Distant Worlds AI

[Icemania]

Test 2: Improved AI
1. Medium Space Port Design changed to have 8 Weapons, 2 Energy, 8 High Tech (the AI builds 2 Energy Research Stations for another 16 Energy Labs) so 100% research potential is used and more Weapons Research
2. Weapons Research Order Improved (had to do it manually as still can't find a way to change anything here)
3. Ship Design Template changed to focused on One Weapon Type
4. Ship Design Template optimised for Size 500 (in this specific test case not a huge change as the Cruiser in Test 1 was Size 465)
5. And Titan Beams are 69% complete!

Here is the resulting Ship Design:

[Icemania]

The patch for 1.9.5.3 is out and quite rightly it has more of a bug fix focus. Time for more AI improvements I hope.

Speak up if you want to support or propose practical AI changes as Matrix are paying attention. Or stay silent, in which case the chances of the changes you want are less.

[Kayoz]

ORIGINAL: Icemania

The patch for 1.9.5.3 is out and quite rightly it has more of a bug fix focus. Time for more AI improvements I hope.

Speak up if you want to support or propose practical AI changes as Matrix are paying attention. Or stay silent, in which case the chances of the changes you want are less.

We've had this cry for better AI being screamed at Matrix since release day. What makes you think now is any different from all the times we've asked/begged/screamed for it in the past?

[Icemania]

Because some of the changes mentioned in this thread have already been implemented.

I don't expect a significant improvement until DW2. But some further incremental changes would be welcome.

The window of opportunity is when we have patches.

Do you have any constructive suggestions?

[BigWolfChris]

Definitely wouldn't mind the AI opened up to being moddable tbf

[Tcby]

ORIGINAL: Icemania

Because some of the changes mentioned in this thread have already been implemented.

I don't expect a significant improvement until DW2. But some further incremental changes would be welcome.

The window of opportunity is when we have patches.

Do you have any constructive suggestions?

QFT

[MartialDoctor]

The AI has been greatly improved from previous versions. It still needs work but I am quite happy with the recent work done.

[Icemania]

ORIGINAL: MartialDoctor
The AI has been greatly improved from previous versions. It still needs work but I am quite happy with the recent work done.

Shh! If you say things like that ... they might not made other AI improvements! [;)]

[BlueTemplar]

What it comes down to is that the human player learns.

The A.I. doesn't.

I think that this is fundamental.

Between 3 of these :
- Good AI
- Scripted AI
- Complex game
You can pick only two.

It's a shame that Codeforce decided to go with a scripted AI (if I'm not mistaken) when the scope of this game would go very well with an AI that actually learns.

These AI do exist, by the way, but they probably require a completely different design.
Examples I can think of :
- Galactic Civilizations 2 AI (as already mentioned).
- Sword of the Stars 1, if I am not mistaken, uses some kind of learning algorithm for the AI ship designs. The problem is that while it seems to learn and adapt to enemy designs, it also seems to promptly forget everything when you start a new game, making it not better than a newbie player in that aspect. There's also the issue that ship and weapon balance is quite different in full real-time 3D tactical combat (that the player will usually use) compared to very simplified "1D" autoresolved combat (that the AI's will use against each other).
- http://arstechnica.com/gaming/2011/01/s ... mpetition/

[Tehlongone]

It's not really a choice between good AI and scripted AI, AI will never really learn anything with the limits to our computer systems. If anyone were to even try that it would need to be a much more limited setting, AI thinking uses CPU cycles and the more factors you want it to consider the more demanding it'll be.

What is good AI anyway? It seems to be a term covering all positive things.

Galactic Civilization 2 AI does not learn anything. It partially adapts to your strategy based on a somewhat limited amount of options, it's decent but not all that bright. I remember beating it consistently on the highest setting where it has outrageous bonuses.

I haven't played SotS1 but I haven't heard anything about it having exceptional AI. If the ship designs work well then Distant Worlds might benefit from something similar, but it wouldn't really be AI.

Some things it could learn from player behavior but it's not a simple matter by any means.

Distant Worlds AI does have some quirks that appear easily fixed, however. I'm not sure how easy it is to modify it though, it just seems some situations could be fixed with simple measures.

[BlueTemplar]

It doesn't matter if the AI learns slowly if it doesn't forget between games. It's even less of a factor if that learning is pooled on a worldwide basis. I thought GalCiv2 AI actually did that?

