AI "cheating" is necessary and welcomed!

[ancient]

i think by now we should be able to make an ai that is good enough to match a human player.. if we cant we should atleast come close with out resourting to cheating..

[Mark_Everson]

polymths, on the "statistical model based on strong human player(s)" I have one word... Yuck.

But that said, my list for Clash AI does include:

Allow the player to give the AIs any arbitrary advantages that they deem necessary

IMO players should be able to pick and choose what they want to help AI with, not have it presented as a big package.

[polymths]

quote: Originally posted by Mark_Everson on 05-03-2001 07:35 PM polymths, on the "statistical model based on strong human player(s)" I have one word... Yuck.

If the AI's development is realistic what does it matter whether this was achieved through statistical modeling or whether it was achieved because the AI actually had to micromanage all units and all details. Isn't the end result what's important?

For example, if the AI's land tiles improve through time through a
"tile maturation process" that is statically modelled after what a human could achieve instead of the AI actually managing each settler and improving its squares one at a time like a human must what does it matter?

[Mark_Everson]

polymths:

I just don't think that you can make it appear realistic. Especially in the military area. As I understand your 'attack patterns' description above, the game would actually Need good AI to execute them! There is no such thing as a limited set of preprogrammed attack patterns that will suffice for most circumstances. And if the military AI is not sufficient, then having the AI armies mature at the same rate as a good players' would do no good! The AI will still lose horribly.

Furthermore, unless your statistical model is very good, you will lose the cause-and-effect relationship that's so critical to have the game feel right. Picture this. I've beaten the Zulus back to one city and am preparing my final attack when that one city builds two wonders, and 12 veteran knights suddenly appear to ham-handedly try and wipe out my army... I fear that kind of stuff is what you'll get. And damningly, if you don't go to those ridiculous lengths then the AI will get pasted anyway.

That said, given the complete inability of at least the previous civ programmers to achieve any semblance of intelligence in the computer opponents something's definitely needed.

Anyway, I don't think this is gonna happen, and I'm not much into purely theoretical discussions...

-Mark

[ancient]

so you realy think the computers citys should just develope themselfs? with out them even having to build workers? thats really stupid.. why not just have it programmed into them to build as many city improvements as you can as fast as they can with the real thing being built in their sity.. it would really work out the same way.. i just dont like the stupid suggestion of the computer getting free irrigation and stuff.. doesnt seem fair..

[polymths]

quote: Originally posted by ancient on 05-03-2001 09:15 PM so you realy think the computers citys should just develope themselfs? with out them even having to build workers? thats really stupid.. why not just have it programmed into them to build as many city improvements as you can as fast as they can with the real thing being built in their sity.. it would really work out the same way.. i just dont like the stupid suggestion of the computer getting free irrigation and stuff.. doesnt seem fair..

What I mean is that the all computer city development should be on statistically autopilot.

Tiles will "age" slowly through time at a realistic rate. Some token workers will roam around to give the illusion that actual workers are working the land but in reality the tiles are simply "maturing" through time by themselves automatically. If these token workers are killed, then the maturing rate will slow until new token workers start roaming the land. But these token workers will not need to be managed at all but will just roam around aimlessly for illusion purposes.

As for city improvements and wonders, these will be awarded to the Civ at certain intervals of time based on the theoretical productive capability of the Civ, assuming that no other Civ already has that wonder.

They key is to develop a good statistical model so that the city improvements and wonders are awarded in a realistic manner.

But all city improvements and land improvements are simply happening in the background with no need for AI decision-making whatsoever.

[polymths]

quote: Originally posted by Mark_Everson on 05-03-2001 08:49 PM polymths: I just don't think that you can make it appear realistic. Especially in the military area. As I understand your 'attack patterns' description above, the game would actually Need good AI to execute them! There is no such thing as a limited set of preprogrammed attack patterns that will suffice for most circumstances. And if the military AI is not sufficient, then having the AI armies mature at the same rate as a good players' would do no good! The AI will still lose horribly...... -Mark

A "statistical" attack would actual not need good AI just realistic statistical analysis. Say you are at war with Zulus. They have a certain number of cities, a certain population, a certain theoretical productive capability, a certain govt, and are a certain distance from you. (Of course the Zulus are well developed because they have been developing "statistically")

Therefore, it is possible to determine that given Y years, it would theoretical be possible for the Zulus to mass N units for a mass attack. At year Y+T, the N units will arrive near your shores. But until year Y+T, no units are actually being manuevered or managed. At year Y+T, a "package" of units be assembled at that point based on the theoretical productive capability at year Y (the time when the fleet was theorectical supposed to be launched.)

