«

Originally Posted by Japo

Dunno, I don't think we need to be so forgiving on a poor AI, it wasn't warranted by processor power alone. And the fact it was substituted for with cheats by the computer opponents. There are older games with good enough AI.

Could you name a few of them?

There are different approaches to AI in 3 Microprose games I love - in Master of Orion 1, in MoM, and in Ufo defense (X-com).

The AI in X-COM is totally pre-defined, mechanic, which is covered by the fact that the opponents are "THEM", do not use the same logic as mere humans. It really works, or worked at the time I played the game.

The AI in MoO is very simple and it is driven by the personnalities and the races. It works surprisingly well (bar a few exploits), mostly due to the fact that the game rules ARE very simple, not complex (like the AI knows exactly where the human fleet is going to attack in 3 turns). They did a very poor job on AI, making it random and intentionnally sub-standard, but dependent on the AI race. It's interesting - it works for me quite well (better than the plastic "super AI" from Galciv). But, in fact, they could do a much better job on MoO if they tried a tiny bit more.

The AI in MoM is the most complex (by several levels) of the three games, but it is found out times and times again. The game simply has incredibly complex rules for its time, several strategical levels. Believe me, this was a lost war from the beginning for the Microprose. The tactical combat itself could use half of the exe (with all the rules, spells, units, combos, exceptions). The omnipresent exceptions are the killer. No way they could put all this inside AND make the game exe small enough to be playable under the machine restriction of the time. Maybe if they changed the memory conception.

But you are right they were probably overwhelmed. And they left a lot of bugs inside as well, so even with the buffs the AI is not good enough. This is a pity.

[Japo]

Originally Posted by kyrubb

Could you name a few of them?

Sim City (although the AI isn't the antagonist)... But most chess games even sub-par ones rank well above Civ in terms of AI.

[Scatty]

Originally Posted by Japo

But most chess games even sub-par ones rank well above Civ in terms of AI.

I would say chess though is more simple than all the possibilities in Master of Magic. While the latter offers nearly unlimited combinations of units, computer personalities for diplomacy (however little of use it is in game), spells and random dungeons (towers / nodes etc.), terrain and ressources layout while the latter three alone can heavily determine how fast and good the opponent can develop, chess offers only, how much? 20 * 20 moves for black depending on how white moves and 20 * 20 the other way around, from which it's just a matter of calculating the possible outcomes and counter-moves and choosing the most optimal outcome for continuing.
While it is getting increasingly complicated soon, I think it's still much more simple than all the available possiblities in Master of Magic.

[Japo]

Originally Posted by Scatty

I would say chess though is more simple than all the possibilities in Master of Magic. While the latter offers nearly unlimited combinations of units, computer personalities for diplomacy (however little of use it is in game), spells and random dungeons (towers / nodes etc.), terrain and ressources layout while the latter three alone can heavily determine how fast and good the opponent can develop, chess offers only, how much? 20 * 20 moves for black depending on how white moves and 20 * 20 the other way around, from which it's just a matter of calculating the possible outcomes and counter-moves and choosing the most optimal outcome for continuing.
While it is getting increasingly complicated soon, I think it's still much more simple than all the available possiblities in Master of Magic.

A chess AI can be as complicated as you want, depending on the moves in advance that you want to calculate, so much that Deep Blue was a dedicated supercomputer. Anyway you're right in part, but that only means that that the AI in MoM or Civ _would_ be good _if_ it took into account all those factors--but it doesn't.

Enemies attack you with their units as soon as they make them regardless of your defences, letting you recuperate, instead of mustering as many forces as necessary before the attack like the dumbest human player does. Computer opponents in early Microprose strategy games have all the same personality: Gengis Khan if they're stronger than you, and Gadafi if they're weaker. They didn't choose good spots for their cities, so much that often I abandoned conquered cities so I could re-found them one spot away. And even the pathfinding algorithm responsible to move your own units with the "goto" function failed very often around gulfs.

I don't doubt Microprose could have done better, but sadly these games were released before they were really ready.

[Borodin]

Originally Posted by Japo

A chess AI can be as complicated as you want, depending on the moves in advance that you want to calculate, so much that Deep Blue was a dedicated supercomputer. Anyway you're right in part, but that only means that that the AI in MoM or Civ _would_ be good _if_ it took into account all those factors--but it doesn't.

It really couldn't, not at the time. The complexity of the rules in MoM had to take into account personalities, separate spellbooks for different leader AIs, many distinct races with their various abilities and drawbacks, many buildings, heroes, all in a bewildering combination that goes far beyond chess or even the Civs. I remember reading through the spellbook and manual when MoM arrived for review and thinking, "AIs with completely different spellbooks? So many choices? This is strategic heaven for fantasy gaming!" Then I saw the other side of the coin when I observed as the AIs fumbled.

Each to their own, but I personally don't think the problem lies so much in poor AI in MoM as in designing a game whose strategic complexity meant the developers bit off far more than computers could chew back in the day. If they'd been able to release MoM as a realtime, pause-able game with the AI thinking while you considered your next move, that would have helped, but I still feel that they really needed far more processing power to get a lot more rule-crunching by the AI done in an appropriate amount of time.

(Parenthetically, Chris Crawford carefully kept the strategic complexity of his games down so the AI could handle everything competently, and he was touted for brilliant AI. What made his games fresh and interesting wasn't complexity, but a careful number of strategic elements that mimicked human emotion, and therefore, behavior.)

[yoga]

Did YOU meet hungry situation?
When your units are starving?
Hmmm

Originally Posted by Borodin

It really couldn't, not at the time.

Originally Posted by Japo

sadly these games were released before they were really ready.

Both are true, I think.
They decided to make a game of epic scope, but the ceiling (I guess) were more the financial side of making it than the machine limitations. Seeing how much bugs they left in the game, the game was finished in incredible haste. In 1.0 they had problems with the sheer game functionality, AI worries being miles away! In 1.2 the enemy stacks started to "not be frozen" on the map. Incredible.

It seems that they just could not afford to go on with 1.3 (or they realized how long it would take to make the game AI significantly better in such a complex game).

As a good comparative measure of how good / apalling AI is in different Microprose games we may use the "cheating" bonus on Impossible (for X-COM it is not really a comparison, because human player never plays the Aliens):

X-COM ............................ 1.4 x normal (stats of the Aliens) - IIRC
Master of Orion I ............... 1.6 x normal (research points) - IIRC
Master of Magic ................ 3.0 x human research, production...

That is quite telling. They had to increase the bonus from 2.0 after version 1.2 was released because the game was still not competitive enough.

---

(As for the chess comparison, Moo is much closer to it (it has only X defined clash positions, Y planets to be settled, X*Y possible movements...). And it was much simpler to make a seemingly good AI, with almost no effort.