Product Ideation in the Age of Artificial Intelligence: Insights on Design Process Through Shape Coding Social Robots

2.1 Current Scenario of GenAI in Design

Creativity can be positioned as a combination of novelty and usefulness (Doshi & Hauser, 2024; Harvey & Berry, 2023). The quality of creative ideas can, therefore, be judged on the level of their novelty and the usefulness to the problem they are addressing to resolve. Research on measuring the impact of GenAI on creativity still needs more development, especially in the disciplinary area of product design. The interaction of organic human creativity assisted by GenAI tools remains largely underexplored. Understanding how such computational tools can be integrated into a classical design process workflow and measuring their influence on the quality and depth of the concepts generated is a subject matter still under study. Studies on the impact of GenAI on general creativity have only recently emerged (Doshi & Hauser, 2024), and studies with a narrower focus on product design specifically have yet to address the potential and limitations of GenAI augmented ideation and co-creation of concepts.

The published literature on the subject argues that the deployment of GenAI increases overall creativity in a sample group of creators but ultimately reduces the results’ diversity (Doshi & Hauser, 2024). Furthermore, it is the creators, weak in creativity, whose output benefits most through accessing GenAI, and conversely, highly creative creators do not benefit from access to GenAI and using it may even degrade their output quality (Doshi & Hauser, 2024; Figoli, Rampino, & Mattioli, 2022).

Current generation, text-to-image, and sketch-to-image GenAI tools cannot generate 3D forms and models, limiting their output to 2D shapes and images. Although text-to-image GenAI tools like Midjourney can generate a vast library of image proposals, they fail to understand domain-specific terms and generate solutions unsuitable for manufacturing purposes (Zhang, Wang, Pangaro, Martelaro, & Byrne, 2023); therefore, they often visualise surprising and unexpected results, which can both detract from or boost the ideation work. In fact, this probabilistic nature of GenAI has been debated in recent literature (Hicks, Humphries, & Slater, 2024), which questioned whether LLM-based GenAI technologies understand any of the output they produce at all, arguing that text-based GenAI applications are “bullshit machines.”

A word of caution to designers considering to deploy GenAI in ideation, which merits attention, is that the dependence of GenAI tools on training data may condition the quality of output depending on the product category being designed. For product categories where a large amount of training data exists, potentially classical products, the probability of encountering creative and innovative proposals is higher than in emerging or futuristic product categories, on which much creative work may not have occurred. The issue of design fixation (Hoggenmueller, Lupetti, Van Der Maden, & Grace, 2023) has been discussed while deploying GenAI for experimenting with creative ideation on social robotics. Design fixation is a “blind adherence” (Jansson & Smith, 1991) to set ideas and concepts that can actually retrogress instead of progressing ideation work to explore new directions.

Regardless of the product category under consideration, the issue of inserting textual descriptions in an appropriately structured fashion, or “prompt engineering,” must be given due consideration to achieve more constructive and valuable results.

With these limitations in mind, due consideration must be given while selecting the GenAI platforms to be used, as they are most likely built on qualitatively different training data (Hoggenmueller et al., 2023) and have different prompt engineering requirements. Nevertheless, due to the recent nature of these software tools, such information remains a “black box” for all practical purposes, and designers must adopt a trial-and-error approach, consequently accepting the probabilistic nature of the output these tools offer.

However, even with these limitations considered, text-to-image GenAI technology has the potential to reshape the ideation work in the concept design phase by acting as a catalyst towards designers through its very quality of generating a numerically vast number of proposals with the inherent element of surprise hard-baked into its mechanics. This probabilistic nature of the GenAI tools introduces the element of serendipity into the ideation workflow, thereby causing unexpected changes in direction due to this very feature.

The resulting benefit to the workflow is the ability to broaden the exploration of possible solutions, which novice designers may otherwise never have considered if working exclusively with traditional tools based on classical methods (Chiou, Hung, Liang, & Wang, 2023). Thus, GenAI assists novice designers in overcoming creative blocks by stimulating their imagination through visual stimuli and providing inspiration.

