This page contains media that is intended to start playback automatically on opening. This may include sound. Your browser is blocking automated playback. Please click here to start media.
Artificial intelligence (AI) has emerged as a transformative force across various domains, and the field of graphic design is no exception. The recent influx of AI tools in graphic design practices has led to an evolving role for graphic designers, requiring them to navigate and adapt to a disciplinary landscape undergoing rapid disruption (Cook & Kwon, 2019; Edberg & Beck, 2020; Hashemieh, 2020; Meron, 2021). As advancements in AI accelerate at a rapid pace, graphic designers’ comprehension and terminology of the topic often struggle to keep up, resulting in misconceptions, concern, and worried fearmongering about the potential loss of jobs, automation, and deskilling of their discipline. To help mitigate this, this exposition proposes a model to instigate a discussion and reflection among graphic designers, to aid in a better understanding of their new role as practitioners in an AI era.
Historically, the use of technology in the production of design and art can be traced back to ancient Greece, where figures like Daedalus and Heron of Alexandria designed machines capable of writing text, generating sounds, and playing music. Throughout history, technological advancements have concurrently evoked both optimism and pessimism among designers and artists. The well-known 19th century Arts & Crafts movement, for instance, was founded in protest against the displacement of humans by industrialization and the ensuing division of the creative and production processes. Arts & Crafts idealized (and, some might argue, romanticized) the innate human creative impulse, craftsmanship, and the meaningful creation of unique, soulful objects. However, later on, one of the movement’s founders, Charles Robert Ashbee (1894), recognized that machines had come to stay and stated, “We do not reject the machine, we welcome it. But we would desire to see it mastered”. Contemporary counterparts to Ashbee’s desire for regulation and judicious use of technology include initiatives such as Tim Berners-Lee’s #ForTheWeb campaign (Cañares et al., 2018) and, most recently, an open letter (Future of Life Institute, 2023) currently signed by over 27,000 AI researchers advocating for a moratorium on AI development.
Coinciding with Ashbee’s acknowledgment of the enduring presence of machines, graphic design established itself as an independent discipline during the late 19th century. Subsequently, the field has undergone successive waves of transformative technologies that have incessantly reshaped its theoretical approaches and practical methodologies. In the 1970s, phototypesetting and desktop publishing gradually replaced hot metal typesetting and paste-up boards and transformed the trade from an analog to a digital undertaking. The proliferation of the internet in the 1990s heralded the transition from static print to dynamic digital communication, forcing designers to embrace working with interaction, interfaces, multimedia, animation, and code. By the late 2000s, smartphones arrived, bringing with them the need to design for new platform affordances and usage patterns. And now, in the 2020s, AI is designated as the next major disruptive technology. Each of these technological advancements has demanded a reskilling of the workforce, necessitating the acquisition of new skills, and embracing novel production and distribution methods. As the Greek philosopher Heraclitus famously said, “Change is the only constant,” and his adage continues to hold for the ever-evolving realm of graphic design.
In recent years, the graphic design trade has witnessed a significant surge in AI-powered tools. Adobe’s Sensei AI framework (Adobe, 2016) was announced in 2016, and since OpenAI first described GPT in a 2018 paper (Radford et al., 2018), text-to-image AIs such as DALL·E (OpenAI, 2021), Midjourney (Midjourney Inc., 2022), Stable Diffusion (Stability AI, 2022), Imagen (Google Research, 2022b), Parti (Google Research, 2022a), and Adobe Firefly (Adobe, 2023) have rapidly evolved over a few years, spanning multiple generations, and profoundly impacted all stages of designers’ workflow, from ideation and sketching to testing and integration. A truly explosive development aptly captured in the title “Today has been a great year in AI” (Khan, 2023).
As AI tools continue to mature and evolve, they will progress along Gartner’s Hype Cycle (Linden & Fenn, 2003), and transition towards a plateau of productivity, unlocking even greater potential for graphic design practitioners. How far along the hype cycle AI has progressed remains uncertain; however, it is evident that there are still significant advancements and discoveries on the horizon that will further amplify the ramifications of AI in graphic design.
