed/chosen achieve tasks set for them by the player including making use of their own sub AIs.

For instance the Air superiority AI, uses a subordinate Scout AI to maintain surveillance over the operational area and assigns appropriate forces to a Strike AI to destroy interlopers. in the old days you would use a patrol of air superiority fighters to do the same job but an AI does exactly the same thing but smarter. Same effect but uses less resources and provides more effective cover without the player having to be distracted from their latest and greatest project.

clever use of strike AIs should allow the player to put together pincer attacks without having to babysit the formations of tanks who inevitably position themselves badly. during that time they can quickly and easily target enemy defensive installations with the appropriate amount of bombers using another strike AI.

Governor AIs could manage construction, manufacturing and maintenance. Strike AI's Assaults, patrol AIs ground control but all of them would just be UI scripts, the same as any patrol or assist command except able to be modified by the player.
Uber doesn't even need to design or implement anything save for the ability to use such scripts. I'm quite sure that there are many players who would gleefully design and implement such assistants which suit their play style and most likely would happily share them with the community.

these aren't units, they are commands like any other, it is the ability to use them effectively against your opponents which will determine their usefulness.

This sounds like (or at least very similar in many aspects) to what a lot of people are calling UI Power, so I guess people debate over the semantics or just how involved it should be: Should the patrol order be able to extend beyond the area assigned? or stick to the area order you specified. Should this order should include the capacity for replacing lost units? or just the units you have assigned to the order? that kind of thing. I like the idea of setting a patrol order that includes synergy in unit types (order attack-aircraft + scout plane as a group causes them to work as a group)

I think the construction part is where most people have the issue, as it isn't about units following your orders intelligently (following appropriate strike ranges, or reacting to the information supplied by other units in their group). Instead it is the AI carrying out things for you. Depending on how something was implemented I could be convinced that it may work but most ideas of how it would work have drawn widespread criticism and I agree with a lot of their concerns. For me I think an extension of FA's template system would work quite well.

This is something I can see going terribly wrong if they were able to be too automated. It would bring about many peoples concerns of getting the best scripts online and watching them go at it without being involved enough. I think that such a thing could be a lot of fun for the modding community and I think that the game should be flexible in terms of modding (mods are something the Uber team seem be very supportive of). However I think it is important for balance that people are provided with a set of balanced universal UI tools at their disposal, otherwise you'd see an AI script arms race or something.

Yeah I think this is where people get tripped up. Some people want full-blown AI automation and call it an AI assistant, and others want powerful UI tools and call them AI assistants too. I think it is more of a sliding scale of ideas rather than the camps of yay and nay that most of these threads tend to divulge into when the term AI assistant is thrown into the mix.

That "strategy part of the game"? You mean the one that you have been repeatedly asked to provide examples of and dodged the question every single time? What are these so-called strategies that people can't do now because the act of queuing a few buildings is so mentally taxing that it consumes their entire minds?

Back this up, since you talk about proof later on. Is your attention so saturated that you literally spend every second of a FA map giving orders and not one looking at things? No.

So if it's just a queue that queues up crappy premade army compositions, how does it even save you any time? Oh no, clicking a few times on a factory and then pressing the "repeat" button is so hard and you spend so much time doing it in every match that it's seriously hampering your ability to play the game so much that you'd rather hand over control of your unit composition to a brain-dead AI so you can sit back and think about abstract "strategy".

Can you see the future? I know I can't. If you're deciding what your base is going to look like before you've even seen the randomly generated map you're about to play on or what your opponent is doing, you deserve to lose.

So, how are players supposed to practice the game when they're not playing it? You've been asked this before and ignored it, but ignoring points seems to be your primary debating tactic, so this isn't exactly surprising.

Assigning appropriate forces to counter an enemy is your job. If you're letting the AI decide what an appropriate response to an attack is, you're not playing the game any more. You claim "defined sets of behaviours", but then you bust out ultra-vague examples like this.