2.2 Novelty of this Research

The authors of this study propose an effective strategy to insert text-to-image and sketch-to-image GenAI tools in the product design process and identify ideation, which occurs during the concept generation phase of any design process model, as the appropriate moment to insert such tools and technologies. By adopting a hybrid approach, combining computational tools like GenAI with classical tools like drawing and modelling, i.e. blending, this research will demonstrate an effective technique to exploit GenAI for achieving quality results.

Insertion of GenAI tools anywhere into the design process is yet a very much emerging area of design research (Figoli et al., 2022). The resulting human-AI collaborative co-design paradigm has been cited as a phenomenon with much promise and potential in existing literature (Figoli et al., 2022). With the emergence of agentic AI technologies and reasoning models in 2025, this human agent and AI system co-design issue will open up even more interesting opportunities.

The strategy of inserting GenAI increases the sheer number of concept designs generated, thus allowing for rapid exploration of a wide range of ideas (Figure 2), allowing the designers to quickly generate multiple form representations by modifying the input text (Kulkarni et al., 2023). AI systems can throw “random visual stimuli” (Figoli et al., 2022), encouraging designers to explore novel and unconventional ideas. The GenAI tools used in this study (Midjourney and Vizcom) can visualise concepts that may be difficult or time-consuming to create traditionally through classical techniques such as drawing or 3D modelling.

However, it is essential to maintain recognition of the limitations of text-to-image GenAI in the ideation phase, as already discussed above. While the technology can generate visually appealing and diverse concepts, it may need more contextual understanding and domain-specific knowledge, i.e. originality, manufacturability, materials, and human–product interaction, among others. This is knowledge that human designers supposedly possess; therefore, this limitation of GenAI validates the presence of the creative human designer in the workflow.

Indeed, as per the definition of creativity used in this article, GenAI only partially fulfils the definition, offering only novelty but completely disregarding utility or usefulness. In this sense, GenAI is a good assistive boost to the already creative human designer, though not a replacement for them. The rigour of the current model of the university classroom studio-based training of product design students focuses quite robustly on the competency to critically assess, dispassionately judge, and unbiasedly qualify concept designs against novelty and usefulness. This capability is combined with soft s’kills based on tacit knowledge, reading social cues, identifying cultural appropriateness, and some hard skills, including understanding manufacturability, technical limitations, and economic viability. It is essential to understand that the current generation of Genai does not encapsulate all the competencies that a human designer is understood to possess through classroom training.

Additionally, the generated concepts may not always align with the specific constraints, requirements, or aesthetic preferences of a given design project. Despite these limitations, studies have shown that using text-to-image GenAI in ideation, designers can explore a broader range of ideas and push the boundaries of their creativity (Paananen, Oppenlaender, & Visuri, 2023). However, to ensure the generated concepts’ quality and relevance, designers must critically evaluate and refine AI-generated ideas with rigorous post-production work, combining them with their own expertise and judgement.

Finally, designing social robots presents a uniquely challenging product category to students because, as a novel emerging product category, the aesthetic quality based on form and composition needs to be carefully balanced with meaning transmission based on language and cultural context to arrive at a design solution with high social acceptability. The lack of existing visual vocabulary to refer to terms of shape codes, pitfalls of social biases, preconceived notions, and the phenomenon of design fixation are threats that can contaminate the results of the GenAI tools by leaking into their training data, thus holding the potential to easily distract a novice designer towards suboptimal solutions.

The designers, therefore, need to cast a very wide net in prompting GenAI during the very beginning of a design process in form exploration exercises where ideation usually occurs, to extract the maximum benefit of this technology. In novice designers, especially those not yet well trained on traditional ideation methods, denying access to GenAI may limit their ideation creativity. The authors have first-hand empirical experience of students struggling with the complexity of finding an equilibrium between aesthetic expression and meaning expression while ideating form, shape codes of products. Furthermore, the iterative and linear nature of sketching and 3D model making is time consuming and may restrict more profound exploration of unconventional form solutions. Introducing text-to-image GenAI tools into the mix may make it possible to increase the divergence of the exploration of conceptual forms, although in theory using GenAI for emerging products may not always be helpful, given that GenAI generates visuals based on training data. Currently, there may not be a lot of existing visual content on social robot designs to refer to when generating proposals.

