Would AI be more competitive?

[henryjai]

In Vicky 1, I don't have much complains about the military AI as it wasn't real bad. But! The AI don't act competitive enough, for example,

1. They never gang on a badboy, the nations with smaller military ability act like cowards even when they have a chance to hit the badboy badly.

2. They are reluctant on protecting their own interests, for example, USA should be trying to protect the countries in America, right now they only protect Mexico and Canada.

3. They sometimes have unrealistic war demands, such as getting a bunch of useless provinces to further cripple their economy (for the case of France)

4. They never try to reform their government

5. They never try to civilize, countries like China are happy with an army of irregulars.

6. Very few countries would really be engaged in a naval race, they are happy with a fleet of raiders in the 1930s.

 

[Alexander Seil]

Well, compare it to the AI in EU3 and Rome. Playing the Timurids and suddenly realizing that you have a land border with Ming is terrifying.

 

[unmerged(138342)]

In Vicky 1, I don't have much complains about the military AI as it wasn't real bad. But! The AI don't act competitive enough

The system of relation points contradicts with the notion of an egoistic self-serving AI. I would really cherish the day when Paradox adopts symmetry in this regard.

 

[unmerged(75409)]

Victoria 1 was programmed with the old engine, that wasn't so good at acting outside of its AI settings. Basically, for Vic1's diplomatic AI you specified a list of countries that the AI likes ("befriend"), a list of countries for which it should issue guarantees ("protect") and a list of countries towards which it should pursue aggressive politics ("war") - i.e., declare war on them if circumstances were favorable. For each country on each of the lists, you could specify a priority between 0 and 200. That was basically all you could do.

The AI was not capable of evaluating the diplomatic impact of most actions outside the direct diplomatic choices. Its fortress build priority was determined by a user definable list similar to the "war", "protect" and "befriend" lists, and did not consider circumstances like friendly relations or alliances. Also it did not understand strategic priorities, for example in HoI2 you could invade Cuba, Newfoundland and Bermuda in preparation for an invasion of the USA, and the US AI remained clueless about your intentions until you actually declared war on them.

From all I have heard, the EU3 AI seems more capable of independent decisions, especially those that have to take diplomatic and economic context into account. However it probably won't be perfect either so don't get your expectations too high.

If the HoI3 system of LUA AI scripts is used then I imagine modders would be able to implement methods by which - for example - the Chinese AI would pursue the step to civilization, i.e. building industry, pushing the military expansion to get the necessary score, etc. Also IIRC the devs said that the missions system from EU3 will be used to guide the AI towards country-specific goals.

However I doubt it will be easy, there is to my knowledge no way to write a script saying "Get 50 econ score, 50 mil score, and research printing press". You have to go into much more detail to actually tell the AI what to do, and fine-tune the LUA scripts to the specific purpose that you have in mind.

 

[henryjai]

I think...the military AI needs one thing, that is to think one more step. They often leave their dug-in positions/fortifications easily. They should think:

"Would the enemy encircle my troops if I moved my troops away from X? (Think also for possible sea-invasions)" - Check if the enemy would have a possibility of cutting you off. If yes, the AI should hesitate to risk his troops unless he could afford losing them.

"Do I need to leave a division in this chokepoint?" - Check if this province is strategically important. A chokepoint hill forest province should have a division left so the enemy won't sit there after you left.

"Is it favourable to move onto that kind of terrain? " - Check the combat odds of counterattacks for the destined province before moving in. I don't move my troops out of my all-owning mountain province and went to battle my enemies at plains!


That is generally what I think to make the AI much tougher than it was in Vicky 1. The Vicky 1 AI isn't a bad one except they never think one more step further.

 

[unmerged(138342)]

In the first days of Vicky I computers were much slower. A good AI is costly, it wasn't really an option having a good AI and still having a playable game.
But now it is another story. We definitelly would expect much more from the AI department.
I remember a few months ago having to bruteforce an MD5 hashed password on my Athlon 64 3000+ processor ~ 1.8 GHz. The program was checking between 7 and 8 million hash codes per second. When I tried it on my friend's computer, which has an i7 CPU, each core running at more than 3 GHz. It was able to check 300 million hashes per sec. I was quite astonished, I did not expect that much of a differnece.

 

[Colon]

In the first days of Vicky I computers were much slower.

Indeed, I originally ran Vic1 on a 700Mhz Duron CPU with 32MB RAM.

 

[Chaingun]

First: I think Vic2 AI will be vastly smarter than Vic1 after a couple of patches, primarily for reasons Leviathan and Sogartar stated.

I think...the military AI needs one thing, that is to think one more step. They often leave their dug-in positions/fortifications easily.

