Defcon Bots!

[Wasgood]

I'm very interested, although I do not have the programming knowledge to do these sort of things.
Could these be used as mods?

[TheRileyDuo]

If there were easy tools to do it with, I would do it. For the moment, though, I don't think I know enough about the code that runs DEFCON to mod the game in any way. Can you guys write a program to help design (at least basic) strategies for an AI?

[bert_the_turtle]

You don't need to know anything about the code that runs DEFCON. The API basically directly mirrors the regular UI. You can query the game state and give orders, you just don't do it by clicks, you do it by calling functions.

[Phelanpt]

I'm very interested, although I do not have the programming knowledge to do these sort of things.
Could these be used as mods?

Not exactly, you still need to have the API installed.
Read the Quickstart guide at Robin's page.
But as long as someone has published a bot, you can try it out for yourself with just a simple command line command (is there a better way to write this? ).

Also, I hope this idea keeps going in Multiwinia, although the API would probably be much more complex. I'd be very interested in trying my hand at MW AI.

[freddie]

Firstly, can anyone decipher exactly what is 'more fun'? Although, clearly, getting your arse handed to you every round wouldn't be it, actually pinning down what a more fun bot would be, if one exists independently of variety, is a study in itself (as for myself, being utterly hopeless at DEFCON, I'd prefer one that didn't launch any of those pesky nuclear weapons at me).

There's a pop sci book called "Critical Mass: how one thing leads to another" (Philip Ball, 2004) which touches on the prisoners dilemma (a trust, defection, personal gain game) and Robert Axelrod's tournaments between set's of algorithms (i.e. bots). He also notes how one of the most successful 'bots' was used to justify the MAD cold war strategy, although the principles of the cold war (and the arsenal) don't hold up with the use of this bot (tit for tat). Applying Axelrod's trials to this would yield a competition trying to minimise the deaths (i.e. more of a rubbishy "real world" gain) and so VERY different winners should emerge... but this is a good echo of what he was doing. The book's really interesting in places, although it's REALLY long winded so just read ch 17 if you're interested in hearing more about this in a pretty easy to digest format.

Just thought I'd repeat stuff that the Introversion social scientists already know.

[Phelanpt]

Firstly, can anyone decipher exactly what is 'more fun'? Although, clearly, getting your arse handed to you every round wouldn't be it, actually pinning down what a more fun bot would be, if one exists independently of variety, is a study in itself (as for myself, being utterly hopeless at DEFCON, I'd prefer one that didn't launch any of those pesky nuclear weapons at me).

Just finished reading Robin and Simon's paper, and they seem to want to do experiments against human players, and have the humans describe what they didn't like about the bot.
Interesting paper, although mostly for an AI crowd.

[The GoldFish]

I would like to see a bot that convincingly only just loses the game to the human player.

[RobinB]

Just finished reading Robin and Simon's paper, and they seem to want to do experiments against human players, and have the humans describe what they didn't like about the bot. Interesting paper, although mostly for an AI crowd.

Yeah, we actually did these tests with a relatively small sample size of 10 novice players, we did a blind test and gave them either the Introversion AI or the (harder) one I wrote. It seems my bot was too hard for them (its still not a very good bot, but as complete novices they felt frustrated). So "more fun" definitely does not mean directly "more difficult to beat", which is kind of obvious I guess, but still having experiments confirm this makes sense.

Defining what "more fun" is is a hard problem actually, both in academia as well as in the game industry. We've got a couple of ideas on that, but nothing which has been verified irl yet. Game designers usually create game ideas instinctively, without a formal concept of fun or suspense. A lot of research left to do!

[Mas Tnega]

I would like to see a bot that convincingly only just loses the game to the human player.

We'll have Peace and Love play it. Problem solved!

[Mono]

Would it be possible to combine the bots with Dedcon and stick them out there on the metaserver for some experience of the real world?

[bert_the_turtle]

No. Dedcon has no idea about the game state and won't be able to drive a bot's interface. You need the bot enabled defcon version, and that is unfortunately officially incompatible with 1.43, so just sticking it out there will not give your bot a single bit of experience

(In practice, it is not incompatible. If you take a hexeditor and edit the version info the bot enabled version to be 1.43, you can join and host games with other 1.43 users no problem. Doing so, of course, is an EULA violation.)

[Wasgood]

Maybe Dedcon is a gateway to allow a "learning bot".

[Pox]

Whoa, I was very interested when I read an article about the collaboration recently (I believe KingAl linked it in another thread...?), but a public API? Want! I eagerly await a python@linux version, would be very very fun to fiddle with.

[Johnis]

I am very happy when i see that defcon is still live and well. I hope to see some kind of patch of course afterwards

[tllotpfkamvpe]

cant start defcon, getting error message "application is configured incorrectly". Does it have to be started thru visual studio?