3.1 Methodology Design

This research is structured as a classroom teaching-based exploratory case study focused on the particular subject of how the ideation work in the concept design phase in a design process is impacted by the introduction of text-to-image and sketch-to-image GenAI tools. Specifically, it examines the potential benefits for students, how their work flow is affected, and the impact on the degree of divergence in ideation. The execution strategy of this research was formatted as design workshops in a practical studio project subject involving Master-level design students. Four workshop editions were conducted at a European Higher Education Institution (HEI) between 2021 and 2024.

The workshop project theme was to shape-code concept designs for social robots with semi-human or pet-like morphologies, with a different specialised subcategory in each workshop edition (companion robot, healthcare assistive robot, educational robot, and general-purpose domestic robot, respectively). The limitation on morphology was imposed because existing literature on social robotics highlights the pitfalls of the uncanny valley phenomenon when approaching humanoid robots (Causo, Vo, Chen, & Yeo, 2016). The respective subcategories were selected by consensus among the co-authors, who agreed on their relevance based on emerging literature in the social robots HRI community.

The first two editions (2021, 2022) employed traditional organic HI ideation methods such as semantic moodboarding, sketching, 3D modelling, and conceptual rendering. The last two editions (2023, 2024) integrated organic HI ideation with synthetic GenAI-assisted ideation methods based principally on Midjourney (n.d.), Vizcom (n.d.), and occasions also DALL-E (OpenAI, n.d.).

The inclusion and exclusion of GenAI in the workshop editions allowed the authors to undertake a comparative analysis of the general process and resultant output occurring in the two test conditions – with and without the injection of GenAI tools in the concept development tasks given to the students.

The decision to insert GenAI into this research on shape coding social robotics was to test the viability of this technology towards concept generation and identify the position of the authors regarding the intense debates surrounding this technology, 2022 onward. The specific tools indicated above were thus selected due to the significant attention in the design community surrounding their emergence in 2022, as well as their commercial availability during this period. Bearing in mind that Midjourney required a subscription to access beyond 50 generation cycles, DALL-E was used as a backup in case the students could not achieve the desired results within the limited access, as DALL-E had free access during this time.

This research used two kinds of GenAI tools: text-to-image and sketch-to-image. Text-to-image tools like Midjourney and DALL.E generate computer-rendered images based on a descriptive text input or prompt. Sketch-to-image tools like Vizcom are particularly targeted towards the product design community and work on generating images based on sketches, which the designer must first upload into the platform. The hand-drawn sketch, combined with a text prompt, which complements the sketch, then triggers the generation of the generative computer-rendered image. Both the typologies of GenAI tools can combine the generated synthetic images with other existing images which the designer may possess (sketch, moodboard, visual benchmark research) and it is possible to generate new images combining these two images, i.e. blend, to generate yet another image combining the visual information and shape codes to achieve a more sophisticated and aesthetically targeted outcome. Vizcom has almost real-time rendering capabilities, allowing students to transform basic analogue line sketches into computer-generated graphics for immediate presentation. Both these types of tools can be used in conjunction or in a complementary fashion, providing the student with a toolkit of GenAI that aligns well with the experimental needs of product design ideation, needing rapid and free-flowing form exploration mechanics. Figure 1 explains the user interface (UI) of Midjourney and Vizcom, explaining the differences between various text-to-image and sketch-to-image operations.

Figure 1

The UI of Midjourney (top) and Vizcom (bottom). Top left, referring to the project Vivo (educational robot), shows Midjourney image generation operation, which works on the Discord server platform through a bot-based interaction. Each iteration of the prompt cycle generates four images. The designer may select the most preferred image to trigger the next iteration cycle. The top right shows the image blending operation in Midjourney. The designer picked an image from a semantic moodboard of “calm” and combined it with the Midjourney-generated synthetic image to achieve blended results. Bottom referring project Hug (general purpose assistive robot) shows the UI setup of Vizcom, the central canvas space is where the base reference sketch of the designer is uploaded, the right column contains the prompt box, below the prompt area is the style reference image upload control which allows a aesthetic influence reference image to be uploaded, and lastly, at the bottom is the slider which adjusts the level of intervention the designer grants to the GenAI to modify the image.