The problem is that it is very hard to formulate such "thoughts" as efficient algorithms with somewhat linear time complexity (time of execution should ideally follow number of provinces, armies, etc). Furthermore, it's nearly impossible to think of all cases. For instance, with your cases I can think of several counter-examples where your logic could become flawed:

They should think:

"Would the enemy encircle my troops if I moved my troops away from X? (Think also for possible sea-invasions)" - Check if the enemy would have a possibility of cutting you off. If yes, the AI should hesitate to risk his troops unless he could afford losing them.

An AI always leaving near cut off is probably open to exploit by human player (fooling AI to leave when human did not intend to attack). Also it's quite hard to formulate an algorithm for "almost cut off".

"Do I need to leave a division in this chokepoint?" - Check if this province is strategically important. A chokepoint hill forest province should have a division left so the enemy won't sit there after you left.

When exactly in mathematical/algorithmic terms do we know if it does? There's a huge amount of ambiguity baked into those sentences.

"Is it favourable to move onto that kind of terrain? " - Check the combat odds of counterattacks for the destined province before moving in. I don't move my troops out of my all-owning mountain province and went to battle my enemies at plains!

You risk getting the AI too stubborn of defending mountain provinces. Chances are it will make either kind of mistake (leaving too soon or too late) depending on what the weights are. Of course, here tactical thinking and being "one step ahead" would help to override flawed strategic assumptions, but it turns out that "one step ahead" is more like analyzing the possibilities of my turn -> your turn, and each turn (which happens to be incredibly short a tick in Paradox games), can involve an enormous number of units because making one move doesn't end the turn (not that'd it matter much with such a high update frequency).

That is generally what I think to make the AI much tougher than it was in Vicky 1. The Vicky 1 AI isn't a bad one except they never think one more step further.

And quite likely they won't under many circumstances, there's no way too do searches like in chess that exposes good tactics at a sufficiently high frequency. You're right about the problem assessment but unfortunately not close to a solution.

 

[unmerged(138342)]

I know there would be a script language exposing some interface for programming an AI, but beeing a programmer myself, I know the performace drawbacks of such a system are big. For example in LUA data duplication due to difference in representation is major problem.
It would be really great to have also native interface for the really hardcore guys in the community. My assumtion is that such an interface would not be left unnoticed and much of the work would be trasfered to the community, which would save Paradox money and we, the gamers, would have a better AI to play against. A good example of such a game is Civilization 4. Although its interface is mostly not documented.

 

[Alexander Seil]

First: I think Vic2 AI will be vastly smarter than Vic1 after a couple of patches, primarily for reasons Leviathan and Sogartar stated.

You can make a game harder just by upping opportunistic aggressiveness. I.e., not just the computer attacking you because you're human, but attacking you because you are already tied up. This kind of behavior is very typical of Rome and EU3 AIs in my experience. Of course a player is then tempted to reload, pre-empt the attack and then go and complain on the forum about how easy the AI is I've just had this temptation in my Epirus game in Rome, but I persevered (I was rewarded with the hated King of Macedonia being deposed in a bloodless military coup and fleeing across the border...right into my dungeon ). The perfidious Illyrians struck me while I was busy protecting the Aetolian League from Macedonians, in the end I've lost my entire experienced army, all my manpower, and have to rely on mercenaries now - I barely avoided ceding any territory or paying tribute because the Illyrians got raped by a massive barbarian invasion while besieging my capital :wacko:

 

[Chaingun]

Hehe, that's somewhat recognizable.

Without reloads, I found EU3:IN AI quite or extremely difficult as certain medium and small sized powers on VHard (nothing will help with difficulty if you play as e.g. France however). I'm no expert of EU3 particulars though.

While on the subject, they added Realistic save mode to Mount & Blade for a reason... Turns out I ended Ctrl+Alt+Deleting myself out of the worst cases regardless.

Game mechanics featuring a large amount of randomness for whether something triggers or not (like a war declaration triggering on 50% of reloads) tend to encourage save reload cheating.

And yeah, it might very well be that the smartest and most opportunistic AI opponent is perhaps not very fun to play against, simply because it will be so dastardly.

 

[Alexander Seil]

I can't say I don't enjoy it, but a lot of players will, indeed, reload or restart when problem arise, so the complaining about bad AI or general ease with which victory is invariably obtained by a human is, in part, indicative of very poor sampling. And it's not just that there is arbitrary randomness - i.e., you might lose an important ally, a heir or a key official to disease, battle, assassins, suicide, etc. in Rome or CK, but also that the player benefits from a save/load function that lets him avoid these - the AI does not.