Just a question as i am about to decide which way to go with my AIs:

Would you prefer a good AI that has a certain style (e.g. a preference for certain units) even if that is a weakness on some maps or would you prefer a "flawless" AI that plays like a "Mr. Mainstream-run-of-the-mill" player?

Character or Strength? And no, both would be ideal, but is normally not possible.

Character, if it's strong in that trait. For example, if I want to practice expanding against an enemy with Air Superiority, an AI that builds a crapload of bombers and attacks my expansions would be really cool.

Character. If I want to fight the 'perfect' AI, I'll fire up a Sorian cheat. I'm not even going to use your AI unless it does something interesting (extreme turtle, t2 rusher, air rusher, anything really), or can outplay a Sorian C (no offense intended ).

If you get a few characters, a random chooser would be fun, I'd rather not always know in advance that I'm going up against an eagle until I scout him or 15 bombers hit my commander .

Gee, thanks for your posts! Yacoub posted in the Sorian AI thread and he wants the same, a themed AI instead of a standard AI. I will focus on this in the next versions of my AIs then. Bean will use much more aircrafts and the Tortuga AI will turn into a really hardcore turtle AI, maybe it will even totally skip building experimentals in favour of T3 artillery and nukes.

Again: Thanks for your honest oppinions, you helped me very much!

It would totally make the experience more fun to have a few 'character' AIs as you call them. Some days one feels like taking the lead and having AIs that support him/her. Other days one might feel like letting a teammate AI take the lead and supporting that teammate AI.

So if we had options like an Air Force AI, a Resource Production focused AI, a Naval AI, a Land war AI, a conventional AI, etc, it would create a variety of ways to team up with AIs and allow the player ways to focus their contribution to the match.

Examples:

One day you feel like being the nuke king and team up with a conventional AI that will competently fight the war just like Sorian but without having to spend resources to develop nukes and experimentals. Instead you would situate yourself behind this teammate AI and focus your resources on nukes and experimentals.

Another time you decide you'd like to focus on the ground war and so you team up with an AirForce AI that takes care of conquering the skies and also assists you by transporting troops you've designated closer to the front lines.

There are a lot of possibilities for teammate or 'character' AIs, just take some time to imagine what you would find helpful in an AI teammate that you truly team up with (and not just play as two separate players with a truce between you).



As noted in the other thread, an example of some of the code changes for the Air Force AI:

-Take away land and sea combat unit production except mobile T1/T2 AA units if applicable (i.e. if the race has useful ones)
-Reduce to a minimum land and sea factories
-Reduce to a minimum land and sea combat units
-Increase air factory focus
-Prioritize support of air factories by engineers
-Increase air unit production
-Develop additional air unit platoons and strategies
-Prioritize code for transporting teammates' units forward into battle
-Develop a 'forward airbase' engineer build script that sets up T2/T3 forward airbases (T2/T3 AA, TMD, refuel/re-arm landing pad) to assist air forces.

If I were doing an Air focused AI it would still have some ground presence, for defensive purposes.

Also, it would have a large sea presence in the form of carrier battlegroups ...

(I'm going to be be checking on scripting to get the AI's to use carriers properly)

i just want an AI that can beat me 1v1 without cheating.

same here pkc.

i think you get an AI to do that it need multiple personalities and super strength

Yeah, for an AI to be a really formidable skirmish opponent in 1v1, it actually needs to have more than one personality so it can adapt to the conditions in play.

That's a fairly steep challenge though, so I'd settle for a few different AI's with their own distinct personalities.

Cool, Sorian's already well along the process to providing that. Then again, if your skill level is very high, what you wish for may never be possible for this game simply because there are so many things a scripted AI cannot account for during battle.

i dont believe a sup com ai will ever beat me 1 v 1 without cheating...

but thankfully there is a magical thing called the internet that allows you to play against people of amazing skill and strength!

i think there should be an AI that copies your every move, wouldent that be interresting, of course you would have to be the same faction (different units move different speeds)

Yay, yes. That would be interesting!

I tried to make my AIs play similar to me, they use the same units and build firebases like i do (you will love the new firebases of my Turtle AI, trust me), but of course they are not as focused and versatile as a human player. But it still is amazing to see an attack done by the AI similar to your own attacks.

--

Giskard was once experimenting with some of this 'Air AI' idea and couldn't get it to work.

He eventually came to the conclusion that to make it work in SupCom, it would have to be a Rebalanced AI. In other words, it would have to get production bonuses for Air and put out a lot more Air power than what any of the other AI's get, in order for it to stay in the game. The Air units would also probably have to be given more hitpoints or make them do more damage more quickly in order to truly have an Air Force in the current SupCom.