With the prospect of significant changes, graphic designers can easily fall prey to metathesiophobia, the fear of change. Some are excessively intimidated by the immense complexity and seemingly magical capabilities of AI to such an extent that they refrain from actively seeking to explore the ongoing transformation. However, engaging in critical reflection and informed discussion on what constitutes professional graphic design practice in the AI era is essential if the discipline is to leverage and enhance its existing discourse and praxis through the new technology. A failure to highlight and justify the value of human-centered professionally executed graphic design would create space for amateur designers to inundate the market with AI-generated shallow, mediocre, generic, and sub-par (potentially counterproductive) design products made using a rapidly growing portfolio of one-click-no-think “intelligent” design tools. This would contribute to devaluing the perception of graphic design’s effectiveness and legitimacy as a human-centered problem-framing and problem-solving activity that requires understanding and skills beyond what AI is capable of.
The prevailing discourse among graphic designers concerning AI is predominantly a binary discussion between two factions referred to by Agüera y Arcas as “technophobic humanists” and “inhuman technologists” (2017). However, a more nuanced, informed, and pragmatically grounded perspective on the role of AI in graphic design is necessary. To facilitate this crucial and timely discussion, this paper proposes a model as a foundation for dialogue among students, educators, practitioners, and scholars in the field of graphic design. The model aims to provide an understanding of the evolving role of graphic designers and argues for their continued relevance in a disciplinary field characterized by the increasing prevalence and diverse applications of AI tools. Moreover, the model offers considerations for graphic designers to effectively utilize AI as a collaborative, augmentative, and value-generating tool. And, this paper posits, it is crucial to recognize AI as a tool, albeit an advanced one. Ultimately, the model strives to foster a nuanced comprehension of the contemporary relationship between graphic designers and AI, with the aspiration of promoting an informed and productive discourse within the field.
Figure 1: The proposed model based on deduction, induction, and abduction.
The model (Figure 1) is based on the intertwined concepts of deduction, induction, and abduction, which are employed in reasoning and inference. Additionally, the model introduces the metaphorical notion of a “disciplinary expertise filter” as well as the established concepts of “black box” and “clear box.” Below, the model’s constituents are presented and elaborated upon.
Deductive design activities are situated on the model’s left side. Deduction employs logic, rules, algorithms, guidelines, theories, and norms to yield specific outcomes. Color theory, principles of composition, typographic treatment, guidelines in old brand books, modern rule-based design systems, and algorithmic-based creative coding are all examples of deductive practices. Early AI research was grounded in deductive principles. Roughly speaking, the notion was that if all available knowledge in a specific domain (e.g., graphic design) could be encoded into a system, it would appear “intelligent.” However, the drawback of this approach was that the systems often appeared rigid and incapable of learning new things unless manually added. Nevertheless, deductive principles are still utilized by AI today.
Inductive design activities are positioned on the model’s right side. Induction employs specific observations to generate general rules. Modern AIs are usually built upon inductive principles, where large datasets serve as the specific observations that form the basis for a model with general (biased) knowledge of the data it is trained on. Well-known examples of inductive AI are the text-to-image tools mentioned in section “The advent of AI in graphic design”. In contrast to rigid deductive AIs, inductive AIs are adaptable and versatile due to their reliance on decision trees, neural networks, and clustering algorithms. Traditional graphic design tasks based on induction include data analysis, gathering inspiration, and trend spotting, among others. Many of these tasks are often considered mundane and labor-intensive, which graphic designers have been using computers to alleviate for years. However, new generations of inductive AIs are capable of handling increasingly complex tasks and drawing conclusions based on ever-growing amounts of data.
