The “lost interest in continuing playing, then halfway trough the game” phenomena

[Ralf]

About halfway trough playing SMAC (this, perhaps, goes for upcoming CIV3 and CTP2 as well) i take a quick look at how my computer-controlled competitors have developed their cities and city-areas. I always seem to discover the following SAD state-of-affairs (i´m playing at Thinker-level and upwards, by the way):

1/ I have about TWICE as many city-improvements per city, as any computer-controlled city.

2/ When i have 2 quality defence-units plus 1-2 fast-moving attack-units garrisoned in each city, the computer-controlled cities, by comparison, have wasted both time and resources producing 6-8+ slow-moving “cost-effective” crap-units per city (many of them wandering around aimlessly).

I do not ask for a game-AI that matches Kasparov. I do not ask for the Holy Grail.
I just want an AI that automatically tries to capitalise on any city-improvement-, or any unit-upgrade possibility that an already achieved technology has to offer. Is this too much to ask for?

The list of suggestions for CIV3 for example, is very impressive – but, if no major AI-changes are made, its gonna be like putting a tiny Volkswagen-engine in a large/heavy Rolls Royce.

You know what, I have noticed this too

Not much to discuss really, the AI MUST be improved.

[Ferdi]

Ralf,

I totally agree with you, but SMAC is a little example among others. Civ II could be incredibly more imposing with just a solid AI.
If I have to choose between civIII(and a crapy AI) and an old civII but with a powerfull AI then I surely take the latter.

[The Joker]

I agree with this completely. The AI must be improved.

I also think the game becomes boring half way through because you can see who will win very early in the game. The only way to improve this part is to incorporate the Rise and Fall of Great Powers and the Make it Harder for Civs to Last ideas from the IC3 list.

[tniem]

Somehow rise and fall is a necessity in Civ III. No nation lasts as the dominant nation in the world for more than a couple hundred years. No nation lasts in exsistence for more than a thousand. China was conquered, Rome, Greece, and others. They came back but after being destroyed as a civ in civilization terms.

If the AI was improved the game would be able to pose as an effective challenge in it's own right without the player having to place limitations on themselves in an attempt to keep the game challenging.

[Steve Clark]

That was why I did not follow up with my initial CivIII Fix: Better AI. I think the two ideas of Making it harder for Civs to last, and Rise and Fall would be enough of a challenge since the burden would be placed on the human players.

[tobyr]

It's very hard to make the AI play really well. Other than opening the API and letting hundreds of players try to improve (the best approach), I suggest an idea that came out of the ToT fantasy game:

In ToT, you can define units and then only let some tribes build them. This means that in the fantasy game, you can choose to play, say, the elves, and then modify the capabilities of units you CANNOT build to make AI tribes much stronger. They will give you a run for your money this way.

you can also fiddle with the tech tree to make sure that early in the game, the AI will have to build the units you most fear: those fast moving attackers.

(The Infidel tribe in the fantasy game does this already, and the AI handles their aggressive, early attacks moderately well.)

CIV 3 better preserve this idea of units that only some tribes may build or own.

- toby


------------------
toby robison
criticalpaths@mindspring.com

[MidKnight Lament]

I know this is off-topic now, but no civ should have pre-defined characteristics for a human player. Any 'specialised' units should be evolved from the choices the player makes with government, leadership style, or with their terrain.

Giving civs pre-defined qualities or units only leads to unbalance. I should have to shape my own civ, not have it shaped for me because I chose Japanese, and therefore get Samurai.

Sorry to diverge.

- MKL

[Lord Maxwell]

On improved AI.

There are several ways of improving the AI in civ type games. One of the simplest to redesign which part of the AI makes which decisins. (All opinions on AI moves from here on come from (perhaps faulty) reverse engineering from how it looks.)
In smac it looks like the lil formers themselves make decisions on what next to build when on auto-improve. This sucks. Cities should place orders with nearby formers for improvements. (Thus the city would actually use the improvement too.)

AI needs to be alot less nervous. It seems to suffer from a more or less constant nervous breakdown in smac, wasting huge amounts of minerals on constantly changing building ques. This has to be lowered.