Shaping the prompt, i.e. prompt engineering, by inserting the appropriate pragmatics (phrasing), semantics (vocabulary), syntax, sentence structuring, and specialised rendering terminology (photorealistic, 4K, wide angle, etc.) is still very much an art form of trial and error, probabilistic in nature and yet an emerging knowledge area. In each of the two editions of the workshops (2023 and 2024), when GenAI was used, a 1 h session was devoted to prompt engineering for aligning the students with the basics of using these tools, before the students began their ideation work.

All the results were captured and arranged for a classroom presentation at the conclusion of each workshop edition using Miro, a collaborative visual boarding platform. Miro permits the simultaneous participation of all the students in the same virtual workspace; therefore, the students can observe and draw inferences from the content posted together with their colleagues. The coordinating teacher created a collaborative Miro classroom workspace that included all the students and all the teachers. The students could post their project development, and the teachers could review and leave their comments. The final project presentations were held in the same studio setting where the workshops were conducted, with care taken to maintain an informal yet professional ambience. Each student received 15 min to present the project development process and the final output by sharing the Miro board and compiling a PowerPoint presentation. A physical scaled model mockup was an optional output that students could build to support their concept design.

3.2 Workshop Design

The workshops were formatted as three-day engagements, with the first day used for theme introduction, technical research, visual research, and selecting three symbolic keywords, as well as semantic moodboarding for each of the three keywords. The second day was divided into two halves, beginning with theory-practical training on GenAI tools and analogue sketching targeted at shape coding the symbolic keywords through the moodboards. Then, GenAI deployment was followed for form exploration. The third day was again divided into two halves, beginning with form detailing and concluding with project communication and presentation. At the end of each of the first 2 days, the authors, in the capacity of supervising teachers, would review the progress of each project in a tabletop studio discussion format.

Given the short timeframe for the workshops, it was decided to adopt the already established double diamond as a reference design process model and focus all ideation and concept generation activity within the initial stages of “discover” and “define” steps of this model (Design Council, n.d.). This decision allowed the workshop to skip the challenges of implementing technical and manufacturing-related concerns typically associated with the later phases of any process model, i.e. “develop” and “deliver” phases in the case of the double diamond model. Thus, the workshop covered both divergent and convergent phases, concentrated in the initial portion of the process. Accordingly, the teachers shaped the student workflow to ensure that time was invested proportionally between divergent and convergent thinking across the three working days. The teachers monitored the time to ensure that the leading divergent portion consisted of collaborative brainstorming, research, and generating low-fidelity ideas through sketching, as well as exploring concepts through visual representations. The trailing convergent portion was dedicated to transforming idea explorations into high-fidelity conceptual renderings, scenario painting, and a concept communication plan.

The workshops were staged under consistent parameters to ensure uniformity of process and facilitate comparison of results between the various editions, with the following salient points:

1. Teachers: Three workshop mentors possessing expertise between 10 and 30 years in product design academia and industry.

2. Students: The class size varied between 8 and 16 second-year Master’s (MA) product design students equipped with four years of undergraduate design degree education but lacking any prior experience or contact with text-to-image or sketch-to-image GenAI software. A total of 36 product design master students participated in the 4 workshop editions.

3. Infrastructure: Provision of a dedicated design studio room outfitted with a comprehensive array of manual and digital design tools for sketching, prototyping, 3D printing, and computer-aided development.

4. Input: Presentation of a design brief shaped as a PowerPoint lecture outlining the current foundation theory on social robotics, problem setting, design challenge, thematic context, objectives, and expected outcome with a specific focus on crafting visual representations of concept designs of social robots appropriately shape coded with semantic keywords (Krippendorff & Butter, 1984; Krippendorff, 2005). In the workshops where GenAI was used, around a 1 h input session on prompt engineering was offered. The design brief is essential to any design project, facilitating and clarifying the scope, aims, and objectives of various parties involved in the development process (Phillips, 2004).

5. Timeline: Each edition was structured in a 3-day schedule encompassing 12 h of teaching contact spread across three sessions of 4 h each, supplemented by approximately 20 h of autonomous student work. During the contact hours, fundamental concepts of shape coding, HRI, and intricacies of social robotics were discussed in group sessions. During the autonomous hours, the students conducted additional research and form exploration individually or in groups of three.

