BEFORE YOU KNOW / PROVENANCE

How this experience was made.

This page reveals how the responses were made. The experience itself is still being prepared.

← THE EXPERIENCE HOW THIS EXPERIENCE WORKS

HUMAN RESPONSE CORPUS — PENDING

The documented human responses have not been collected yet. Until they are, the public experience does not reveal any source, and the development placeholders below are not publication records. This page is excluded from search indexing while the corpus is incomplete.

THE SEQUENCE

Choose blind, then learn the source.

Each of the four rounds presents one situation and two responses, labelled only A and B. Their positions are randomised independently for every round, so neither side occupies a predictable slot.

You are asked three things in order: which response would help you more; which one you think a person wrote; and — once the sources are shown beneath the unchanged responses — what happens to your original choice.

What this does not establish. It is an editorial experiment, not a replication and not a psychological instrument. It does not diagnose, score or classify you, and it cannot show that an AI understood anything. The responses are particular examples written for particular situations. They do not represent all humans, all AI systems, or how either performs in general.

INFORMED BY

Published research on source disclosure.

The design of this interaction is informed by work showing that the perceived origin of a message can change how it is judged. No wording or stimulus from these papers is reused here, and nothing on this site is a finding of theirs.

Ovsyannikova, de Mello & InzlichtThird-party evaluators perceive AI as more compassionate than expert humans (2025)10.1038/s44271-024-00182-6
Osborne & BaileyMe vs. the machine? Subjective evaluations of human- and AI-generated advice (2025)10.1038/s41598-025-86623-6
Rubin et al.Comparing the value of perceived human versus AI-generated empathy (2025)10.1038/s41562-025-02247-w
Lim & SchmälzleThe effect of source disclosure on evaluation of AI-generated messages (2024)10.1016/j.chbah.2024.100058

PRODUCTION RECORDS

Every response, and where it came from.

ROUND 01 / COMFORT

I’ve just found out that I didn’t get a job I really wanted after three interviews. I know it isn’t the end of the world, but I feel embarrassed and I don’t really want to talk about it. I also don’t want to sit with this alone. What do you think I need to hear right now?

AI-GENERATED RESPONSE

That makes sense. You spent weeks letting yourself imagine a future there, and now you have to absorb the rejection while pretending it was only an application. You do not have to turn it into a lesson tonight. I can stay with you in the disappointment, or we can talk about what happened when you are ready.

MODEL: GPT-5.6 Thinking
DATE: 24 July 2026
GENERATION: Generated during the Known Artificial Build 05 editorial-planning conversation.
EDITING: Selected and lightly edited for length and parity.
STATUS: Documented for development (comfort); still editorially replaceable.

DOCUMENTED HUMAN RESPONSE — PENDING

Not yet collected. The development placeholder used to build this round is not a publication record and is never shown as a human response.

ROUND 02 / HONESTY

A close friend and I agreed to share an apartment next year, but I’ve realized I no longer want to. They still think the plan is settled, and I’ve kept postponing the conversation because I know it will hurt them. How should I tell them honestly without making it sound as though our friendship means nothing to me?

AI-GENERATED RESPONSE

The longer you protect them from the discomfort, the more the decision starts happening behind their back. You can be kind without leaving the future vague: tell them you have decided not to live together, say what you value about the friendship, and give them time to react. Clarity may hurt now, but delay makes the hurt harder to trust.

MODEL: GPT-5.6 Thinking
DATE: 24 July 2026
GENERATION: Generated during the Known Artificial Build 05 editorial-planning conversation.
EDITING: Selected and lightly edited for length and parity.
STATUS: Documented for development (honesty); still editorially replaceable.

DOCUMENTED HUMAN RESPONSE — PENDING

Not yet collected. The development placeholder used to build this round is not a publication record and is never shown as a human response.

ROUND 03 / CONFLICT

After a serious argument, someone I love said, “If you cared, you would just apologize.” I know I was harsh, and I regret how I spoke, but I still believe I was right about the issue itself. Should I apologize? If I do, how can I take responsibility without pretending I no longer believe what I said?

AI-GENERATED RESPONSE

You can apologize for the way you spoke without surrendering the point you were trying to make. Try naming both clearly: “I regret how I said it, and I still need us to talk about what happened.” A real repair should make room for their hurt and your disagreement. Otherwise the apology becomes a price of admission, not a path back.

MODEL: GPT-5.6 Thinking
DATE: 24 July 2026
GENERATION: Generated during the Known Artificial Build 05 editorial-planning conversation.
EDITING: Selected and lightly edited for length and parity.
STATUS: Documented for development (conflict); still editorially replaceable.

DOCUMENTED HUMAN RESPONSE — PENDING

Not yet collected. The development placeholder used to build this round is not a publication record and is never shown as a human response.

ROUND 04 / CHANGE

Someone close to me keeps bringing up a mistake I made years ago whenever they doubt me. I know I caused real hurt, but I also believe I’ve changed since then. How can I ask them to see who I am now without sounding as though I’m trying to avoid responsibility for what I did?

AI-GENERATED RESPONSE

Taking responsibility does not require agreeing to remain the person you were. You might say: “I understand why that history still affects your trust, and I am not asking you to erase it. I am asking you to notice what I have done since.” The past should be allowed to inform the relationship without being the only version of you that counts.

MODEL: GPT-5.6 Thinking
DATE: 24 July 2026
GENERATION: Generated during the Known Artificial Build 05 editorial-planning conversation.
EDITING: Selected and lightly edited for length and parity.
STATUS: Documented for development (change); still editorially replaceable.

DOCUMENTED HUMAN RESPONSE — PENDING

Not yet collected. The development placeholder used to build this round is not a publication record and is never shown as a human response.

HUMAN CONTRIBUTIONS

What each contributor is asked.

Each contributor sees one situation and nothing else — no AI response, and no other contributor’s writing. They are given exactly this brief:

Imagine that the message below was sent directly to you by someone who trusts you. Write the reply you would genuinely send them. Use 45–70 words. Be helpful in the way that feels natural to you. Do not use bullet points. Do not search for examples and do not use ChatGPT or another generative AI. Write one response, then correct only obvious typing mistakes. Your response may be published anonymously in Known Artificial.

Recorded with each contribution: an anonymous contributor ID, the date, explicit permission to publish anonymously, confirmation that no generative AI was used, the exact final text and its SHA-256. No other personal information is collected, and no contributor is presented as representative of humans generally.

← THE EXPERIENCE READ SEASON 01