[Urban Ranger]

Ralf,

You seem to have a number of misunderstandings of how AI can work.

"By comparison, the game AI (or any silicon-based intelligence for that matter) is 100% dead and non-experiencing (of course). It lives in a 2-dimensional "flat" world, figuratively speaking - by that i mean it cannot possible "overview" anything."

This is definitely not true. There are various ways to program some sort of overview into the AI, for example, using overlayed weighted fields of influence.

There are also heuristic algorithms that learn from past experiences and evolutionary algorithms that could come up with some amazing constructions.

It's not easy, but it's possible to write competent computer players without resort to cheating.

[ancient]

how bout using a learning script for the ai, it could acces itself then find out what it needs to do more of next time, that way the more you play the gaem the deadlier your opponent would be..

[ancient]

quote: Originally posted by polymths on 05-03-2001 09:55 PM What I mean is that the all computer city development should be on statistically autopilot. Tiles will "age" slowly through time at a realistic rate. Some token workers will roam around to give the illusion that actual workers are working the land but in reality the tiles are simply "maturing" through time by themselves automatically. If these token workers are killed, then the maturing rate will slow until new token workers start roaming the land. But these token workers will not need to be managed at all but will just roam around aimlessly for illusion purposes. As for city improvements and wonders, these will be awarded to the Civ at certain intervals of time based on the theoretical productive capability of the Civ, assuming that no other Civ already has that wonder. They key is to develop a good statistical model so that the city improvements and wonders are awarded in a realistic manner. But all city improvements and land improvements are simply happening in the background with no need for AI decision-making whatsoever.

if the computers tiles ages how will they be able to build roads in between citys? and anyways if the cpu gets to have free tile improvements theyll have more time to build buildings and unit giving them an advantage

[Gramphos]

quote: Originally posted by ancient on 05-04-2001 06:51 AM how bout using a learning script for the ai, it could acces itself then find out what it needs to do more of next time, that way the more you play the gaem the deadlier your opponent would be..

What if you use different strategies every time you play. That would realy make the AI more stupid as it tries to get better according to your last stategy all the time.

I say don't try to make the AI lern your gameplay that would just confuse it (or delay the game another five years ).


[This message has been edited by vgriph (edited May 04, 2001).]

[Russian King]

In civ2: this was a GIGANTIC problem!!!!!

"The comp should be smart enough"-good point, but not the case, in civ 2 the comp built much more than you could build in the ancient age whether You cheat or not . Thats not acceptable. The comp should be given NO head start and shouldnt cheat at all: it just doesnt make it better or even more challenging.
YES it doesnt make it more challenging, it just makes it look dumb and WEAK!
and the fact that its so powerful-what would you think of Bill Gates if his mom made the whole company work and he was siting wasting the money?
Sorry, Bill, no offence, your a good guy.

[Cyberbugs]

I have posted this in a previous, quite small thread, so I may as well repeat it.

What I would like to see, is that the advisors "cheat".
By that I mean that they sometimes provide you with useful inside information. For example the military advisor may say: "Many nuclear missiles are being produced by different civilizations. I suggest you build some anti-nuclear defence."
This way the advisors don't just provide you with things that you are already aware of, like "your naval fleet is too small".
However, the inside information shouldn't be *too* revealing, of course, but it would very much enhance the usefulness of the advisors.

[Gramphos]

quote: Originally posted by Cyberbugs on 05-04-2001 11:22 AM What I would like to see, is that the advisors "cheat". By that I mean that they sometimes provide you with useful inside information. For example the military advisor may say: "Many nuclear missiles are being produced by different civilizations. I suggest you build some anti-nuclear defence." This way the advisors don't just provide you with things that you are already aware of, like "your naval fleet is too small". However, the inside information shouldn't be *too* revealing, of course, but it would very much enhance the usefulness of the advisors.

This isn't actually what is discussed in this topic, but you make a point. This may be a new effect to Espionage.

[Ralf]

Mark Everson quote:
Most badly in the assumption that an AI can never take anything but an exhaustive search of possible actions at the lowest-level view. This just isn't right.

You perhaps misunderstood my intention with that post.

