even if there were no negotiations (imho) risk is more boring than diplomacy because diplomacy requires such a higher understanding of tactics than risk. in risk it seems like a more russian army approach by trying to overwhelm the opponent with nothing but more and MORE armies.

an AI might do ok in the no press games we sometime play here.

I didn't say risk wasn't boring, but only that if there's no prospect of joint action or alliance, you won't be able to do very much as you have to defend all your SCs almost all the time. If there's an Austrian in Serbia, a Turk in Bulgaria, and a Russian in Rumania, what do you do? If you trust one over the other, you can support his probable move, but without negotiations there's way to differentiate. The only way to settle it would be to program the AI to always prefer attacking the strongest player, but I think that will end up in the long term in a dynamic stalemate.

so nobody has played the diplomacy PC game? i wouldn't recommend it but it may get better in the future.

its old fashioned..... and sadly there a a few flaws...sometimes it freezes in between turns... and its ALL backstabbing, no REAL aliances for more than one move

why would we play pc version when there's phpdiplomacy.net... i'm confused

I've got the (newish) Paradox PC Diplomacy game on order (it was remaindered on eBay at about A$5 so I thought I'd see what it's like). As regards the older MicroProse game, the consensus seems to be that it's a realistic simulation of a game against six students from a "special" school.

Basically you play with six arguing Smart Jasons. (No offense)

Sorry to Samael, the Paradox version is also severely limited by the AI quality.

The Paradox version has also flaws on the adjudicator. At times the AI is capable to use armies to support fleet's moves to water provinces (I saw it happening 2 times).

Gobbble: Playing 6 Smart Jason's would be impossible - playing 3 in one game is hard enough. Having said that he said he'd stop and I haven't seen any negative traces of him since.
All the computer versions I've seen are flawed as it is hard to teach a computer that you can build up trust. As you say they just stab madly!

figle, would it be hard to code a simple AI for non press games?

Very hard.

You'd need some 'trust' variables, and actions will change these values. But it's hardest to get the computer to play sensibly. It seems they don't know how to support holds and moves properly.

I think for non press games a decent AI could be made, for there isn't much trust to be built.

It would be hard to do yes, but I had considered taking a little fork off and having some fun trying to write an AI that would do things slightly logical instead of the CD players. The reason I say slightly logical is that actually if they don't always do what you'd expect they could be very potent. If there was a fully solo game then you could write a good AI for it (if you were very patient!), but as soon as any communication comes in the task becomes hell - if it's only as simple as getting the computer to recognise when you support-hold one of its units and are trying to make a hint.

I definitely don't think AI and Diplomacy mix well; it's hard to think of a game worse suited for AI. phpDip especially wouldn't handle an AI well, since it's on shared hosting and AI requires a lot of computer power.
Specifically most Diplomacy AIs will have the computer enter large permutations of possible orders that enemies could enter, and possible orders that it could enter, and adjudicate these permutations to figure out the best choices.

phpDip's adjudicator isn't slow, but being in PHP and being complex more emphasis was placed on maintainability and correctness than speed, so an AI trying lots of move possibilities would sap a huge amount of computing power.

I think AI for phpDip can definitely be completely ruled out :-(

A little AI managing civil disorder countries (in games which start with less than 7 people) could be made. Nothing more than some standard moves and then holding.

For example here at tournaments if there is a table with 6 players the cd country is turkey and it does the following: Con->Bul, Ank->Con, Smy->Ank then support holds until it's dead (in 1901 it builds an army in Smy). I believe something like this could be done.

The CD country would be Italy, but if you think about it, the country is in civil disorder, so military coordination isn't reasonable, even for a support hold.

Which is how the rules are written anyway...
I think that an AI would be disadvantaged in no-press games as there is still communication isn't there? (Limitations of phpDip aside, I was just talking in general) If you did have a lot of time, you could build a complex interface where you can converse with the computer your intended moves goals etc... and it would build a trust rating and compare the likelyhood of you stabbing with the advantage of stabbing you... this would be open to flaws... but it's a start.

AI for diplomacy will only work once the level of AI engineering passes the Turing Test. Since that hasn't happened yet, I'm won't be playing AI diplomacy.

Theoretically, a functional AI for a game even as complex as Dip wouldn't be as difficult as it may seem. From a programming standpoint, I've found it's often helpful to differentiate between AI and SI (Simulated Intelligence). The latter would be similar to other NPC's in games, like Chess, Civilization, Poker, etc. Just think of how far ahead the typical human player forecasts, and shoot for that. Since the NPC isn't going to be doing your dishes or asking you about your day afterwards, you really don't need something that could pass the Turing Test. All you need is enough logical variables, along with some slight randomness, to simulate intelligence in the context of the game tactics.

The tricky part, as pointed out, lies in the press. But a simple way around this would be to do what others have done in the past with negotiable NPC's; if a country is run by an NPC, replace the chat text field with a complex series of selectable communication options. This would eliminate the need for any kind of intelligent language parsing. Then the NPC decides on a response based on another large set of factors; unit placements, observed aggressiveness of the human player, trustworthiness, randomly-generated personality attributes, etc. Though this does amount to a large number of variables, if coded properly we're only talking about a fraction of a second's difference in terms of processing time. Math functions in general really haven't been much of a speed issue since the advent of the math co-processor. What will slow down a script's execution is excessive database queries, external socket connections, complex image manipulation, etc. But a few lines-- or even a few thousand lines of additional ifchecks and variable calculations should not make any noticeable difference.

So I guess to summarize, it certainly can be done, but it would require a lot of alpha testing (aka "trial and error") to get it just right.

why not just play with people

sean, i played a few times using the paradox one. it was god awful for anything other than tactics. i went through the horribly long "negotiation" tutorials only to find that the ai never seemed interested in my proposals and would often repeatedly suggest the same stupid proposal (me giving them support somewhere for nothing) over and over after i already rejected it. totally ridiculous.