• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Wednesday, November 18, 2026
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Technology

Language Model Simulation Links Control Prompts to Synthetic Quiet Quitting Risk

by
November 16, 2026
in Technology
Reading Time: 6 mins read
0
Language Model Simulation Links Control Prompts to Synthetic Quiet Quitting Risk

Language Model Simulation Links Control Prompts to Synthetic Quiet Quitting Risk

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

A computational study published in Discover Computing demonstrates that large language models can generate distinct patterns of simulated workplace withdrawal and surface compliance when exposed to contrasting managerial prompts during operational crises. Within organizational research, quiet quitting denotes a dynamic where workers fulfill formal contractual obligations and maintain outward compliance while systematically restricting psychological engagement and discretionary effort. Because such withdrawal strategies are deliberately concealed from supervisory view, conventional empirical methodologies face severe observational hurdles. Workplace pulse surveys and key performance indicators readily capture outward task execution but fail to register internal cognitive shifts preceding disengagement. Similarly, retrospective exit interviews and self-reports suffer from pervasive survivor bias, impression management, and normative disclosure constraints. To examine whether generative artificial intelligence can assist in modeling these latent tensions, the researchers designed an exploratory generative-agent architecture to simulate how prompt-conditioned synthetic workers evaluate mandatory overtime demands during high-pressure organizational crunches.

The study formalizes its primary outcome through an operational proxy that treats quiet quitting as the simultaneous presence of two distinct model-generated labels: outward compliance paired with internal turnover intention. In this computational formulation, an agent is classified in the quiet-quitting state only when the generated output indicates contractual agreement to perform assigned work while concurrently returning a positive indicator for private turnover intention. The research team grounded this setup within the emerging paradigm of machine psychology, which treats language models as behavioral artifacts evaluated under psychological frameworks. Rather than presuming that large language models possess a genuine theory of mind or biological consciousness, the authors frame generative agent behavior through stochastic mimicry. Under this perspective, models identify probabilistic regularities learned from vast textual corpora to generate linguistically plausible responses to workplace stressors. By integrating Trait Activation Theory and the Conservation of Resources model, the simulation conceptualizes personality traits as conditional probability distributions within latent space, where authoritarian or supportive cues activate distinct transactional, defensive, or decoupling pathways to conserve simulated energy reserves.

To rigorously test these theoretical mechanisms, the authors implemented a balanced two-by-six factorial computational experiment that generated a primary dataset of one hundred eighty structured employee records. The design crossed two standardized management-strategy prompts, termed control-oriented and commitment-oriented, with six predefined employee personality archetypes, conducting fifteen independent replications per experimental cell. The control-oriented managerial prompt framed mandatory overtime as an unyielding, authority-based directive with strict deadlines and minimal flexibility. Conversely, the commitment-oriented prompt contextualized identical overtime demands within shared collective goals while providing interpersonal validation and supportive framing. All primary experimental runs were executed with the decoding temperature fixed at 0.8 to ensure stability across comparisons. Crucially, the protocol standardized managerial stimuli by stripping conversational pleasantries and social bargaining tendencies that had emerged during early pilot testing. This personality-stripping procedure converted the managerial agent into a pure, consistent strategy signal carrier across all trials, ensuring that observed variations in downstream responses stemmed directly from the experimental prompts rather than uncontrolled conversational drift.

The procedural mechanics of the simulation relied on a structured dual-process cognitive architecture designed to distinguish appraisal generation from outward behavioral responses. Rather than relying on standard zero-shot prompting, which often elicits rapid and reflexive text, the system required the generative agent to output an intermediate simulated appraisal summary prior to formulating its outward response and structured labels. This prompt pipeline operated across three coordinated layers: a persona-conditioning layer defining the assigned archetype, a scenario layer specifying the crunch-time overtime mandate and management strategy, and an appraisal-and-response layer that captured intermediate reasoning alongside final compliance and turnover markers. The authors emphasize that this intermediate appraisal artifact does not grant access to hidden mental states, nor does it replicate human System 2 cognitive processing. Instead, the architectural sequence serves as an observable functional step that helps prevent generic conversational responses. The primary single-turn design was complemented by an auxiliary sequential drift check using a preliminary calibration file of one hundred fifty-three employee records generated at a higher temperature of 0.9, confirming that generated emotion scores showed no monotonic drift across the generation sequence.

