Question to devs about game AI

[MakerLinux]

How deep is the AI of the game? Like, are Xenon just preprogrammed to spam certain sectors and try a fixed invasion patter or they can plot some tactics, like distracting the player in one sector while they conquer another sector, or using false flags or devious plans? It would greatly help immersion if Xenon could opt for a little more creativity...

And the factions too. In my gameplay, I am being the double-sided agent, I sell ships and weapons to all factions, I do mission for HOP, PAR and the Argon with no consequences. Shouldn't they he smarter? In their war, are they capable of more convoluted decisions and tactics, like what I asked about Xenon?

[MakerLinux]

No response? Is my question that uninteresting? I think this ithe answer would be quite valuable for the gameplay, to the point that this information should be on the manual.

Should I assume then that the AI is barebones and just follows a predetermined script? If so, is there an opportunity for modders that want to make a smarted and more creative AI -- I mean, are there API bindings for improving the Xenon/factions behavior, initiatives and reactions?

[Tomonor]

Chillax, brolax.

The AI behind the factions do follow a pattern to some extent (especially since race relations are locked), but there are different stages of the wars (with nifty details such as reconing/scouting, deploying forces, battle, armistice, etc) and if a faction gets lucky they can take over control of the sector indeed.

That being said, there already two mods posted in this forum that tempers with the war logic and and its AI.

As for Player induced consequences and dynamic diplomacy, nope, that never was part of the X games thus far. I can see why it would be mandatory to have in X4, though. But you could also think of the Player's faction like a Corporation that resides between factions and battles - supplying them without consequences, because that's how shady business is like.

[Fenris321]

Seriously? Give them some time (more than 9 hours). They have more jobs to do than just reading the forums (if that's even in the job description). Egosoft devs are among the best if not the best at responding to forum posts that I've seen, and I'm on a ton of gaming forums. Just give it some time before breaking out the attitude.

[Imperial Good]

The AI is pretty much like any normal RTS AI. It has no deep learning functionality and no neural networks. Rather it uses a set of hard coded rules and responses.

[MakerLinux]

Seriously? Give them some time (more than 9 hours). They have more jobs to do than just reading the forums (if that's even in the job description). Egosoft devs are among the best if not the best at responding to forum posts that I've seen, and I'm on a ton of gaming forums. Just give it some time before breaking out the attitude.

You're right that I was too anxious to get a response, but there was no ill attitude towards the developers. I re-asked just because I saw a couple of developer responses in other threads in the meantime.

[Alan Phipps]

Asking questions of the devs after lunchtime on a Friday is never a good idea either. (Well, not unless you include pizza bribes.)

[Ingr4m]

When it comes to Xenon behaviour, it seems there is not much strategy going on. At least in my game, though I'm only 30 hours in and haven't participated in any war event. All I've seen is that they do build defense stations here and there while the owners don't care. In some sectors they spawn periodically to ensure a permanent thread, other than that it's mostly random spawn and attack behaviour.

[CaptainX4]

How deep is the AI of the game? Like, are Xenon just preprogrammed to spam certain sectors and try a fixed invasion patter or they can plot some tactics, like distracting the player in one sector while they conquer another sector, or using false flags or devious plans? It would greatly help immersion if Xenon could opt for a little more creativity...

And the factions too. In my gameplay, I am being the double-sided agent, I sell ships and weapons to all factions, I do mission for HOP, PAR and the Argon with no consequences. Shouldn't they he smarter? In their war, are they capable of more convoluted decisions and tactics, like what I asked about Xenon?

I think you have hit the big red button. AI is not and was never a strong side of egosoft so dont expect miracles. the current war efforts are still scripted on the x3ap level or maybe even worse, pilots are like mad cows and im nearly sure these things are here to stay just let it go, get in a ship and keep flying around because thats is the only thing atm this game is good at

[Falcrack]

Asking questions of the devs after lunchtime on a Friday is never a good idea either. (Well, not unless you include pizza bribes.)

