Super-Transcend Levels

[jed]

My philosophy is that there are so many possible advantages that one can give the AI, and so many handicaps you can give the human, that it must be possible for everybody to find a challenge even if they find transcend level too easy. I have been working on play-balancing a set of beefed-up AI factions to play against, and generally tweaking things to help the AI. I hereby present my favorite way of playing.

First the settings. I use the Alien Crossfire world with standard rules. I made one change to alphax.txt: I make Industrial Automation of pre-requisite of mines. This makes the AI build forests instead in the early game.

I personally find the Monsoon Jungle the most unbalancing aspect of the game, so I make my own maps and delete it. You might not know how to do this (it is not obvious) so here is an explanation: Go into the scenario editor, and you will see that first of all you can edit the map world parameters from there. I choose all standard parameters except ocean coverage which I set low, because the AI plays badly when it needs to get across water. I then hit the fast generate random map button (button 0) and keep going until I see a map that looks like it has the outlines that I like (I look for one with interesting continents that are all connected, even if by threads). When I get a map that looks good, I find the Monsoon jungle. You need to get the cursor on it, then go to the "remove landmark" menu item, and remove it. If you don't have your cursor on it, you can't remove it. I save a few maps like this. Don't worry, you'll never remember what you saw.
(I also change the islands parameter in alphax.txt to "9" from "36", to reduce the number of islands in the maps, but you can get good maps without doing this.) I think it is much more satisfying to play on good maps.

OK, now for the beefed-up factions. Remember these are for the AI to play, not you! I tried to keep each faction true to the philosophy of the original SMAC 7, and I play-tested them against each other to make sure they were balanced. You may not know how to do this either, so here is an explanation:
Start a game, with the first faction on the list. Make the first move for your faction, but don't hit the make move button (I presume you use the option of hitting this button before you move--if you don't start using it now.) Open the scenario editor. Go to the switch sides menu item and switch to another faction. When it asks, answer you want to let the computer play for you. The screen will go dark. Light it again using the reveal map menu item. Hit "Move complete" and then just hit return at every announcement. You can watch the AI play itself. And make sure your preferences are set for fast movement for all moves. Very instructive if you want to improve or balance the AI, but kind of disillusioning if you think the ordinary factions are any challenge.

This is getting long, so I will post the
factions in a new message.

[This message has been edited by jed (edited May 13, 2000).]
[This message has been edited by jed (edited May 13, 2000).]

[jed]

Now, how do you play with these factions? Well, I'll just make some suggestions. Not much challenge to play as one of them--might as well play on citizen level if you're into that type of thing. Challenge level #1 is to take an ordinary faction against the beefed up factions. Challenge level #2 is to take a weakened faction--I like playing as the peace-keepers, but with -1 Growth instead of -1 Effic, and with no bonuses or free techs at all, but with the ability to go police state. If that's not tough enough, just keep adding handicaps--say -1 in 8 different SE categories and 110% research costs. Sooner or later, you should find your match.

Or, if you think the beefed-up factions are too beefed up, take away techsteal, or free formers or whatever. I think the game is most enjoyable when you don't know whether you'll win or not.

[jed]

OK, on to the factions. First of all, some global comments. I gave every faction 2 techs, +3 or +4 SE bonuses, a free facility, a free former, an impunity to its SE preference, the techsteal ability, at least talent/6 and an assorted bonus or two. I kept all the agendas and aversions as usual, but I did change some of the other interests and emphasises to make the AI factions play more consistently with their advantages. I tried to keep them consistent with the feel of the original SMAC 7. The techsteal ability is there to make sure that wars between the AI are not a complete boon to the human. I gave everbody an interest in exploration because it makes the AI play better.

Believers:
Explore, Conquer. Aggressive. Emphasis: Probe
Social Psych, Applied Physics
Free Rec Commons
+2 Support, +1 Morale, +1 Probe
Impunity Fundamentalist
FANATIC %25 Attack Bonus
Talent/6, Techsteal, Free Former
Comments: I made these factions with the faction editor, sometimes editing the faction.txt files by hand. The faction editor lets you know which bonuses use up one of the 8 available "slots". Thus you need to use FANATIC bonus, which doesn't use a slot, instead of 125% Attack strength which does. Also note that Miriam will do research right away. The 10 years handicap she usually gets is from RESEARCH -2 (* 5 years/ SE point).

Gaians
Explore,Discover,Build,Conquer.
Erratic. Emphasis: Planet.
Biogenetics, Centauri Ecology
Free Recycling Tanks
+2 Planet, +2 Efficiency
Impunity Green
+1 Fungus Nutrients
Talent/6, Techsteal, free former
Comments: Switched them from pacifist, and switched from just explore. They are not shy about using their native life form enhanced army.

Hive
Explore, Build. Erratic. Emphasis Police.
Doctrine Loyalty, Progenitor Psych.
Free Perimeter Defence
+2 Industry +1 Growth
Impunity Police State
125% Defense strength
3 bonus before needing hab complex
Talent/6, techsteal, free former
comments: A kinder gentler hive, which works more in building up its hive-like industry, with a strong defensive shell. (Changed from aggressive to erratic and took away conquer.) Make sure you say -3 for pop bonus rather than +3, which is a negative.

Morgan
Explore Build, Pacifist. Emphasis Economy
Industrial Base, Industrial Economics
Free Energy Bank
+1 Economy, +1 Industry +1 Probe
Impunity Free Market
100 extra energy credits, +1 Commerce
Talent/6, techsteal, free Former
Comments: Impunity free market seems huge and it is, but Morgan was probably the weakest AI faction originally. Now he's just another tough guy.

Peace-keepers
Explore Discover, Erratic, Emphasis Growth
Biogenetics, Ethical Calculus
Free Recycling Tanks
+1 Growth +1 Support
Impunity Democratic
*2 Votes, 2 Pop bonus before hab complex
90% research costs
Talent/4, techsteal, free former
Comments 90% research costs instead of +1 research because I ran out of slots. Note Talent/4 instead of talent/6 like everybody else. Can't give them free children's creche. Anybody who gets free children's creche takes over, but especially the peace-keepers.

Spartans
Explore, Discover, Conquer. Aggressive
Emphasis Support.
Doctrine Mobility, Applied Physics
Free Command Center
+2 Morale, +1 Police, +1 Research
Impunity Power
Free Prototypes
+1 Research/Base
Talent/6, techsteal, free former
Comments:
Switch them to aggressive to take advantage of their strengths. Because of a bug, a "free" command center is not free--it costs 1 energy credit per turn. To compensate them, I gave them 1 extra research point/base. This works out well. Think of it as a Spartan tax to fund military research. I also switched them to emphasis support, so that they will try to eliminate the biggest weakness of their army.

University
Explore Discover, Pacifist, Emphasis Research
Information Networks, High-Energy Chemistry
Free Network node
+2 Research +1 Economy +1 Efficiency
impunity knowledge
techshare 3 (gets any tech 3 other factions have)
talent/6, techsteal, free former
Comments: Gave them high-energy chemistry to help their early defense with a 3-armor. Switched them to pacifist--they should not want to pick fights.

[Basil]

Just wanted to say that I found the advice useful and the factions seem well-designed, although I'm not sure I'll use both those faction variants and the AI-friendly map any time soon . So far I've only had time to try watching the AI play against itself... that was an education.

[joer]

I read more and more posts about how to make the game harder by altering the setup. Here's my $0.02 on those.

I am sure that there are many ways to make me loose and the AI win. Hey, it happens often enough on standard Transcend level, so there's no need for me to freak with the text files to give the AI super-bonuses.

However, what I would really like to see is an AI that is smart enough (yes, I stll want believe that I stands for Intelligence, and not Idiot) to be challenging, but not impossible to beat on even ground and over a longer period of time. Without it having bazillions of bonuses and me setting up the situation that screws me from the start. Without it having to rely on cheats and me having to restrict myself in every way, unable to use some of the great possibilities the game offers.

