 When Help Threatens and When Users Over-Rely: Proactive and Reactive Help from Intelligent AgentsMarc Goutier Publications of Darmstadt Technical University, Institute for Business Studies (BWL) from Darmstadt Technical University, Department of Business Administration, Economics and Law, Institute for Business Studies (BWL) Abstract: The increasing integration of intelligent agents (IAs) into knowledge work has reshaped human-AI collaboration. Contemporary IAs can initiate interaction themselves and offer proactive help from IAs before users request it, whereas reactive help from IAs remains tied to user prompts (Baird & Maruping, 2021; Berente et al., 2021). This shift is consequential. Users often fail to request help from IAs even when help from IAs would improve their results (Fuegener et al., 2021, 2022; Kraus et al., 2021), yet identical help from IAs can be welcomed or resisted depending on who initiated it, triggering self-threat, fostering over-reliance on help from IAs, or undermining Complementary Team Performance (CTP), i.e., the joint performance of a human and an IA that exceeds what either achieves alone (Fuegener et al., 2021, 2022; Hemmer et al., 2023). While IS research has extensively theorized the content and quality of help from IAs, it has examined willingness to accept help from IAs, reliance on help from IAs, and collaboration outcomes largely in isolation and under the implicit assumption that help from IAs is broadly beneficial (Baird & Maruping, 2021; Li & Karahanna, 2015). This dissertation therefore positions who initiates help from IAs as a design variable for agentic IS artifacts and examines, across five peer-reviewed articles (A–E), how proactive compared to reactive help from IAs shapes users' appraisals, coping responses, and collaboration outcomes. Theoretically, it draws on the framework of Human–Agent Interaction (Baird & Maruping, 2021), Self-Affirmation Theory (Steele, 1988), Appraisal Theory (Lazarus, 1991), and research on CTP (Fuegener et al., 2021, 2022; Hemmer et al., 2023). The first part of the dissertation (Articles A–C) addresses users' psychological reactions to proactive help from IAs and the conditions under which they arise. Articles A, B, and C show that proactive help from IAs increases perceived self-threat and reduces willingness to accept help from IAs, with self-threat serving as the mediating mechanism. This effect varies under three boundary conditions, namely perceived control (A), identity-relevance of the task (B), and relative competence of the IA (C). Theoretically, this contributes in three ways. The first contribution is to analytically separate the initiator of help from IAs from its content and to establish initiative as a psychologically consequential property of human-AI collaboration rather than a neutral efficiency feature. The second contribution is an extension of SAT from interpersonal helping (Craig et al., 2019; Harari et al., 2022) to human-AI collaboration, identifying relative competence of the IA as a boundary condition with no clean interpersonal analogue, since competence asymmetry between humans and IAs can be a structural rather than incidental feature. The third contribution introduces self-threat as an identity-protective mechanism that operates independently of the instrumental quality of help from IAs, such that users may decline objectively useful help from IAs to safeguard their self-concept (Sherman & Cohen, 2006; Steele, 1988). Practically, these findings imply that proactive help from IAs should be designed as an optional suggestion rather than a correction ("suggest, do not apply"), that reactive help from IAs should be the default for identity-relevant tasks, and that users should retain control over proactivity itself. The second part (Article D) shifts attention from willingness to accept help from IAs toward reliance on help from IAs and over-reliance on help from IAs. Using a mixed-methods design, Article D shows that proactive help from IAs increases both reliance on help from IAs and over-reliance on help from IAs, even when users abandon correct answers of their own to follow incorrect help from IAs, and identifies three appraisals through which users evaluate help from IAs, namely perceived reliability and trustworthiness of the IA, emotional and social dynamics, and the cognitive impact of help from IAs. The contribution is to establish willingness to accept help from IAs and reliance on help from IAs as related but theoretically and empirically distinct coping responses, and to extend Appraisal Theory (Lazarus, 1991) to human-AI collaboration as a lens that explains why users may devalue their own correct judgments when following proactive help from IAs, beyond performance-based accounts. Practically, this reframes the design objective from maximizing reliance on help from IAs to calibrating it and recommends deliberate evaluation moments such as showing reasoning, displaying confidence signals, or prompting users to commit before suggestions are revealed, particularly in high-stakes domains such as healthcare, finance, and legal services. The third part (Article E) addresses the design question that follows. Since neither proactive nor reactive help from IAs alone is sufficient to improve collaboration outcomes, Article E introduces Intelligent Helping, a design feature that assigns proactive and reactive help from IAs based on the relative performance of human and IA. Comparing intelligent with random treatment allocation, the article shows that this alignment improves CTP while preserving human control. This generates three contributions. The first reframes the goal of reliance on help from IAs from maximization to calibration, such that users rely on help from IAs when the IA is the more capable party and retain independent judgment when they are. The second operationalizes this calibration as a concrete design feature that combines proactive and reactive help from IAs rather than treating them in isolation, providing actionable design guidance for agentic IS artifacts. The third resolves the apparent tension between high levels of help from IAs and human control by showing that both can be achieved jointly. Practically, this means proactive help from IAs is useful where the IA outperforms the user, and reactive help from IAs is useful where the user outperforms the IA. Beyond the individual articles, the dissertation contributes to IS research on human-AI collaboration by positioning who initiates help from IAs as a first-class design variable in research on agentic IS artifacts, by integrating Self-Affirmation Theory and Appraisal Theory within a process-oriented framework that treats willingness to accept help and reliance on help from IAs as sequential but partly independent responses, and by introducing Intelligent Helping as a theoretical principle for combining proactive and reactive help from IAs. This contribution becomes especially consequential with the rise of agentic AI. As initiation moves from the user to the agent by default, the appraisal and coping dynamics examined here determine whether agent-initiated action is experienced as supportive or controlling, whether reliance on help from IAs is appropriately calibrated, and whether human control is preserved. The dissertation thereby complements the predominantly capability- and governance-oriented discourse on agentic AI with a behavioral foundation for design, and supports an augmentation-oriented perspective in which agentic AI succeeds not by maximizing autonomous action but by offering help from IAs that strengthens human judgment rather than supplanting it. Future research should examine how users' appraisals and reliance on help from IAs evolve over time, how the identified boundary conditions interact in complex settings, and how Intelligent Helping performs in field and multi-actor environments. Ultimately, this dissertation advances a more nuanced theoretical understanding of human-AI collaboration: not as a question of whether humans or IAs perform better in isolation, but of how the design of help from IAs shapes the psychological, behavioral, and performance dynamics through which humans and IAs work together. Date: 2026-10-15 Downloads: (external link) Related works: Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text Persistent link: https://EconPapers.repec.org/RePEc:dar:wpaper:162811 Access Statistics for this paper More papers in Publications of Darmstadt Technical University, Institute for Business Studies (BWL) from Darmstadt Technical University, Department of Business Administration, Economics and Law, Institute for Business Studies (BWL) Contact information at EDIRC. |
Is your work missing from RePEc? Here is how to contribute.
Questions or problems? Check the EconPapers FAQ or send mail to .
EconPapers is hosted by the
School of Business at Örebro University.