6. Output: Twenty-seven social robot concept design proposals were presented in total. Each project was presented through a 15 min PowerPoint summarising the entire design process and supplemented by the collaborative Miro board, which presented the more elaborate version visualising the entire visual research, semantic moodboarding studies, shaped coding sketches, form studies, GenAI prompt engineering with highlights for successful and unsuccessful phrases, GenAI visual output, blending exercises and final design renders. The work was always developed individually, except for the fourth edition, where students worked in groups of three.

7. Evaluation: The three authors (teachers) collectively judged the creative quality of each project and attributed verbal critiques (crit) for each presentation. The critiques were followed by a group discussion, during which students and teachers debated the final results. These group discussions enabled critical analysis, comparison, and constructive feedback on the diverse concept design strategies. No quantitative or qualitative grade was awarded; therefore, these workshops did not impact the students’ semester academic results. To make a judgement, the authors considered four factors:

   1. The appropriateness of the overall concept for the suitability towards the intended social robot subcategory (companion, healthcare assistive, educational, or general purpose domestic), considering the theoretical knowledge on social robotics shared at the beginning of the workshop;

   2. Quality of the process visualisation and evidence of the ideation and concept generation work on Miro and PowerPoint;

   3. Overall novelty of the proposed design concept in terms of innovation; and

   4. Resolution of the final proposal in terms of form, mechanics, HRI, detail design, scenario graphical quality, and presentation communication.

8. Consent: The student’s consent to publish and share the output of their work was not an issue, since the guidelines of the HEI in which this activity occurred stipulated that all content generated within the premises remained the intellectual property of the institution and could be used for research and publication purposes. The data protection, personal identity, and privacy of the students were ensured by not sharing any photographs of the general classroom environment with visible faces in this reporting.

4.1 Differentiating Organic and Synthetic Creativity in Ideation

Integrating computational tools of synthetic creativity like GenAI into design curricula is a relatively new development, and both educators and students are still navigating the opportunities and challenges associated with their practical implementation. The authors’ empirical observations indicate that students often need very well-developed written communication skills to interact effectively with GenAI tools to avoid suboptimal results and encountering frustration. Current classroom teaching and orientation of product design education in the European context, at least, is primarily calibrated towards stimulating and shaping organic creativity. The introduction of synthetic creativity through GenAI computational tools is still unmeasured and therefore largely not yet controlled or monitored.

Organic creativity must defend and transparently expose the logic and reasoning behind every creative decision implemented by the designer along the entire ideation phase of the overall design process, while arriving at a specific conceptual proposal (Figure 2). On the contrary, text to image GenAI tools at the moment of developing this research do not expose the logic and reasoning behind the 2D images and designs they generate from a prompt, such tools are still very much a black box tool in the overall picture of a design process in the minds of many in the design community (Figure 3).

Figure 2

Project Cuteo (pet robot) – Miro board demonstrating the overview of the classical methods workflow from the first workshop (2021) edition.

Figure 3

Project Vivo (educational robot) – Miro board demonstrating the overview of GenAI computational methods workflow and the resulting pure synthetic ideation using Midjourney from the third workshop (2023) edition. The analysis of text prompts and the resulting images they generate can be seen, with yellow highlights indicating text prompts aligned with the target semantic shape codes and red highlights indicating text prompts deviating from the target semantic shape codes.

Speaking in terms of ideation and concept generation, where this research focuses, the stark differentiation between expressing organic and synthetic creativity causes disharmony in the process workflow of the students. When the students attempt to merge the GenAI synthetic creativity based on black box training data with their spontaneous organic creativity based on analogue sketching and visualisation work, a cognitive dissonance occurs. Students must transition from evidencing their thinking from lines, sketches, and form composition explorations into written words and precisely imagined descriptive word (adjective) details. Lines and sketches are drawing-based visual artefacts on which product design students receive a strong training base, while a prompt is a written language-based artefact based on reading and writing skills, on which design students are not usually well trained.

In traditional methods based on organic creativity, the initial visual exploration of ideation often begins with low-fidelity doodling or rapid sketches when the designer still has no clear vision of the final solution and is actively trying to concretise that vision by drawing. However, in the case of the computational methods based on GenAI synthetic creativity, the initial visual exploration required the designer to already hold some concrete vision in terms of shape, geometry, proportion, and context to generate valuable content.

