Measuring Creativity in the Age of Generative AI: Distinguishing Human and AI-Generated Creative Performance in Hiring and Talent Systems

A new research paper from MIT, presented at the this http URL conference, tackles the critical challenge of evaluating human creativity in an era dominated by generative AI. Authored by Yigal Rosen and Ilia Rushkin, the study argues that as large language models (LLMs) like GPT-4 and Claude enhance baseline output quality, they create ambiguity in talent assessment. The observable artifacts in hiring tasks—such as written proposals or creative briefs—may now be partially or fully AI-generated, making traditional evaluation methods obsolete.

The paper's core contribution is a reconceptualization of creativity as a distributional and process-based property. The researchers introduce a quantitative framework that operationalizes creativity as 'novelty in synthesis,' measured through idea generation and transformation within high-dimensional embedding spaces. This method aims to capture distinctions that simple quality assessments miss. Their empirical evaluation demonstrates that these metrics align with human intuitive judgments while revealing a key structural shift: creative output in AI-augmented environments tends toward a bimodal distribution.

This finding has profound implications for competitive strategy and organizational leadership. The study concludes that in the age of generative AI, the primary signal of human creative capability is no longer output fluency or polish—qualities that AI excels at—but rather distinctiveness and the ability to synthesize novel connections under shared constraints. This shifts the focus in hiring and talent systems from judging a finished product to understanding and measuring the creative process itself.