In the model, abductive design activities are positioned above deduction and induction. This is not to imply that abduction should be perceived as a “higher level” activity than deduction and induction; all three are formally distinct and not reducible to each other. Abduction complements deduction and induction by creatively inferring design solutions from observations and contextual cues, often described by designers as using their “common sense,” “intuition,” or “gut feeling.” Abduction is currently considered a significant blind spot for AI. Previous attempts at abductive AIs in the 1980s, 1990s, and most recently in the 2010s failed and were abandoned as they could not capture true abduction, but were merely variations of deduction and induction (Larson, 2021). A substantial portion of graphic designers’ work occurs in open-ended scenarios, where abduction plays a central and irreplaceable role. Broadly speaking, creativity can be viewed as a form of abductive inference. When engaging in creative processes, graphic designers make imaginative leaps, combining existing knowledge, observations, and contextual cues to generate novel and plausible ideas or solutions. The placement of abduction at the top of the model is deliberately chosen to reflect graphic design as a fundamental human-driven activity, where humans—with their empathetic and creative abilities—serve as the interface to the surrounding world, encompassing both the objectives and applications of the graphic designer’s work. Simultaneously, the model also positions humans as the initiating factor in any human-machine collaboration.
The model introduces the concept of a metaphorical “disciplinary expertise filter.” Positioned between the model’s abductive layer and the deductive/inductive layer, this filter serves two functions: i) outwardly describing and qualifying the abductively identified tasks that the graphic designer wishes to accomplish using AI (see “Automation” section), and ii) inwardly abductively evaluating and either accepting or rejecting the results that the graphic designer receives back from AI (see “Augmentation” section). The imaginary density and thickness of the filter can be seen as the graphic designer’s professional knowledge of the practice of their field. Amateur designers without formal training or knowledge of the various aspects of the field have a thin and low-density filter that relatively unobstructed and uncritically allows all input (automation) and output (augmentation) to pass through. Professional graphic designers have a thick filter, developed through their experience and intuition, which rejects both input and output that do not meet their standards in terms of aesthetics, functionality, consistency, homogeneity, identification, conveying the right message, legibility, functionality, applicability, reproducibility, flexibility, as well as a host of other criteria employed to assess the quality and effectiveness of graphic design products.
The model employs the term “automation” as a general expression for an activity initiated by a graphic designer as a result of abductive reasoning and intended to be outsourced to either a deductive or inductive AI process. Before initializing any automation, the graphic designer must utilize their expertise in the field combined with their knowledge of the AI tools they wish to employ, to formulate the desired task in a format amenable to machine processing (e.g., an input for a process, a prompt, a piece of code, a rule in a digital brand book). This process is both creative and abductive. The quality of the implemented automation depends on the thickness of the “disciplinary expertise filter” it must first pass through. An amateur with a thin filter may indiscriminately initiate poorly formulated automation tasks. In contrast, a professional with a thicker filter can launch more precisely formulated automation tasks grounded in solid disciplinary knowledge and experience. The proverb “a question well asked is a problem half solved” aptly applies in this context.
In the model, the outcome of any initiated automation is described as “augmentation.” This term encompasses the general idea that every automation process will augment the work of a graphic designer. This augmentation can take various forms, such as general enhancement of tasks, generating new ideas, faster execution, rapid iteration, efficient exploration of solution spaces, more precise insights, pattern recognition in collected data, etc. The result of automation also needs to pass through the graphic designer’s disciplinary expertise filter. The same pattern observed during automation applies here (see previous section): The thin filter of an amateur allows any result to pass through uncritically. In contrast, the professional, due to their strong disciplinary grounding, is more inclined to critically assess the outcome of an automated task, allowing only results of sufficient quality to pass back through the disciplinary expertise filter to be incorporated into the ongoing work.
