The holy grail for agressive AI code

I don't want to sound too harsh, but could you please give some useful proposals as to how to improve the AI rather than just bashing? I think it is quite obvious that the AI is by no means competitive but nagging does not improve anything.

My suggestion: if you know someone capable of AI programming try to convince him/her to pick up the challenge. You could even dive into the topic yourself...

Another point to consider: not everyone likes to fend of a endless stream of enemy ships from the beginning on, so what is considered a "challenging" AI is also subject to personal preference.

regards
Arne

PS: no offence intended

I agree with Dizzy.

You could try AI team mode with the current version (v1.14) of the Balance Mod. Medium galaxy, medium number AIs, no neutrals, low AI bonus, 500,000 starting tech points and 2000 racial points would be a decent setup.

Space Empires Depot | SE:V Balance Mod

Since they won't make a treaty with you and they won't be bogged down by conflicts with other AI empires, most of them will be at war with you shortly and since they'll likely be stronger via their treaties with each other, they should be more aggressive in attacking.

Space Empires Depot | SE:V Balance Mod

I enjoy a good scrap against the AI.
I'm playing Kwoks Balance Mod 1.14/1.66
Medium AI, Team Mode, High AI bonus, hard difficulty and I have just managed to fend off 3 fleets of 40+ ships from 3 different races.

The AI is fine, not everyone wants to be mired in battle all the time.
___________________

@Dizzy: please calm down mate. What I wanted to say is that constantly complaining about a problem seldom contributes to solving the latter.

I am no *real* programmer either (studying avionics) but the one thing I understood well is that AI programming is a huge undertaking that is easily underestimated by non-professionals. Just have a look at other computer games. Although graphics are constantly improving the complaints about the AI are the same since 15 years: too predictable, only completive when cheating, ...

After all please consider that it is quite obvious that MM isn't a company like Ubisoft or EA Games - it is one sole programmer who brought the whole thing up, not some hundred.

come on, keep smiling

regards
Arne

AI is indeed an massive, difficult undertaking. But the problem isn't simply making an AI capable of fighting. It's making an AI that is fun to play against. Challenging is much easier than fun.

For example, the AI can micromanage every single ship, unit, and planet without becoming bored or distracted. By this trait, it can win by piling up sheer numbers and pushing endlessly until the human collapses. But is that fun?

I believe that the type of AI most fun to play against isn't one designed to always win, but one designed to be adaptive. It's no fun having an enormous fleet that can't be touched pounding away at your empire till the end of time. It is, however, fun if the AI can adapt to your strategies and counter them, meaning you always have to keep improving and changing. To me, that would be fun, regardless of winning or losing.

However, adaptivity is among the most difficult of things to achieve in programming, especially in the context of an AI.

-----
A project a day keeps the deadline away.

Aggressive AI? You seek to have the satisfaction of being utterly demolished?

Team Mode game, Intel allowed, High AI bonus (or better yet, customise just how much "High AI bonus" is - it's 2 lines in Settings.txt), No Neutral Empires.

Naturally, go for High Difficulty, medium Paradise galaxy, medium number of empires. While you're at it, make it a simultaneous turn game too.

Now... making a fun AI... that's an entirely different question. But if you just want to have a masochistic experience, then jack up the AI bonus alongside the above settings.

---Sig---
Playing Touhou games (Go to this Wiki if interested. No, nobody else is that good/insane as that replay). No rush for SE

Dizzy ever consider multiplayer? No matter how good an AI is a Human opponent will always be better, more annoying, and genocidal.

The problem isn't making the AI hard; that's easily done by turning all the game settings against the human player, as others have mentioned here. I think what's desired is an AI that *actually attacks*. In all my games I'm still not sure if I've suffered a single attack of any significant force on my homeworld system from an AI player. They seem simply unable to cope with travelling through several systems with a fleet for an attack. The only way this seems possible is if you're either in the same system as one of their colonies or in a system next to one with an AI colony.

This really wasn't anywhere near as big a problem in SEIV, unless my memory has failed me.

Anyway, what this means, in effect, is that the player is given just about all the time in the world to research, build up and secure his defenses, before launching an attack of his own with better designed ships.