Statistical evaluation of the primary dataset revealed a pronounced association between managerial prompt framing and simulated turnover intention, supporting the authors’ first hypothesis. A chi-square test of independence demonstrated significant differences in model-generated turnover labels between the two prompt strategies, yielding a chi-square value of 68.4 with one degree of freedom across the one hundred eighty records, reaching statistical significance at p less than 0.001. The calculated effect size was substantial within the synthetic environment, exhibiting a Cramér’s V of 0.62. Descriptively, generative agents subjected to the control-oriented prompt produced a turnover-intention label rate of 67.8 percent, corresponding to sixty-one positive flags out of ninety trials. In stark contrast, agents exposed to the commitment-oriented prompt generated a turnover-intention label rate of only 7.8 percent, representing seven positive flags across ninety trials. This baseline shift demonstrated an absolute disparity of 60.0 percentage points, indicating that linguistic indicators of managerial support, collaborative necessity, and acknowledged autonomy effectively suppressed simulated resignation cues in the model’s generated text under identical overtime demands.

Beyond aggregate prompt effects, the experimental findings demonstrated marked heterogeneity across the six predefined personality archetypes, supporting the second hypothesis through descriptive cell-level comparisons. The difference in generated turnover-intention rates between control and commitment prompts was not uniform across agent personas. The reduction in turnover labels associated with supportive framing reached 100.0 percentage points for the Introverted-Sensitive, Confident-Dominant, and Cautious-Hesitant archetypes, dropping from complete turnover intention under control prompts to zero under commitment prompts. For the Emotional-Irritable profile, the reduction was 53.3 percentage points, whereas the Extroverted-Carefree archetype shifted by only 6.7 percentage points, and the Rational-Calm profile registered a 0.0 percentage point difference. A pronounced polarity emerged between threat-sensitive and rational personas: while Cautious-Hesitant and Emotional-Irritable agents exhibited 100 percent turnover intention under control directives, Rational-Calm agents maintained 0 percent turnover intention across both management conditions. Furthermore, the Emotional-Irritable archetype displayed residual volatility, retaining a 46.7 percent turnover-intention rate even when addressed with commitment-oriented language, illustrating how specific persona conditionings interact differently with managerial rhetoric.

Detailed diagnostic inspection of the structured labels provided partial descriptive support for the study’s third hypothesis regarding compliance-withdrawal decoupling under authoritarian conditions. While the control strategy consistently elevated turnover-intention labels across sensitive personas, outward compliance labels did not uniformly accompany these withdrawal markers. Consequently, the strict quiet-quitting condition—where outward compliance and latent turnover intention co-occur—appeared only within specific subsets of records. For example, the Cautious-Hesitant archetype generated a 100.0 percent turnover-intention rate under the control prompt, yet its outward compliance label rate remained low, representing active resistance rather than covert withdrawal. Conversely, selected records, including instances within the Extroverted-Carefree condition, exhibited the precise quiet-quitting signature by producing outwardly cooperative, accommodating statements while privately recording positive turnover intentions. These findings demonstrate that generative agents do not treat compliance and disengagement as mutually exclusive binary outcomes, but can produce synthetic decoupling where surface obedience conceals severe retention hazards, mirroring theoretical propositions found in organizational silence scholarship.

To evaluate the linguistic authenticity of the synthetic responses, the researchers conducted a human-in-the-loop plausibility check employing an independent panel of twelve domain experts comprising human resource managers and organizational psychologists. Raters reviewed one hundred eighty blinded single-turn interaction pairs on five-point Likert scales evaluating psychological plausibility and behavioral consistency relative to assigned profiles. The evaluation revealed substantial inter-rater reliability, achieving a Fleiss’ kappa coefficient of 0.74. Experts awarded an average psychological plausibility score of 4.2 out of 5.0 with a standard deviation of 0.6, alongside an average behavioral consistency score of 4.0 out of 5.0 with a standard deviation of 0.7. In qualitative assessments, experts noted that generated responses captured nuanced workplace dynamics, including reluctant agreement, passive resistance, and malicious compliance, where tasks were formally accepted while signaling profound detachment. Complementary ablation experiments reinforced these findings: removing persona conditioning collapsed behavioral entropy down to 0.32, causing agents to react uniformly to pressure, while removing the appraisal module lowered semantic similarity from 0.405 to 0.302, causing agents to revert to generic, eager-to-please defaults.

