Since the 21st century, with the rapid development of information technology, the mode of social production and the basic structure of human cognition have undergone profound changes. The spread of Artificial Intelligence (AI), Big Data, Cloud Computi...
Since the 21st century, with the rapid development of information technology, the mode of social production and the basic structure of human cognition have undergone profound changes. The spread of Artificial Intelligence (AI), Big Data, Cloud Computing, and Generative Algorithms has gradually transformed cognitive activities—previously dependent on experience and intuition—into data-based and algorithm-based processes, and rapid changes are continuously taking place.
The design field is one of the areas where these changes appear most prominently. The introduction of technologies such as Computer-Aided Design (CAD), Deep Learning, and Generative Adversarial Networks (GANs) has greatly contributed to improving the efficiency of the design process and expanding the diversity of outcomes. At the same time, designers are confronted with the need to redefine their cognitive roles across the processes of creative ideation, problem structuring, and solution convergence in an environment with increased information overload and complexity. Ultimately, the paradigm shift from “intuition- and experience-centered” to “data- and algorithm-centered” not only changes the operating principles of design practice but also poses new tasks and challenging demands for the entire design education system.
From a macro perspective, major countries recognize the convergence of artificial intelligence and design as the core of future competitiveness and are actively promoting it at the national strategic level. In the case of the United States, the National Artificial Intelligence Research and Development Strategic Plan (2016); Korea’s Artificial Intelligence Industry Development Strategy (2016); and China’s Next-Generation Artificial Intelligence Development Plan (2017) commonly designate the design field as a key area for deep application of artificial intelligence technologies. Following such designations, universities and industries at the micro level are developing multidisciplinary educational models that integrate cognitive science, psychology, and computer science into design education, establishing new educational systems aimed at cultivating talents equipped with both creativity and technical capabilities. However, within these changes, important practical and theoretical issues remain regarding how to maintain the creative subjectivity of designers, and how to reallocate traditional approaches and AI-based approaches within educational systems. These issues are expected to be addressed through continuous AI-based approaches in the future.
Due to these environmental changes, existing “AI × Design” research has accumulated certain achievements in tool utilization and case analysis; however, research that comprehensively explains the interactions among cognitive mechanisms, learning behaviors, and instructional strategies remains insufficient. Many studies focus on superficial changes—such as platform updates or curriculum revisions—and systematic discussion on how changes in cognitive structure should be reflected in educational decision-making has not been sufficiently conducted.
In the bibliometric analysis conducted in this study, although the knowledge network of this field is compartmentalized by research themes, the structural characteristic of low interdisciplinary connectivity was confirmed. Through a preliminary review aimed at structural strategy analysis, data construction for 225 representative papers was performed, and keyword co-occurrence analysis was comprehensively conducted on the research sample. As a result, a total of 236 nodes and 536 links were identified, and the overall network density was 0.0193, indicating that although individual research topics are accumulating, integration between topics remains low. The Modularity Q value, which represents the quality of clustering, was 0.7749 (>0.5), showing high homogeneity within clusters but weak connections between clusters.
From the perspective of educational research, existing studies mostly remain at describing the introduction of tools or changes in assignment types. Research that models how learners’ cognitive processes change after the adoption of artificial intelligence, and what behavioral and thinking characteristics appear in the four phases—problem structuring, generating, evaluating, and converging—is very limited. Accordingly, there is also a lack of sufficient empirical evidence to establish systematic instructional strategies that reflect cognitive transitions of learners.
To compensate for these research gaps, this study establishes three interconnected research objectives.
First, at the macro level, bibliometric visualization techniques are used to systematically identify the knowledge structure and thematic context of the related field, and to identify core themes that repeatedly appear as well as relatively weak points.
Second, at the meso level, grounded theory is applied to derive the key processes and major/initial categories of design cognition under AI intervention, and to construct a cognitive mechanism map that enables comparison and contrast with the traditional paradigm.
Third, at the micro level, by integrating AHP (subjective evaluation) and CRITIC (objective evaluation), a comprehensive weighting methodology is used to quantitatively evaluate the relative importance of cognitive elements; furthermore, through a comparative experiment conducted with a traditional group and an AI-assisted group, the temporal dynamics of the two paradigms are compared. Based on this, a teaching strategy and learning outcome framework composed of “problem externalization–meaning transformation–evidence-based evaluation–stable convergence” is proposed.
This study first used CiteSpace, setting “Artificial Intelligence”, “Design Cognition”, and “Design” as the core search groups, and collected a total of 425 initial documents from the Web of Science. After applying criteria excluding non-empirical research such as conference papers, news articles, and book reviews, the data were verified, and 225 papers were confirmed as the final analysis sample. The selected documents secured nearly 100% coverage in major fields such as abstract, DOI, academic discipline, and research area, ensuring the reliability and completeness of the analysis. Examining the annual publication trend showed that related research has increased rapidly since 2020, reaching peaks in 2022 and 2024, confirming that research on this topic has accelerated in recent years.
The keyword network analysis revealed 236 nodes and 536 links, with a network density of 0.0193. In the cluster analysis, 12 core clusters were extracted, and the Modularity Q value of 0.7749 indicated that structural consistency within clusters is high, but connectivity between clusters is limited.
Next, by applying grounded theory and category construction methods, the major procedures of the traditional design process and the AI-assisted design process were derived. A total of 15 formal participants were secured through purposive sampling and snowball sampling, covering various subfields such as industrial design, interaction design, fashion design, and UX design. Data collection combined semi-structured in-depth interviews (30–60 minutes), operational demonstrations, and behavioral path observation to ensure diversity and reliability of the data. In the analysis procedure, open coding, axial coding, and selective coding were applied step-by-step to construct a multi-layered structure expanding from initial categories (F#) to major categories (D#) and core categories (A#). This category system served as the basis for constructing indicators and distinguishing dimensions in subsequent quantitative evaluation.