4.2 Semantic Moodboards to Shape Coding Through Organic vs Synthetic Creativity

Once the students narrowed their problem definition to a specific pain point that could be addressed through social robotics, they identified a product opportunity gap on which to focus their ideation. The next step required them to begin shape coding exercises to create a visual representation of their solution, which would eventually concretise into their concept proposal. Semantic moodboarding was selected as the method to teach shape coding skills in this workshop series and therefore became the central departure point for all students as they launched their concept design journey. A semantic moodboard visually represents a target symbolic adjective (keyword) through visual details of products drawn across various categories. The challenge to students was that the morphology and details of their social robot must transmit any visual adjective they select, i.e. friendly, trustworthy, and cute.

Transitioning the semantic moodboard into form proposals is thus a critical step. Each student had to select three symbolic adjectives to attribute as target characters for their respective social robot. The products to insert into the moodboard were selected after careful consideration, filtering only details of products, avoiding any poetic compositions, abstract art, natural organisms, or objects. The background was kept neutral in a solid colour devoid of any foreign objects. The retention of colour in the selected images was allowed. Figures 4 and 5 demonstrate the variation in the creative paths between traditional and GenAI approaches, which the novice designers typically adopted, to arrive at a concept design proposal.

Figure 4

Project Cuteo (pet robot) – Semantic moodboard to concept design demonstrating steps of concept generation through organic creativity. (Clockwise, top left) Moodboarding of three target symbolic adjectives for character attribution – cheerful, affectionate, helpful. (Top right) Classical ideation method using analogue sketching. (Bottom right) Scenario representation with Photoshop. (Bottom left) 3D modelling to 3D rendering, presenting the final concept design proposal.

Figure 5

Project Vivo (educational robot) –semantic moodboarding to concept design with GenAI demonstrating steps of concept generation through mixed methods of blended synthetic and organic co-creation. (Clockwise, top left) The semantic moodboard I created organically attributes the symbolic adjective – “interactive”. (Top right) The second moodboard II created synthetically with Midjourney, experimenting blending combinations of two images from the first moodboard I and deriving a series of synthetic abstract forms. (Middle right) The third moodboard III created with Midjourney, blending the first moodboard I with a tripod-shaped web camera to explore functionality. (Bottom right) The fourth moodboard IV created with Midjourney, blending an exemplar social robot generated through Midjourney, as shown in Figure 3, bottom left, with the web camera used in moodboard III. (Bottom left) Organic sketches refining the form proposed in moodboard IV. (Middle left) Final 3D renders resulting from a 3D model built by the design student in Rhino. This process is an example of co-creation with GenAI, combining organic and synthetic creativity.

In the traditional path of organic creativity based on classical methods represented in Figure 4, the design student picks specific details from exemplar artefacts of the moodboard and tries to transform the visual geometry into a conceptual product form using extensive sketching.

In the computational path of pure synthetic creativity based on GenAI, represented in Figure 3, traditional drawing-based creativity has no input. In this condition, a designer has no opportunity to leverage their drawing and 3D modelling training for achieving high-quality outcomes, but instead must master written text prompt inputs, which now hold a vital position in achieving visual mastery of concept form proposals.

For novice designers, another significant challenge with pure synthetic creativity is that the direct link between the moodboard and drawing is broken, as no detail is lifted from the moodboard for transformation into a form exploration sketch.

4.3 Limitations of GenAI Synthetic Creativity

The authors observed that as the workshop progressed after several cycles of generation, a precise dilution of form diversity started to appear on the Midjourney platform. Although the output of the generated form proposals was quantitatively immense, the same could not be said of the qualitative diversity (Figure 3). Midjourney is a text prompt-based system that does not give the designer much control over the final output apart from selecting the most relevant generated and running another cycle of generations based on the selected proposal. This black box approach can derail the ideation, often distracting novice designers. Blending can increase the level of intervention.

4.3.1 Insights from Student Projects

Upon comparing the results obtained between ideation occurring from pure synthetic creativity based on GenAI prompting (Figure 3) and blended creativity merging synthetic and organic co-creation methods (Figure 5) of moodboarding, drawing, 3D modelling, and GenAI prompting, one can see the qualitative difference in form resolution, conceptual sophistication, and consideration for HRI.

