중앙대학교 인공지능인문학연구소

HK+인공지능인문학

학술지지난 호 보기

지난 호 보기

eISSN: 2951-388X
Print ISSN: 2635-4691 / Online ISSN: 2951-388X
제목[인공지능인문학연구 23권] What Agentic AI Products Must Avoid-Zombie Tasks, Fake Work, and the Collapse of Trust_안광섭2026-08-31 16:44
작성자 Level 10
첨부파일인공지능 인문학연구 23_06.Ahn,Kwangseob.pdf (20.74MB)

1. Introduction

2. Theoretical Background and Conceptual Definitions

3. The Single-Player-but-Not-Solitary Loop

4. Zombie Tasks: When Delegated Work Reveals Itself as Fake Work

5. Conclusion


Collaborative software has traditionally been thought to require single-player value: one user must benefit before others arrive. This study argues that in agentic artificial intelligence (AI) products, the thesis is necessary but insufficient. Because the user delegates work to a non-human counterpart and must decide what can be entrusted to it, the individual-use phase is a relational loop rather than an isolated utility loop. The study proposes the single-player-but-not-solitary loop and specifies credible presence: the quality by which an agent’s activity can be verified as work rather than merely appearing to be so. Drawing on Luhmann’s account of trust as complexity reduction and distinguishing trust from calibrated reliance, it defines the zombie task, a delegated task whose resource consumption outlives the intention that justified it, and analyzes its two subtypes and three strata of loss. Four design principles emerge from this analysis: responsible stopping, transparent state-sharing, ownable artifacts, and auditability.

중앙대학교 인문콘텐츠연구소
06974 서울특별시 동작구 흑석로 84 중앙대학교 310관 828호  TEL 02-881-7354  FAX 02-813-7353  E-mail : aihumanities@cau.ac.krCOPYRIGHT(C) 2017-2023 CAU HUMANITIES RESEARCH INSTITUTE ALL RIGHTS RESERVED