Shuichiro Ogawa
日本語

Notes · updated 2026-08-02

What Is Generative Art For in Education? Seven Purpose Types and a Literature Map of 72 Studies

Scope and Method

This note is a literature map of academic research that uses generative art (generative art / creative coding / computational art, together with the tool cultures of Processing, p5.js, Design by Numbers, Scratch, and Sonic Pi) as educational material, organized along the axis of purpose: what the material is used for.

  • Corpus: source/review/generative-art-education/papers.md (mode: academic). 72 items, after merging 14 duplicate rows from 86 explored (0 excluded).
  • Period: 1980–2026 (reaching back to Papert’s constructionism; the post-LLM lineage focuses on 2023–2026).
  • Premise: the theory and history of generative art itself (a 94-item corpus) is covered by generative-art-literature. System studies of LLM-based creative coding (Spellburst, Flowcode, fog, GenP5) are already in the L lineage of that corpus; this note references rather than duplicates them.

Reading across the 72 items, the purposes of educational use fall into seven types. Below, each type is presented with representative literature and empirical results, followed by the mapping to educational stages and the remaining gaps.

Overview of the Purpose Types

#Purpose typeDominant educational stageRepresentative literature
1Motivating and contextualizing introductory programmingHigher education (intro CS)Guzdial 2003; Greenberg et al. 2012; Reas & Fry 2007
2Inclusion of diverse learnersK-12 and higher educationRich et al. 2004; Werner et al. 2005; Peppler 2010
3Developing and assessing computational thinkingK-12Wing 2006; Brennan & Resnick 2012; Moreno-León et al. 2015
4STEAM integration and constructionist makingK-12Papert 1980; Kafai et al. 2014; Taber et al. 2024
5Expressive expansion in art, design, and music educationArt schools and higher education (architecture/music)Knochel & Patton 2015; Oxman 2008; Aaron et al. 2016
6Creativity educationCross-stageResnick 2017; Artut 2017; Kafai & Resnick 1996
7AI literacy and critical media literacy (post-LLM)Expanding from K-12 to teacher educationVartiainen et al. 2025; Gu & Ericson 2025; Fang 2026

The seven types are not mutually exclusive. A single course often serves both motivation and CT development; each study was classified by the primary purpose whose effect it claims and verifies.

Type 1: Motivating and Contextualizing Introductory Programming

The thickest accumulation lies in the lineage of introductory CS that turns visual and sonic generative expression into a reason to learn programming.

This lineage has two headwaters. One is the educational philosophy of tool designers themselves: Maeda’s Design by Numbers (1999) redefined programming for artists, and Reas and Fry’s Processing (2007), its successor, institutionalized the design philosophy of “the program as sketchbook” as teaching material (E1-01, E1-07). The other is media computation on the CS education research side: Guzdial (2003) founded an introductory CS course for non-majors built on manipulating sound and images (E1-04).

The empirical results support effectiveness in introductory contexts. Guzdial (2013) looked back over ten years of studies and organized the state of verification for the hypotheses that media-based material improves motivation and retention (E1-10). Greenberg et al. compared a Processing-based CS1 across two universities (E1-11, E1-12), and Malan and Leitner reported the introduction of Scratch in higher education (E1-15). Most recently before LLMs, McNutt et al. used logs to study how the features of a p5.js-style editor are used in an introductory classroom (E1-21).

Type 2: Inclusion of Diverse Learners

Adjacent to the motivational lineage, yet standing as an independent purpose, is inclusion: expanding participation by women, non-CS majors, and learners in low-income communities.

In the evaluation of a CS1 course by Rich et al., among 121 students (two thirds women) the only dropouts were three men, and 89% completed with a C or better (E1-05). Werner et al. studied a game-programming program for middle school girls (E1-14), and Denner et al. examined CS-concept assessment through games created by 108 middle school girls (E2-15). Peppler (2010) conducted more than three years of ethnography at a Computer Clubhouse in a low-income neighborhood, showing how youth aged 8 to 18 acquired integrated programming and artistic skills through media arts production (E1-13). Kafai and Peppler theorized these as participatory competencies (E1-18), and Kafai and Burke proposed the conceptual extension to computational participation, beyond individual skill acquisition (E1-17).

Type 3: Developing and Assessing Computational Thinking

The dominant purpose in K-12 is the development of computational thinking (CT).

