Detail View

Augmenting Player Experience through the use of Generative AI in Open-World Games

Citations

WEB OF SCIENCE

Citations

SCOPUS

Metadata Downloads

DC Field Value Language
dc.contributor.advisor 김선준 -
dc.contributor.author Taehun Kim -
dc.date.accessioned 2026-09-01T19:32:37Z -
dc.date.available 2026-09-02T06:00:32Z -
dc.date.issued 2026 -
dc.identifier.uri https://scholar.dgist.ac.kr/handle/20.500.11750/60803 -
dc.identifier.uri http://dgist.dcollection.net/common/orgView/200001010790 -
dc.description Generative AI, Open-world games, Player experience, Technology probe, Co-design workshops -
dc.description.abstract Open-world games promise wide action spaces, but many commercial titles still rely on pre-authored runtime responses that break down when players attempt actions designers did not anticipate. Generative AI offers a pos- sible alternative: flexible, context-sensitive responses produced during play. Prior work has primarily examined generative capability in games; this thesis focuses on player requirements for open-world contexts. We ask three questions: (RQ1) what limitations and unmet needs do players experience in current open-world games? (RQ2) how do players expect those needs to be addressed when generative AI enters open-world environments? and (RQ3) how can those requirements be organized for system design? To answer these questions, we built ExtrA (Extendable Architecture Agent), a Minecraft 1.21.4 technology probe that combines Gemini 3 Flash Preview (gemini-3-flash-preview) with a Model Context Protocol (MCP) world-editing executor. ExtrA allowed participants to configure, inspect, and test generative-AI subsystems dur- ing live play. Around this probe, we ran a three-stage qualitative study: semi-structured interviews with 22 open- world players, ExtrA probe sessions with 15 of those participants, and four co-design workshops with 14 of those 15 players where they reasoned about future open-worlds at the system level. Participants' expectations extended beyond improved NPC dialogue. They described worlds in which player actions leave lasting marks, entities carry social histories, and small interventions can ripple through ecology, economy, factions, and narrative. They also wanted the creation of in-game items, skills, quests, rules, and story events as part of play, with ways to preview and adjust their effects, and they expected the system to allocate detailed simulation to what currently matters to the player, including nearby objects, ongoing quests, recent ac- tions, and socially important entities, rather than maintaining uniform fidelity across the entire world. Based on these findings, we derive seven structural constraints in current open-worlds (C1-C7), seven match- ing player requirements for generative AI (R1-R7), and four design implications (DI1-DI4): design generative AI as world infrastructure (DI1), support in-game creation of content and rules (DI2), schedule simulation fidelity by proximity, importance, and relevance (DI3), and personalize fairly within shared worlds (DI4). The thesis con- tributes ExtrA as a technology probe, the C1-C7 / R1-R7 classification, and the DI1-DI4 design implications. Together, these contributions reposition generative AI in open-worlds from a content-production tool to world- scale interaction infrastructure for requirement-driven world change. Keywords: Generative AI, Open-world games, Player experience, Technology probe, Co-design workshops|오픈월드 게임은 넓은 행동 공간과 플레이어의 자유를 약속하지만, 현 상용 타이틀의 런타임 응답 대부분은 사전에 작성된 스크립트에 의존하기 때문에 설계자가 예상하지 못한 행동에는 충분히 반응하지 못한다. 대규모 언어모델과 에이전트 아키텍처를 포함한 최근의 생성형 AI는 유연하고 맥락에 부합하는 런타임 응답을 제공할 잠재력을 지니지만, 기존 연구는 주로 생성형 AI가 무엇을 만들어낼 수 있는지를 보였을 뿐, 플레이어가 어떤 한계를 보완하기를 기대하는지에 대한 체계적 이해는 부족하다. 본 연구는 다음 세 가지 연구 질문을 다룬다. (RQ1) 플레이어가 현재 오픈월드 게임에서 경험하는 한계와 충족되지 않은 요구는 무엇인가, (RQ2) 생성형 AI가 도입될 때 플레이어는 그러한 요구가 어떻게 해소되기를 기대하는가, (RQ3) 그러한 요구사항을 시스템 설계를 위해 어떻게 조직화할 수 있는가.
본 연구는 3단계 질적 연구 설계를 따른다. 먼저 22명의 오픈월드 플레이어를 대상으로 반구조화 인터뷰를 진행하여 반복적으로 등장하는 한계와 기대를 수집하였다. 다음으로 Gemini 3 Flash Preview(gemini-3-flash-preview)와 Model Context Protocol(MCP) 기반 월드 편집 실행기를 결합한 Minecraft 1.21.4 기반 기술 프로브인 ExtrA를 구현하고, 15명의 참여자가 실제 플레이 중에 생성형 AI 서브시스템을 직접 구성하고 시험하도록 하였다. 마지막으로 그중 14명이 4회의 코디자인 워크숍에 참여하여 미래 오픈월드의 시스템 수준 동작을 함께 구상하였다.
분석 결과 플레이어들은 더 많은 대화나 더 많은 자동 생성 콘텐츠가 아니라, 행동의 흔적이 시간이 지나도 남는 지속적 결과(persistent aftermath), 사회적으로 맥락화된 NPC와 개체, 시스템 간 인과적 연쇄, 상황 기반 안내와 보이는 월드 상태, 플레이어가 만든 아이템, 스킬, 퀘스트, 이야기의 수정 지원, 동적 시뮬레이션 충실도, 공정한 공유-월드 개인화를 기대한다는 점이 드러났다. 이로부터 현재 오픈월드에서 플레이어가 경험하는 7가지 제약(C1-C7), 그에 대응하는 생성형 AI 요구사항 7가지(R1-R7), 그리고 이를 시스템·인터랙션 차원으로 옮기는 4가지 설계 함의(DI1-DI4)를 분류 체계로 정리하였다.
본 연구는 생성형 AI를 단발적 콘텐츠 생성기가 아니라 월드 규모 상호작용 인프라(world-scale interac-tion infrastructure)로 재정의한다. 핵심은 시스템이 얼마나 많은 콘텐츠를 만들어낼 수 있는가가 아니라, 플레이어의 의도가 어떻게 해석되고, 그 해석이 어떻게 월드 상태로 전환되며, 그 변화가 플레이어에게 어떻게 이해 가능한 방식으로 드러나고 다시 수정될 수 있는가이다. 본 연구는 (1) 요구사항 도출을 위한 ExtrA 기술 프로브, (2) 플레이어 경험 제약 분류(C1-C7), (3) 생성형 AI 요구사항 분류(R1-R7), (4) 4가지 설계 함의(DI1-DI4)를 학술적 기여로 제시한다.
핵심어: 생성형 AI, 오픈월드 게임, 플레이어 경험, 기술 프로브, 코디자인 워크숍
-
dc.description.tableofcontents List of Contents
Abstract i
List of Contents iii
List of Tables v
List of Figures v