The model also employs the established concepts of ‘black box’ systems and ‘clear box’ (sometimes also referred to as ‘white box’) systems. The term ‘black box’ is used to describe an opaque system or process where the internal workings are hidden, and only the input-output behavior/interface is observable. Many modern AI design tools are commercial, proprietary, and closed ‘black box’ systems accessible through a basic interface hiding a level of complexity that even their creators may not be able to fully explain. ‘Clear box’ describes the opposite; transparent systems where the internal mechanisms and logic are accessible and understandable to observers. This is applicable, for example, in algorithm-based systems that utilize an (often manageable) set of rules formulated by humans. It may seem reasonable to categorize deductive AIs as ‘clear box’ systems and inductive AIs as ‘black box’ systems, but that would be incorrect. Some deductive AI systems may involve complex algorithms or intricate decision-making processes that make them less transparent or easily understandable. Similarly, certain inductive AI systems can provide insights or explanations about their generated predictions, making them more interpretable in nature. It is more accurate to evaluate the nature of an AI system on a case-by-case basis.
While acknowledging the inherent fuzziness of this generalized distinction, the model attributes the term ‘clear box’ to the automation of well-defined (mostly deductive) tasks, often carried out by relatively simple AI tools. Conversely, the term ‘black box’ is assigned to (mostly inductive) AI tools characterized by extensive complexity or opacity in their rules, datasets, models, or internal mechanisms, making the underlying rationale behind their outputs difficult to elucidate. The reason for incorporating these two contrasting concepts into the model is to help dispel the notion of AI technology as being “magic,” as suggested by the British science fiction author Arthur C. Clarke, famous for his famous third law stating that “[…] any sufficiently advanced technology is indistinguishable from magic.” By including these concepts in the model, it aims to challenge the perception of AI as a mystical or inexplicable technology and instead encourages a more informed and nuanced understanding of its capabilities. Often, many simple tools mistakenly labeled as “AI-powered” are merely mundane algorithms that generate their output using pre-made template designs, combinatorics, pre-defined flows, or deterministic rules without any intelligent intervention from the machine. Conversely, genuine AI tools are the result of immensely complex and sophisticated computations… but they are not magic.
With the model presented, the question arises, “What can it tell us?” This section offers six possible perspectives that graphic designers can utilize, using the model as a reference, to engage in discussions and reflections on their evolving role within a rapidly changing professional landscape.
In its entirety, the model reflects the praxis of graphic designers, with typical tasks and disciplinary terms distributed based on their underlying reasoning. The modus operandi of graphic designers relies on frequent and rapid shifts between deductive, inductive, and abductive tasks and states. They engage in analyzing existing knowledge, observing visual cues, and employing intuitive leaps to (re)frame problems and generate innovative ideas. They deduce specific design decisions from established principles, induce general trends from research, and abductively infer unique solutions that address specific problems. This constant interplay of states also encompasses an alternation between “creative” and “non-creative” tasks. Analyzing a large amount of data, adhering to layout rules described in a brand book, or setting up a page layout according to defined constraints does not necessarily stimulate the creativity of graphic designers. The most stimulating tasks for graphic designers, where they perceive themselves as most creative, typically lie in the abductive realm, where inductive and deductive inputs are integrated, synthesized, and interpreted.
The model demonstrates how graphic designers need to adapt to collaborating with AI through an ongoing dialogue involving automation and augmentation (see previous sections). However, it is important to note that this dialogue is always initiated and evaluated by the graphic designer. Meron (2021) highlights how creatively driven professional graphic designers generally welcome the automation of laborious tasks. AI tools are typically much more efficient, precise, and fast (though also biased and uncritical) in performing deductive and inductive tasks. Therefore, it makes sense to outsource these tasks to the computer, freeing up more time for the graphic designer’s abductive activities. The new AI/human partnership in graphic design will likely give rise to a wide range of predicates such as “AI-assisted,” “AI-made,” “AI-generated,” and “AI-powered.” These attributions are a natural consequence in a time where the much-hyped initials AI are beneficial for marketing purposes. However, as the collaboration between AI and humans consolidates, it will become implicit to mention that AI has contributed to the design process. After all, modern graphic design products are not labeled “computer-made.”