Usually, in my games, the AI starts very well, since it does not have a lot of options in the early game. So it can crank out pods and militairy units much faster than me. Later in the game, it bites itself in the butt, because it simply doesn't realize what it is doing. Excess production of militairy units which just stand around to eat up the remaining production, lousy SE settings that kill off any research and inability to effectively use air power are just three major examples. I would like to see an AI which is a challenge even in the late game, but being able to beat with just that bit of better tactics and strategy that divides the human from the computer. And I would like to have that experience on Librarian level, of course (where the AI and the human are on 'even ground').

I realize more and more that my own late-game skills suck badly. Why? Because, way too often, by that time, I have decided that the game is already over. There is no toe-to-toe competition between me and the AI anymore, it's just squeezing out a little time advantage, which is not what I am looking for in the game. Once in a while, I would like to see a game where I have been doing reasonably well in the early and mid-game, haven't lost anything crucial to the AI, but it still was just a tad better at expansion. Good enough to keep me trying hard and keep me on my toes.

Ok, ok, so this is a purely whiney post, which will not change the damndest thing. And maybe I am the only one who feels that way. But, for me, a challenging enemy is vastly more satisfying than a challenging (not to say unfair and unbalanced) setting. Am I alone out there? Or does anybody else feel simular?
-joer.

[Kinjiru]

Joer-

I do not think you will find a single person on these forums that would disagree with you. The trouble is, it cannot happen.

There is no such thing as artifical intelligence. Computers do not think, they simply execute programs and churn through algorithms. They do not learn.

SMAC has been out for two years now. The humans playing the game have learned how the AI plays and learned from each other. If boards like this one did not exist, I am sure that I would still be stuggling to beat Transcend level. But I have learned from other players through the discussions we have had here.

The computer has no such advantage. So, we are left with doing the kinds of things described in this post to make the AI 'better'.

If you truly want to see games play out like you described, please please please pick up a book on C++ and start learning. The gaming community would benefit greatly from a programmer who is more concerned with making better AI and less eye-candy.

[Kinjiru]

Jed-

How do you set the faction preferences for Explore, Conquer, etc. I remember doing that once and I thought it was through the scenario editor. Is this the only to set those parameters?

[WhiteElephants]

I've heard a similar complaint about the AI cheating to win on another forum about another game and someone responded with something to this effect: "The day the AI doesn't need to cheat to win is the day a computer will be your boss".

So, boys and girls, the morale of this story is be carefull what you wish for because you might just get it.

[Basil]

Kinjiru - those faction preferences are in the faction text files. You can change them in either the faction editor or by editing the text files directly (in, say, notepad).
EG:
The Lord's Believers, The Fundamentalist, Believers, M, 2, Miriam, F, 1, 1, 0, 0, 1,

The 1,1,0,0,1 are the faction preferences. 1 - aggressive (-1 would be pacifist, 0 erratic), 1 - interested in power (conquer), 0 not interested in knowledge, 0 not interested in wealth (building), 1 interested in exploring. To make the changes effective in only a particular scenario, store the changed faction files in a scenario folder. For more information look at the file "faction.txt".

Joer, Kinjiru - I think the "cheating" of the AI's in many games is due to the economics of game development, not what is technically possible. Coding a production advantage for the AI's which varies by the game's difficulty level is quite easy, while making the AI understand game features is hard work, and the better you want the AI to understand the feature the more work it is. Realistically, you (the designer) are looking at a trade-off between how good the AI will be and how many features you will have in the game. And no matter how good you think your AI is, it will almost always make sense to give the players the option of giving the AI advantages simply because that is so easy to code.

Having said that, I agree that the limitations of the AI in SMAC/X Joer mentions are major problems .

[joer]

More idle thoughts on a lazy Saturday morning...

Anyways, I will not bore you with abstract comparisons between executing an algorithm, learning and thinking. I think Papa Turing said most of what there is to say, and, hey, that was over 30 years ago. Look up his stuff if yer' into that and haven't already.

However, if 'intelligence' is used in a more pragmatic, practical way (as in: simply making better choices), then even a simple algorithm can do better in certain areas than the AI is doing now.

