As Generative AI (GenAI) technologies evolve at unprecedented speed, so do the challenges of understanding their functions, limitations, and potential. This study investigates how visual metaphors act as cognitive tools that make abstract GenAI concepts more accessible by linking them to familiar visual representations, allowing us to see the unseen. Grounded in conceptual metaphor theory (Lakoff & Johnson, 1980), dual-coding theory (Paivio, 1991), and visual literacy theory (Avgerinou & Pettersson, 2020), it examines how metaphors shape perceptions and subsequently the intended use of GenAI within educational and professional contexts.
Through a targeted review of recent academic literature (2021–2025), the study identifies the most prevalent metaphorical framings of GenAI and organizes them along a continuum from non-anthropomorphic to anthropomorphic representations. By mapping this spectrum, the resulting work highlights how metaphor selection influences the ways GenAI is conceptualized and discussed. The study results also informed a subsequent empirical, interdisciplinary inquiry (Karakitsou et al., 2025) that focused on examining faculty’s GenAI perceptions through visual metaphors and projective techniques.
According to Avgerinou and Pettersson (2011; 2020), visual literacy encompasses visual communication, language, learning, perception, and thinking. Within this framework, visual metaphors function as pivotal mediators that translate abstract concepts into tangible imagery, thereby facilitating meaning-making, comprehension, and engagement (Avgerinou, 2011; Reese & Bendito, 2003). Informed by dual-coding theory (Paivio, 1991) and the working memory model (Baddeley & Hitch, 1974), these metaphors enhance learning through the coordinated processing of verbal and visual information. In educational contexts, the intersection of visual literacy and metaphorical reasoning enables learners to interpret and navigate complex digital environments, positioning visual literacy as a foundational dimension of GenAI literacy, defined here, following Bozkurt et al. (2024), as the integrated knowledge, skills, and critical understanding required to engage with GenAI technologies effectively, ethically, and responsibly.
Metaphors have long played a vital role in educational technology by helping explain abstract or unfamiliar concepts through comparisons with familiar ideas (Rodriguez & Dimitrova, 2011). Typical examples, such as scaffolding in instructional design (Avgerinou, 2011) and ecosystems in EdTech (Weller, 2022), illustrate how analogy helps educators make sense of complex systems. Furthermore, Mason (2018) identifies metaphorical categories, including tools, journeys, and construction, each facilitating understanding of and communication about technology’s functions in education. Yet, as Weller (2022) cautions, overused analogies, such as "disruption," can frame technology in overly commercial or deterministic ways. More precise and context-sensitive metaphors, by contrast, enhance conceptual clarity and encourage informed decision-making. As GenAI becomes increasingly embedded in education, the metaphors used to describe it determine whether it is perceived as an empowering aid or as a challenge to traditional learning practices.
The current study emerged from the initial phase of a larger project that focused on the exploration of faculty perceptions and use of GenAI (Karakitsou et al., 2025). This first phase involved a targeted review of recent academic literature (2021–2025) that examined or employed metaphors to explain or frame GenAI. Searches were conducted in Scopus, Web of Science, and Google Scholar using terms such as Generative AI (GenAI), (visual) metaphor, (Gen)AI imagery, anthropomorphism, and education. Only peer-reviewed sources were included.
Each text was analyzed for the dominant metaphorical framings of GenAI, grouped into two main categories: anthropomorphic (e.g., assistant, collaborator, intern) and non-anthropomorphic (e.g., black box, digital plastic, calculator for words). Guided by the conceptual metaphor theory (Lakoff & Johnson, 1980), dual-coding theory (Paivio, 1991), and visual literacy theory (Avgerinou & Pettersson, 2020), the present study established a continuum of metaphors that, in total alignment with Oster et al.’s anthropomorphism spectrum (2025), shaped the theoretical and visual framework later used in Karakitsou et al.’s research (2025).
The analysis of the literature revealed two dominant trends in how GenAI is framed: one that anthropomorphizes GenAI by assigning it human-like qualities and agency, and another that deliberately avoids anthropomorphism, emphasizing its mechanical, probabilistic, or systemic nature. These metaphorical framings form a conceptual continuum ranging from black box (Ganesh, 2022; Nerlich, 2024), funhouse mirror and digital plastic (Roe et al., 2024), to collaborator (Mollick, 2024a), assistant (Mollick & Mollick, 2023), and intern (Mishra, 2023; Mollick, 2023).