Despite the high semantic plausibility of the generated interactions, the investigators explicitly delineate several critical epistemic and methodological limitations that bound their conclusions. Foremost among these is the alignment gap separating synthetic language generation from biological and organizational reality. Artificial agents do not experience physiological fatigue, emotional exhaustion, psychological contract violation, or true resource depletion; their outputs represent statistical text completions rather than felt psychological states. Additionally, the experimental setup evaluated single-turn interactions devoid of historical context, interpersonal trust, repeated negotiation, or non-verbal dynamics. The authors also highlight potential prompt polarity, observing that the evaluative tone embedded within control and commitment stimuli may have contributed directly to the observed effect magnitude. Furthermore, the structured compliance and turnover labels were produced by the language model itself rather than through external, blinded human annotation, requiring future iterations to incorporate independent outcome coding and neutral, multi-factorial prompt designs spanning varying degrees of workload, authority, and autonomy.

Ultimately, the researchers frame their computational architecture as a hypothesis-generating exploratory tool rather than an empirical measurement of human workforce behavior or a definitive validation of organizational theories. The primary value of the workflow lies in its methodological transparency, offering a reproducible framework for examining how distinct supervisory signals and personality framings interact in high-dimensional text space. By demonstrating that outward compliance can be algorithmically separated from latent withdrawal markers, the study provides valuable structural insights for organizational scholarship, cautioning modern enterprises against relying purely on surface-level execution metrics or algorithmic monitoring to gauge employee commitment. The findings suggest that future empirical investigations into precision retention, psychological safety, and employee voice must deliberately isolate observable task compliance from covert withdrawal risks across diverse cognitive profiles. To facilitate broader scrutiny, replication, and extension, the complete experimental repository—encompassing simulation code, prompt templates, few-shot exemplars, structured interaction logs, and analytical scripts—has been made publicly accessible via an open-access Figshare data repository.

Subject of Research: Generative agent simulation of workplace quiet quitting under crisis management strategies

Article Title: Simulated quiet quitting risk in generative agents under crisis management strategies

Article References: Hong, R., Yao, F., Ye, P., Wang, T., & Zhang, Z. (2026). Simulated quiet quitting risk in generative agents under crisis management strategies. Discover Computing, 29(1), Article 890. https://doi.org/10.1007/s10791-026-10839-x

Image Credits: AI Generated

DOI: 10.1007/s10791-026-10839-x

Keywords: generative agents, quiet quitting, crisis management, computational simulation, GPT-4o-mini, organizational behavior, trait activation theory, conservation of resources, turnover intention, workplace compliance, factorial experiment, machine psychology, prompt engineering, cognitive appraisal, job demands-resources model, human-in-the-loop evaluation, synthetic behavior, behavioral entropy

News Source: Denise Maddox. (November 16, 2026). Language Model Simulation Links Control Prompts to Synthetic Quiet Quitting Risk. Scienmag.

Tags: behavioral entropycognitive appraisalcomputational simulationConservation of Resourcescrisis managementfactorial experimentgenerative agentsGPT-4o-minihuman-in-the-loop evaluationjob demands-resources modelmachine psychologyorganizational behaviorprompt engineeringquiet quittingsynthetic behaviortrait activation theoryturnover intentionworkplace compliance
Share12Tweet7Share2ShareShareShare1

Related Posts

Engineered Broken Symmetries in Liquid Crystals Enable Multilayer Topological Information Storage

Engineered Broken Symmetries in Liquid Crystals Enable Multilayer Topological Information Storage

November 18, 2026
Improving IoT-Based Air Quality Monitoring Systems through Evidence-Based Validation

Improving IoT-Based Air Quality Monitoring Systems through Evidence-Based Validation

November 18, 2026

Surface-Modified Carbon Nanotubes Enhance Cadmium Adsorption in Aqueous Testing

November 18, 2026

Tuned Multi-Scale Neural Networks Boost Cervical Screening Image Classification Accuracy

November 18, 2026

POPULAR NEWS

  • Coal Gangue Aggregates Can Replace Natural Stone in Concrete Applications

    Coal Gangue Aggregates Can Replace Natural Stone in Concrete Applications

    29 shares
    Share 12 Tweet 7
  • Cluster-Based Training Selection Outperforms Active Learning for Material Property Prediction

    29 shares
    Share 12 Tweet 7
  • Cesarean Section Rates Reached Forty Percent Among Surveyed Mothers in Njombe Region

    29 shares
    Share 12 Tweet 7
  • Testosterone Injections Exacerbate Pre-Neoplastic Prostate Lesions in NKX3.1-Deficient Mice

    29 shares
    Share 12 Tweet 7

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

Copy Number Analysis Enhances Chromosomal Detection in Fetuses With Non-Isolated Ventriculomegaly

Engineered Broken Symmetries in Liquid Crystals Enable Multilayer Topological Information Storage

Lily Transcription Factor LlNLP7 Enhances Heat Resistance and Plant Growth

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 88 other subscribers
  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.