مدل‌سازی غیرخطی تعامل عوامل اخلاقی، فرهنگی و ساختاری بر نیت افشای دانش حساس با روش سطح پاسخ: مقایسه انسان و هوش مصنوعی

هدف: با توجه به افزایش اهمیت شفافیت، پاسخ‌گویی و مقابله با فساد در سازمان‌های دولتی، افشای آگاهانه دانش حساس به‌عنوان یک رفتار داوطلبانه و اخلاق‌مدار در بطن خط‌مشی‌گذاری عمومی اهمیت یافته است. این پژوهش با هدف مدل‌سازی تاثیر توامان و غیرخطی سه متغیر «تمایل اخلاقی فردی»، «فرهنگ اشتراک دانش اخلاقی» و «موانع ساختاری اشتراک دانش» بر «نیت افشای دانش حساس» و مقایسه الگوی انسانی با مدل هوش مصنوعی گروک-3 در بین کارکنان ادارات خدماتی استان لرستان انجام شد.روش پژوهش: تحقیق حاضر از نوع کاربردی- توصیفی و آزمایشی است و با بهره‌گیری از طراحی مرکب مرکزی و روش سطح پاسخ انجام گرفت. جامعه آماری شامل 700 نفر از کارکنان ادارات خدماتی استان لرستان و حجم نمونه 385 نفر بود که با روش نمونه‌گیری تصادفی طبقه‌ای انتخاب شدند. ابزار گردآوری داده‌ها پرسش‌نامه سناریو محور با 15 ترکیب سه‌سطحی بر اساس و مقیاس لیکرت 7 درجه‌ای برای سنجش نیت افشای دانش بود. روایی محتوایی با نظر پنج متخصص و پایایی با ضریب آلفای کرونباخ 86/0 تایید شد. داده‌ها با استفاده از نرم‌افزار دیزاین اکسپرت نسخه 13 و برازش مدل درجه دوم تحلیل شدند.یافته‌ها: نتایج نشان داد هر سه متغیر مستقل در سطح 1 درصد اثر معناداری بر نیت افشا دارند. ضریب مربعی «تمایل اخلاقی فردی» منفی و معنادار بود که نشان‌دهنده اشباع اثر آن در سطوح بالا است، در حالی که «فرهنگ اشتراک دانش اخلاقی» و «موانع ساختاری» اثری خطی و مستقیم داشتند. مدل هوش مصنوعی نیز روند مشابهی را نشان داد ولی ضریب خطی «تمایل اخلاقی» را بیش از دو برابر تخمین زد. در شرایط بهینه (تمایل=7، فرهنگ=7، موانع=1)، نیت افشا در مدل انسانی برابر 55/5 و در مدل هوش مصنوعی برابر 7 محاسبه شد که حدود 26 درصد تفاوت را نشان می‌دهد.نتیجه‌گیری: نیت افشای دانش حساس متاثر از ترکیب اخلاق فردی، زمینه فرهنگی و موانع ساختاری است و در شرایط بهینه نیز به‌طور کامل تحقق نمی‌یابد. یافته‌ها بیانگر آن است که هوش مصنوعی در شناسایی روندهای کلی عملکرد قابل قبولی دارد، اما نمی‌تواند پیچیدگی‌های روان‌شناختی و اجتماعی کارکنان را به‌طور کامل بازنمایی کند.اصالت/ارزش: این پژوهش نخستین مطالعه ایرانی است که با استفاده از ترکیب روش سطح پاسخ و مدل زبانی گروک-3 به تحلیل غیرخطی و مقایسه انسان–هوش مصنوعی در حوزه افشای دانش حساس می‌پردازد. استفاده هم‌زمان از سناریوهای طراحی‌شده، مدلسازی سطح پاسخ و الگوریتم‌های زبانی پیشرفته، مرزهای رایج روش‌شناسی در تحلیل‌های رفتاری را توسعه می‌دهد و زمینه تدوین خط‌مشی‌های هوشمندانه‌تری را برای ارتقاء شفافیت سازمانی فراهم می‌سازد.

nonlinear modeling of the interaction of ethical, cultural, and structural factors on the intention to disclose sensitive knowledge using response surface methodology: a human versus ai comparison

