Algorithmic oversight saps gig workers' cooperative engagement: perceived algorithmic control heightens feelings of being treated as a tool, cutting willingness to sustain value co-creation—especially among workers already averse to algorithms.

Main Finding

Perceived algorithmic control (PAC) — workers’ perception of algorithmic monitoring, normative guidance, and behavioral constraints — reduces platform workers’ sustainable value co-creation behavior (SVCB). This negative effect operates because PAC increases workplace objectification (WO; workers feeling like replaceable/instrumental “tools”), which in turn lowers SVCB. The indirect (mediated) effect of PAC → WO → SVCB is stronger for workers with higher algorithm aversion (AAV), i.e., dispositional negative attitudes toward algorithms.

Key Points

• Theoretical framing: draws on Objectification Theory (WO as an outcome of instrumentalizing work contexts) and Trait Activation Theory (AAV as a latent trait activated by algorithmic cues).
• Hypotheses tested:
  • H1: PAC negatively affects SVCB.
  • H2a: PAC positively affects WO.
  • H2b: WO negatively affects SVCB.
  • H2c: WO mediates PAC → SVCB.
  • Moderation: AAV amplifies the PAC → WO → SVCB indirect effect.
• Empirical results: All hypotheses supported. Higher PAC → more WO → less SVCB; the mediated path is stronger when AAV is high.
• Conceptual nuance: WO is distinguished from organizational dehumanization by focusing on the worker’s instrumentalized self-understanding (replaceability, controllability) rather than an external denial of human attributes.

Data & Methods

• Design: Two complementary empirical approaches:
  • Two-wave field survey of platform workers (n = 285). Measures collected across two time points to reduce common-method bias.
  • Scenario-based experiment (manipulation: PAC experimental vs. control; n = 216) to test causal effects of PAC.
• Samples: Platform workers in gig-economy contexts (e.g., delivery, ride-hailing) — study conducted in China (data subject to ethical restrictions; not publicly available).
• Measures: validated self-report scales for PAC (normative guidance, tracking/evaluation, behavioral constraint), workplace objectification (WO), sustainable value co-creation behavior (SVCB), and algorithm aversion (AAV).
• Analysis: mediation and moderated-mediation analyses (regression-based / path-analytic tests across survey and experimental data) showing significant direct and indirect effects and significant moderation by AAV.
• Limitations noted by authors: reliance on self-report measures (though partially mitigated by two-wave design and experiment), sample/geographic concentration, and constrained public data access.

Implications for AI Economics

• Short-term efficiency vs. long-run value creation: Heavy algorithmic control can raise operational efficiency but reduces workers’ willingness to invest discretionary effort that generates sustained service quality and platform reputation — a negative externality on long-run platform welfare and user experience.
• Worker heterogeneity matters: Algorithm aversion (AAV) is an important moderating trait. Platforms that treat all workers identically risk larger drops in SVCB among workers with higher AAV, which can affect service consistency and retention heterogeneously across the workforce.
• Principal–agent/incentive design: Algorithmic governance that relies solely on monitoring and penalties may fail to elicit intrinsically motivated co-creation. Incentive systems should incorporate autonomy-supporting mechanisms, feedback channels, and recognition to sustain SVCB.
• Policy and governance: Findings support regulatory and platform-level interventions that increase transparency, human oversight, appeals/override mechanisms, and safeguards for worker health and autonomy to reduce WO and its negative behavioral consequences.
• Directions for platform strategy and measurement:
  • Invest in design features that reduce perceived objectification (e.g., explainable recommendations, worker input on rules, variable autonomy modes).
  • Segment policies by worker AAV profiles and test adaptive governance (more human-centered interaction for high-AAV workers).
  • Track longer-run metrics (retention, service quality, complaints, platform-level customer lifetime value) to quantify trade-offs between algorithmic control intensity and sustainable co-creation.
• Research implications for AI economics: results highlight socio-psychological channels (objectification, aversion) through which algorithmic systems generate behavioral externalities. Economic models of platform governance should incorporate behavioral heterogeneity and non-pecuniary motives (autonomy, recognition) to predict optimal algorithmic design and regulation.

Suggestions for future work (concise): - Use longitudinal administrative/behavioral platform data to estimate long-term causal effects on orders, retention, and quality. - Experiment with alternative algorithmic governance designs (transparency, opt-in autonomy levels, human–AI hybrid allocation) and measure SVCB and platform-level returns. - Cross-country samples to test cultural moderation and regulatory effects.