Growth/Industy/Tech vs. World Size/Difficulty

[Bblue]

Gnome: Unless I missed something... (It has been a while since I looked at this mind you) But at Librarian settings on a standard or larger map (or a custom map)then the comp should be 'even'.

As far as AIs that don't 'cheat' the only one I own is for an older WWII wargame (V for Victory series, a top quality game) the difficulty levels in it were represended by improved AI play (no cheats) and as far as an AI goes, it did fairly well. Not nearly human, but it would give you a good challenge every now and then.

[David Byron]

I had a quick look at this the other day. I think the error in finding the formula is of thinking that the bonuses for map size and for the AI's difficulty level are of the same +/-10% type as the normal industry and growth levels. They aren't.

On a size tiny map you have 8 columns to fill but it is still 100%. Each column is 12.5%. Consequently (with rounding down) a +10% bonus doesn't register as one less column. It's 2.5% short. On the other hand a 10% penalty is rounded down to a 12.5% penalty of one whole column extra.

In effect then it works out that your first +10% is ignored but after that it's normal. Penalties are normal too. Now for AI on the lower levels I *assume* it works the other way. If they have over 10 columns then each column is worth less than 10% so a +10% is a column off but a -10% is TWO columns extra for the poor old AI. This could be tested with an AI Pirates' Growth on the easy levels but I haven't done so.

Because I'm a mathemetician not an engineer....

[gnome]

Another good example of a game where the AI didn't cheat is SSG's Warlords series. It was actually a pretty tough nut to crack on Warlord level (the highest).

I was addicted to Warlords II until Civ II came along; mostly because Warlords II is a military-only game and the set of units doesn't change over time. But as far as the challenge of the game, it was superior.

[David Byron]

The theory I mention above seems confirmed except I haven't tried it on different map sizes. With Transcend level / unsurpassed the basic # columns is 5 and 5 9instead of 7/7 for transcend level alone). Thus each one column represents 20%. Positive bonuses less than 20% (ie the odd bonus levels) are ignored / rounded down. Thus:

Pirates with no SE bonus (-1 growth) have 6 columns.

Miriam with no SE bonus (+0 growth) has 5 columns.

Yang with no SE bonus (+1 growth) has 5 columns.

Caretakers on Democracy (+2 growth) have 4 columns.

Usurpers on Democracy (+3 growth) have 4 food columns.

Caretakers as above with Crech (+4 growth) have 3 columns.

Similarly for Industry:

Spartans with no SE bonus (-1) has 6 columns.

Everyone else with no SE bonus has 5 columns -- including Yang with +1 industry.

I assume therefore that a pop boom will occur for anyone with +6 growth irrespective of the number of food columns this would be because of difficulty level, size of map or "unsurpassed" status.

[Tau Ceti]

Indeed, that formula even explains all the values listed in my Nov 23 post. And I just tried starting a game on Talent (which gives -1 to the AI), and Santiago, with -1 Industry, did drop the expected two columns to 13.
[This message has been edited by Tau Ceti (edited February 03, 2000).]

[Bblue]

Thanks David!! I think you solved this one!! Now is only 2 questions I can think of left in the Growth/Industry bonus question.

1) Is the 'Unsurpassed' bonus always a 2 column reduction or does it vary according to how unsurpassed you are? (my guess is that it is constant at -2)

2)Transend(-3) + Tiny(-2) + Unsurpassed(-2) = 3 columns, each equal to 33 1/3% so a 40% (+4) bonus should drop the columns down to 2. I have never caught an AI at 2 columns (3 columns is the least I've ever noticed ) So is there a limit at 3 columns? Yeah, Yeah I know I'm the only person that likes to play on tiny maps

[David Byron]

What remains to be seen is if the size column modifier is cumulative in an additive or multiplicative manner with the difficulty modifier and the unsurpassed modifier. ie would it be 3 columns or 4 for tiny / transcend / unsurpassed?

[Bblue]

*Bump*

[Helium Pond]

Bblue, thanks for bringing this thread to my attention. I'm trying to figure out how to reference it in the list I'm making (over in the "Difficulty levels explained?" thread). I don't want to mention map sizes or unsurpassed bonuses, because I just want to keep the focus on difficulty settings. This isn't always strictly possible, because some things depend on both, such as the bureaucracy penalty. In that case, I just used a sample value from a standard planet as an index of comparison.

In this case, I think first I need to know the "basic" number of minerals/line for each difficulty level, assuming no SE modifiers apply--which is something I'd want to put on my list anyway. Do you know these numbers?

[Bblue]

Assuming Standard map size the 'base' number of minerals per line and the base number of nutrients per line is 10 (AI and human players). This is modified by difficulty as follows for the AI players (humans are not modified). [This is modification due to difficulty level only, no SE, map size, or unsurpasssed bonuses included]

Citizen -3
Specialist -2
Talent -1
Librairian 0
Thinker +2
Transend +3

or expressed as nutrients/minerals per line:

Citizen 13
Specialist 12
Talent 11
Librairian 10
Thinker 8
Transend 7


PS. These numbers are only correct for maps of standard or larger size and custom size maps (tiny and small maps have additional modifiers)

------------------
"Power does not corrupt; it merely attracts the corruptable"


[This message has been edited by Bblue (edited May 01, 2000).]

[Helium Pond]

Thanks for simplifying this very complex thread for me. I'll add that info to my list.

[Ned]

This thread illustrates a game design strategy along the lines of the following:

To increase the difficulty level, the designers gave the AI increasing cheats rather than increased time to think before it moved.

This is almost like buying a chess game where the designers employed cheats to increase the difficulty level.

Regardless, many of us have expressed frustration at the stupidity of the AI across the board. This thread proves, in my opinion, that it does not get smarter when one increases the difficulty level. This is unfortunate.

[Sikander]

Ned,

It seems AI is the constant bane of the computer gamer. Still, I don't think it's fair to compare SMAC to chess. Chess is a game where intense research has only recently made AI a match for the best players, and only then at great expense (including research staff). Let's face it, Fireaxis isn't IBM, and your desktop is not Deep Blue.

Multiply this limitation by the inherent difference in the complexity of the game. Chess has 64 squares, all of which are identical, while SMAC has hundreds or thousands of squares, with numerous terrain types, which can change during the game. In chess only one of your 16 pieces can move in a turn. In SMAC, all of your (potentially hundreds) units can move every turn, which combined with randomization of combat results raises the number of possible outcomes by orders of magnitude. Thus it is nearly impossible for the AI to look even 1 turn deep, there are too many variables.

Finally, consider that SMAC is still in large part CIV, which was a miniscule little DOS program. The system works, but it would be interesting to see a new champion paradigm. Perhaps a game designed to simplify the task for the AI combined with a beefed up AI design could give the player a battle without resorting to various cheats.

[Ned]

Still, Sik, it is somewhat disturbing to realize that the Computer AI is NOT planning overall strategy while the player is taking his or her turn. Strategy is the weakest part of AC's AI.

Ned

[Sikander]

I totally agree Ned. The AI is not playing the same game that we are, and it's analysis of the current situation is cursory and short sighted. We are left with multiplayer to test ourselves until we see a significant challenge from the AI.