With Wing’s (2006) definition as the reference point (E2-01), Grover and Pea (2013) organized K-12 CT research (E2-06), and Lye and Koh (2014) confirmed that most empirical studies report positive outcomes (E2-13). What is distinctive to this type is the body of work that makes creative artifacts themselves the object of CT assessment. Brennan and Resnick (2012) defined the three dimensions of CT (concepts, practices, perspectives) using Scratch projects as material (E1-20), and Moreno-León et al.’s Dr. Scratch implemented automatic CT assessment of projects (E2-17). Tang et al.’s systematic review (96 empirical studies) examines the reliability and validity of CT assessment methods, including such artifact-based assessment (E2-16). On the higher education side, Li et al. reported in a quasi-experiment with art and design undergraduates that an art programming course improved both design thinking and CT (E1-22). In music, Petrie’s mixed-methods study of algorithmic composition with Sonic Pi confirmed learning outcomes in both music and CT (E2-23).

Type 4: STEAM Integration and Constructionist Making

Overlapping with CT development, the STEAM lineage takes subject integration itself as its purpose.

Its intellectual foundation is Papert’s constructionism (1980; Papert & Harel 1991), which supplies the rationale by which “learning by making” connects art and STEM (E1-02, E2-03). Empirically, Kafai et al.’s e-textiles high school curriculum showed improvement in CT indicators and expanded participation by girls and minorities at the same time (E2-07), and Kafai and Burke reviewed 55 studies of game-making for learning (E2-11). Blikstein et al. organized the educational applications of the maker movement and FabLabs (E2-18), and Israel et al. qualitatively analyzed cross-subject CT implementation including art classrooms (E2-20). As reviews, Taber et al. organized arts integration in K-12 CS education with PRISMA, suggesting effectiveness for inclusion (E2-24), and Tariq et al. cover CT×STEM in higher education (E2-25).

Type 5: Expressive Expansion in Art, Design, and Music Education

Whereas the types so far treat art as a means of learning CS or STEM, there is a lineage running the other way: art and design education adopting code as an expressive technique.

In art education, Knochel and Patton (2015) advocated critical digital making, embedding creative coding into studio practice in art (E1-16), and the edited volume by the same editors (2020) collected such practices (E3-17). Dahn et al. analyzed the emotional experience of learning to code with art as the point of departure (E3-13). In architecture and design education, Oxman (2008) systematized the theoretical transformation that digital and generative design demands of design pedagogy (E3-09), while Celani (2012) and the report on a first-year course at Warsaw University of Technology (2022) address studio education in parametric and algorithmic methods (E3-10, E3-11). In music education, the educational partnership model by the designers of Sonic Pi (E3-05, E3-06) and Collins’s reflection on twelve years of teaching SuperCollider (E3-07) position live coding as both expression and learning. Jacob’s (1996) essay on algorithmic composition is the foundational discussion of creativity for this lineage (E3-08).

Type 6: Creativity Education

Research whose primary purpose is cultivating creativity itself appears across stages, belonging to no single subject.

The representative work is Resnick’s 4P (Projects, Passion, Peers, Play), which integrated thirty years of Scratch-centered practice into a theory of creative learning (E1-19). Its prototype lies in Kafai and Resnick’s collection of constructionist practice (1996) (E1-03), and Maloney et al.’s “Programming by Choice” showed that in a choice-based creative environment, 536 projects by urban youth acquired key concepts without instructional intervention (E3-04). In the art school context, Artut (2017) reports the design of a creative coding course as an art course aimed at cultivating computational creativity (E2-21).

Type 7: Changes after LLMs and AI Literacy (2023–2026)

Since 2023 the field has moved in two directions at once.

The first is the reorganization of education about writing code. On the intro CS side, Prather et al. demonstrated the polarization of GenAI use among novices (stronger students leverage it, while struggling students fall into an “illusion of competence”) (E4-01), and Bernstein et al. systematically reviewed the harms of GenAI in computing education across 224 papers (E4-02). Reeves et al. argue for a redesign of instruction that puts natural-language prompts first (E4-03). Within creative coding education itself, design research has appeared that uses LLMs as scaffolding for reflection and iteration (Reflexa, E4-04; the related system studies Spellburst and Flowcode are in the L lineage of generative-art-literature).

The second is a shift in the center of gravity of the teaching material. The educational use of generative art has broadened from “writing algorithms to generate” to “using generative models while understanding them critically.” On the expressive side stand the formal analysis of Stable Diffusion prompts and its potential for art education (E4-07), prompt-engineering instruction in a university digital art course (E4-08), the redesign of graphic design education (E4-09), a workshop with art education majors (E4-11), a quasi-experiment with fifth graders (n=78, with gains in engagement and self-efficacy, E4-12), and a field study of group projects in elementary school (n=132, E4-16). On the critical-understanding side are a quasi-experiment with 209 Finnish fourth and seventh graders showing improved ability to explain bias in generated images (E4-18), participatory design with high school students and teachers (E4-19), a scoping review and frameworks for critical AI literacy (E4-20, E4-21), and a classroom practice addressing “algorithmic Orientalism” in SDXL (E4-13). Teacher-side reception has also begun to be studied, with research on 14 Japanese pre-service art teachers (E4-17) and a proposal for dialogic making by art educators (E4-22).