I studie computer science, and I agree with all said above

to say it bluntly and short AI programming is difficult.

Game AIs always cheated by scripted events ( missions ) extreme boni and scripted build and action orders. Most RTS and other games make it easy on the AI with small numbers of unit types, research options and clear mission targets, any turn based game got the advantage of having time to calculate it's result but doesn't work any different then most rts games, at least in my oppinion.

any other kind of game really don't need that kind of sophisticated ai to be challenging, more than that, if shooting or other kinds of none rpg battle ( speaking rule based ) the developer has to cheat in favor of the player who doesn't want to play against aim-bots.

chess computers if you had something like those mind have a very limited set in possible moves, every piece is predefined and they still cheat with the brute force of calculational power .. they just calculate as many moves ahead as possible and chose the move which will most likely lead to Not Losing ... note that Not Losing .. not winning.
They are by far not creative.

Other games I often hear in comparison are GalCiv2 and Civ IV ( which I personally don't like ) I never played GalCiv long enough to notice the quality of the AI, but I found the Techtree and planetary development to be not very satisfying and generic .. in other words very AI friendly ( I don't even know how combats are fought in gal civ so I can't comment on those )

Civ IV is just number crunching like a chess computer, the battles are random dicethrows with modified values, the techtree is one straight line, city developing and colonisation is plain easy and the AI is regularly outteched by any logical thinking being ( I know to expert Civ players so I learned from watching ) and turn times are really kept slow ( still the game had bigger tech requirements than SE:V to begin with, the first few patches where performance only )

one of the most important things you need for good ideas are ingenious programming and time
the two ressources that get less and less valued in modern software engineering .. you see its now called engineering and developing .. no more inventing and innovation.
games became like cars. or like movies with the acompanied merchandise hype.

but enough bashing maybe there are a few solutions

first things first .. the build order technique is quite a good one to get the ai up and running. equip the ai with a few good build orders that are executed and it should plan its empire well.

then teach the ai the research rush, which should make it more aggressive in a way, that it actually gets its hand on a few good weapons.
teach it choke-point defense and include a small route planning algorythm. this should really not be that hard .. the route planing for ships is quite good ( anybody noticed or took that for granted ? .. played a rts lately )
combine it with the systems to avoid list ( you could know which systems .. or at least warppoints you are crossing .. ) so there is no more senseless wasting in single ships
and add the victory conditions to the mix so the ai knows what to go for .. or oppose.

another hard part would be to teach the ai to counter human design .. there are two ways,
use the build in rock-paper-scissor design ( rockets-pd-guns ) or the good old copycat ( if my enemy builds fighters .. i copy his design and try to build more .. i am an AI i can manage better than him

the hardest part would be .. teach the ai good management

but in my oppionon aaron is doing a damn good job, he published a massivly complex game as an indipendent minor games manufacturer .. which is compared with million dollar A-Class games.

This is like comparing b-movies with hollywood productions
a big budget alone doesn't buy a quality movie, or game .. ( we wouldn't be here if it would ) but some things need money and time

Remember : Pillage first! then burn.
- Cpt. Kaff Tagon

I remember something about SEIV that made it a challenge. If you held onto a wormhole, the enemy would attack it with ever increasing fleet sizes until it broke through, but all the time it would expand in the other direction also.

If they could somehow get the AI to remember how many ships were in a specific sector (i.e on the other side of a wormhole) and then use that information to build a fleet that would combat it, then it would be a big jump for the AI.

Your lord and master (below Foamy) LordHavoc

Largest obstacle to "take this fleet and just beeline towards that big enemy colony over there" approach (which is what people seem to want) is the Supplies logistics. Maybe we should switch on the AI script cheats such as "all vision" and "infinite supplies" so it can do these things. It's down to the AI modders' decision though, whether they want to 'cop out' like that or try and do things clean.