Attacks could be alot more efficient if attack decisions were given over to a central authority, which deisgned a plan and followed-up on it. (Gathered units, made a feint and then a strike.)

AI needs to be easily patchable. Real easily patchable. Like an ai patch every month for a year, constantly adding to gameplay. Hire a guy to improve it for a year, it will pay off. (This is NOT to say that an initially bad AI is excusable.)

The AI should be able to analyze human moves and learn from em. If this isn't already slated then it's too late. Maybe for Civ IV. (This could easily be done if users are willing to have pretty huge swap files, and after finished playing let the computer crunch numbers for an additional five minutes. Found good moves and tactics could be sent to a central db and incorporated in downloadable patches. Probably most likely to be seen first for the final assault stages on a city. Easy to see patterns, easy to evaluate. Could also easily be done for city spacing.)

[Grier]

I always hated when I got half way through the game and realised that I was definatly going to win and was just going through the motions.

[Adm.Naismith]

That's a basic concept: a player enjoy a game when the other player is good enough to make the challenge worthy (fun to win, excusable to lose).

"Better AI" is a common statement in every press release of a incoming game.

SMAC AI try to react to player move: attack it with lot of planes and it will build units with AAA special, use Mindworms often and it will add Trance instead.

But that's (almost) all. No mass attack, often five to ten transport ships wandering without escort nor units aboard. Lot of useless route or Mag tube, not enough proper city development.

I (as lot of others) already suggested to try to develop an AI that try to learn some tactics playing against other (Human or not) players. I don't think lot of players can really make good use of open source AI, because I know that programming AI is not an easy task (if really it was, first developers team to make a good AI would sell its game as sweets).

I really have more hope into learning and exportable AI. I'm not dreaming of AI that really learn, but of AI that learn to mimic the human player tactics, then use them in advance, not simply reacting.

Let's do an example (in SMAC terms, because I'm still playing it). An AI lose three games with a human good player. Playing, it learn that he/she end bulding mainly forests, one borehole for city, one-two defensive units per base, lot of Chaos Rovers, then it stacks them mainly three at a time before attacking a base. No artillery units (useless).

Understanding a tactic is not an easy task, but something can be done with some number crunching, IMHO.

Next game it will try to counterattack on a similar (mirroring) way. Now I can see you: what's the point playing against myself style? Right, BUT NOW the human player can export on Internet FTP AI site the resulting logic. I can download an AI resembling his/her game style and let it play against me.
That will be quite different from my game style, not as good as playing against real human player, still a reasonable compromise if I can't play on-line with the real general

Ok, a learning/mimic AI is still not easy, BUT can be done with some success, AFAIK, at least using some player help into the learning process (I'm thinking about what I know of "Creatures" game, may be I'm wrong).

A large "downloadable trained AI" site can make wonder on the replayability of game.
Firaxis can keep up some "Automated Challenge" where seven different AI play one against the others on a Firaxis Net, until the better players gain better rate (and some extra training, too). It's common in Chess game to let different AI players to fight to help AI programmer to learn better tactics.
It could be more difficult still not impossible to make something on this line with CIV AI.


------------------
Adm.Naismith AKA mcostant

[tniem]

I like the idea of uploading and downloading mimicing AI's. Especially if it would also include it learning what you like and not only mimicing it but also defeating it. If you love using rovers than the AI would set traps for your rovers. I'm not sure the best defense for SMAC for the rovers but the AI would implement it. Then they would hit you with a strong attack that takes advantages of rovers weaknesses.

[Earthling7]

These little AI files sent to FIRAXIS would have to be compiled into one downloadable file, unless Civ3 accepted an "AI files pool", a folder containing multiple AI files. It would then be able to pick one file for that specific game, or search the files for strategy...

Something like that???

[Sirotnikov]

Ahm...
I'm sure I already posted in this thread. but way is may message not here?


anyway, what I wanted to say is:
Ralf, you're right.

[Adm.Naismith]

Earthling7, I'm not sure about the needs to compile the "AI Rules". I bet it depends by how the AI engine will work.