And again, you claim well-defined behaviours, and yet here you have the AI picking targets, assigning bombers to targets using the vaguely specified property of "appropriateness" which needs to take into account each target's position, its hitpoints, the position of defenders, the chance of a bomber getting through AA, the expected damage of each bomber and target positioning for maximising splash damage, and positioning forces (presumably relative to some enemy which the AI will be reacting to). This is not a UI script, this is an AI that's doing information processing and decision making. It's playing the game for you.

yup. It simplifies number crunching and gets my forces murderizing more effectively. if murdering the T2 land defences is what i need,I want it done. I want to say take these fifteen bombers kill those things as fast as possible and do it just before Hewey, Dewey and Louie get there with their tanks. Then i want Hewey to move north and clean up the enemy power supply, dewy to head west and hunt their production facilities and louie to hunt down every last engineer it can catch.

As far i'm concerned my job is to win the game according to the rules, using the tools at my disposal to their fullest potential. not to babysit stupid forces who cant manage to swamp a front line without someone holding their hand. If using fifteen different subordinate AI scripts but together by some guy in Sweden allow me to murderize more effectively and easily i will.

An alternate view is that no level of AI would take decisions away from the player. Enabling the AI to do something is a decision. Instead complicated AI is bad because while when a player enables it they don't know the effects of the decision they just made. This goes back to my definition that acceptable AIs are those that can be exactly defined in a relatively short algorithm.

For example auto-kite is simply "Periodically move to the closest point which puts the nearest enemy at max range". I think everyone can understand what they means so when they give a unit an auto-kite command they know what is going on. It needs a few quantifiers such as ignoring air units but that definition is otherwise fine. On the other hand a complex base AI would have a ridiculous number of cases to be usable by good players.

If you are worried about development time and noob-trap-AI then we just need powerful UI scripting and an open environment. Then players can mess around with automation and we will end up with an extended UI of the useful pieces of automation. We're at a point that AI can't competently make the important decisions in RTS so I think it would work fine.

@silverdude You miss the point. You talk about picking targets and assigning responses "appropriately". What's your bomber target appropriateness algorithm? Or your air superiority response appropriateness algorithm? It's all very well to talk about a magic AI that does all that and has no drawbacks, but actually implementing it is another matter. All the good UI suggestions on this forum have been able to at least write pseudo-code that demonstrates how such an improvement would work. All the "AI takeover" side have been able to provide is vague waffle about it somehow reducing micro through unspecified automation mechanics.

I dropped my vote on limited number of assistants because i think it is an interesting idea worth pursuit because this is a game, but just like resource rich and other built in mods in supreme commander i think it should be enable/ disable in game lobbies. so if you want it for a fun game turn it on for standard ranked like games leave it off

An AI assisting the Human would help react better to situations on different planets while you deal with any one.

The only other solution would be 2-3 monitors with 2-3 screens each. Which will limit the people who would play this game to people with 2-3 monitors and 2-3x the computer specs.

I dont agree with that line of thought at all. There is no need for AI's to help with this, because if the help is needed, the game is just doing it wrong. Sure at some point the game's world will be so big that it is simply not controllable by human players anymore, but that should just be the maximum size of maps that normally are played. There is no point in increasing the maps sizes just to let an AI play on the additional planets.
With a good user interface it is definitely possible to manage a whole solarsystem of a few planets and thats just the maximum that makes sense to support. "Support" meaning to plan the gameplay and user interface for. Technically ofc there is no reason not to support as many planets as the server can handle.

There are not many players who actually use 2 screens in SupCom, because it turns out that looking back and forth between them is something that is not more effective than simply moving the game's camera. I myself use 2 screens and I found myself rather alone in discussion about "2 screens or 1?" in a german forums a few years ago. I therefore doubt that more screens equals easier control over more battlefields. The bottleneck is your and mine brain, not the number of screens.

That's just... just... the only thing that came to mind is off :|

This isn't about letting the AI play the game for you, it's about letting small discrete 'ai' control small things like automatic resource acquisition or basic skeleton fire/defense bases. Hell, maybe even army control such as target prioritization so you can send an army to attack a base then go do something else knowing that the army will definitely prioritize that defense structure that just got built. Or the other way around, allowing you to direct a battle while the ai builds that skeleton base.