Simple example: Terraforming:
if my square is rolling and at least moist, put a farm, road and solar collector on it. If it's rocky, put a mine and crawler on it. If it's anything else, put a forest on it.

Now, this is surely not the best and most efficient example in all situations, but there are situations when the AI doesn't even realize this simplicity and will put mines on rolling, rainy squares.
There's hundred of more examples of these 'simplicities' that I can think of. Now, I do not know the AI engine of SMAC. I'd love to take a look. But I still think that making it better in certain areas would have not been too hard.

Theoretically, with enough effort, the AI could 'learn' as well, by adapting to the playstyle of the human player and storing how it has lost in previous games: Does he love rover-rush strategies? Beeline for ECM units. Is he obsessed by air units (like me): beeline for Advanced militairy algorithms.
Or the AI could even try out switching between different SE settings and store which have worked well and which have not.

Oh, yeah, and I did read C/C++ books before. No, honest! K&R is, apart from these forums, currently about the only thing I read (yeah, yeah, pathetic me). I even read articles on gaming theory. The simple terms that come up there, such as 'looking ahead of what the opponent can do' doesn't seem even considered in the SMAC AI engine. I'm sure that this game would be much different if all of the Hive's masses of totally obsolete defenders suddenly got upgraded to highest armor AAA garrisons once a chopper comes in their range.

I think your "eye-candy" remark is an important one, and I agree wholeheartedly. However, I do wonder if the majority of the gaming community agrees. Gaming magazines seem to have the tendency to base their judgement strongly on how cool a game looks, and less how much in-depth strategy it offers on higher levels (compare articles on Civ:CTP vs. ACT). I have read that a substantial part of Firaxis budget went into the mini-movies. Now, those are nice, but if I want to watch a movie, I go to the theatre. In the end, I get much more out of a good AI.

While I have currently very little time to get any deeper into the open code that some turn based strategy games use (hey, I gotta play MP ACT), I have made the silent vow that I will apply at Firaxis for a job as AI programmer once I got my degree. So what if they laugh me out and reject me? At least I can say then that I've made a try...
-joer

[Ned]

I think the AI is particularly weak on strategy. During the "dead time" that takes place while the Human is moving, the AI should plot and replot possible changes to its strategy to exploit advantages or weaknesses it sees in the enemy position. But from my observation, it seems to do nothing. Therefore, since it must respond to the Human's move within a reasonable amount of time after the Human is finished, it must act on certain rules it was given by its creators. It cannot play through multiple senarios and see which outcome is better. It doesn't have time.

So, ultimately, the game has no strategy and is highly predictable.

For some reason, the CIV I AI seemed on the whole to be better at strategy. I wonder why?

[Lord Maxwell]

Kinjiru, offcourse you can make a "true" ai for a specialized field. You can (ahem) easily teach it to reason about a limited field and then weigh pros and cons based on some chosen evaluation algorithm.
If you do it with a neural net you won't even fully understand how or why it reasons as it does, only that it has been trained to do so.

In games like smac a layered ai would be best. Using qualitative reasoning at higher levels to decide upon overall strategies based on imprecise and non-closed information sets.
Then when the meta-strategist knows what it wants to do it leaves over to lower level ais. (Say in smac it decides to invade an enemy, using a nettle on the east side and a serious drive on the northwest. It sets a few "gather" areas where units meet. Then it leaves this info to the war-ai, which reasons with much less imprecision, decides on units to use, both based on calculation and heuristics.)

The real sign of a "learning" ai is the ability to find and evaluate (and later use) new heuristics. (Rules-of-thumb for you with low ai knowledge.)

And to make the game more interesting I have an idea. Once you get a game to evaluate the player, find new heuristics and incorporate them, then you add a nifty thing, the ability to upload heuristics to a central data-store, and get new ones from there. (An apolyton forum for the game basically.)
For the first generation of fluid heuristics I'd say that a human would need to atleast occasionally witness the process, but it can most likely be fully automated.