Positioned as the initiating factor at the top of the model, it is the graphic designer who chooses to utilize specific AI tools in their design process. The power balance between humans and machines is clear: AI works for us, not the other way around. This echoes Ashbee’s recurring desire (see “Background” section) that we must control the tool; the tool should never control us.
According to the Oxford Dictionary, a tool is defined as “a thing that helps you to do your job or to achieve something.” It is easy for amateur designers to “achieve something” with a tool that suggests how it should be used. However, it requires much more from professional designers to “do your job” as their profound expertise necessitates a comprehensive comprehension of the AI tool to meet their expectations for the resulting product. However, this encapsulates the essence of craftsmanship: possessing extensive knowledge of the subject matter and exhibiting mastery over the tools at hand, including AI.
The model indicates how collaborating with AI requires an understanding of how to interface with the new technology; how to create “good” automation that results in “good” augmentation. This necessitates a reskilling of graphic designers. The “Future of Jobs Report 2023” (World Economic Forum, 2023) forecasts that the three main areas of reskilling for graphic designers are the ability to work with AI and Big Data, analytical thinking, and creative thinking. The first focal point revolves around AI technology itself. The latter two focal points, analytical and creative thinking, already form an integral part of the graphic designer’s existing practice. However, as AI finds its way into the disciplinary methodology, a reevaluation of its constituents becomes necessary.
The model introduces the notion of a “disciplinary expertise filter” (see previous “Disciplinary expertise” section). As described, the metaphorical thickness of the filter depends on the graphic designer’s disciplinary knowledge and experience. Thus, the importance of ongoing research conducted by graphic designers, whose perspectives often fade in a scientific environment with a reductionist view of graphic design (Cook & Kwon, 2019), cannot be overstated. Design schools need to adapt their curriculum and teaching methods to reflect the increasing use of AI as assistants and inspirational co-creators. In an era where AI tools empower everyone to create what appear to be professional graphic design products, strengthening the theoretical and practical aspects of the graphic design discipline is essential to counterbalance amateurishness. The thickness of their respective filter will differentiate the AI-assisted work done by professionals and amateurs, respectively, which pragmatically translates into their ability to critically evaluate the outcomes of their AI collaboration through a strong professional lens.
The model suggests how human dominance over AI in the abductive domain (see previous sections on “ Abductive tasks” and “Disciplinary expertise”) will ensure that human graphic designers are not replaced by machines anytime soon. The ability of humans to blend soft skills such as intuition, empathy, sympathy, and emotional intelligence with their experience and professional expertise is an invaluable factor in the abductive process referred to as “creativity”. The current state of research on true abductive AIs indicates that we as humans will maintain our patent on the field for some time. Paula Scher, a graphic designer famous for, among other things, quickly sketching the Citicorp bank logo on the back of a napkin, is often credited with the quote: “It took me a few seconds to draw it, but it took me 34 years to learn how to draw it in a few seconds.” The accumulated experience, knowledge, and intuition encapsulated in Scher’s original idea and the accompanying sketch cannot be transferred to any AI. However, once an idea is conceived, collaborating with AIs can make the process of exploring, developing, and refining it more efficient, inspiring, and stimulating.
A recent popular meme read: “To replace graphic designers with AI, clients will need to accurately describe what they want. We’re safe.” However, as argued in this paper the following caveat should be added: “…only as long as we strive to understand AI and learn how to use it to our advantage.” The model proposed in this paper aims to contribute to this endeavor, providing an offset for discussion and exploration of the evolving role of graphic designers in the era of AI. The author hopes that this model will assist in fostering a deeper understanding and productive engagement with AI among scholars and practitioners within the field of graphic design.
The author affirms that there are no conflicts of interest pertaining to this research study. The study was carried out independently, without receiving any external funding. The author is associated with The Danish School of Media and Journalism, and the affiliation has had no impact on the design, implementation, or reporting of the study. The opinions, findings, and conclusions presented in this paper solely reflect those of the author.