Human-Centered Design of AI Decision Support: Algorithmic Vigilance and the Role of AI Literacy

Open Access
Article
Conference Proceedings
Authors: Julia Czerniak-Wilmes, Lukas Messmer, Leonie Däullary, Teuta Salii, Philipp Rüegg, Frieder Loch
Abstract

Artificial intelligence is increasingly embedded in workplace decisions, yet practitioners oscillate between algorithm aversion and automation bias. Algorithmic vigilance, i.e. calibrated, critically reflective reliance on AI, is the normative design target, but it remains unclear what it looks like in everyday professional work and which conditions support it. Existing design principles largely stem from laboratory studies and treat AI literacy as the main individual moderator.This exploratory study combines interviews with 13 practitioners from construction, production planning and school administration with the Meta AI Literacy Scale (MAILS). Interview statements were quantitized into ordinal indicators and related to MAILS scores (Spearman's ρ). Practitioners unanimously retained decision authority and verified AI outputs. AI literacy tended to go with more specific transparency requirements (ρ = .44), whereas task complexity and decision criticality shaped explanation demand (ρ = .76 and .81) and automation affinity (ρ = −.55). AI literacy was unrelated to emotional security (ρ = .19) and emotion regulation (ρ = −.03); self-efficacy was associated with more systematic verification (ρ = .59).We derive five hypotheses and an extended model of algorithmic vigilance that adds task complexity as a second dimension next to AI literacy and locates emotional security in system design. The design implications are role- and task-sensitive transparency, active competence support and error-type-differentiated confidence communication, implemented through confirmation gates, calibrated confidence signals and configurable autonomy. A confirmatory study will test the hypotheses.

Keywords: AI Literacy · Algorithmic Vigilance · Automation Bias · Algorithm Aversion · Task Complexity · MAILS · Mixed Methods · Quantitizing

DOI: 10.54941/ahfe1008302

Cite this paper
Downloads
0
Visits
1
Download PDF

More from this volume

← Modeling Individual Differences in Motivational Responses for Personalized Support of Sustained Behavior Change toward Well-BeingWhen Helpful AI Feels Too Trustworthy: Engagement and Reliance in Design Ideation →
View all articles in Human Factors in Design, Engineering, and Computing →