Who Leads the Design? The Architect or the Artificial Intelligence

When the “Editor” Becomes Less Creative Than the Human Working Alone

There is a paradox that troubles every architect who has begun using generative tools in the studio: the more polished and complete an AI-generated output appears at the outset, the more the designer senses that the work “does not belong to them.” Two controlled poetry-writing experiments, led by researcher McGuire and colleagues, found that participants who received a complete poem from an AI system and then merely edited it produced work that professional poets rated as less creative than that of participants who wrote entirely on their own, without any machine assistance. This finding, though drawn from literary practice, raises a fundamental question for the architectural studio: is the difference between “an architect who leads and directs the tool” and “an architect who merely corrects what the tool has produced” a matter of degree, or a matter of kind?

The answer, according to a cluster of recent experimental studies, is that the difference is fundamental — and its effects extend into five sensitive dimensions of architectural practice: the designer’s confidence in their own abilities, their engagement with the task, their sense of ownership over the work, the quality of their cognitive involvement, and the fidelity of the design to cultural identity when the work involves reviving traditional crafts and motifs.

Editor Versus Partner: Why Sequencing Makes All the Difference

The central idea advanced by researcher Rafner and colleagues in their foundational paper is that the essence of creative collaboration with machines lies in the balance between automation and human control — a balance that must preserve “the creative intent of the human” rather than letting it dissolve into machine output. This is precisely what McGuire and his team tested in practice. In their first experiment, participants were placed in the role of “editor,” receiving a finished piece from the AI and then revising it; this group produced the weakest creative results among all conditions, while the AI working entirely on its own, without any human intervention, ranked last of all. But in the second experiment, when the researchers redesigned the interface so that the participant wrote the opening line and chose the overall concept, then exchanged turns with the AI in an equal dialogue resembling a “roundtable discussion,” the gap between those who worked alone and those who collaborated with the machine disappeared entirely — and this collaborative approach clearly outperformed the “post-hoc editing” approach.

This finding has a direct translation into the architectural office. When a design partner is tasked with reviewing a massing diagram or a facade proposed by a generative tool, they fall into a psychological “anchoring” trap — unconsciously tending to adopt the machine’s initial decisions rather than challenging them, which narrows the space for genuine initiative. But when the architect begins by defining the design concept and the functional and site-specific constraints themselves, then uses the tool as an interlocutor offering alternatives that the architect evaluates and directs, the resulting creative output returns to — and potentially exceeds — the level achieved by unassisted work.

Testing Confidence: Does the Tool Build a More Confident Architect, or a More Dependent One?

Measuring “creative self-efficacy” — an individual’s belief in their capacity to produce innovative work — revealed the clearest gap between the two modes. In McGuire’s experiments, this measure was notably lower among those who played the “editor” role compared to those who designed alone or collaborated with the machine from the outset, and statistical analysis confirmed that this decline in self-confidence was the mediating link explaining the weaker creativity ratings among the “editing” group. The researchers attribute this to the fact that those who set their own creative direction — rather than merely reacting to a finished output — retain their autonomy and intrinsic motivation.

A broader survey study, conducted by researchers Zhang and Xu on 348 Chinese university students, documented a troubling paradox: as the frequency of generative-tool use increased, students’ apparent self-confidence and sense of competence in problem-solving rose, but this rise was accompanied — with even greater statistical strength — by deepening technological dependence on the tool itself. Projected onto architecture, this finding suggests that young students or architects who grow accustomed to constant reliance on generative image and drawing tools may feel increasingly confident in their skills, while their actual capacity for independent design deteriorates once the tool is withdrawn — a fundamental challenge for architectural education programs that have begun integrating these tools into the studio without clearly defined boundaries of use.

Who Owns the Design? The Question of Psychological Ownership in the Age of the “Artificial Partner”

Researchers Xu, Cheng, and Kuzminykh raise a striking point: the more a generative tool appears “personified” — perceived by the user as an independent partner with its own agency rather than a mere instrument — the less the human feels their actual contribution to the work matters, and the more their psychological ownership over the final output diminishes. Their explicit recommendation to designers is to reduce the apparent “human” traits in these tools’ interfaces and to maximize the user’s space for autonomy, in order to preserve the sense that the work carries their imprint.

