AI questions

[Edwin P.]

1. In SC1 you could abandon Egypt and the AI would not invade and you could leave the UK bereft of any troops and send your entire navy to the Mediterrean and the AI would not invade.

2. In SC1 the Axis would never invade Norway or Sweden.

3. The AI would never attack Denmark on turn 1.

4. The AI always uses the same basic strategy to defend France, never evacuates units to the UK and never scapping a few ships to purchase a French armor unit or more corps or even a HQ unit.

5. The UK AI never attacks Ireland (a standard human allied strategy)

Are any major improvements to the strategic AI planned?

[ April 15, 2004, 07:04 PM: Message edited by: Edwin P. ]

[J P Wagner]

I sure hope so...there definately needs to be a more aggressive AI, especially, as you mentioned, in the MED....as the Axis, to virtually leave Italy undefended, throughout the entire war, without the Allied AI even taking a sniff at it, must be changed....

[Night]

The A.I in SC1 was very sub-par.

One of the main problems I brought up on the old forum was it's failure to make a strategy to defeat enemy armies, such as laying traps, using faints, etc. etc. and focusing mostly on keeping to taking cities. This made it very easy for you to smash an enemies forces and make him unable to do anything but continue to lose more forces.

the A.I. in SC2 should reconize the importance of destroying enemy forces and preserving it's own. It is a much more effective strategy.

For example the Germans failure to take Stalingrad (as a city) wasn't as damaging as the fact that they lost the whole army group there.

[Edwin P.]

I would allow the AI to know whether there are any units in a country.

Example:

No Units in Romania then the AI might consider a sea invasion from a Russian port. (I did this to a recent opponent). AI moves to coastal hex Turn 1. Invades and captures capital Bucharest on Turn 2 and Gains much needed plunder. Another strategy that the AI never uses.

I think that there are about a dozen routines that could be written for the AI to take advantage of these cases, of course the AI would not execute these routines every time, otherwise the enemy would never leave an opening.

1. Russian Bucharest Invasion.(if no units in Romania)

2. Allied Libyian Invasion.(if no units in Libya)

3. Russian Swedish Invasion. (if no units in Sweden)

4. Allied Roman Invasion (if no units in the Boot)

5. Axis Canadian Invasion (if no allied units in Canada)

6. Allied Lisbon Invasion (if no Axis units in Axis controlled Lisbon)

7. .................

They for the AI is to move to a coastal hex where it cannot be spotted before invading.

[pzgndr]

Are any major improvements to the strategic AI planned?

Yes, but ... It's going to be a while before much happens. Not clear at this point what exactly that may be. I'm sure Hubert has his programming ideas, and he's fully aware of the demand for a more robust AI. It's just too soon to say.

As I see it from my limited perspective, there are at least three parts to the AI puzzle. One part is the generic multi-level fuzzy-logic AI that is more or less hardwired into the code and needs to be flexible enough to accomodate not only SC2 but all of those other custom games folks are going to create. So there will probably be some tradeoffs between the "ideal" for SC2 and "other" games. We'll have to accept that.

Part two is the new scripted AI events we will be able to edit, and that is the biggest planned improvement. What format those will be and what capablities/limitations there may be between our scripts and the generic AI engine remain to be seen. I'm hoping we CAN include many of the various proposals to get the AI to behave in cunning and nefarious ways.

Part three would be the various FoW game options that help the AI behave smarter (or at least help compensate for known weaknesses WITHOUT cheating or unfair advantages like +2! Experience).

There are probably other aspects to consider as well, like how to get the AI to "think" about long-term research and diplomacy and such. Rest assured these are all on the radar screen. Let's keep our fingers crossed and hope for the best.

[Edwin P.]

Thanks for the update on the AI and the potential for a cunning and nefarious AI.

[Edwin P.]

One item that has always bugged me is that the AI does not respond to an Axis naval strategy.

For example - if the Axis has built loads of Subs the AI will never build a naval force to counter it. Thus I can build a wall of Axis subs to cut off the British Isles from America.

Perhaps something as simple as a ratio of Allied Warships to Axis Warships in the Atlantic might trigger a Control the Ocean Strategy.

If the Axis Navy to Allied Navy Forces in the Atlantic is greater than 1.5 to 1 then the US AI will execute a control the seas production strategy until the ratio is less than 1 to 1.

What might this production strategy be for the AI?

1. Build Cruisers

2. Build Bombers

3. Recall Units from the Pacific (a strategy available only to the AI?)

4. Research Naval Bombers

Of course the AI will need an tactical strategy to go along with this.