Assessing students' creativity often involves examining how many ideas they generate and how original those ideas are. Yet final answers alone may not fully reveal how ideas develop or what students can achieve with appropriate support. A student whose initial performance appears unremarkable may reconsider a problem and develop a more refined solution with guidance. Examining this process could help educators recognize creative potential that might otherwise go unnoticed.
Researchers at East China Normal University, working with collaborators at the Shanghai Innovation Institute, have developed CREATIVA, an interactive creativity assessment system powered by large language models. A preliminary study published in ECNU Review of Education presents the system's theoretical foundations, design, and pilot findings. Corresponding author Jiajun Guo led the project's conceptualization and overall direction. By recording students' problem-solving processes and their responses to AI support, the research team seeks to explore an approach to creativity assessment that places greater emphasis on developmental potential.
Traditional creativity assessments commonly examine the characteristics of students' outputs, such as the number and originality of their ideas. Although these measures capture performance, they offer less insight into how students understand problems, adjust their thinking, and develop their ideas with support. CREATIVA adopts a "process-dynamic" approach that incorporates the development and refinement of ideas into the assessment process. Its design draws on the concept of the "Zone of Proximal Creativity," which focuses on the creative performance students can demonstrate with appropriate support and individual differences in how they respond to that support. By comparing initial solutions with revisions made after receiving guidance, the system seeks to provide a richer basis for understanding creative potential.
Built on the educational large language model InnoSpark, CREATIVA comprises four stages: problem discovery, problem identification, solution generation, and solution refinement. In a scenario involving an "AI companion," for example, students first identify challenges worth addressing, select a central problem and analyze it from multiple perspectives, propose solutions, and refine them in response to feedback. The task thus follows a continuous process from discovering problems to developing and improving solutions.
The AI assumes different roles throughout the assessment. It initially provides background information and task guidance to help students understand the scenario. During solution refinement, it offers graduated prompts that encourage students to consider feasibility, potential risks, and areas for improvement. Students then revise and submit their final solutions, allowing the assessment to capture how they respond to support and develop their ideas.
In an initial evaluation, 29 undergraduate students enrolled in a psychology course completed CREATIVA alongside two existing creativity measures: the Alternative Uses Task and Realistic Presented Problems. Five graduate students independently rated participants' CREATIVA responses using the assessment framework. Across the 12 scoring dimensions reported in the study, consistency coefficients ranged from 0.80 to 0.94, indicating a high degree of consistency among human raters. The researchers also found associations between selected CREATIVA indicators and scores on the existing creativity measures. These findings provide preliminary psychometric support for the system, suggesting that some aspects of the performance it captures are related to those assessed by established assessment tools.
The team also piloted CREATIVA with elementary, junior high, and senior high school students, adapting task scenarios and levels of guidance to their curricular content and background knowledge. One notable observation was that some students who performed less well during the initial problem discovery stage submitted higher-quality solutions after receiving AI support. This suggests that students' responsiveness to guidance may reveal aspects of their creative potential that are less apparent in their initial performance.
These findings remain preliminary. The school pilots do not establish that the system produces lasting improvements in creativity or that it has been broadly validated across educational settings. Further research is needed to examine its applicability across subjects and learning contexts and to determine how its assessment results can inform instructional support and talent identification.
CREATIVA offers a pathway toward creativity assessment by bringing greater attention to the process through which students develop their ideas. By examining how students respond to guidance and revise their ideas, the researchers hope to help educators gain a fuller understanding of both learners' demonstrated abilities and the creative potential that appropriate support may help bring out.
Journal
ECNU Review of Education
Method of Research
Data/statistical analysis
Subject of Research
Not applicable
Article Title
CREATIVA: A Preliminary Report on a Process-Dynamic Creativity Assessment System Powered by Large Language Models
Article Publication Date
16-Sep-2026
COI Statement
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.