This intersects with findings from another research team led by Westphal, who tested across three experiments — involving nearly 480 participants combined — the effect of granting users the ability to adjust and control the output of an intelligent system. The result was that the ability to modify a system’s recommendations tangibly increased levels of trust, understanding, and intended compliance with the output. In other words, control over the decision is not a technical detail — it is the essential condition for a designer’s sense that the work is theirs. When a firm’s workflow is structured so that the architect always retains the option to modify, reject, and direct — rather than simply accepting or rejecting outright — the firm protects one of its most valuable professional assets: the sense of authorship.

When to Give the Machine Free Rein, and When to Rein It In: A Lesson from the Phases of the Creative Process

Among the most precise findings offering a practical resolution to the “who leads” dilemma is work by a research team led by Huang, conducted through qualitative interviews, two laboratory experiments, and an online experiment involving 198 participants. The conclusion of this work is that collaboration with AI during the idea-generation phase enhances “cognitive flexibility” and breaks habitual thought patterns, producing more novel and inventive ideas. But during the refinement and elaboration phase, the tool’s most effective role shifts to reducing the “excess cognitive load” that burdens the designer, thereby raising the level of feasibility and coherence in the final concept — without necessarily increasing its novelty.

The practical implication of this for the architectural studio is clear: AI should not be treated as a single tool used the same way throughout every design phase. In an early brainstorming session, when the team is searching for a bold conceptual direction for a building, the tool can be given greater freedom to propose unexpected possibilities. But once the project enters detailed development — coordinating structure with space, aligning facades with municipal requirements — the tool’s more appropriate role becomes helping the architect manage informational complexity, rather than proposing new conceptual alternatives that might disrupt the decision-making path the human has already charted.

When Heritage Is Reformulated Mechanically: Who Protects the Identity of the Motif?

In the context of architecture and design connected to cultural heritage — such as reviving traditional craft patterns and motifs within contemporary products and facades — the question of “cultural fidelity” becomes the most critical criterion of all. A study by researchers He and Tao on Chinese “blue calico” cloth, a craft registered as intangible cultural heritage, developed an enhanced AI model that achieved tangible improvements in transferring traditional color gradients and linear patterns onto modern products. Yet the researchers themselves caution that the field as a whole suffers from “distortion of cultural features and weak technical convertibility into commercially successful products,” noting that only a small percentage of traditional-craft digitization projects in developing countries succeed in reaching actual markets.

Another striking finding, from researchers Lu, Song, and Zhang, reveals that a single generative model exhibits different cultural tendencies depending on the language used to query it — when queried in Chinese, it tends toward a more holistic, socially interdependent mode of thinking compared to when queried in English. This has direct implications for any architectural firm using these tools to reimagine a local heritage motif: phrasing the request (the prompt) in the region’s own language and cultural framework is not a linguistic detail but an actual steering mechanism, one capable of tilting the output toward the desired identity or away from it.

In a parallel behavioral study, researchers Bui and Filimonau found that images generated by AI, when presented without disclosure of their machine origin, appear significantly more “authentic” to viewers than when their AI origin is disclosed — whereas human-made images are rated as more authentic precisely when their human origin is disclosed. This finding places architects and designers before a genuine ethical-marketing dilemma when using AI to restore architectural or craft heritage: transparency is professionally required, yet it may weaken the public’s perceived authenticity of the resulting product.

The Complete Image Constrains Imagination: A Lesson from Chinese Ceramic Design

Among the most practically useful findings for studio designers comes from an experiment by researcher Zhu, in which novice designers drew inspiration from Liao Dynasty Chinese ceramics. The study found that presenting partial images with low visual detail, accompanied by narrative texts bearing a moderate metaphorical distance from the final form, stimulated significantly more creative and culturally rich designs — while presenting the complete, clear heritage image led to what is known as “design fixation,” in which the designer’s imagination becomes confined within the literal form presented to them.