To evaluate the relative importance of the categories derived from grounded theory, a comprehensive weighting procedure combining AHP and CRITIC was performed. In the AHP stage, “exploring cognitive pattern changes of designers in the AI environment” was set as the highest-level goal, and a hierarchical structure (core categories–major categories) enabling comparison between the traditional paradigm and the AI-assisted paradigm was constructed. The expert group consisted of nine members, and pairwise comparisons were conducted using Saaty’s 9-point scale. All judgment matrices were verified by calculating the Consistency Index (CI) and Consistency Ratio (CR), and only matrices satisfying CR ≤ 0.10 were included in the final analysis. Matrices showing inconsistency were revised and adjusted through repeated feedback to secure consistency.
In the CRITIC stage, the discriminative power of each evaluation indicator was calculated using standard deviation, and conflict was assessed by computing correlation coefficients among indicators. After normalization, the objective information content of each indicator was derived and weights were calculated. This method provides objective weights based on data, complementing AHP, which depends on expert judgments. The integrated weighting results showed that AHP and CRITIC commonly identified “evidence-based evaluation,” “semantic coherence,” and “convergence stability” as key factors, indicating that AI introduction forms a balanced mechanism that not only accelerates generation but also strengthens governance and control capabilities.
To identify the cognitive time allocation structures of the traditional design group and the AI-assisted design group, a comparative experiment was conducted. Both groups performed a product concept design task with the same goal and under the same experimental environment, with differences only in the tools and collaboration methods used. The traditional group used a total of 5,400 seconds (90 minutes), and the AI-assisted group used 75 minutes as the basis for analysis.
For the traditional group, the time proportion by core category was highest for A3 (alternative construction and iteration) at 41.3% (2,228 seconds), followed by A2 (creative association and inspiration) at 21.9% (1,180 seconds), A4 (reflective evaluation and subjective judgment) at 19.3% (1,043 seconds), and A1 (problem recognition and requirement identification) at 17.6% (949 seconds). At the major category level, B13 (16.4%/883s), B8 (11.2%/604s), B12 (10.7%/577s), and B11 (9.4%/506s) accounted for high proportions, forming a closed iterative loop of “structural expansion–framework formation–core line modification–alternative comparison.”
In contrast, the AI-assisted group exhibited a distinct pattern in which the cognitive structure was heavily concentrated on “human-led convergence and governance” rather than on divergent phases. D10 (manual optimization) accounted for 17.0% (763s), and D13 (critical review) for 12.8% (576s). The governance axis (D13·D15·D14·D16·D18) accounted for 38.8% of total time. Early divergence (C1: D1–D4) and generation-centered adjustments (C2: D5–D7) appeared as “short and condensed repetitive adjustment loops.” The overall structure showed an “early compression–mid joint construction–late reinforcement” pattern, with the majority of time invested in decision-making (C3: 37.2%) and metacognitive regulation (C4: 38.8%). These results suggest that AI intervention induces a reinforced cognitive structure centered on convergence, judgment, and governance rather than divergent exploration.
In summary, the main findings of this study are as follows:
First, AI intervention did not weaken the designer’s agency; rather, it acted as a factor that promoted the reorganization of design cognitive structure. Although the AI group had a shorter total working time than the traditional group (approximately 75 minutes vs. 90 minutes), the cognitive focus clearly shifted. The initial preparation stage and divergent generation process were compressed, while evidence-based evaluation, critical review, and passive governance in the mid- and late-stage processes accounted for a large proportion of time. This shows that AI does not simply replace design thinking but readjusts the designer’s role from a “creative producer” to a “core agent of evaluation, regulation, and governance.”
Second, through grounded theory analysis and AHP–CRITIC-based comprehensive weighting, it was confirmed that “evidence-based evaluation” and “convergence stability” emerge as the most critical cognitive elements in the AI environment. The weights of these factors were significantly higher than those in the traditional intuition-dependent process, indicating that cognitive mechanisms are being reorganized from generation-centered to evaluation- and verification-centered processes after AI intervention. This result goes beyond the limitations of existing research, which mainly focused on tool usage or case description, by integrating qualitative analysis and quantitative analysis to empirically present the priorities of “generation–evaluation–governance.”
Third, by establishing a comprehensive research pathway connecting “bibliometric–qualitative–quantitative–experimental–transformational” methods, this study proposes a new methodological framework for design cognition research. The bibliometric analysis identified the macro structure of the research field, the qualitative interviews and weighting analysis produced a systematic cognitive framework, and the comparative experiment empirically verified the temporal dynamics of key cognitive factors. This multi-layered, evidence-based approach represents an important academic achievement of this study and suggests a replicable and expandable analytical pathway for future interdisciplinary research.
Based on the above findings, the following three suggestions were derived:
First, design education should focus not only on improving tool usage skills but also on strengthening critical evaluation and governance capabilities. It is necessary to set questioning ability, explanatory ability, judgment ability, and evidence presentation ability as core learning objectives.
Second, teaching and learning processes should systematically strengthen evidencing and process recording, and by establishing version tracking and standardized evaluation mechanisms, ensure the verifiability and re-confirmability of learning outcomes.
Third, research and practice should actively expand interdisciplinary integration and cross-cultural comparative studies so that the field of “AI × Design Cognition” can move beyond parallel and fragmented development toward systematic convergence.