As of what a "good" AI is, I have a (perhaps unconventional) viewpoint that it should be compared to the performance of a real player. A good AI is an AI as strong as a good player, when playing on a level ground. So if you're a good player and playing a game with 20 good AI's, and neither you, nor the AI's have any kind of bonuses nor handicaps, then on average you should lose 20 times out of 21. (Here you can guess why developers usually don't bother to even try to make good AI's)

[Shark7]

Even the researchers who spend millions of dollars to figure out how to make a true learning AI haven't made a lot of progress yet. In the end, the AI is nothing but code, and it is limited by the code. Sure they can 'learn' to do simple repetitive tasks (manufacturing robots), but making an AI that can reason as complexly as a human brain is still a ways off I think.

[Tehlongone]

To some degree perhaps, it made little difference however, and would not be feasible in a game like Distant Worlds. It never learned anything that I noticed about my strategy or at least it had no clue how to beat it. I think it was a good AI (with issues) but the game also seemed a bit simpler.

With your definition of a good AI then I can say with great certainty that no 4X game has ever had a good AI. The most you can hope for is that it makes no clear mistakes and uses a sensible strategy. Distant Worlds is actually not that bad in this regard, though it has issues.

[BlueTemplar]

Well, they've managed to get a pretty good AI in the Starcraft AI competition, didn't they?
Of course it helps when the game is real-time and requires fast reflexes, since the AI can be much faster than the player can ever hope to be and can micromanage every unit at the same time...

[Shark7]

ORIGINAL: BlueTemplar

Well, they've managed to get a pretty good AI in the Starcraft AI competition, didn't they?
Of course it helps when the game is real-time and requires fast reflexes, since the AI can be much faster than the player can ever hope to be and can micromanage every unit at the same time...

That sounds like it is less of a 'good' AI, and more of a human limitation.

The DW AI doesn't cheat, gets no bonuses, and follows the same rules as the player. It's one of the few that doesn't cheat which might make it seem weak, when in truth it is the fact that your aren't playing against an AI that has clear advantages to make up for its weaknesses. And while it still has the advantage of being able to manage all its units all the time, any player that leaves automation on gets the exact same abilities with the AI helper.

[Cauldyth]

Let's also remember that many strategy games have a dedicated AI programmer, at least for part of their dev cycle. Now let's remember that DW has a programmer, period. With that in mind, it's remarkable it has an AI at all! [;)]

[Tehlongone]

Well Starcraft isn't 4X, it's a pretty simple game and like you said much of the difficulties in that game comes from doing everything fast enough. If the game proceded at snail pace in each fight you'd have time to outsmart the AI better.

Anyway I look forward to the day AIs get smart enough to outwit an average human in a halfway complex game. It'll probably be a few decades, though..

[ParagonExile]

ORIGINAL: Tehlongone

Well Starcraft isn't 4X, it's a pretty simple game and like you said much of the difficulties in that game comes from doing everything fast enough. If the game proceded at snail pace in each fight you'd have time to outsmart the AI better.

Anyway I look forward to the day AIs get smart enough to outwit an average human in a halfway complex game. It'll probably be a few decades, though..

Galactic Civilizations 3 is introducing AI's which are capable of adapting human ships designs and strategies to its current situation, and it can apparently search Steam's network for more. If it turns out well, I hope it becomes the industry standard.

[Icemania]

I posted some links on Page 1 regarding what Galactic Civilisations 3 has in mind. While I prefer the design of Distant Worlds, if they only achieve a fraction of their vision, I'll be spending more than a little bit of time playing it ... and hopefully getting slaughtered sometimes due to AI strategies from other players that are superior to my own. I really hope this is in the base design for Distant Worlds 2!!!

[BlueTemplar]

GalCiv3 is very promising indeed :
- The strongest 4X AI ever made (probably).
- At least some form of tactical combat (won't probably be nearly deep enough to my liking, but anything deeper than GalCiv2 is already an improvement).
- A powerful ship designer, but also one where you don't need to spend way too much time on purely cosmetic aspects :
http://www.twitch.tv/stardock/b/526718527

Now, that is still not enough for me to shell 93€ on the Founder's Elite Edition, but I will be following the development of the game with attention.