Why I'm leaving Medium: AI policy
An author quitting Medium over its permissive stance toward AI-generated content sparks a broader debate about how generative AI is reshaping online writing and creative work. Commenters argue over whether AI assistance enhances expression or erodes authenticity, how to keep AI crawlers from scraping personal blogs, and whether the economic gains from automation will justify potential job losses and degraded information quality. Many see a return to self-hosted blogs and human‑curated spaces as a way to preserve trust and personality in what they fear is becoming an increasingly AI-saturated web.
Scope of Objections to AI on Medium
- Many agree that AI-generated “slop” (SEO’d listicles, shallow tutorials) is already degrading search results and Medium’s quality.
- Some defend a personal, absolutist rejection of AI in creative work (text and even images) as a matter of values and association: once AI touches a piece, trust in the whole drops.
- Others see this as overreaction: AI is “just a tool,” akin to spellcheck or an editor, and usefulness depends on how narrowly it’s applied.
Scraping, Robots.txt, and Publisher Control
- One camp argues that all public sites will eventually be scraped; robots.txt is voluntary and often ignored.
- Another claims major AI crawlers and CommonCrawl currently respect robots.txt, with legal reinforcement in the EU’s text/data-mining rules.
- Perplexity is cited as an example of apparent robots.txt-ignoring behavior; others say this was a misunderstanding about user-initiated fetching rather than crawler training data.
AI, Work, and Society
- Analogy debate: opposing AI is compared to an ice delivery worker trying to ban refrigerators; critics call this dismissive of livelihoods.
- Some worry AI will accelerate a long-running pattern where automation removes lower-skill jobs, leaving many unemployable without a clear safety net (UBI, etc.).
- Others argue that increasing automation should mean more leisure, but skeptics note productivity gains haven’t reduced working hours historically.
Use Cases: Helpful vs Harmful
- Medical diagnosis is a key fault line:
- Pro-AI commenters cite studies of imaging systems with high accuracy, and are fine with AI-assisted doctors.
- Skeptics question real-world performance, demand human final judgment, and some would switch doctors if AI tools were used at all.
- Distinction is drawn between:
- AI as upstream assistance with human review.
- AI as the final decision-maker, which many reject.
Alternatives to Medium and the “Dead Internet”
- Suggested alternatives: Ghost, WriteFreely, Bear, static-site generators (Hugo) with hosting via Netlify/Cloudflare/etc., classic Blogspot/WordPress, or fully self-hosted HTML with RSS.
- Several foresee large platforms turning into bottomless pits of AI content; self-hosted blogs plus RSS are proposed as a refuge, albeit with less reach.
Tone and Readability of AI Text
- A common, specific dislike is the “AI voice”: verbose, polite, repetitive, and often off-topic.
- Some argue people only notice bad AI; well-done AI text may already be widespread and invisible.
- Others emphasize that they want direct human thought, not an LLM’s rephrasing of someone’s ideas.