This suggests that AI tools generating a “complete and ready-made” version of a given heritage motif for the architect may in fact be less stimulating to creativity than a tool that offers partial cues or narrative fragments about the craft, then leaves room for the architect’s own imagination to build upon them — bringing us back to the core idea of this article: that the human must begin, not merely correct what the machine has already finished formulating.

✦ ArchUp Editorial Insight

The debate over whether architects should lead or merely edit generative tools is not a design conversation; it is a labor-economics conversation. Firms adopt “editor” workflows because they compress schedule and fee pressure, not because they produce better design authorship — the research cited simply confirms what procurement logic already dictates: speed rewards acceptance, not interrogation. Cultural-heritage digitization projects follow the same incentive structure, where market-conversion targets favor complete, licensable outputs over partial, ambiguous ones, even when completeness is what suppresses interpretive depth. What emerges as “design ownership” or “cultural fidelity” is therefore not a creative variable but a downstream artifact of who controls timeline, disclosure, and compensation. The architecture — original or derivative, faithful or diluted — is simply where these upstream governance decisions become visible.

References

Lee, B.C., and Chung, J. “An Empirical Investigation of the Impact of ChatGPT on Creativity.” Nature Human Behaviour, 2024.

Rafner, J., Beaty, R.E., Kaufman, J.C., Lubart, T., and Sherson, J. “Creativity in the Age of Generative AI.” Nature Human Behaviour, 2023.

McGuire, J., De Cremer, D., and Van de Cruys, T. “Establishing the Importance of Co-Creation and Self-Efficacy in Creative Collaboration with Artificial Intelligence.” Scientific Reports, 2024.

Zhang, L., and Xu, J. “The Paradox of Self-Efficacy and Technological Dependence: Unraveling Generative AI’s Impact on University Students’ Task Completion.” The Internet and Higher Education, 2025.

He, J., and Tao, H. “Applied Research on Innovation and Development of Blue Calico of Chinese Intangible Cultural Heritage Based on Artificial Intelligence.” Scientific Reports, 2025.

Lu, J.G., Song, L.L., and Zhang, L.D. “Cultural Tendencies in Generative AI.” Nature Human Behaviour, 2025.

Xu, Y., Cheng, M., and Kuzminykh, A. “Am I Working with ‘Someone’? Psychological Ownership as a Lens for Designing Agentic GenAI.” Proceedings of the IEEE International Conference on Collaborative Advances in Software and Computing, 2025.

Grassini, S. “Computational Power and Subjective Quality of AI-Generated Outputs: The Case of Aesthetic Judgement and Positive Emotions in AI-Generated Art.” International Journal of Human–Computer Interaction, 2024.

Peng, Y., Wang, X., Chen, J., et al. “The Impact of AI-Supported Instructional Interventions on Students’ Creative Thinking: A Meta-Analysis of 56 Experimental Studies.” Journal of Research on Technology in Education, 2025.

Wang, Y., and Xue, L. “Using AI-Driven Chatbots to Foster Chinese EFL Students’ Academic Engagement: An Intervention Study.” Computers in Human Behavior, 2024.

Huang, S., Long, L., Zhu, Y., and Zhu, J.N.Y. “Human–GenAI Collaboration Across Creative Phases: Cognitive Mechanisms Shaping Novelty and Usefulness.” International Journal of Information Management, 2026.

Westphal, M., Vössing, M., Satzger, G., Yom-Tov, G.B., and Rafaeli, A. “Decision Control and Explanations in Human-AI Collaboration: Improving User Perceptions and Compliance.” Computers in Human Behavior, 2023.

Bui, H.T., Filimonau, V., and Sezerel, H. “AI-thenticity: Exploring the Effect of Perceived Authenticity of AI-Generated Visual Content on Tourist Patronage Intentions.” Journal of Destination Marketing & Management, 2024.

Zhu, Y. “Comparing the Effects of Different Types of Cultural Inspiration on Design Creativity.” The Design Journal, 2020.