Paper Number
ICIS2025-2450
Paper Type
Complete
Abstract
Generative AI (GenAI) based tools are becoming integral to software development workflows, offering substantial potential for productivity gains but simultaneously posing significant risks through “hallucinations”—the generation of outputs that appear plausible yet are fundamentally incorrect. We study the double-edged sword presented by GenAI, by examining the impact of hallucinations on task performance for both individual developers and teams. Drawing upon experimental findings, we confirm the detrimental effect of hallucinated GenAI responses on performance. Crucially, this negative impact is demonstrably mitigated by higher levels of developer expertise, suggesting that expertise provides a buffer against misleading AI suggestions. Furthermore, our results reveal a greater negative effect of hallucinated code within team settings. These findings underscore the necessity for developers to cultivate strong critical evaluation and verification skills, highlighting that human expertise remains a key factor in successfully navigating the complex landscape of risks and rewards associated with GenAI adoption in modern software engineering.
Recommended Citation
Currim, Faiz; Srinivasan, Karthik; and Tripathi, Arvind, "The Impact of Generative AI Hallucination on Coding Performance: The Moderating Role of Expertise and Group Dynamics" (2025). ICIS 2025 Proceedings. 27.
https://aisel.aisnet.org/icis2025/gen_ai/gen_ai/27
The Impact of Generative AI Hallucination on Coding Performance: The Moderating Role of Expertise and Group Dynamics
Generative AI (GenAI) based tools are becoming integral to software development workflows, offering substantial potential for productivity gains but simultaneously posing significant risks through “hallucinations”—the generation of outputs that appear plausible yet are fundamentally incorrect. We study the double-edged sword presented by GenAI, by examining the impact of hallucinations on task performance for both individual developers and teams. Drawing upon experimental findings, we confirm the detrimental effect of hallucinated GenAI responses on performance. Crucially, this negative impact is demonstrably mitigated by higher levels of developer expertise, suggesting that expertise provides a buffer against misleading AI suggestions. Furthermore, our results reveal a greater negative effect of hallucinated code within team settings. These findings underscore the necessity for developers to cultivate strong critical evaluation and verification skills, highlighting that human expertise remains a key factor in successfully navigating the complex landscape of risks and rewards associated with GenAI adoption in modern software engineering.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
12-GenAI