Offcourse, this doesn't at all touch upon the fluffy debate on whether or not computers can ever truly think, as they are silicon based. (And created by man, not god, as the argument so often goes.)
If it can reason about it's field of expertise, find a new solution not hardcoded into it, evaluate and store information for later use, perform as well or better than a human then I darn well say it can.

(On a side note, it shouldn't even be very hard to create good game ai, considering some of the truly wicked problems that ai can solve, like sinter plants, decaffeination and cement kilns. (All examples where you don't get immediate feedback on changes in variables, and where risks are large, and where in the last five years computers have consistently outdone humans for both safety and productivity.)

[Biddles]

What is really needed is an 'open AI'. By this I mean an AI that a player can come in and add extra algorithms to.

For example: In the morgan AI you would put something like "if WEALTH option is available then do not select FREE MARKET".
As a player, if I am playing morgan than I have realised that while morgan might be biased towards FREE MARKET, it is better to choose WEALTH and get the +1 energy per square and then choose GREEN economy. Why not make it so that we can tell the AI to do the same. It might be difficult for people who don't know about programing (sadly myself included) but I am sure there would be people out there who would be wiling to make AI files and make them available on the firaxis (or apolyton) site.


------------------
- Biddles

"Now that our life-support systems are utilising the new Windows 2027 OS, we don't have to worry about anythi......."
Mars Colonizer Mission

[jed]

Hey, I'm glad my post got some response. It looked like it was going to be ignored.

I now believe that my factions are best balanced for a "large" rather than "standard" map, so I'm playing on that setting.

One other important comment that I did not mention originally. If you are watching the AI play itself, it will go much faster if you do not watch every faction move every turn. If you really want to move forward fast, switch to the faction that has the fewest units, and turn off the omniscient view, and you'll soon be cranking through the "move complete"'s. Of course, you lose the big picture this way, but it goes faster. I don't really watch beyond 2200, though, so I can't guarantee that my factions are well-balanced for the middle or late game.

About the wish for "real" AI, rather than tough settings... Of course that would be great, but it's not going to happen any time soon. I prefer to focus on the advantages of SMAC/X. The game has great atmosphere. It has really strong gameplay. Better than any other computer game. For me, the main thing is to have that atmosphere and to have a challenge. Something to think about. If I win too easily, it ruins it for me. So for me, it's much more fun to play against unnaturally strong factions in the way I described. Only problem is this week I was too busy to play. :-(

[jed]

I've been experimenting some more. My goal was to get 7 balanced super-factions on standard-sized maps (but hand-crafted to remove the Monsoon Jungle and get nice continents as described above), where each one of them built up a reasonable empire. I started with the super-factions described above, then took away techsteal, made each of them interested in all of Explore, Discover, Build and Conquer, and made their aggressiveness levels as follows:
Believers, Hive, Spartans, Peacekeepers:Erratic
Gaians, University, Morgan: Pacifist.

These settings seem to work very well. If you have ever looked at the ordinary factions using the scenario editor, you know most of the time they do not properly terraform or build base facilities. My super-factions now *all* do a *much* better job at this, terraforming and building quite sensibly and going for secret projects early on.

I switched everybody to E,D,B,C because it made them much more flexible. Given that my super-factions already had enhanced personalities, I don't think they need this to give them their personalities.

I thought that maybe switching the ordinary factions to E,D,B,C would also improve them, but it actually didn't seem to help as much as it does for the super-factions. Not sure why.

I toned down the aggression levels, because otherwise the factions get involved in too many stupid costly early wars with each other, effectively knocking 2 or 3 of them out of contention. The Spartans and Believers
still pick enough fights.

I have tried playing with weakened peacekeepers against 6 super-factions, and it is really quite pleasant. In combination with the fact that you will have a hard time becoming number 1 or 2 on the power graph, so the AI's won't hate you, the lessened aggression makes for a more peaceful, builder-friendly game, with more "sane" diplomacy. Now this won't be to everybodies taste, but I enjoy it, and I suspect many other people are looking for this type of game as well.
[This message has been edited by jed (edited June 01, 2000).]