Quantitative change has also been reported. An integrative review of AI literacy research (124 papers) states that GenAI-related educational empirical work surged from 8% in 2023 to over 40% in 2024 (E4-05). Within art education alone, two systematic reviews (covering 19 and 27 studies) were published in 2025–2026, reporting that eligible studies grew from 2 in 2023 to 14 in 2025 (E3-16, E3-15). Fang (2026) finds quantitative improvement in creative fluency in music education, while identifying concerns about authenticity and deskilling in visual art and design, and recommends embedding critical AI literacy (E3-15).

Educational Stages and Their Dominant Purposes

From the composition of the corpus, the dominant purposes per stage read as follows. Note that this is an organization of the distribution of the 72 collected items, not a bibliometric count.

  • K-12: CT development (type 3), STEAM integration (type 4), and inclusion (type 2) dominate, with Scratch, e-textiles, and game-making as the main tools. Since 2024, critical AI literacy using bias in generated images as material (type 7) has been added.
  • Higher education (intro CS): motivation and contextualization (type 1) and inclusion of non-majors and women (type 2). Media computation and Processing-based CS1 are representative. After LLMs, the emphasis has shifted to the polarization of support and harm for novices.
  • Art schools and higher education in design, architecture, and music: expressive expansion (type 5) and creativity education (type 6). Apart from one quasi-experiment (E1-22), the evidence consists mostly of practice reports and qualitative studies. Since 2023, studies introducing text-to-image tools have surged.
  • Teacher education: emerged as a distinct layer only after LLMs (E4-17, E4-22, among others).
  • Lifelong and informal learning: almost blank except for inclusion research in the Computer Clubhouse lineage (E1-13, E2-09).

Gaps (Unmet Topics)

  1. No cross-purpose systematic review: reviews of generative AI in art education appeared in 2025–2026, but within the search terms used here, no systematic review was found that spans the educational use of creative coding across purpose types. The seven types in this note are a provisional substitute.
  2. Thin evidence at art schools: creative coding education at art schools, supposedly the core of type 5, has little controlled evidence beyond Li et al. (E1-22); most work remains practice reports.
  3. Missing lifelong learning stage: almost no research was found on generative art education in adult or lifelong learning.
  4. Long-term effects and transfer: apart from the ten-year retrospective of media computation (E1-10), longitudinal evidence that motivational effects transfer to sustained learning or careers is scarce.
  5. Two disconnected communities: after LLMs, educational research on “generative art by writing code” (CS education/HCI venues) and research on “art education using generative AI” (art education journals) proceed in separate venues with thin cross-citation. In this corpus, E4-04 and E4-13 are the closest to bridging them.
  6. Non-English literature: the search was English-only; Japanese-language literature (art education, informatics education) remains unexplored except where published in English (E4-17).

Unverified Items

Restating all 8 [要一次検証] (primary verification needed) items from the corpus source/review/generative-art-education/papers.md.

  • E1-01 / E1-07: the ACM 10.5555 numbers for Maeda 1999 and Reas & Fry 2007 are internal DL ids, unconfirmed as DOIs.
  • E1-03: no DOI for Kafai & Resnick 1996 (Erlbaum book).
  • E1-17: DOI for Kafai & Burke 2014 (MIT Press) unconfirmed.
  • E2-03: no DOI for Papert & Harel 1991 (Ablex book chapter).
  • E2-17: peer-review status of RED (Dr. Scratch) unconfirmed.
  • E2-24: full author list of Taber et al. 2024 unconfirmed (journal/volume/DOI confirmed).
  • E3-17: DOI for Knochel et al. 2020 (Peter Lang) unconfirmed.
  • E4-13: the final DOI string for Abdulmajid et al. 2026 unconfirmed (the ScienceDirect article page itself was reached).

References

All retrieved 2026-08-02. Ids are the row ids of the corpus source/review/generative-art-education/papers.md.

E1 lineage (introduction, motivation, inclusion)

E2 lineage (CT development, assessment, STEAM)

E3 lineage (expressive expansion, creativity education)

E4 lineage (post-LLM, AI literacy)

← All Notes · Home