發表評論-心理與性-萬維博客-萬維讀者網（電腦版）

【摘要】在人才選拔流程中，人們採用了多種多樣的工具；隨着人工智能（AI）技術的不斷演進，AI解決方案的開發已逐步滲透至員工選拔的特定領域，尤以申請人簡歷篩選及非同步視頻面試環節為甚。然而，利用結合機器學習的AI模型來解讀基於文本的評估中心（AC）模擬測試作答內容，這一領域目前仍鮮有人涉足。本研究的主要目的在於評估某款AI模型的聚合效度與效標效度。本研究的次要目的則是探究：相較於人類評估者（其評分常受限於“範圍受限”效應），該AI算法在對書面模擬作答進行評分時，是否能更充分地利用評分量表上的分值區間。研究人員利用AI技術對15,000名領導者的基於文本的AC模擬作答輸出進行了分析——這些文本數據總量高達3300萬字——並從中識別出了38項勝任力。隨後，該AI模型被用於對來自三個獨立領導者樣本的模擬測試結果進行評分，以此評估其聚合效度與效標效度。研究結果顯示，該模型的聚合效度介於0.63至0.73之間，效標效度則介於0.51至0.54之間。在評分範圍的利用方面，標準差指標顯示，該AI模型所使用的評分分值區間較人類評估者更為寬泛。本研究基於三個獨立樣本所呈現的實證結果表明：首先，AI算法在對書面文本進行評分時，其評分方式與人類評估者具有高度相似性。其次，AI算法在充分利用可用的全部分值區間方面表現得更為出色。該AI模型在各項指標上似乎已達到與人類評估者比肩的水平——其準確率與召回率指標均高達0.91——這預示着AI模型在未來有望輔助甚至替代人類評估者開展工作。

【關鍵詞】人工智能，領導力勝任力，選拔與評估，領導力發展，人工智能與機器學習

[Abstract] A wide range of instruments is employed within the talent selection pipeline, and, with the progressive evolution of artificial intelligence (AI), the development of AI solutions has made inroads into certain areas of staff selection, notably the processes of reviewing applicant resumes and conducting asynchronous video interviews. However, the use of AI models with machine learning to interpret text-based assessment center (AC) simulation responses has remained largely unexplored. The main aim of this study was to assess the convergent and criterion validity of an AI model. A secondary objective of the study was to see if the AI algorithm utilized more scale points in rating written simulation responses compared to human assessors’, whose scores typically suffer from range restriction. AI was used to analyze the text-based AC simulation outputs of 15,000 leaders, comprising 33 million words, and 38 competencies were discovered. The AI model was then used to score the simulation results of three separate samples of leaders to assess its convergent and criterion validity. The results showed convergent validities ranging from 0.63 to 0.73 and criterion validities ranging from 0.51 to 0.54. In terms of range utilization, the standard deviation indicated that the AI model utilized a wider range of scores than human assessors did. Empirical results presented in this study across three samples suggest that AI algorithms can score written text in a similar way as human raters. Second, the AI algorithms are better at utilizing the full range of scores available. The AI model seems to be on par with the human rater, with accuracy and recall metrics at 0.91, indicating the possibility of augmenting or replacing the human assessor.

[Key words] artificial intelligence, leadership competencies, selection and assessment, leadership development, artificial intelligence machine learning

論文原文：Bronkhorst, P. V., & Becker, J. (2026). Use of artificial intelligence in leadership competency development and selection: An empirical study. Consulting Psychology Journal, Volume 78, Issue 1, Pages 1–26. March 2026. https://doi.org/10.1037/cpb0000288

（翻譯兼責任編輯：MARY）

（需要英文原文的朋友，請聯繫微信：millerdeng95或iacmsp）