"A Human-Level Tactical Artificial Intelligence at Ligny" Topic

"The seeds of Napoleon's defeat at Waterloo were sown two days earlier at his victory near Ligny. Napoleon needed to surround and completely remove the Prussian army as a viable force on the battlefield. Instead, they escaped to Wavre in the north and resurfaced at the worst possible time on Napoleon's right flank two days later at Waterloo.

MATE1) 2.0 is now capable of analyzing the battle of Ligny, June 16, 1815 from both the Blue (L'Armée du Nord on the offensive) and Red (Prussian on the defensive) positions. (MATE is the AI behind General Staff: Black Powder…"

Main page

link

Armand

This is so interesting and stimulating that you get a big DANKESCHOEN.

(smile)

Armand

Call me when it can do a backhand blow.

It does show how very, very difficult to program this kind of thing is. As far as I can see, for the AI to work, someone has first to assign subjective victory points to various objectives and units, and determine victory conditions in terms of those points. This gives it the type of data it can work with, but it then proceeds to plan on the basis that destroying units and accumulating victory points achieves victory. The obvious question is, Well, does it? If its Ligny plan succeeded in destroying x number of units, has it won what a historian would recognise as a victory? Presumably if it destroyed an unengaged Landwehr unit off to the east, that would count for as many points as if it had destroyed one in the centre of the line. If it does not win equal points, did someone subjectively program those units with different values, or can the AI work this out for itself? And what if it destroyed one in the centre of the line, yet the line still held?

Then there's the question of reactions. Everybody has a plan till they get punched in the mouf, so I wonder how this AI handles the enemy unexpectedly attacking, instead of sitting on the defensive as planned?

There seem to me to be too many variables for this ever to work in the way it's being done. Kudos to him for the effort but programming a chess computer would have to be easier: victory is always by checkmate, it's IGO-UGO and all possible moves are known in advance. None of that is true here.

I would have thought some sort of Monte Carlo analysis would work better, where the AI works out every possible outcome from the starting positions thousands of times, then picks the strategy than wins most often. You wouldn't just go for the tallest spike in the output, you'd go for the tallest spike that's surrounded by a lot of other spikes. So even if you miss the outcome you were aiming for, you get one that's still not bad.

Thanks…

Armand

The victory conditions include fixed numbers of each unit type killed.
The AI doesn't decide what those conditions are. The scenario designer sets them.

I wrote the army, map and scenario editors for General Staff. I'd have written the game as well but I don't have much spare time nowadays. I also have significantly different ideas on how the game should work to Ezra.

"Then there's the question of reactions. Everybody has a plan till they get punched in the mouf, so I wonder how this AI handles the enemy unexpectedly attacking, instead of sitting on the defensive as planned? "

If you know you are playing against an AI, once you get a 'feel' of how it plays, you can usually lure it into traps and crush it (though it might eventually learn… ;-)