purpose: public sector organizations today operate under intensifying demands to uphold ethical standards, mitigate corruption, and ensure public trust. whistleblowing—the deliberate, principled disclosure of sensitive internal information such as financial irregularities, misconduct, or policy violations—serves as a vital corrective tool. however, the decision to blow the whistle is not driven by a single stimulus; instead, it emerges from a complex nexus of factors. an individual’s moral inclination, shaped by personal values and integrity, provides the foundational readiness to report wrongdoing. simultaneously, the prevailing ethical knowledge sharing culture within an organization either encourages or stifles such disclosures by signaling whether employees will be supported or ostracized. finally, structural barriers—bureaucratic hurdles, opaque reporting procedures, fear of retaliation, and lack of legal safeguards—directly impede the act of disclosure. while prior research has examined these dimensions in isolation and largely through linear models, the present study addresses a critical gap by modeling their joint, potentially non linear influences on whistleblowing intention. moreover, the advent of advanced language models, such as grok 3, affords a unique opportunity to compare algorithmic predictions against actual human behavior. accordingly, this research aims to (1) develop a comprehensive, non linear quantitative model capturing the combined effects of moral inclination, ethical culture, and structural barriers on whistleblowing intention among public service employees in lorestan province, iran, and (2) evaluate the comparative predictive accuracy and limitations of human survey data versus grok 3’s outputs under identical vignette scenarios.methodology: an applied, descriptive experimental field study was undertaken, employing central composite design (ccd) in conjunction with response surface methodology (rsm) to explore main, interaction, and quadratic effects. from a population of 700 service sector employees, a stratified random sample of 385 was determined via cochran’s formula, ensuring sufficient statistical power. data collection used a scenario based questionnaire featuring fifteen unique vignettes, each systematically varying three independent variables—individual moral inclination (a), ethical knowledge sharing culture (c), and structural barriers (b)—across low, medium, and high levels. participants rated their likelihood to disclose sensitive knowledge on a seven point likert scale (1 = “definitely would not” to 7 = “definitely would”). content validity was confirmed through expert review by five scholars in knowledge management and organizational behavior, and internal consistency reliability was established with cronbach’s alpha = 0.88 in a pilot test of 30 employees. in parallel, the same vignette scenarios were input into grok 3 to generate ai based intention scores. four regression frameworks—purely linear, two way interaction, quadratic (second degree), and cubic (third degree)—were fitted for both human and ai datasets using design expert 13. model selection criteria included coefficient of determination (r²), adjusted r², anova f tests, lack-of-fit tests, and sum of squared errors (sse). the quadratic model demonstrated superior explanatory power (r² = 0.95 human; r² = 0.97 ai) with non significant lack-of-fit and was selected for detailed analysis. a composite utility function was then applied to pinpoint the optimal factor combination that maximizes whistleblowing intention.findings: the human‐based quadratic model explained 95% of variance in whistleblowing intention, with highly significant coefficients. moral inclination (a) exhibited the most pronounced positive linear effect (coefficient = 0.96, p < 0.01), highlighting its central motivational role. its accompanying negative quadratic term (–0.76, p < 0.05) revealed a saturation threshold beyond which additional moral motivation produced diminishing increases in disclosure intent—an effect seldom captured in linear analyses. ethical knowledge sharing culture (c) registered a robust positive linear coefficient of 0.95 (p < 0.01) across all levels, with no evidence of saturation, signifying its consistent enabling function. structural barriers (b) exerted a significant negative linear effect (–0.59, p < 0.01), indicating that each incremental barrier unit steadily suppresses willingness to report. under the optimal conditions (a = 7, c = 7, b = 1), predicted human intention reached 5.55 on the seven point scale with a composite utility of 0.87. grok 3’s quadratic model paralleled these trends but with distinct magnitudes: a’s linear coefficient was 1.64 (p < 0.01), b = –0.60 (p < 0.01), and c = 0.80 (p < 0.01). notably, ai identified significant interactions: a×b (–0.375, p < 0.05) suggested that high moral drive combined with strong barriers markedly dampens intention, while a×c (0.225, p < 0.05) signified synergistic gains when moral drive aligns with a supportive culture. the ai quadratic a² term (–0.43, p < 0.05) reaffirmed saturation in moral motivation. grok 3’s optimal predicted intention soared to 7.00, approximately 26% higher than human respondents, reflecting ai’s lack of psychosocial risk aversion and highlighting the complexity of real‐world disclosure decisions.research limitations/implications: the cross‐sectional vignette approach, though experimentally robust, cannot fully emulate the emotional stakes, group dynamics, and organizational politics of authentic whistleblowing. the geographic focus on a single province reduces the generalizability to distinct cultural, regulatory, or institutional contexts. self‐report measures risk social desirability bias and cognitive fatigue, especially over multiple scenario evaluations. grok 3’s opaque proprietary architecture precludes in‐depth understanding of how it weights and processes scenario information. future research should incorporate longitudinal designs tracking actual disclosure behaviors, broaden samples across diverse settings, integrate objective whistleblowing records, and explore moderating influences such as employees’ perceived organizational support and individual risk tolerance.practical implications: this study furnishes a multi‐lever, evidence‐based roadmap for public sector management. first, cultivate intrinsic moral motivation via scenario‐based ethics training, peer support networks, and visible ethical leadership—but calibrate intensity to avoid motivational saturation. second, strengthen an ethical knowledge‐sharing culture by embedding transparency in organizational mission statements, celebrating exemplary whistleblowers, and maintaining open communication channels for ethical concerns. third, dismantle structural barriers through simplified, confidential reporting procedures, robust anti‐retaliation policies, and clear legal safeguards. while ai models like grok 3 can support scenario planning and policy simulations, they must complement—rather than replace—human judgment, particularly where psychosocial and cultural nuances dictate employee risk calculations.originality/value: this research represents the first iranian application of an integrated ccd rsm experimental design combined with a cutting‐edge large language model to analyze whistleblowing intentions. departing from conventional linear regression, it uncovers novel saturation dynamics in moral motivation and maps intricate factor interactions. the introduction of a utility optimization metric offers actionable guidelines for calibrating moral, cultural, and structural interventions. by juxtaposing human survey data with ai predictions, the study elucidates both the promise and the limitations of ai in ethically sensitive organizational contexts, thereby advancing methodological frontiers in the study of organizational ethics and public governance.