---Sig---
Playing Touhou games (Go to this Wiki if interested. No, nobody else is that good/insane as that replay). No rush for SE

There are several problems. There's the vision and supplies problems, like Psieye mentions. In addition there's a problem where the AI sends his fleet to attack, but then identifies a threat to one of his own colonies. This can lead to the AI aborting his attack, prioritizing defense instead. And shouldn't he? After all, there's little point in laying waste to enemies if you are dying as well. One solution would of course be to split the fleet into several, which is currently something the AI can do. But then you encounter other problems where the AI's fleet is suddenly too small both for defense and attack.

One solution might be to have the AI concentrate much more on warp point defenses? Not sure if it would matter though, since when players attack they do so in force, and if you've got to spread the defense over several warp points then it's unlikely you can stop a concentrated attack.

Anyway, the reason all of this becomes a bigger problem in this game than in many others is because the travelling distances are so long, and so much happens in few turns. Combine this with Fog of War and Finite Supplies and you suddenly have an AI that really has no idea what to do. The AI might send a fleet on a mission, but when it gets there many turns later there's a good chance the whole scenario has changed.

In short, there's just *way* too much for the AI to consider, and having the AI cheat and see everything( i.e. no fog of war for AI empires ) might be the only possible way of making the game even remotely challenging in single-player.

The biggest gripe I have about the AI is when I send them a gift of technology or a planet, they always seem to turn it down. Even mostly harmless things like Storage tech.

Also if I offer them a totally lopsided treaty in the computers favor (IE I share all tech and they share no tech with me), they turn that down as well.

Now, I don't know how MM has organized his code internally but it would seem logical to me to steal some concepts from the object oriented programming world. Simply treat ships as objects. Give them properties and methods appropriate to their needs. I have done some programming in the past both in the old school methodology and in the object oriented philosophy. Treat ships like cockroaches. Cockroaches survive not because they are smart but because they are extremely well suited to their environment. They suck at doing advanced calculus(like most of us)but are experts at finding food and avoiding capture.

Ants have brains the size of grains of sand but yet can find food, follow paths, find their way home and interact with others. I submit that ants are vastly more 'intelligent' than the AI in this game at this time. Sometimes being 'intelligent' means being very good at a limited range of activities. This strategy seems to work OK for ants and cockroaches. View ships as being individuals within a larger community. If it works for ants..........

It's not that simple. At all. You can't just go "object oriented." It's got to be planned out from the very beginning otherwise it will be faster just to rewrite it ALL.

A cockroach, written object orientedly, is *much* more advanced than your average chess master program. Its world is way more complex. It need to see things, differentiate patterns & movement. It need to analyze what it see and what it hears and see if there is danger. It's a lot more than just "scatter if you see light." We're not programming a simple flocking algorithm. The problem you need to object orient the whole world, as well as the diplomacy, resource management, etc, etc ,etc.

OO works well in a Real Time setting, such as the tactical combat. Flocking will work, and individual ship behavior when broken formation, on when and how a ship should accel/decel to get into the range as specified by its set strategy. But that would be about it.

Now if you're talking about it would be easier to manage/change/update the code base, that's a different debate. But developing AI using OO is counter productive in a turn based game, simultaneous or not. Using a "fuzzy" expert system is much much simpler in comparison. However the variables are so numberous. Just to figure out how large a fleet should be sized is hard enough, where the fleet should gather, even cheating (IE, if the AI can see all the data, with or without sensors).

True to the core Hexxon,
just google for robot-soccer .. the things those little shoe-boxes can do are really amazing ..
and a good example for what Hexxon said .. it's way to much information for a game to cope with.

But OOP would still be a good thing for the number crunching part of the AI.
( but aaron is using OOP .. I am quite sure of it, with the modding capabilities and modularity of SE, I don't think it could be done another way .. he just choose to put his data storage in plain text )

expert systems would be it to rework the game ai ( hmm.. i wonder what you could call the ministers .. )

Remember : Pillage! then burn.
- Cpt. Kaff Tagon

You're on about artificial life and swarm intelligence. Not sure how you would consider applying this to the game however, there are just too many variables at the moment for this approach to work as far as I'm aware. So by all means, go ahead, try that is the whole aim of that area of research but I don't think that the current limitations found on this area will allow that approach to be successfully applied.

