Is human-like and well playing contradictory for Diplomacy bots?

摘要

This paper presents a non-player character (NPC, bot) for the strategy game Diplomacy. The bot is able to communicate with other players and thus shows a human-like behavior. We investigate how far the playing abilities can be improved without corrupting the human-like behavior. Is there a trade-off at all or do these skills complement one another? Different versions of the bot are tested against other bots and humans which requires means to automatically measure believability. We derive such a measure after a general approach and apply it for monitoring the believability criterion while improving the playing strength of our bot.