But, if you can rebalance a single AI and just give that AI the Air perks while leaving the rest of the AI balanced as it is, then Giskard was thinking that a true Air AI could indeed be created and it could actually be made to win against the other AI's and even against a human, but it would take actual rebalancing of that particular AI for that AI only, in order to make a SupCom Air AI possible.

--

Moe also speculated that it would take rebalancing work and/or additional AI work in order to keep a conventional (Nuke-less and T4-less) AI competetive against other AI's. Sorian seemed to think that his AI would scale up and just compete if the nukes and T4 were removed. I think it depends upon the map.

Sorian has even said that he will make conventional versions of his AI's for us to play with in 2.5, if he is reminded, which we like to do once in awhile.

--

But, a rebalanced AI that gets production perks for air or for sea or for some such might be another 'character' that some would like to see, and might be the only way to get something like and Air Force out of SupCom.

im for an AI with strong character but is still fairly strong. Also, say, if you manage to make, say an air ai and a ground ai, that they would compliment each other, covering each other's weaknesses.

It's inevidable that 'unbeatable' AIs will be with us soon, so strength should not be the only trait an AI has (except for AI vs. AI competitions). Even experts like Unconquerable will eventually fall to an AI due to several factors:

* AIs can precisely monitor econ and spending at all times
* AIs can precisely monitor and control ALL units at ALL times
* AIs can react instantaneously to any event, no matter how minor
* AIs are capable of perfect timing and coordination
* AIs never get tired or lose concentration

We humans gloat about our creativity and adaptability but creativity isn't much use against an opponent who is never 'surprised' in a game with fixed rules and limits. Also attacking an AIs weakness is less effective if the AI itself is adaptable or simply performs 'random' actions occasionally to disguise its true behaviour.

The problem simply boils down to time. At standard game speeds human thinking and coordination are simply no match for what an AI is potentially capable of. Even if our brains could keep up we just can't click that fast. As processor speeds grow this gap will only increase.

The only thing keeping us competitive at the moment is that most AIs tend to make gross mistakes or follow obvious or rigid patterns. As these mistakes are resolved we will have to rely more and more on AI personalities and deliberate nerfs to keep them entertaining.

So yeah, go with personality.

Yes, pre-3251, Sorian was actually getting his AI's so that I couldn't beat them, unless I could get them on the right map and predict or know what they are going to do.

After 3251, the AI's were broke on maps like Four Corners and a few other maps, and I had no problem beating them.

I don't know if Sorian ever figured out the Four Corners thing and who broke that map for the AI or if he figured out a way to fix it, but sometimes it's also a map thing.

It doesn't take much nowadays for the AI or a computer to surpass my reflexes, but when I can beat a 'hard' AI with relative ease, then I know that something is either wrong with the map or wrong with the AI itself.

Most of the time, they get the AI to win simply by making it outproduce you. It doesn't take much AI skill or programming to do that. The impressive thing happens when it beats you out of sheer skill and strategy, which usually still isn't the case most of the time. Most of the time the AI wins because it outnumbers you or has greater fire power or outproduces you or researches faster than you do.

The pre-3251 Sorian AI's were usually outproducing me and also of course getting to T4 before I did. I just usually don't win if the AI outproduces me or gets ahead of me in the tech tree and out-researches me.

But, an AI that just plain beats you out of pure cunning and skill still doesn't seem to come along all that often. Most of the time if there is the appearance of cunning or skill, it's because the AI establishes a better position on a map and sets an ambush there or builds a firebase there, and most of the time, that is luck, and not good programming, although the most recent pre-3251 Sorian AI's were doing a very good clearly programmed and skilled building of forward firebases in the most choices and most central of positions often within range of your base, and in that case, the AI deserved its win, no question!

It was a trick that the pre-3251 Sorian AI's were able to repeat. It would get position on you and build a firebase there, and then you would start to lose, and it happened more than once or was more than simply a piece of luck or a one-trick pony. It could actually do it on more than one map and in more than one game.

considering i have not been beaten by a non-cheating sorian, and i get my arse handed to me online, i seriously doubt an AI will ever compete with someone like unconquerable.

this is easily the hardest game to design an AI for. too many strategic options to "teach" an AI how to play. still, i'm very greatful for sorian et al for their hard work. and love the challenge the C versions provide.

A sufficently well-written AI could out-micro any human player. We're talking about never bottoming out and never wasting a single point of mass. Would the economic advantage be enough to beat some of the best players? Probably not. But that's where a strategic layer comes in.

well, get writing it then mate! :p