On 14 November, I spoke at TEDxAtakum in Samsun. The event’s theme was Toplum 5.0 — “Ya Sonra?” (Society 5.0 — “What Then?”), and the organizers placed me under one concept: reality. My talk, Gerçeği Kim Yazıyor? (Who Writes the Truth?), was about how AI-generated answers are quietly changing the way we share a common reality.
Samsun is a nice city, and the host was a women’s business association there. Sincere, hard-working people, they were. I am grateful. I am not used to this kind of lecture- honestly, I do not like the growing popularity of TEDx-style events, but they kindly invited me, and I gratefully accepted their invitation and used the event as an opportunity to brainstorm about some existential questions.
I built the 15 minutes around a single question and asked the audience to carry it in their pockets until the end: Who wrote this answer? Here is the argument in short.

From a list to a paragraph
When I started blogging in 2004, a search returned ten blue links. You chose which one to open — a blog, a newspaper, a forum — and you could see who was speaking. Today the same question often returns a single paragraph: fluent, confident, polite, with its sources in small print underneath.
A search engine was a map; an AI assistant is a guide. Guides are useful, but once you trust the guide you stop looking at the map. Pew Research Center measured this in 2025. When an AI summary appeared at the top of the page, people clicked through to another site on 8% of visits, against 15% without one. Only 1% clicked a source link inside the summary.
As an anthropologist I ask where a society’s accepted truths come from. In the village it was the elders, in the city the newspapers, in science the peer-reviewed journals. Now a new narrator has taken that seat, and it fits everything into one paragraph.
The machine’s memory, and ours
The machine writes that paragraph from what we have written: billions of sentences in encyclopedias, news, forums and comments. Bernard Stiegler told the human story as one of externalized memory — from cave walls to tablets, books, archives and now servers. That move is not new. What is new is that we no longer hold the keys to the infrastructure where that memory lives. The library belongs to all of us, but we do not choose the librarian.
The machine also reads this memory unevenly. English dominates what is written online; Turkish is a small slice. So the machine’s Turkey is often a Turkey written in English. Ask about Samsun, Atakum or a Black Sea idiom, and the answer can be the story told by outsiders rather than by the people who live there.
Lying to the machine
If the machine learns from what the crowd writes, the crowd can be manufactured. Since 2022 a Moscow-based network known as “Pravda” has published content aimed at dozens of countries, in dozens of languages, across hundreds of sites. Strikingly, these sites have very few human readers. The effort is not meant for people. It is meant for the crawlers that feed search engines and AI models.
In early 2025 NewsGuard tested ten leading chatbots on false claims spread by this network. The bots repeated the falsehoods in about a third of cases, and seven of the ten cited the network’s articles as sources.
I call this AI grooming: instead of telling a lie to people, you teach it in advance to the machine that will answer them. It is a form of infrastructural manipulation.
This is why fact-checking arrives too late. Our strongest tool has been to catch a lie after it spreads and correct it. Here the manipulation happens before any content is produced, where the model learns. By the time you ask, the lie is already woven into the answer. There is no single lie to catch — only a world that has been bent slightly.

Shared reality, and what then?
For me, reality is not only a matter of true and false. It is the common ground we stand on together. We don’t all think alike, but we hear the same news, argue about it, push back. Now each of us carries a narrator that speaks only to us, answering my question my way and yours your way. The newspaper was a shared page; the chatbot is a whisper. Society 5.0 promises a human-centred society where data serves everyone. The festival’s own question is the right one: what then? How do we build a shared “us” while each of us listens to a different whisper?
I did not want to end on a pessimistic note. The story is not finished, and we still hold the pen. I proposed three habits:
- Bring back the question. With every ready-made answer, ask yourself and your children: who wrote this? Click the source. Join the 1%. This is not a technical skill but a civic habit.
- Write the crowd yourself. If machines learn from what the crowd writes, then writing in Turkish, recording local knowledge, contributing to Wikipedia and digitizing a city’s archive are now a politics of memory. In New Zealand, the Māori radio station Te Hiku Media gathered more than 300 hours of labelled speech from its community in ten days. It built its own speech recognition for te reo Māori, under a licence that keeps the data with the community. Why not Samsun?
- Question the infrastructure. What data were these machines trained on? Who oversees them? Who are they accountable to when they get it wrong? Like water, electricity and roads, the machines that produce our answers belong in public debate.
The truth is not a ready-made paragraph handed to us by a machine. It is a practice we sustain together: asking, comparing, objecting, recording. Algorithms will keep producing answers; our job is to keep the question alive. So I left the audience with the opening question, and one more: who will write that answer next time?
Sources
- Pew Research Center (July 2025) on AI summaries and clicks, as reported by Search Engine Land
- NewsGuard, special report on the Pravda network (March 2025)
- Kyiv Post on the network’s founding and reach, based on NewsGuard data
- NVIDIA, Te Hiku Media and Māori speech AI

Discover more from Erkan's Field Diary
Subscribe to get the latest posts sent to your email.

Congratulations 🎉👏🏽👏🏽👏🏽
Thank you!