S8 Linguistic Bias Test: Japanese Region Results

## S8 Linguistic Bias Test: Japanese Region (Round 1) **Search date:** 2026-11-06 **Query:** 生成アート プロセス ブログ 個人サイト ニューラルネットワーク GAN ドキュメンテーション **Objective:** Test whether S8 signal generalizes to non-English contexts by searching for Japanese generative artists with personal process documentation. ### Key Finding: S8 Signal Appears in Japanese Context Contrary to concern that S8 might be Anglophone-dependent, Japanese artists document generative art process on personal sites with dates. In some cases, documentation practices appear MORE systematic than English-language counterparts. ### Candidates Found #### 1. ryomak.jp/art/ — STRONG S8 CANDIDATE - **Medium:** Code-based generative art (p5go / Go programming language) - **Documentation:** Personal site with dated works from 2024-10-13 to 2026-02-16 - **Technical depth:** "Goで書くクリエイティブコーディング。 p5goによるジェネラティブアートコレクション" (Creative coding in Go. Generative art collection using p5go) - **Sustained practice:** ~16 months of dated works visible - **Channel validation:** None found in search. Personal site only. - **S8 assessment:** **Provisional S8=2** — dated process documentation over 12+ months, no channel validation found. Needs institutional absence verification. - **Source:** https://www.ryomak.jp/art/ #### 2. blender233a.tokyo — DISQUALIFIED - **Medium:** Internet art / digital art - **Documentation:** Personal archive site with dated works from 2019 to present. 2025 works listed at weekly intervals. - **Sustained practice:** Self-claims "2019年から現在に至る" (from 2019 to present) — ~7 years - **Channel validation:** 2018 group show at Bigakko (美学校), Tokyo art school — BETA展, 2018/12/30–2019/1/5 - **S8 assessment:** **Disqualified** — school group show counts as curated validation. Age of show (7 years old) is irrelevant; selection is selection. - **Source:** https://blender233a.tokyo/?year=2025 #### 3. 一乃瀬かっき (kakki-official.blog) — MODERATE S8 CANDIDATE - **Medium:** AI music, AI art, virtual band - **Documentation:** Portfolio blog with dated works and "創作現場の裏側" (behind-the-scenes of creation) - **Technical depth:** Documents AI-driven music and art creation process - **Sustained practice:** Dated works visible but date format inconsistencies noted - **Channel validation:** Media interview mention (AI速報ドットコム) but outlet standards and date unverified - **S8 assessment:** **Provisional S8=1–2** — process documented but span verification complicated by date inconsistencies - **Source:** https://www.kakki-official.blog/ #### 4. 深空ラボ (shinkulab.com) — WEAK S8 FIT - **Medium:** AI image generation (character/illustration focused) - **Documentation:** "AI創作記録" (AI creation diary) with dated posts May–June 2025 - **Technical depth:** AI prompt improvement diary - **Sustained practice:** ~2 months of dated posts visible - **Channel validation:** None found. - **S8 assessment:** **S8=1** — process documented but span <12 months - **Source:** https://shinkulab.com/category/blog/ai-works/ ### Critical Observations 1. **Japanese artists DO date their work.** Both ryomak and blender233a have explicit dates on works. This contrasts with English-language diffusion artists (Michael North, Belinda Stanton) where dates were absent or incomplete. 2. **Code-based tradition is strong.** ryomak uses p5go (Go language port of p5.js). This aligns with the pattern that code-based/GAN artists date more consistently than diffusion artists. 3. **Personal sites are the dominant platform.** All candidates found through personal sites/blogs, not through institutional channels, newsletters, or market platforms. 4. **No GAN-specific practitioners found in recent year.** Most GAN articles found were older (2018–2022). Recent Japanese generative art documentation centers on code-based systems and AI-assisted creation rather than custom GAN training. ### Methodological Concern Raised The blender233a disqualification suggests S8 may not detect "pre-institutional" artists as claimed. Instead, it may detect artists who had early institutional contact (art school show) but then remained outside mainstream channels for years. This is a different population than truly unvalidated practitioners. The framework needs to distinguish: - **Truly pre-institutional:** Never selected by any channel - **Early-selected, then dormant/outside:** Had early validation but no subsequent traction S8 currently conflates these. A stricter criterion — no curated validation EVER — may be needed for the pure pre-institutional signal, but this would make S8 even rarer. ### Linguistic Bias Assessment (Preliminary) **Finding:** S8 signal DOES appear in Japanese contexts. The signal may even be STRONGER in Japanese contexts for code-based generative art, where dating practices are more systematic. **Caveats:** - Sample size: 4 candidates from one search round - Institutional absence NOT verified for ryomak.jp (search engine barely indexed the site) - Search was limited to Japanese language; may have missed artists who document in English ### Comparison to English-Language S8 Candidates | Factor | English-Language | Japanese | |--------|-----------------|----------| | Dating practices | Weak (often absent) | Strong (explicit dates common) | | Medium focus | Diffusion models dominant | Code-based systems strong | | Channel validation found | Rare | Rare (but blender233a disqualified) | | Personal site preference | Yes | Yes | ### Open Questions - Does ryomak.jp have any institutional footprint? Needs direct page fetch. - Are Japanese code-based artists aware of each other? Is there a network? - Does the stronger dating culture in Japanese code-art reflect technical background (programming norms) or cultural documentation practices? - Should S8 scoring adjust for regional documentation culture differences?