Paper Type

ERF

Abstract

Artificial intelligence (AI) has become an omnipresent support technology for a variety of decisions. Yet people's beliefs about how well AI suits a task, their perceived task-AI fit, may diverge from actual task-AI fit. This divergence creates miscalibrated reliance on AI. When perceived fit is too high, users over-rely on AI, which can lead to inferior outcomes. When perceived fit is too low, users dismiss useful AI guidance missing out on better decision opportunities. Both of these outcomes represent decision vulnerabilities. Drawing on task-technology fit (TTF) and trust in automation theories, we propose a model with two antecedents of AI reliance mediated by trust in AI: perceived AI capability and perceived self-capability. Reliance on AI is hypothesized to be associated with decision vulnerability (deviation from optimal decisions) moderated by the level of actual task-AI fit. Our paper concludes with the description of a proposed behavioral experiment to test the hypothesized relationships.

Paper Number

1465

Comments

SIG HCI

Share

COinS
 
Aug 15th, 12:00 AM

Task AI Fit – Misplaced Trust, Miscalibrated Reliance and Decision Vulnerability

Artificial intelligence (AI) has become an omnipresent support technology for a variety of decisions. Yet people's beliefs about how well AI suits a task, their perceived task-AI fit, may diverge from actual task-AI fit. This divergence creates miscalibrated reliance on AI. When perceived fit is too high, users over-rely on AI, which can lead to inferior outcomes. When perceived fit is too low, users dismiss useful AI guidance missing out on better decision opportunities. Both of these outcomes represent decision vulnerabilities. Drawing on task-technology fit (TTF) and trust in automation theories, we propose a model with two antecedents of AI reliance mediated by trust in AI: perceived AI capability and perceived self-capability. Reliance on AI is hypothesized to be associated with decision vulnerability (deviation from optimal decisions) moderated by the level of actual task-AI fit. Our paper concludes with the description of a proposed behavioral experiment to test the hypothesized relationships.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.