[SURVEY] Do you really care about the A.I.?

[norvandave]

I won't buy the game unless it has a good AI.

[Hentzau]

Originally posted by norvandave:

I won't buy the game unless it has a good AI.

Ditto! Might as well be using vassal, cyberboard, or ADC2 otherwise. Plus I'm sort of anti-social. lol

I'm hoping that SC2, with its associated editor, is sort of a replacement for those, but with AI.

[pzgndr]

A "good" AI is in the eye of the beholder. SC2 will have an AI, so it's a moot point to argue about whether it should have one or not. And no need for discriminating solo gamers to be concerned about whether to keep SC2 on the shopping list. I am very optimistic that between editable scripted events and editable AI scripts, players will have a challenging computer opponent. Just how variable and random it may be, or could become, remains to be seen.

I've commented before about SC1, about how the AI is very good at what it does but is incomplete in many ways - Mediterranean strategy for both Axis and Allies, Battle of the Atlantic strategy, etc. The new AI Planning scripts should address most all of the significant strategies and some viable alternatives. Where some weaknesses may exist, we should be able to compensate with event scripts. Various event triggers could be more or less random, resulting in interesting and variable gameplay. I'm expecting to have a lot of fun with all this, as is everyone else.

[Codename Condor]

You are all forgetting this is a 1-man-PC-game, and by the looks of it, its gonna be quite complex (much more complex than SC), no way AI is gonna kick asses here.

In my opinion, there's no way to get a challenging AI, so the option lies in making possible several AI advantages, fog of war issue like JJ stated, better rolls, more supply, more strengh,...and of course it will need more scripting because in a few days there are gonna be some interesting moves the AI should know how to counter.

Would be good to have a huge community here with fellows with good programming capabilities to make the game better.

[Mockylock]

I agree that it almost impossible to create a good AI in a complex setting.

That is why you need to have the AI learn from experience/already played games (as stated in another thread). If not, it's just a matter of time before our 'intuitive brains' have learned the underlying AI function, and then the AI is toast for good. The better the implemented AI, is just a matter of delay on our learning of the AI (in the no-learning case).

Therefore there is only one way as I see it - give the AI information on what strategies leeds to victory, and take actions accordingly (which is log the games, and do pattern recognition on the strategies etc). So, when a new strategy paradigme between players occurs which will lead to a lot of victories, the AI will also learn it and (in time, based on the frequency of observations) react accordingly. It is very complicated to implement yes, but the theory is there. And I personally would really like an AI that evolves over time. Not saying the AI would be perfect - but definitely better - and most important more intuitive and better suited to react on new strategies (and current strats)

I am no AI expert, but I beleive the concept is already implemented in shoot 'em up games and in games like AOE (but there the AI only learns in-game (I think) and that is a different story.

Sorry if there are some overlap with the other thread.

[pzgndr]

In my opinion, there's no way to get a challenging AI

It all depends on what you expect. If you want a truly random human-like computer opponent, then you're not likely to see it. And if the AI did choose a low-probability off-the-wall strategy and handed you a less-than-satisfying game experience, then what good would that be? If you want a tough opponent that challenges you to meet the victory conditions (ie, provides you a good chance to lose the game) then I believe that's achievable.

And I personally would really like an AI that evolves over time.

I'll play devil's advocate here for a moment. A problem with this is that the game itself and also the players are evolving over time. So any early head-cracking strategies that take advantage of game bugs or other weaknesses probably will not be appropriate later. It's important to get all the game mechanics and play balance issues resolved first, and that will take time. After that, the AI can be fine-tuned as necessary.

[JerseyJohn]

[Mockylock]

quote: And I personally would really like an AI that evolves over time.

I'll play devil's advocate here for a moment. A problem with this is that the game itself and also the players are evolving over time. So any early head-cracking strategies that take advantage of game bugs or other weaknesses probably will not be appropriate later. It's important to get all the game mechanics and play balance issues resolved first, and that will take time. After that, the AI can be fine-tuned as necessary.

Yes I agree on that - only games played after patches have been applied should be logged. Imagine if all top player games after SC1 patch 1.07 were logged and analysed - I am pretty sure AI performance would increase significantly. Even player styles could be found, and used by the AI to counter a specific player. Etc..

Edit: SC1 was really great, the best game I have ever played, maybe except civilization. So I will buy SC2 no matter what. Very good job Hubert.

[ December 06, 2005, 09:05 AM: Message edited by: Mockylock ]

[Codename Condor]

Originally posted by Mockylock:

Haha, u havent played Rambo, have u? his mind is so chaotic in the battlefield the AI would go crazy playing against him. ROFL.

[Mockylock]

Originally posted by Codename Condor:

</font><blockquote>quote:</font><hr />Originally posted by Mockylock:

Haha, u havent played Rambo, have u? his mind is so chaotic in the battlefield the AI would go crazy playing against him. ROFL. </font>

hehe no, I havn't played Rambo. But actually the AI could benefit, seeing that Rambo is totally random, it would maybe develop a more cautious playing style against him?

The point is, to have the AI act as the human brain, reacting to pattern recognition in the observed environment.

To Hubert and alike -> Neural Networks, Evolutionary algorithms etc...

[Edwin P.]

1. Writing a learning AI is very time consuming.

2. Though it could be done in a simplified fashing by having the AI track which combination of strategies lead to the fastest surrender of France, for example. It would then be more likely to select these strategies in the future.

Example:

Strategy 1: AI attacks Low countries and takes Paris in 12 turns in 3 games

Strategy 2: AI attacks Denmark then Low Countries and takes Paris in 7 turns in 3 games.

The AI would thus be more likely to choose Strategy 2 in future games. If over time the success of this strategy declined it would increase its chances of selecting another strategy.

The key issue with a game like SC2 is that there are so many variables to track.

Does one only track the countries one conquers before taking France or does one also track which units one produces or how many units defend France.

3. It is even simpler to give the AI a range of strategies to select from without progamming the AI to learn from its actions (and for most game oriented AI's their learning ability is quite limited) and to update this over time as new patches are released and players write new AI scripts.

4. HC has said that the AI should be able to do a few things well. I agree, if it can do so in an unpredictable fashion.

[ December 06, 2005, 01:11 PM: Message edited by: Edwin P. ]