Suggestion: Genius Level AI - Expert Level AI with a Random Bonus

[Edwin P.]

I proposed a new AI Level called Genius where the AI receives a randomly determined bonus.

Thus not only does the AI fight better but it receives a game changing bonus that varies from game to game and makes a game against the AI a bit more unpredicatable.

Axis AI Genius Bonus may include:

a. Germany receives 3 Free tech chits or

b. Germany starts with an extra 3 subs or

c. Germany starts with 3 Rocket units and level 2 Rockets or

d. Germany starts with 3 diplomacy chits or

e. Spain (or Finland and Sweden) joins Axis after Axis Minors Join

f. Italy starts with Gun Laying Radar Level 1 (as they shared tech with Germany) or

g. Neutral Turkey annexes Iraq, stopping one lend lease route to Russia or

h. Italy starts with 1 HQ unit and 1 Engineer Unit or

i. Axis spy steals UK codes, Axis AI spots all UK Naval Units for xx Turns or

j. Germany receives a bonus Engineer and Paratrooper Unit at start or

k. Germany starts at a full war economy or

l. Axis AI receives 2 of the above bonuses.

Allied AI Genius Bonus may include

a. US starts with 2 bonus HQ Units or

b. US enters war with 2 Carriers from their Pacific fleet or

c. US enters war with 2 bombers or

d. US (or USSR) readiness jumps an extra 10%, so it enters the war sooner than expected or

e. UK receives 3 bonus corps in Egypt - ANZAC, South African and Indian or

f. USSR Units start three hexes back from the border with Germany or

g. Turkey Joins Allies after Axis takes Rostov or

h. Ultra - Allied AI Can spot all Axis Naval and Air Units for xx turns or

i. UK starts with bonus Engineer and HQ unit or

j. Allied AI receives 2 of the above bonuses.

Of course, the bonus(es) should vary by scenario.

[ May 16, 2004, 12:50 PM: Message edited by: Edwin P. ]

[pzgndr]

It's kinda difficult to have a decent discussion about the AI since nobody but Hubert really knows what goes on inside that black box. And nobody but Hubert really knows how SC1 interfaced with the AI. So without knowing exactly what the issues are, it's difficult to propose solutions.

With AI scripting in SC2, this should all become more transparent. We might still not understand what goes on inside the generic AI black box, but at least we'll see the campaign script inputs and judge results. Hopefully the scripting will allow us to establish some historical priorities here and there.

Remember also that we will want this generic AI to work for alternate games, and be able to respond smartly to our own ahistorical strategies. So I think it's more important to get the AI to "think" well enough on its own regarding all sorts of combinations involving research, diplomacy, military production, offensive and defensive strategy, the full spectrum of land/air/sea operations, etc. etc. That's a tall order to fill. Too many contraints and limitations on the AI's "free thinking" may just cause it to do stupid things. That "law of unintended consequences" is pretty inconvenient, yes?

[Edwin P.]

I agree, without knowing about the AI its hard to make a recommendation.

I was thinking more along the lines of being able to script a randomly determined bonus that the AI would get at a specific AI intelligence level in a particular scenario so you would not know exactly what you were facing, as the AI (in SC1)tends to follow the same build and combat strategy in a scenario.

For example at Genius AI there would be a series of text lines for the AI to select a bonus for the AI in each 1939 scenario, for example: 01 Axis [Ger Subs +3 Units], 02 Axis [spain Joins Axis Turn 23], 03 Axis [Ger Rocket +3 Units], 11 Allied [uS Readiness +10%], 12 Allied [uS Bombers +3 Units], 13 Allied [uS Bomber Tech +2, US Long Range Tech +2]

One improvement I hope to see is an AI that will choose between a wider range of comparable strategies and internally track which ones are more successful, so it can adapt to different games. For example if in Scenario 1 a higher level of research (or more armor units) pays off then it will tend towards using that in future games in Scenario 1. If it does better with no research in scenario 4 then it will place a lower priority on research when playing scenario 4.

[ May 16, 2004, 02:41 PM: Message edited by: Edwin P. ]

[pzgndr]

internally track which ones are more successful, so it can adapt to different games

This would be nice. Probably beyond current scope and capability, but a neat idea. Besides, if Rambo ever met an AI that could truly adapt to his game, then where would we be?

If the AI scripts allow us to set certain parameters such as strategic versus operational versus tactical offense/defense "values" (weighting factors, etc.) then we should at least be able to experiment with different values and judge whether AI performance is better or worse. Sort of like tuning your lawnmower with a rusty screwdriver based on how the engine sounds. Not very scientific, considering all of the variables and randomness in a given game, but hopefully we'll be able to optimize AI performance.