Paper Number

ICIS2026-3381

Paper Type

Short

Abstract

Employees often must verify AI-generated outputs that, although produced quickly, are not always easy to interpret, trace, or defend. This short paper conceptualizes verification burden as a distinct job demand that arises when employees must check, validate, correct, and justify AI-generated outputs before final use under retained human accountability. Drawing on agency theory, we propose perceived output opacity, perceived output fallibility, perceived output responsibility, and perceived error consequences as antecedents of verification burden. Using the JD-R model, we link verification burden to emotional exhaustion, reduced end-user performance, and lower AI continuance intention. We propose a two-phase design to refine the construct qualitatively and assess its discriminant validity and nomological position quantitatively. This offers a more precise account of AI-enabled work, in which output generation may be automated while verification remains human work and accountability remains with employees.

Comments

03_Future_of_work

Download

Share

COinS
 
Dec 14th, 12:00 AM

Verification Burden in AI-Enabled Work: When Delegated Output Generation Leaves Accountability with Employees

Employees often must verify AI-generated outputs that, although produced quickly, are not always easy to interpret, trace, or defend. This short paper conceptualizes verification burden as a distinct job demand that arises when employees must check, validate, correct, and justify AI-generated outputs before final use under retained human accountability. Drawing on agency theory, we propose perceived output opacity, perceived output fallibility, perceived output responsibility, and perceived error consequences as antecedents of verification burden. Using the JD-R model, we link verification burden to emotional exhaustion, reduced end-user performance, and lower AI continuance intention. We propose a two-phase design to refine the construct qualitatively and assess its discriminant validity and nomological position quantitatively. This offers a more precise account of AI-enabled work, in which output generation may be automated while verification remains human work and accountability remains with employees.