And putting a soft limit on map size in a game built to begin with 40-person specialty mega-battles and then scale into the future is just... against what their stated goals are? Right? They've said they want it to scale over time, and over time means it will happen. Just look at TA, and it wasn't even built to scale so incredibly much, as far as I know. At some point we'll have 40 AI running in a moons of jupiter map on our personal computers, and I'll bet you you'll want help somehow or quit out of frustration since your attention can't be 20+ places at once. If I was a game designer, I would probably rather it wasn't the 'quit out of frustration' one, which is also probably why they've already included multi-human one-force control.

I believe the reason very few people use multi screens in SupCom is two-fold:
1. Computer power
2. Window control
It was annoying to actually try to do anything with the second screen. If not impossible. It was a glorified minimap (albeit a good one) that used a whole screen. It was not really useful for control, which was what I believe most would have used and wanted it for, and the small minimap was generally better at conveying the same information without switching screens and with a much smaller fps footprint.

PA will supposedly fix some problems by have the same gui controls for every window, meaning your main viewpoint won't have to move to where a window is in order to do anything important there. FPS will always be a problem as rendering all those different views will be hard, but at least they will be more than a security camera control-wise!

-----------------------------------------------------------------
Now, for in general. Others have done this, but I'll do it too.
Of the arguments against it, I've seen two main ones. It appears to be either "it'll make players stupid" or "I'll be losing control".

For the first one, I might put stock in it if it was in any way quantifiably true. Or even important for the majority of players. It has been said pretty often that a good number of players never play online, much less in any competitive capacity. Just look at Steam, only 4.2% of CupCom2 players have even won a ranked match. 0.4% go on to win 25. I have the "win an online match achievement", which is at ~30%, and I don't even remember ever playing online vs a human. SupCom2 was pretty much made for competitive play with its smaller maps and faster-moving gameplay.
For the majority of us who just want to play the game and not learn the ins and outs of the smallest mechanics, I think allowing us to have another tool at our disposal isn't going to affect you very much. And for those who decide they want to go online and beyond, well, go ahead and tell them it sucks for online 1v1 play AFTER you've tried it and NOT before. And please, please, please don't just tell them they suck and are stupid for using it. Nothing makes me quit an online game faster than getting yelled out of a match. One of many reasons I don't care about online-only MOBAs. Thick skin and all that, but I don't play games to get maliciously told I'm stupid and a waste of air.

For the other, don't use it. Seriously. I don't know why that's even being insinuated. It's absurd to think that because the option exists, you have to use it. That's like saying you have to use strat view in SupCom to know what's going on, even though you can just use the minimap. And we all know strat view turned out to be generally loved, right?

A decent argument is that it'll take dev time/money to add it in as a feature of the base game. How much, I don't know, but probably substantial. If the devs don't want to worry about it now or ever that's their call and I'll still be totally fine with it. In this case though I'd like if it was possible to add in with the modding support. I, personally, would love client-side helper-ai scripting. Something the server doesn't have to run other than processing inputs, as the ai could only do things the player can but faster and autonomously.

-------------------
And jurgenvonjurgensen, seriously, stop demeaning the people who do want it. It's one thing not to agree. It's another to call the other person stupid and mock him for things like spelling mistakes. Even if I agreed with your stance, I'd be hard pressed to agree with your actual arguments because I wouldn't want to be associated with your attacks on people.

Yes! Swoon—massive single player vs AI skirmishes. <3

We have already discussed the question of whats a task so small that it can be done by an AI on A LOT of pages. Short conclusion of my standpoint: Letting an AI lay out a base is too much, letting an AI control units in any way that contains decisionmaking is too much. All these things mean basically: The AI plays the game, the player watches. Which is wrong.

There is 2 things wrong with this statement:
1.) I never said the game is supposed to have any kind of limits. All I said was that the focus of developing should be on reasonably sized maps and gametypes. The games balance and user interface should be designed with rather normal settings in mind (1v1, 2v2, etc.) cause those are the settings that most people play. Sure the engine will hopefully be build so well that in 50 years we might be able to play on 1000 planets with 500 players, but it is totally unreasonable to try and design the games balance or user-interface for such settings.
2.) PA is not built to begin with 40-person specialty. In fact the 40 player things seems to me more like a highly experimental idea that Uber wants to try. It is by no means the kind of gametype that is supposed to be standard.

I actually use my 2nd screen as a minimap while I am zoomed in on the main-screen to place buildings and stuff. It is fine for that, but the help is minimal, yes.

Here I really need to write something selfish, arrogant and whatever else negative you might name it -my apologies-: Why care for offline playing people? Why give a damn about them?
Most of them -if not all- play so slow and unrefined -since they dont have real opponents to practice with-, that it is an outright joke to go and actually change the game for them. Changing games so they are fit for these "casual" players is what is breaking so many more complex game-genres. Also PA has been crowd-funded as a game that is aiming for more multiplayer than singleplayer. Sure Skirmish-players are there, but if your only opponent is the game's AI you just dont need any kind of refined user interface anyway, cause the opponent-AI for skirmishes will be so bad that you will be able to beat it blind and left hand only after a while anyway.

Fine, dont learn them, like I said: vs the AI or vs some friends, who have never played the game online themselves you wont need those small mechanics anyway.
No need for any kind of change to the actual game.

I said that to myself when SupCom2 was announced to have a not-streaming eco, a techtree and super big sized units on tiny maps. SupCom2 sucks

Well yeah I would not use it anyway, cause it will be useless OR it will be so terribly good that it breaks all what RTS stand for, breaking PA. Well I really dont think that the last case will happen. Either they just dont try to do this AI-stuff or they will fail at it, creating some useless features.

So this is quite important:

Yes damn extremely substantial. I have never even seen any kind of AI in any RTS that was actually able to do anything decent looking beyond 100000 APM micro in starcraft. Before the flow-field stuff came up even simple pathfinding of hugher groups of units was a task very poorly implemented in many RTS.
I actually tried out sorians AI for SupCom the other day, because somebody wrote that it is so good. tbh -sorry sorian-: It sucks. Sure it beats the normal ai with ease, I let it play a 30 minutes game vs the normal ai. But it is so afar away from any kind of useful AI-Helper that it is a joke the even ask for one. Any kind of base-layout or tactical unitcontrol is so hard to do that it would probably be more suited to be the topic of a scientific paper than an idea for a actual RTS.

Colin I agree with everything you said, you hit it out of the park this time....except for this;

It's not really Fair to Sorian, he didn't create a new AI, merely enhanced and optimized the current AI, yes it's not a great brand new type of RTS AI, but as you said it is better than what FA had stock. Sorian had to work within the system and limits already present for the stock AIs.

Arguably as a better than average player your somewhat biased against Stock AI opponents already because of how easy they are for you.

But still, great points on designing for competitive play, it's one of those things a lot of people don't even think to consider, but it's so true, like in SCII the balance is 100% geared towards 1v1, and the game is very different competitively in a 2v2(or 3v3 or 4v4) setting especially when you consider different races and how they can combine to cover each others weakness.

Mike

Now don’t read this as me defending all the concepts that the AI Assistant supporters have suggested; as I have stated above I have come to have reservations and would like to see some specificity and restraint in the AI camp. But in relation to its feasibility, surely Uber could implement the same logic/response appropriateness algorithm that enemy AI utilise when you are facing them in single player skirmishes? It doesn’t strike me that the AI would have to be completely rewritten for an assistant (how such a thing should be implemented is of course up for debate). Enemy AI has to make decisions based on the resources it has available at the time and what it knows about you, then implement a strategy. Whilst it doesnt hold a candle to a human it doesn’t seem an impossible task to impliment it? Whether such a thing should be implimented and the impacts on gameplay I think is where the debate is at.

I agree with this, if any AI assistant were to be implimented it would need to be transparent for the player (like your auto-kite example). I guess those are considered to be more Unit or group AI than an AI assistant, as AI assistant brings to mind a computer ordering units around as well as you.
I think that the issue with the concept is less about whether an AI could do it (after all we face enemy AI in single player matches and they can perform actions like make bases etc) and more whether such a thing should be implemented. I think your argument goes a long way to explaining why; specifically “they don't know the effects of the decision they just made”.

I agree with the idea of target prioritisation when your units are set to attack, the idea that units when sent into a base will prioritise the destruction of an attacking unit or base defence prior to attacking something like a power plant. Self-preservation first unless the user specifies specifically to attack a radar installation/power plant/anything that doesn’t constitute a present threat to the unit. Again I feel this falls more under smarter units than Assistant AI (wow this really needs to be broken down into more than one topic).
People do seem dead against the idea of AI’s making fire/defence bases – I agree to an extent as it would encourage lazy play. I like the idea of user built base templates for this reason as it requires the players thought, but it lets automation take care of the micro management. I don’t however think it is impossible to set up as some people seem to think. Again, enemy AI in skirmishes can do it. So it doesn’t seem impossible to implement. I just wonder at what it would mean if it were.

This here is an important aspect for me. As I’ve mentioned above, if they are allowing a situation where more than one player can manage an army, would it be such a stretch to allow that to be an AI for those games? We can already play with AI allies in games. It strikes me if there were three friends wanting a 2v2 match that one person could volunteer to go with the AI assistant as their co-op buddy. As Uber will be working on an AI for offline skirmishes it doesn’t see seem like it would take massive amounts of developers time to enable a player to limit what resources the AI had access to and dynamically set what type of AI it was (rush, defence etc).

@ Cola_Colin - I think you have nailed it on the head in relation to what size of match the game should be optimised for - it seems to be something people frequently confuse with what the maximum size of map a game should allow. I dont think people have to worry that the game wont support massive maps. Uber seem committed to having the capacity of hardware be the only limiting factor.

I personally don't see how it would hurt newbies. There will be a skill level where, if anything, without getting to see an AI start you off, you will try this game and immediately put it down as not for you. A lot of people put off RTS because they arent born with an inherent database on how to play them.

I think players watching resource management and build order (and think about it, the AI could even be programmed to build structures but require you to click the completed structure to set its placement) would learn that much of the game if they were starting with no RTS experience.

That could easily give them bread and butter, and make matches with other opponents who are somewhat fresh to RTS more mild-mannered and casual while still being interesting widespread warfare, with 2 AIs building and fighting as the players try to make smaller strikes to capture or destroy certain targets.

While we are discussing this, an assistant AI would have to be more limited than a full working opponent AI sightly anyway. For instance, they shouldn't try to imperialize resources or planets on their own if they are player-assistance. The player should have to make that tactical choice. The player should however, be allowed to order the units to move to the spot and then let the AI set up shop with the units you sent to the resource/planet. (opponent AIs should be intellegent enough to imperialize reasonable targets, like discovered resources if within his economy to fortify it, or planets if within his economy to establish a base on it)

This is an interesting idea, kind of like a tips option. Im reminded of a chess computer I once had that you could get hints from, and it would present you with options to try and the possible repercussions. Obviously a lot simpler than a full blown RTS, but I feel an apt analogy.

I think this will draw a bit of ire, and I do worry about the trend in games of 'dumbing things down", pretty much what most people disliked about supcom2.

Another interesting idea which limits the issues brought up with runaway idiot AI's. The idea that you would have to give it the okay to proceed with a proposed action. The issue I see with it is that pop ups are irritating and unless one could set what the AI notified you of it runs the risk of being like clippy in MS word.

[Edit] Having read your post again I see you make no mention of pop ups. I dont know where I got that from. So are you envisioning this as a "consult AI" option with the option to allow it to do that one action?

For example the AI may then remind the player that there were mass points scouted eariler that the player had forgotten to utilise or some such, and upon saying okay and selecting engineers for it to use it then sends them to that location to build on them?