Awomosu – Frameworks Extracted

Source: Awomosu, Abi. (2026, June 14). Writing was never a test of who could think. Why the people who were never heard are the ones this AI medium needs. Substack. https://substack.com/home/post/p-201968216

1. The core reframe

Tool vs. medium. A tool is picked up and put down; a medium reorganises the people inside it (cf. print, broadcast, smartphone). Tool-framing produces only two positions — use it / refuse it — neither of which addresses how to operate inside something that's already shaping you.

The amplifier thesis. The model doesn't author, it amplifies whatever it's pointed at. Pointed at nothing (no specified voice), it defaults to the trained-on average — which she calls the "Empire voice" (corporate/editorial/publishing register). Pointed at a writer's own corpus, it amplifies that writer, sometimes into caricature. The "AI voice" people recoil from is therefore not a machine sound but a familiar institutional register, finally audible at volume.

2. The "-speak" taxonomy

A historical sequence of adaptations writers make to whatever controls access — each prior one forward-running (stretching toward legibility for a gatekeeper), with flaw-speak as the sole backward-running exception (shrinking to seem less capable).

-speak	Medium / gatekeeper	Adaptation
Publisher-speak	Print	Standardised spelling; invented "illiteracy" as a category
Broadcaster-speak	Radio/TV	Flattened accent and story shape toward a neutral, universal register
Txt-speak	Mobile/SMS	Compression to fit character limits (ur, c u l8er)
SEO-speak / portal-speak	Search engines	Keyword-stuffing for the crawler
Algo-speak	Social platforms	Ciphered words (unalive, seggs) to dodge moderation/shadowban
Flaw-speak	AI detectors	Deleting em-dashes, inserting typos, dumbing down — writing worse to be believed human
Soul-speak (proposed alternative)	—	Writing from one's own observations/feelings first ("soul before silicon"); using AI to listen/research, not to generate first

3. The voice/ear grid

Two axes determine outcome — not "how much AI was used":

• Voice (x-axis): did the writer bring something only they could say — specificity, lived detail, an angle that can't be generated?
• Ear (y-axis): can the writer hear when the output has flattened toward the average, and push back?

Resulting quadrants:

• High voice + high ear → "alive." E.g., the voice-note writer: dictates their own thinking, AI only transcribes/formats. Ear fully intact because nothing of the writer's content was generated.
• Low voice + low ear → "slop." The generate-and-paste, content-engine pattern — produced overwhelmingly by people who could already write but switched off the internal reader for speed/volume.
• High voice + low ear (yet) → "Lost Treasure." A new or non-linear writer with real substance who can't yet hear when the machine has smoothed them. Framed as a waiting room, not a verdict — the ear is the most learnable part of the system.
• Low voice + high ear → (implied, not separately named) a skilled writer with a sharp ear but nothing distinctive to say — competent but generic.

Key claim: the variable that matters is never the dial between "more machine" and "more human" — it's whether the internal reader was in the room.

Practice spectrum (within "high ear" usage)

• Voice note — dictate, AI transcribes/structures only; cleanest use, especially valuable for dyslexia/executive-function barriers
• Argue-with-it — draft first, use AI as adversarial sparring partner; no AI sentences survive into final text
• Scaffold and abandon — AI builds an outline that's then argued against/discarded; listening happens in the reaction
• The iron rule — AI for research/fact-checking/reading only; prose held sacred
• The polish — AI smooths a finished draft; high risk of sanding off the writer's texture unless the ear catches it
• The twin — model trained on the writer's own corpus to amplify their voice; highest ceiling, highest vigilance cost (risk of self-caricature)

4. Bibliography

Books cited

Apprich, Clemens; Chun, Wendy Hui Kyong; Cramer, Florian; Steyerl, Hito. Pattern Discrimination.

Buolamwini, Joy. Unmasking AI.

Fisher, Max. The Chaos Machine.

Freire, Paulo. Pedagogy of the Oppressed.

Gray, Mary; Suri, Siddharth. Ghost Work.

Havelock, Eric. The Muse Learns to Write.

Jarvis, Jeff. The Gutenberg Parenthesis.

Jensen, Derrick. A Language Older Than Words.

Kockelman, Paul. Last Words: Large Language Models and the AI Apocalypse.

Lanier, Jaron. You Are Not a Gadget.

McGilchrist, Iain. The Master and His Emissary.

McLuhan, Marshall. Understanding Media: The Extensions of Man.

McLuhan, Marshall; Powers, Bruce. The Global Village.

Pasquinelli, Matteo. The Eye of the Master: A Social History of Artificial Intelligence.

Scott, James C. Seeing Like a State.

Shlain, Leonard. The Alphabet Versus the Goddess.

Vallor, Shannon. The AI Mirror.

Vara, Vauhini. Searches: Selfhood in the Digital Age.

Wolf, Maryanne. Reader, Come Home: The Reading Brain in a Digital World.

Yunkaporta, Tyson. Sand Talk: How Indigenous Thinking Can Save the World.

Yunkaporta, Tyson. Right Story, Wrong Story.

Peer-reviewed / empirical research cited

Agarwal, Sachin; Naaman, Mor; Vashistha, Aditya. (2025). AI suggestions homogenize writing toward Western styles and diminish cultural nuances. ACM CHI. https://doi.org/10.1145/3706598.3713564

Atari, Mohammad; Xue, Mengchen; Park, Peter; Blasi, Damián; Henrich, Joseph. (2023). Which humans do LLMs resemble? Harvard University working paper.

Doshi, Anil; Hauser, Oliver. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances. https://doi.org/10.1126/sciadv.adn5290

Draxler, Fiona; et al. (2024). The AI ghostwriter effect: When users do not perceive ownership of AI-generated text but self-declare as authors. ACM Transactions on Computer-Human Interaction. https://doi.org/10.1145/3637875

Fan, Yizhou; et al. (2025). Beware of metacognitive laziness. British Journal of Educational Technology. https://doi.org/10.1111/bjet.13544

Liang, Weixin; et al. (2023). GPT detectors are biased against non-native English writers. Patterns. https://doi.org/10.1016/j.patter.2023.100779

Padmakumar, Vishakh; He, He. (2024). Does writing with language models reduce content diversity? ICLR. arXiv:2309.05196

Sharma, Mrinank; et al. (2024). Towards understanding sycophancy in language models. ICLR. arXiv:2310.13548

Shumailov, Ilia; et al. (2024). AI models collapse when trained on recursively generated data. Nature, 631, 755–759.

Steen, Erin; Yurechko, Kathryn; Klug, Daniel. (2023). You can (not) say what you want: Using algospeak to contest and evade algorithmic content moderation on TikTok. Social Media + Society. https://doi.org/10.1177/20563051231194586

From the author's own book

Awomosu, Abi. How Not To Use AI: 50 Contrarian Principles for the Imagination Age. (Cited laws: 2, 3, 8, 30, 38, 46, 47 — covering medium-vs-tool, listen-before-generate, full-spectrum thinkers and AI replacement fears, "ancestral" framing of AI, emotional vocabulary, writing-to-resonate, and content-vs-connection.)