The one the good Captain allready mentioned.
Free the AI from chewing up itself and it will come for you.

and maybe you should have noticed by now .. that this single piece you want doesn't exist. Just think of the steps you make when you are waging war against an opponent in SE:V and write them down on a sheet of paper in a form that the stupidest, thickest and slowest person you can think of, whose only ability is to be insanly good at math, can follow to be a worthy opponent in your eyes.

when you done that you are on the foot of the mountain that is AI programming

Remember : Pillage! then burn.
- Cpt. Kaff Tagon

Just because you program something in C++ or whatever popular OO language, pop all your datasets into classes, doesn't really make things object oriented. Been trying to teach my juniors for years - just because things are in classes doesn't mean they are properly constructed as objects. Its the properly thought out set of methods (more importantly interfaces) will truly make them oo. But again I'd like to emphasize, that a lot of projects are easier if they are not built as true objects. Using an OO Language to help out with encapsulation, memory management. But once you start doing polymorphisms you really have to plan well. Looking at the way how "strategy" section work, you can sort of see tell-tale signs of a polymorphism exorcise gone wrong - where the setting makes no sense for fleet, taskforce, and ship defaults. This is not any sort of bash on MM, I'm still shocked and amazed that one person did all this and time to even wrote a rudamentry AI.

Regardless of how an AI is put together, it needs a large, strong set of helper functions in order to work well. Have you ever use "Explore" and "survey"? That kind of stuff will go a long way and a much more tackle-able goal. Get the moders a bunch of good, working helper functions and let them figure out how to massage the AI. To make them seem more... adaptable is more key to fun than anything.

For now, we play multiplayer and use "AI" as speedbumps to your real opponents.

There is no 'Holy Grail'. What there needs to be are a few extra AI functions so that the modders can finish what they've started writing. That or a willingness to abandon the "no AI cheats" stance and perhaps, decide to have some suicidal AI who don't care they'll die so long as they get an attack in (which is actually a viable solution given an impossibly high AI Bonus). Either way, it is not 'a single piece of code'.

---Sig---
Playing Touhou games (Go to this Wiki if interested. No, nobody else is that good/insane as that replay). No rush for SE

It is not a problem to make the AI attack, it's easy to do. But is it good enough just to attack without consideration of the target's strength? Should the AI just attack and ignore it's defensive obligations? Remember, any AI empire usually has more than just the human player to worry about.

The Balance Mod AI is designed to be become more aggressive when it knows it's ahead of the other races in score. Good human players can almost always get ahead assuming an equivalent start, so most of the AI players it will encounter will react more defensively. Of course, it's not completely static and if there's an opportunity to attack, an AIs will consider if it looks favorable.

One game option could be to provide 5,000,000 starting tech points, while taking fewer yourself. I've found the AI can become quite tenacious under these circumstances. One downside is that the AIs also tend to attack each other a lot, which can be a drag for human players in terms of their action and turn times. This is why I usually suggest team mode to players looking for more defensive action because the AIs are freed up of their other strategic obligations and can focus their forces against the human player, so you see more sustained attacks because the AI can direct most of its forces at you. You don't necessarily need to use any bonus for the AI in this case, because their partnership treaties with the other AIs fill this void.

Space Empires Depot | SE:V Balance Mod

I looked at the AI code in this game (thankfully it's possible to edit it). AI is kinda crazy, it doesn't try to min/max anything at all. Of course, that way it sucks even economy-wise. And it's hard to make a good one given available tools. I mean, that interpreter doesn't support nested lists (or multi-dimensional arrays). Why game creator didn't use something like Python or Lua? Why some weird custom script engine? Lua should be much faster than any home-made script language. C++ is even better, like in Civilization 4.

I have an idea to help the AI with the supplies problem. But I really don't know how it is going to work and if there will be any other problems.

I have just started a game and I have an additional tech area as a racial trait, that I'm using to test new components I created. I think I will add a copy of the supply/ordnance components in this tech area and increase the amount of supplies/ordnance in the standard components. I will use only the copied components and nothing will change for me, but the AI will have lots of supplies and ordnance.

Do you think it would help the AI?

Ouch. I think I'll give it a shot.