My intention was/is NOT to imply that a better Civ-AI only can be created if one takes the most exhaustive search of possible actions, at the lowest possible level, each and every turn (and since that is practically impossible in a random-map mega-optional game like Civ; any hopes of a better AI is utterly futile. So lets all forget about it).

Above was NOT what I meant. I only stated that IF one stubbornly daydreams about a "humanoid passed-the-Turing-test" kind of AI that learns/re-values/re-adjusts its choices and strategies dynamically and automatically each and every turn - including how to move around AI-units with 100% human-player efficiency - THEN one can only be sure to achieve such lofty (and unrealistic) goal by executing that very exhaustive search of possible choices, at ALL levels at every turn - and several turns deep. Just like a high-ranking AI is forced to do in chess, by the way.

HOWEVER - as I mentioned further down in my previous post: One is that above comparison with chess is - to a certain degree, misleading. Chess is in some ways a very different beast than Civilization...
Well, it certainly is! Civ is MUCH more "wrong-move forgiving" then chess. Also, one dont necessarily have to analyse several moves ahead in civ, in the same abstract forehead-vrinkling ways as one is often forced to in chess. In this sence, playing Civ is MUCH more simple and straigthforward.

Also, as you mentioned yourself - there are many AI shortcuts and methods that one can apply. I have read through your "AI-list", and it all make sense. Cant argue against anything you suggest. I will only emphasize some of my own ideas:
[*] Most importantly is of course that Firaxis from the very start, tries to design and implement ideas in as "AI-friendly" ways as possible. Many added new ideas works for example like "double-edged swords", in the sense that they act equally as much on the human player, as on the AI-civs. But, because the human player always is prone to be the more active and expansionistic one - the burden often lies much more on the HP and then one the AI-civs. This is something the game-designer can (and should) exploit.
[*] The use of player-editable AI-build and AI-prioritize template-lists. I have often wanted to be able to "spoon-feed" the AI-civs through the text tweak-files, thus saving them from their own stupid selves as much as reasonably possible. For example by: Specify editable templates of what they should upgrade/build; units, CI:s and Wonders. Exactly which units & CI:s - how many/and in which order. Specify an indevidual forced-upon number of AI-cities each AI-civ MUST prioritize. Specify in which order AI-tech advances and what government-forms each AI-Civs should follow/prioritize. Specify how & how much AI-city areas should be cultivated. Specify how AI-cities should exploit excess wealth. Specify prefered Gold/science allocation, and so on, and so on...

Above AI-build and AI-prioritize templates means that instead of forcing the AI to “evaluate” and “choose” in which order this or that tech-advance, city-improvement and even single unit should be build – the whole AI evaluating process is BYPASSED as much as possible.
Instead the AI follows blindly a pre-made setups (templates) until something triggers it to temporarily take over direct command. You can compare it with a blind man being given exact blind-read instructions how to move around inside an unknown building. He can follow this pre-made template meticulously and rather effectively (with minimal of independent analyzing). If he never the less bump into something – he temporally put his blind-read map in his pocket, and starts to feel ahead and analyze the situation. Soon feeling confident, he can take the map up again…

Above, has to a certain degree already implemented in most strategy-building-games of course. But I want it to be done much more thoroughly and player-accessibly/tweakably then ever before in Civ-3.
[*]Ah! My lovechild! My favourite Civ-AI shortcut idea: The need for pre-game calculated AI-city placements, with all potential AI-city locations already invisibly layed out (like yet unlit red diodes that punctures a electric standalone touristmaps on big railwaystations), in conjunction with the map-generation. On manually created world-maps, these potential/invisible AI-city locations can be manually pin-pointed all over the map.

On which potential/invisible AI-city location (or unlit red diode) each AI-civ chooses to build its first capitol city, may very well be totally random as long as the starting-locations are reasonably far away from each other. Also, in which direction each AI-civ chooses to expand may also be random, or then they start to get contact with each other; according to varying non-foreseeable in-game factors.
It is however, then Civ-3 scenario-creators starts to make use of potential AI-city placements on their own maps, that this idea really showes off its advantages. Suddenly the Civ-playing scenario-creator gets a whole new time-dimension under his fingertips. Not only can he now exactly pin-point ideal potential AI-city locations (and distances between them): But he can also determing how each AI-Civ should expand into map-sections thats not yet uncovered in the beginning of the scenario - if they want to have that control, of course.

Read more in this thread: http://apolyton.net/forums/Forum6/HTML/002067.html (never mind the title, read on).