Critics such as Hunger (2023) argue that anthropomorphic metaphors perpetuate GenAI hype, exaggerating intelligence and autonomy, while Mills and Angell (2025) caution against the metaphor of hallucination, which wrongly implies consciousness. By contrast, alternative framings such as stochastic parrot (Bender et al., 2021), digital plastic (Roe et al., 2025), and brain without a mind (Cao & Dede, 2023) seek to underline GenAI’s generative yet non-conscious nature. Anderson (2023) critiques metaphors such as tool and collaborator for oversimplifying GenAI’s cognitive functions and reinforcing the false perception of autonomy. Roe et al. (2024) analyze metaphors including black box, funhouse mirror, and echo chamber, which foreground opacity, distortion, and bias, highlighting the need for critical reflection on transparency in GenAI. Similarly, Ye and Li (2024) identify journey, war, human, and object as metaphors in the European Union AI Act (EU AIA), revealing how regulatory discourse mirrors both human aspirations and fears.
Anthropomorphic framings personalize and moralize technology, positioning GenAI as capable of intention or judgment; non-anthropomorphic framings depersonalize GenAI, casting it as data-driven and procedural. Collectively, these discussions suggest that metaphor selection profoundly shapes public and academic understanding of GenAI, influencing levels of trust, skepticism, and ethical concern.
Within the identified continuum, anthropomorphic metaphors tend to simplify complexity through familiarity, fostering accessibility and emotional resonance, while often misrepresenting GenAI as sentient or autonomous. Terms such as assistant, collaborator, and co-worker (Mollick, 2024a,b) suggest partnership and reliability while masking the human control and algorithmic constraints behind the technology. Conversely, non-anthropomorphic metaphors like black box, mirror, map, or digital plastic (Roe et al., 2024) emphasize interpretive opacity or transformation, aligning with efforts to promote transparency and critical distance. Yet these metaphors also frame GenAI as inaccessible or alien, potentially dampening users’ curiosity and agency. This tension between personalization and abstraction, which reflects deeper societal narratives about innovation, control, and accountability, mirrors educators’ ambivalence about whether to view GenAI as a learning partner or a computational instrument.
From an educational perspective, this analysis underscores the importance of metaphor awareness as a dimension of Critical GenAI Literacy, that is, “the ability to critically analyse and engage with AI systems by understanding their technical foundations, societal implications, and embedded power structures, while recognising their limitations, potential biases, and broader social, environmental, and economic impacts” (Roe et al., 2024, p. 2). By recognizing how language and imagery shape perceptions of GenAI, educators and learners can develop a more informed, ethical, and balanced understanding of GenAI.
Karakitsou et al. (2025) applied these theoretical insights by using visual metaphors to elicit faculty attitudes toward GenAI. Their findings mirrored the literature’s duality (optimism about creativity and productivity alongside concerns over autonomy, dehumanization, and ethics), reinforcing that metaphorical framings shape both cognition and discourse in higher education.
These complementary studies demonstrate how metaphors can advance Critical GenAI Literacy by revealing implicit perceptions through visual methods like metaphor elicitation and projective techniques. They also underscore the need for deliberate use of metaphor in teaching, communication, and policy: visual and linguistic framing should clarify rather than mystify GenAI, temper hype, and promote informed engagement. Moreover, the continuum of metaphors identified here provides a practical framework for designing instructional materials, visual communication, and faculty development initiatives that cultivate reflective and ethical integration of GenAI in education.
This study contributes to ongoing discussions on perceptions of GenAI by demonstrating how visual metaphors shape our understanding. By examining the current literature on GenAI visual metaphors, the study offers a novel approach to unpacking GenAI discourse and fostering more critical, informed engagement with emerging technologies. By analyzing the use of visual metaphors, we can enhance GenAI literacy, avoid misconceptions, and equip educators and learners with the criticality necessary to navigate the evolving digital landscape.
Paivio, A. (1991). Dual coding theory: Retrospect and current status. Canadian Journal of Psychology, 45(3), 255–287.
EdTech Archives