AI Query

[gavril]

I'm really pleased to hear about the new game. I suppose I'm also a bit wary, because I found FoG really disappointing in solo mode, with a pretty dumb AI. I know btw that that wasn't everyone's experience, just my own take on it!
From what I've seen so far, inc the streamed demo, it certainly looks as if the Pike & Shot AI will provide a tougher challenge. Just wondering if anyone who's helped write or beta test the game can say anything about how they found the AI?
Thanks in advance,
Jay ('Gavril').

[Galangalad]

Well, the AI beat me four times and I could only win Ravenna.

If you are tired fighting the AI, try other players. You can create multiplayer challenges or accept challenges

[gavril]

Well, the AI beat me four times and I could only win Ravenna.

If you are tired fighting the AI, try other players. You can create multiplayer challenges or accept challenges

Thanks for that, sounds like the AI may be tougher then?

Re multiplayer, that's not really for me, or at least it wasn't with FoG. My take on it - which is no doubt an idiosyncratic one - is that multiplayer games are too hard to win against players who have made a detailed, almost forensic study of the program and its eccentricities; and that it ceases to relate in any meaningful way to the historical warfare that is supposedly being modelled. I enjoy going up against a tough AI; I don't enjoy being slaughtered by people who've devoted most of their free time to finding out how to squeeze the last ounce of advantage out of a rather whimsical computer program...

[Schweinewitz]

@ gavril:

What I saw in Surtur's revealing video and read in several reviews the AI seems to be quite capable, and this is good news.

You might check the articles below for more impressions:

http://www.wargamer.com/article/3674/fi ... and-musket
http://www.pockettactics.com/news/ios-n ... pike-shot/


@Galangalad:

Solo player here, and I think I'm not the only one. One I day I got really fed up by hearing the same sentence in those endless FOG discussions over and over again: "Try multiplayer!" In the end it was only a weak excuse for FOG's inferior AI. And right now I'm getting a kind of allergic reaction from reading something similar here. - I honestly hope that this game's AI is much better than FOG's; but please understand that I'm a bit wary too ... Well, at least it's a kind of relief to hear that you've been beaten by the AI several times.

[gavril]

@ gavril:

What I saw in Surtur's revealing video and read in several reviews the AI seems to be quite capable, and this is good news.

You might check the articles below for more impressions:

http://www.wargamer.com/article/3674/fi ... and-musket
http://www.pockettactics.com/news/ios-n ... pike-shot/


@Galangalad:

Solo player here, and I think I'm not the only one. One I day I got really fed up by hearing the same sentence in those endless FOG discussions over and over again: "Try multiplayer!" In the end it was only a weak excuse for FOG's inferior AI. And right now I'm getting a kind of allergic reaction from reading something similar here. - I honestly hope that this game's AI is much better than FOG's; but please understand that I'm a bit wary too ... Well, at least it's a kind of relief to hear that you've been beaten by the AI several times.

Thanks for the links mate, very helpful. Totally agree with your comments about Multiplayer! I think that if FoG had been designed with a challenging AI it would have become an all-time great, and a money spinner for Slitherine.

[Galangalad]

Be sure that the AI will be up to the task for you single player gamers. I usually don't play against random people either so I hear what you are saying.
I am not too good at the this game but I can usually defeat a bad AI.

[gavril]

Be sure that the AI will be up to the task for you single player gamers. I usually don't play against random people either so I hear what you are saying.
I am not too good at the this game but I can usually defeat a bad AI.

That all sounds very promising! Likewise the reviews. It sounds as if the game is very mod-able too and no doubt that will help. I suspect I'll shell out for the core game and keep my powder dry (geddit???) until I've worked out how well the core game plays solo...

[rbodleyscott]

From the beta board:

the AI is the best I have ever met in a wargame

I'm curious what people's impressions are. The AI, on the middle setting, is beating me more often than I beat it. That's rare. Am I just not good at this game, or is the AI superior to that of other games?

Except for the tutorial, I am currently losing every battle I play.

Part of it is no doubt learning the mechanics, I am struggling with the mental challenge of the timing of charges, distance need to "prep" the enemy etc from years of gaming in a hexagonal frame of mind to squares!

I have not yet had the AI do anything "bad" with its units which makes for no freebie kills.

I appreciate that the AI uses the same rules as us, which was my impression. That just emphasizes the quality of the AI.

the AI ... plays very well. It is very good at finding and attacking your flanks, for example, and I have never seen it doing anything stupid.

[gavril]

Thanks for that Richard. Glad to see how deeply involved you are with the development of the game. I know you set high standards, having had the dubious pleasure of being trounced by you in a Berkeley Hordes tournament a few years ago!

[IainMcNeil]

Richard is not just deeply involved he coded the game entirely! It uses our BA engine but everything beyond that was designed and coded by Richard.

[rbodleyscott]

Richard is not just deeply involved he coded the game entirely! It uses our BA engine but everything beyond that was designed and coded by Richard.

Though not, of course, the graphics, which were done by Rob Graat.

[Adraeth]

Concerning the AI during the beta test i made tons of skirmishes against it, i must admit it is really difficult to win against AI.