Asking devs anything, at any time, is always hit or miss, usually miss!

[MakerLinux]

When it comes to Xenon behaviour, it seems there is not much strategy going on. At least in my game, though I'm only 30 hours in and haven't participated in any war event. All I've seen is that they do build defense stations here and there while the owners don't care. In some sectors they spawn periodically to ensure a permanent thread, other than that it's mostly random spawn and attack behaviour.

I know it's an illusion - the illusion of the presence of a reasoning mind - but it is part of what adds depth to a game. That's why I would love to get this added to this game, I think it would fit perfectly, either via egosoft themselves or modders. Even apparently frivolous things like making the enemy mock you, remember you and challenge you (like in Shadow of War with its Nemesis system) helps making you try to apply an instinctive "theory of mind" to guess what the enemy is thinking or planning. Makes the game feel more real, more grounded.

[Socratatus]

I have come to the realisation that, these days, `AI` does very little. they could make the AI better, but often it`s just programmed routines designed to do just enough to make the Player feel like he has an interesting opponent/Ally.

There`s nothing more I`d love than an indepth AI, one that had objectives and goals with little tricks against its opponents- in fact all else could take a step back if AI made massive strides in learning and using tactics...

I used to always ask and even demand great AI for many games, and saw little. I gave up when I read several Devs say it wasn`t worth the `effort`. Which basically meant not enough people cared to see a good AI as long as they could get flashy graphics and play Online. We are the few who`d like a great AI.

I know good AI is hardwork, but that`s why it`d be so great to have it.

[deadmoomin]

I think AI is like the frontier of gaming. Major developments in audio technology ended many years ago. Most people barely perceive a difference between 16, 32, and 64 bit audio. Audio-tech such as Binorial, Surround Sound, "environmental audio textures", etc are all cherries on the cake but nothing majorly impactful to your audio experience.

Similar thing with graphics, the jump from 2D to 3D in gaming was probably the last major, revolutionary, milestone. Since then it has been a simple matter of increased 3D definition, more and more objects, increased resolution, nicer physics etc. But the perceptible increase in "improvements" to visual graphics are also slowing down. Many games these days have "great" graphics. Good graphics today are an expected norm.

But AI, AI is the new frontier. It has been ignored (relative to the other areas) because gamers have been more easily impressed by audio and graphics. Besides that, the average gamer doesn't even really understand what "good AI" is and, if he has a sense of it, he's unable to verbalize it enough to write forum posts and complain about the awful standard of "Path Finding", "Tactical and Strategic AI". The average gamer doesn't even appreciate that AI is capable of a lot more than is currently found in most games.

The vast majority of games are very lazy with AI. They prefer to focus on Audio and GFX design because AI is hard. Good AI requires a lot of unique skills, not just technical, but also psychological understanding of humans, historical, and strategic/tactical (for combat and resource games) understanding. It requires a lot of knowledge which most game companies do not have - because they themselves do not value.

It's sad, our AI in games is stuck in late 1990s early 2000 tech levels. Yet good AI in games (particularly single player games) has the most potential to deliver an outstanding, deep and engaging gaming experience.

[Socratatus]

Good AI requires a lot of unique skills, not just technical, but also psychological understanding of humans, historical, and strategic/tactical (for combat and resource games) understanding. It requires a lot of knowledge which most game companies do not have - because they themselves do not value.

You touch it with a needle.

And yes, AI progress is still stuck in the late 1990s.

[tero]

Imroving the Ai is also the most difficult one. For gfx and audio the hard work is done by the soundcard and gfxcard. But for the AI the hardware won't help you.
All work has to be done by the CPU, That also means that the AI has a bit of impact on the performance. It's often better to simplify the AI in favor for good FPS. Especialy when a lot of entities needs to be controlled by the AI.
The X games are a good example for that because a single task may require just little CPU time but when you multiply that with the number of entities in the whole game it becomes quite noticable.
Other games have it a bit easier. I.E. when only a dozen enemies near the player needs to be controlled and everything else lay in wait (ie FPS shooter). These kind of games can implement a far better AI without a big performance loss.
There are some tricks you can use to save CPU time, ie X simplifies the AI for oos. Which is not always a good thing.
With powerful multicore CPUs this has become less an issue but not every player has one of the bigger ones. So devs can't fullly use the potential yet.

