Incorporating Trust and Trickery Management in First Person Shooters
Authors
Nielsen, Niels Christian ; Oddershede, Henrik ; Larsen, Jacob
Term
4. term
Education
Publication year
2004
Abstract
Denne afhandling undersøger, hvordan menneskelignende karaktertræk kan bygges ind i computerstyrede holdkammerater (bots) i team-baserede førstepersonsskydespil (FPS). Vi bruger det populære spil Counter-Strike som testmiljø. Først designer vi en grundmodel, hvor en bot kun fokuserer på holdets samlede succes. Derefter udvider vi modellen med to personligheder: en trickster (en snyder), som indimellem lyver for egen vinding, og en holdorienteret bot, der konsekvent handler til fordel for hele holdet. For at opdage trickstere indfører vi tillidsstyring: Hver bot fører tillidsscorer for de andre og kombinerer egne erfaringer med omdømmeoplysninger fra holdkammerater for at vurdere, om information er troværdig. Vi gennemgår udfordringerne ved at bruge tillidsstyring, når en gruppe skal træffe fælles beslutninger og ikke kun individuelle. Til sidst undersøger vi, hvordan trickstere kan forsøge at omgå disse mekanismer, hvilket vi kalder Trickery Management: Tricksteren prøver at gætte holdkammeraters tærskelværdier for tillid for at afgøre, hvornår det kan betale sig at lyve.
This thesis explores how human-like character traits can be built into computer-controlled teammates (bots) in team-based first-person shooter (FPS) games. We use the popular game Counter-Strike as a testbed. We first design a baseline bot that cares only about overall team success. We then extend this to support two personalities: a trickster that sometimes lies for personal gain, and a team-oriented bot that consistently acts for the team’s benefit. To identify tricksters, we add trust management: each bot keeps trust scores for teammates and combines its own experience with reputation reported by others to judge whether information is reliable. We examine the challenges of using trust management when a group must make collective decisions rather than only individual ones. Finally, we study how tricksters might evade these defenses, which we call Trickery Management: the trickster tries to infer teammates’ trust thresholds to decide when lying is worth the risk.
[This abstract was generated with the help of AI]
Documents