I see no big problems implementing an "export /import AI learned rules" into the CIV III menu. They can use any file format they need!

Tniem, I'm afraid that an AI developing proper "counter move" is not a realistic hope today.

Any decent game can use some "pre-programmed" tactic that can be more or less difficult to master on first bunch of game you play.

Then a human player learn how to turn around that obvius predefined order and gain an easy win.

What I suggest is that you can have lot of different tactics available, that are easily build by others gamers via the "learn by mimic" concept.

May be you don't end with some wonderful AI, really mastering every available game tactics (better if not, IMHO, I like to win at the end ).

Simply you have, any few games, the opportunity to match against different tactics that you must learn how to properly counter.

If that's not replayability, what's then?

------------------
Adm.Naismith AKA mcostant

[MidKnight Lament]

Shogun: Total War is supposed to have an AI that learns your tactics, and moves to counter them in the next game. How cool would that be if it worked with Civ?! Probably wishful thinking, but I'm prone to dreaming now and then.

- MKL

[Gord McLeod]

Computer programs that learn strategies and tactics from human players are not exactly new. It's very common these days... one of the first games I remember that featured it was Descent, though I doubt very much that it was actually the first. I don't see any particular reason that CivIII couldn't have a genuine learning system.

[Adm.Naismith]

Gord Mc Leod, aehm, well about

quote: Computer programs that learn strategies and tactics from human players are not exactly new. It's very common these days...

I'm afraid "I must dissent"

That's common hype by marketing. E.g. please note Firaxis claim on official Alien Crossfire site:

quote: BENCHMARK AUTOMATION Intelligent city governors (...)

If you played the game you know governor are almost intelligent as stones.

IMHO the actual games don't learn enough. But I know that game publishing returns are becoming more interesting, so it's more money for differentiate products, more money to pay last research in graphics engine and maybe some also on IA engine ...

I'm hoping.

------------------
Adm.Naismith AKA mcostant

[Gord McLeod]

quote: Originally posted by Adm.Naismith on 04-15-2000 06:35 PM That's common hype by marketing. E.g. please note Firaxis claim on official Alien Crossfire site:

Yes and no. There are marked differences in what 'learning systems' can mean... I'm something of a hobbiest in the field, so I know that it is possible and not TOO difficult, but it depends on the results you're expecting and what you personally define a "learning system" to be. If you want something that learns as fast and as well as a human being... well, we're a LLLLLLOOOOOONNNNNNGGGGGG way off from that point yet. And, I'll admit, games as complex as Civ and SMAC can make the task both harder and easier, simultaneously.... But, I maintain, it is very possible. Not very EASY... but nobody said it would be easy, did they?

quote: If you played the game you know governor are almost intelligent as stones.

No argument here...

quote: IMHO the actual games don't learn enough. But I know that game publishing returns are becoming more interesting, so it's more money for differentiate products, more money to pay last research in graphics engine and maybe some also on IA engine ...

Exactly - they don't learn *enough*. They do learn, though, and if you can get them to learn the right things at the right times and at the right pace, you don't need to make them geniuses in a human scale.

[The Joker]

I actually think the SMAC SI was a HUGE improvement compared to Civ2. Not only did it react to your attacks (as Adm.Naismith described), it also made real attacks in stead of just sending units against you one at a time and it built cities that actually had a significant amount of improvements in them (Morgan especcially).

I am not saying it was perfect, cause it was not so at all, but it was an improvement, and if the gap between SMAC and Civ3 is as large as the one between Civ2 and SMAC I don't think we'll have too much to worry about.

Of cause there are lots of things that can be improved. Lord Maxwells idea of having the automated formers and cities work together is sort of a must. I tryed to automate my formers as much as possible, but I still had to check if my cities were actually using the improvements made. I know that I would be more succesful if I micromanaged my formers, but that is a price I think is worth paying for having a funnier game.

Naismiths idea with a mimicing AI is just great. Not only would it improve the AI, it would also make it sort of possible to play against the best players around (via their AI).

You should be able to have several AIs downloaded at the same time, and choose which one to play against for each game. You should also be able to let them play against each other so they could get better and better.