I. INTRODUCTION 1
1.1 The Appeal and Structural Limits of Games 1
1.2 Open-world Games and Adjacent Genres 1
1.3 Generative AI as an Approach to Open-world Limits 2
1.4 Research Questions 3
1.5 Contributions 4

II. RELATED WORK 5
2.1 Experiencing Open-worlds: More Than Map Size 5
2.2 Procedural Content Generation and Player Adaptation 6
2.3 Generative AI in Games: From Dialogic NPCs to Runtime World Change 7
2.4 Mixed-Initiative Generation and Human-AI Interaction 9
2.5 Synthesis: The Gap Between Generative Capability and Player Requirements 11

III. METHOD 12
3.1 Study Design Overview 12
3.2 Technology Probe: ExtrA 12
3.2.1 Participant-Facing Interaction Layer 13
3.2.2 Subsystem Creation· 14
3.2.3 Subsystem Activation 15
3.2.4 Minecraft Integration and World Editing 16
3.2.5 Multi-Agent Component Structure 17
3.2.6 Auxiliary State and Feedback Components 18
3.2.7 System Specifications and Operational Characteristics 19
3.2.8 Multi-Agent Pipeline and Agent Roles 21
3.2.9 Agent Personas and Prompt Specifications 22
3.3 Participants 24
3.4 Procedure 25
3.4.1 Stage 1. Pre-study Semi-structured Interviews 25
3.4.2 Stage 2. Technology-probe-based Contextual Inquiry 27
3.4.3 Stage 3. Co-design Workshops 27
3.5 Data Analysis 29
3.5.1 Researcher Positionality 32

IV. FINDINGS 34
4.1 Current Open-worlds Look Updated but Do Not Feel Alive 34
4.2 Players Wanted Socially Contextualized Entities, Not Just More NPCs 35
4.3 Players Expected Small Actions to Cascade Across Systems 35
4.4 Players Wanted Situation-aware Guidance and Visible World State 36
4.5 Players Wanted to Create Content and Rules During Play 36
4.6 Players Did Not Expect the Entire World to Run at Uniform Fidelity 36
4.7 Players Wanted Personalized Narrative Without Sacrificing Shared-World Fairness 37
4.8 Classifying Constraints in Current Open-worlds 38
4.9 Requirements Identified Before the Probe 40
4.10 Requirements Refined Through Probe Sessions and Workshops 41

V. DISCUSSION 43
5.1 Design Implications 44
5.1.1 DI1. Design Generative AI as World Infrastructure 44
5.1.2 DI2. Support In-game Creation of Content and Rules 45
5.1.3 DI3. Schedule Simulation Fidelity by Proximity, Importance, and Relevance 45
5.1.4 DI4. Personalize Fairly Within Shared Worlds 46
5.2 Player-facing Visibility of World Changes 47
5.3 Reflections on the Study Process 47

VI. LIMITATIONS AND FUTURE WORK 48
6.1 Limitations of the Study Design 48
6.2 Future Work 49

VII. CONCLUSION 51

REFERENCES 52

요 약 문 (Summary in Korean) 56
-
dc.format.extent 57 -
dc.language eng -
dc.publisher DGIST -
dc.title Augmenting Player Experience through the use of Generative AI in Open-World Games -
dc.type Thesis -
dc.identifier.doi 10.22677/THESIS.200001010790 -
dc.description.degree Master -
dc.contributor.department Department of Electrical Engineering and Computer Science -
dc.contributor.coadvisor Jean Y. Song -
dc.date.awarded 2026-08-01 -
dc.publisher.location Daegu -
dc.description.database dCollection -
dc.citation XT.IM 김881 202608 -
dc.date.accepted 2026-07-21 -
dc.contributor.alternativeDepartment 전기전자컴퓨터공학과 -
dc.subject.keyword Generative AI, Open-world games, Player experience, Technology probe, Co-design workshops -
dc.contributor.affiliatedAuthor Taehun Kim -
dc.contributor.affiliatedAuthor Sunjun Kim -
dc.contributor.affiliatedAuthor Jean Y. Song -
dc.contributor.alternativeName 김태훈 -
dc.contributor.alternativeName Sunjun Kim -
dc.contributor.alternativeName 송진영 -
Show Simple Item Record

File Downloads

  • There are no files associated with this item.

공유

qrcode
공유하기

Total Views & Downloads

???jsp.display-item.statistics.view???: , ???jsp.display-item.statistics.download???: