What It Is
A nonfiction listicle structured as an April Fools game: eleven “AI features” that were secretly just humans, two of which are fabricated. Readers guess the lies. The conceit creates engagement while delivering a serious argument about AI labor.
The Argument
AI products are systematically marketed as autonomous when they depend on underpaid, invisible human labor. The “Mechanical Turk” pattern — humans doing the work that press releases claim doesn’t exist — runs through the history of commercial AI. The piece documents it case by case, then discloses its own AI-assisted production, implicating itself in the dynamic it critiques.
The Eleven Items
Real cases:
- Amazon Just Walk Out — ~1,000 workers in India reviewing footage; 700/1,000 transactions required human review vs. target of 50
- X.ai “Amy Ingram” — AI scheduler backed by human trainers like Willie Calvin reading and writing emails
- Expensify SmartScan — receipt images posted to Amazon Mechanical Turk, transcribed manually for cents per task
- Content moderation — human moderators in Kenya, Philippines, India reviewing worst content; OpenAI paid <$2/hour; PTSD rates high
- Cruise “driverless” cars — remote human operators took control 2-4% of rides
- RLHF — thousands of human evaluators training LLMs; CWA survey: 66% waited for tasks unpaid; 23% had employer health insurance
- AI customer service — decision trees; human agents receive transfers but still ask you to re-explain everything
- Google Search Quality Raters — 16,000 humans evaluating results with 170-page manual; 719,000+ tests informed by their ratings
- AI art generators — trained on artists’ work without consent; class-action lawsuits; Copyright Office: purely AI-generated work can’t be copyrighted
- Amazon Mechanical Turk — named after the hoax intentionally; median wage ~$2/hour; only 4% earning above minimum wage
The meta-item (#11): The piece itself — written with Claude, reviewed and restructured by the author. “A collaboration between a language model and a guy in a room in Michigan.” The final item implicates the piece in the dynamic it spent 1,500 words critiquing.
What Makes It Work
- The game structure (spot the 2 lies) creates genuine engagement — readers have a reason to read carefully
- The Mechanical Turk historical frame (250 years, same scam, better branding) elevates it from a list to an argument
- The self-disclosure at the end is the sharpest move — it doesn’t let the reader or author off the hook
- Sourced throughout with footnotes; this is serious reporting in listicle form
Connection to HDftS
The fiction and nonfiction are two angles on the same question. HDftS satirizes AI over-autonomy (AIs doing too much, developing their own logic). This piece reveals AI pseudo-autonomy (AIs hiding humans behind the curtain). Together: AI is neither as autonomous as feared nor as autonomous as marketed.
What It Leaves Open
- Who are the “2 lies”? (Not revealed in the clipped version — would need the full published piece)
- The RLHF item is the most politically charged; could be expanded into a standalone piece
- The AI art / copyright question is unresolved legally and will develop
- The self-disclosure (#11) raises a follow-up: does the human-AI collaboration model change the ethics of this, or just make it more honest?
Sourcing
Extensively footnoted. Key sources: The Information (Amazon Just Walk Out), TIME (OpenAI/Kenya), CWA Union survey (RLHF workers), Carnegie Mellon study (Mechanical Turk wages), multiple class-action filings (AI art).