Pure synthetic ideation suffers from design fixation, where a very narrow interpretation of what a social robot form could be is possible. For novice designers not trained to spot design fixation, they may get distracted or convinced of the limited visual vocabulary of the GenAI. Figure 3 illustrates that even after several cycles of prompting, the final results are repetitive, indicating a circular workflow with no evident progress.

This is the case of text-to-image GenAI like Midjourney, while in the case of sketch-to-image GenAI like Vizcom (Figure 6), the issue of design fixation is not such an immediate concern. This is because Vizcom was more custom designed for product design applications, and the interaction mechanics are based on the insertion of an original sketch or drawing from the human designer, which is then modified or completely replaced with a new synthetic image. The human designer always maintains control over how much interference Vizcom is allowed; therefore, they can actively intervene to limit the creep of design fixation.

Figure 6

Project Hug (general-purpose assistive robot): Miro board demonstrating overview of mixed methods workflow from the fourth workshop edition (2024) based on drawing, paper prototyping, and Vizcom GenAI. (Clockwise, top left) Ideation with sketches and paper prototyping. (Top right) Sketch to GenAI transition with Vizcom by merging the sketch with synthetic rendering emerging from prompts (refer Figure 1). (Bottom right) fine tuning the concept design by tweaking the prompts and experimenting altering details. (Bottom left) The final concept design proposal rendered directly in Vizcom. In Vizcom, the transition from hand sketching to a high-fidelity 3D visualisation is much more fluid, and the boundary between the organic and synthetic creativity is more porous. The human designer has much more control.

In contrast, the blended creativity approach breaks the design fixation pitfall. Figure 5 evidences how inserting classical methods can recover the ideation once a designer detects repetitive or stereotypical proposals from GenAI. In Figure 5, it is possible to see the implementation of abductive thinking by the design student in the step of exploring moodboard III, when the decision to insert a tripod webcam morphology into the form generation mix is evidenced. The designer in this case is considering issues related to the stabilisation and deployment of the robotic body, with the possibility of reducing the volumetric footprint when not in use. The workflow in the case of blended creativity is much more linear, approximating a classical pattern of ideation. GenAI application in this scenario is dosed and carefully applied only in specific moments of ideation, typically in early stages of ideation, then the application of classical methods of drawing are relied upon to derive a well resolved, unique and innovative robotic concept breaking the stereotypical image of social robots.

5.1 Necessity to Train Critical Analysis for Leveraging GenAI in Future

The emergence of “reasoning” model LLMs such as DeepSeek R-1 and OpenAI o3 in 2025, which allegedly can undertake critical analysis and think more like a human breaking down a complex problem into simpler constituent components (Woodle, 2025), could potentially offer a qualitatively different kind of output once they are deployed into creative applications such as art and design. A significant potential disruption in a product designer’s ideation workflow could occur if reasoning models could be presented with a problem-setting scenario in textual format, and based on which they could generate divergently original and practically suited visual proposals. The designer would then adopt the role of a critical thinker needing to dispassionately evaluate whether the reasoning AI synthetic proposals are superior to their own organic proposals.

Therefore, the authors assess that a critical analysis of AI-generated outputs needs to become a fundamental aspect of studio-based product design education. Designers must rigorously scrutinise GenAI outputs to avoid being misled by visually appealing, yet impractical designs produced. This evaluative task necessitates thoroughly analysing AI-generated proposals for feasibility and intended functionality. The critical judgement ability to evaluate hedonistic aesthetic attractiveness and pragmatic design viability will become an even more fundamental product design skill after the mass adoption of GenAI. It will underpin the relevance of maintaining a human product designer throughout the design process, especially within the iterative ideation phase. This condition will ensure that GenAI functions as a supplementary enhancement to the human designer’s skill set rather than a substitute.

Educators must also engage in ongoing professional development to stay abreast of GenAI technologies, given the rapid pace of development. By fostering a culture of knowledge-sharing among higher education institutions, educators can collaboratively work towards developing robust classroom teaching frameworks that effectively integrate Gen AI tools into the twenty first-century design education curriculum.

Finally, the ethical question of AI alignment and addressing potential biases in GenAI imagery, which may impact the perception and mental models of often young novice designers in formation, will be ever more important as this technology becomes widely adopted.