The AI itself plays like a very good opponent on a tabletop wargame, for sure humans are humans but you won't be disappointed by this AI

[gavril]

Concerning the AI during the beta test i made tons of skirmishes against it, i must admit it is really difficult to win against AI.

The AI itself plays like a very good opponent on a tabletop wargame, for sure humans are humans but you won't be disappointed by this AI

The more I hear, the more I like! I'm going to be buying it for sure! A decent AI and a flexible editor is all I need to keep me happy

[gavril]

I've been playing the game since it was released, off and on, and I have to say that I'm really enjoying it. The AI is giving me a tough fight even on the default setting, and as someone wrote elsewhere the game is very unforgiving of mistakes. Hats off to Richard Bodley Scott and the development team for producing a really challenging game with depth, subtlety and a lifetime of modability built in! Thankyou guys!

[rumguff]

People need have no concern about the quality of the AI, its kicking my ass most times on the second difficulty level of five. Glad to hear its not simply low user IQ on my part!

Can the developers comment at all on the class of algorithms used for the AI? Certainly there seems to be a lot of processing going on behind the scenes since there are many pregnant pauses in the game play that cannot be graphics related. I also wonder what the split between tactical and strategic decision making the AI uses; I have not played enough games to get a sense for its understanding of strategy, or whether there is a difference in the AI's approach between random and historical scenarios.

I have also read that the difficulty settings don't change the AI quality, merely the balance of forces in the scenario.

[rbodleyscott]

Can the developers comment at all on the class of algorithms used for the AI? Certainly there seems to be a lot of processing going on behind the scenes since there are many pregnant pauses in the game play that cannot be graphics related. I also wonder what the split between tactical and strategic decision making the AI uses; I have not played enough games to get a sense for its understanding of strategy, or whether there is a difference in the AI's approach between random and historical scenarios.

Broadly there are two distinct levels of algorithm.

The lower level AI algorithms control the behaviour of each individual unit within the context of the "orders" specified by the higher level of algorithms: e.g. which enemy unit to choose as their optimal target, how far to move towards them (or away from them) and the avoidance of putting themselves in positions where they could be flanked.

The higher level algorithms control the "orders" that each section of the army (team) will use: Teams vary in the historical scenarios, but in skirmishes they consist of main infantry body, each cavalry wing, each light troops wing, cavalry reserve and centre light troops.

In most of the historical scenarios, the AI controlling the team "orders" is scripted according to the historical situation, and will issue different orders to each team during the course of the game according to the actions of the player. Mostly the teams have activation stages, and certain conditions will activate the next stage (or one of a number of branching stages) with orders to match.

In the skirmishes, and in a few of the historical scenarios, the AI uses a function called AI_Masterplan() which decides the team orders on a turn by turn basis depending on the battlefield situation. It works on the basis of assigning the player army to teams similar to those of the AI side. However, it reassigns the player teams each turn. So, for example, if you moved most of your right wing cavalry over to your left, the AI would reassign them to your left wing cavalry team and plan accordingly.

The team orders specify when the team should advance, where they should advance to, how single-minded they should be in moving towards that destination, and what stance they should take against various opposing unit types (light troops, non-light infantry, non-light cavalry) along the way and when they get there.

I have also read that the difficulty settings don't change the AI quality, merely the balance of forces in the scenario.

That is correct, although it isn't always the balance of forces. In Lutter, for example, the difficulty level controls how demoralised the Danish forces are at the start of the game. (On the 2 easiest levels they are not demoralised at all).

We wanted the AI to be as good as we could make it, so it would seem a shame to nobble it for the lower difficulty levels. And it would be hard to make it better than "as good as we could make it" for the higher difficulty levels!

[rumguff]

Thanks Richard, fascinating answer - I'm really glad I asked the question now. Hopefully this info might lead to my defeating the AI more often.

One more question if I may - were the AI settings hand-tuned play testing or was it trained automatically using labelled data (e.g. selecting the AI parameters based on how well that set of parameters reacted to a suite of battlefield situations)? If its the latter it may take human players a long time to get as good as the AI.

[rbodleyscott]

One more question if I may - were the AI settings hand-tuned play testing or was it trained automatically using labelled data (e.g. selecting the AI parameters based on how well that set of parameters reacted to a suite of battlefield situations)? If its the latter it may take human players a long time to get as good as the AI.

No, the upper level of algorithm is a hand-crafted rules-based system - obviously informed by play-testing. Though most of the low-level stuff analyses the odds using the actual combat resolution algorithms - which are not simple. That and checking for flank threats requires a fair amount of processing.

[gavril]

Like Rumguff says, fascinating answer Richard!

TBH the experience with Pike & Shot is sooooooooo much better than playing FoG against the AI. Is it verboten to ask if you'd consider - at some stage - working on an ancients or medieval offering based on the Pike & Shot engine?!?

[rbodleyscott]

Like Rumguff says, fascinating answer Richard!

TBH the experience with Pike & Shot is sooooooooo much better than playing FoG against the AI. Is it verboten to ask if you'd consider - at some stage - working on an ancients or medieval offering based on the Pike & Shot engine?!?

I come from a background of Ancient/Medieval wargaming so it does seem likely. What gets done first, however, has not yet been decided.