Just my 2 cents

[Techedge]

Simplifying (trying not to oversimplify)

First of all you need to understand that A.I. in games is just a set of predefined algorithms that has little to do with "real world" A.I. In this post I'll use A.I. with it's proper sense.

Purposed/narrow A.I. has been real since many year but has been used with specific hardware on specific matters (science, medicine… or chess and recently go ). However programming such an A.I. takes a considerable amount of time, specific skills and dedicated hardware.

This said, going past a (predefined) set of algorithms has not been an option for game devs, because it was just inconceivable to devote such amount of time (and skills, and hardware) for just ONE game (because, you know, Deep Blue can't play go, as AlphaGo can't play chess). This could be (is?) the reason because we haven't seen any progress since 1990s: devs just stopped improving their predefined set of algorithm when they thought it was an acceptable balance between their time and the challenge offered.

The "recent" advent of machine/deep learning, however, could be game changing (pun intended), because they could boost the development of gaming A.I. if some (big?) software houses were to invest in it. Why big houses? Because I see programming a "real" A.I. for just one game not so profitable in terms of devs time (let alone specific skills), at the moment. Maybe in the near future the creation of specific frameworks will reduce this time, thus permitting little software houses to access A.I. creation for their own games without the need of specific skills in the field of A.I.

Thus given, computational power shouldn't be a problem, because you don't really need A.I. to run on your hardware and respond in real time to your input. You just need A.I. to learn from player inputs from all over the games played in the world and produce responses to those inputs.
Then servers could update the game predefined algorithms (asynchronously, but think this as a continuous stream of small updates).
This makes sense if you think of the two cases:
A.I. has already processed a similar case (from you or another gamer), than a proper algorithm has been already submitted to your game to respond to your input;
A.I. hasn't processed a similar case yet, than your strategy is winning, but the new situation is then submitted to the A.I. which processes the input and elaborates a proper response algorithm: next time your strategy will be countered (and the same will happen to other gamers).

This could also operate in real time via cloud computing, but in my opinion this would not be the best solution (taking into account costs and benefits).

TL;DR
In my opinion game devs (not only Egosoft) won't use "real world" A.I. until a groundbreaking way of using it will come up (if ever, but I'm confident it will). So don't expect many improvements on the "game" A.I. in the near future.

[Damocles_]

The scientific and commercial approaches to AI (namely deep learning algorithms / pattern matching) are not really applicable to game ai, thats the main reason why its not used.

Those algorithms solve very specific problems that would not help making an artificial opponent appear smarter or more lively in a game. (outside of games like chess and go).

In terms of the ai-toolbox, games need more things like pathsearching, behavior trees, goal planners, (and especially good tools for developers.)
Not something where neural nets would be very useful.

[Techedge]

The scientific and commercial approaches to AI (namely deep learning algorithms / pattern matching) are not really applicable to game ai, thats the main reason why its not used.

Those algorithms solve very specific problems that would not help making an artificial opponent appear smarter or more lively in a game. (outside of games like chess and go).

In terms of the ai-toolbox, games need more things like pathsearching, behavior trees, goal planners, (and especially good tools for developers.)
Not something where neural nets would be very useful.

Yes, that was the reason I pointed out the difference between "A.I. in the proper sense" and "A.I. as a set of pre-programmed algorithms" in games. And for the same reasons I don't see any near future improvement in that field, except with a groundbreaking new ways to use machine learning. My supposed way of using it was just a mere speculation to underline how computational power is not the actual limit in the improvement of A.I. in games, while the time (and clearly skills) of the programmers is.