Be more specific about the "etc"? #45

lemoer commented 2026年03月26日 18:05:00 +01:00

First of all, thanks for the idea and spec.

Cited from the README.md:

The presence of this file indicates that the content of your site was generated by you, not AI. It's fine to use AI tools that assist you with spellchecking, grammar, formatting, etc. Just make sure to be clear and transparent about how AI is used, and consider publishing an /ai slashpage with your policy

We were just discussing what is behind the word "etc". It is is not very clear to me.

Let's imagine the following exemplary scenarios:

1. You write a text by hand and tell an AI to reduce the text to half of the content.
2. You write down some bullet points and let an AI formulate whole sentences out of it.

In both cases, I would say, one could argue that AI was used as "a tool that assists you". The ideas/content comes from a human and the AI was only used to "make a nice" text out of it. So one could argue that both scenarios are part of "etc".

On the other hand, one could also argue that this is more than "spellchecking, gramar, formatting", which would then question what "etc." means.

Also, if a human reads what the AI has produced and takes responsibility of it. Does this make a difference?

Maybe some clarification would be helpful?

First of all, thanks for the idea and spec. Cited from the README.md: > The presence of this file indicates that the content of your site was generated by you, not AI. It's fine to use AI tools that assist you with spellchecking, grammar, formatting, etc. Just make sure to be clear and transparent about how AI is used, and consider publishing an /ai slashpage with your policy We were just discussing what is behind the word "etc". It is is not very clear to me. Let's imagine the following exemplary scenarios: 1. You write a text by hand and tell an AI to reduce the text to half of the content. 2. You write down some bullet points and let an AI formulate whole sentences out of it. In both cases, I would say, one could argue that AI was used as "a tool that assists you". The ideas/content comes from a human and the AI was only used to "make a nice" text out of it. So one could argue that both scenarios are part of "etc". On the other hand, one could also argue that this is more than "spellchecking, gramar, formatting", which would then question what "etc." means. Also, if a human reads what the AI has produced and takes responsibility of it. Does this make a difference? Maybe some clarification would be helpful?

robida commented 2026年03月26日 22:34:23 +01:00

To be honest I added that "etc" because I don't know the answer to your question. But I also think it doesn't matter — let me try to explain why.

A person will add a human.json to their website because they want to declare that they created the content. It's a way of saying to the world:

I wrote this myself, and I'm proud of it.

The problem is that "I created this myself" is very subjective. If they used an old-fashioned spellchecker, or used Microsoft Word's grammar check, does that count as being human-created? What if they asked Claude to proofread the text, and incorporated some of the suggestions? Where do we draw the line — and do we draw the line based on algorithmic complexity, on environmental impact, on degree of copyright infringement? One person might be OK with someone using a traditional spellchecker on their blog post, but not with them running it by ChatGPT, even though they're both just algorithms with different complexities. Because of this, I don't want to be the AI police and define the exact criteria needed for someone to have a human.json file on their website.

Personally, I wouldn't want to read a blog post that was generated by AI from a few bullet points. I would be slightly more comfortable with a blog post that was shortened in half by AI, assuming the author's original "voice" was still there. On the other hand, I would be fine with a non-native speaker using AI to translate their whole blog post to English, and I would be perfectly fine with someone that needs an eye-tracking communication device using AI to write sentences from bullet points. But that's just me.

How do we encode that in a JSON file? We can't. Which is why the recommendation is "try to not use AI, and if you do, be transparent about it". And which is why I think anyone who wants to publish a human.json file should just go ahead and publish one. I'm not going to police anyone.

I also think it doesn't matter much at the end of the day, because the usefulness of the protocol depends on the people you decide to trust. If you're completely against AI you will likely only trust sites of people who are also against AI, and they will likely be vouching only for similar people. So your web-of-trust becomes one of similarly minded people. Likewise, if you want to trust someone who uses AI to generate paragraphs from bullet points, your web-of-trust will likely have people that have a lax approach to AI, and what it means for content to be human-created.

With all that said, I think we can update the spec to say something like:

The presence of this file indicates that the content of your site was generated primarily by a human: you. Any use of AI is assistive only, and is described in accompanying documentation (consider publishing an /ai slashpage with your policy).

To be honest I added that "etc" because I don't know the answer to your question. But I also think it doesn't matter — let me try to explain why. A person will add a `human.json` to their website because they want to declare that **they** created the content. It's a way of saying to the world: > I wrote this myself, and I'm proud of it. The problem is that "I created this myself" is very subjective. If they used an old-fashioned spellchecker, or used Microsoft Word's grammar check, does that count as being human-created? What if they asked Claude to proofread the text, and incorporated some of the suggestions? Where do we draw the line — and do we draw the line based on algorithmic complexity, on environmental impact, on degree of copyright infringement? One person might be OK with someone using a traditional spellchecker on their blog post, but not with them running it by ChatGPT, even though they're both just algorithms with different complexities. Because of this, I don't want to be the AI police and define the exact criteria needed for someone to have a `human.json` file on their website. Personally, I wouldn't want to read a blog post that was generated by AI from a few bullet points. I would be slightly more comfortable with a blog post that was shortened in half by AI, assuming the author's original "voice" was still there. On the other hand, I would be fine with a non-native speaker using AI to translate their whole blog post to English, and I would be perfectly fine with someone that needs an eye-tracking communication device using AI to write sentences from bullet points. But that's just me. How do we encode that in a JSON file? We can't. Which is why the recommendation is "try to not use AI, and **if you do, be transparent about it**". And which is why I think anyone who wants to publish a `human.json` file should just go ahead and publish one. I'm not going to police anyone. I also think it doesn't matter much at the end of the day, because **the usefulness of the protocol depends on the people you decide to trust**. If you're completely against AI you will likely only trust sites of people who are also against AI, and they will likely be vouching only for similar people. So your web-of-trust becomes one of similarly minded people. Likewise, if you want to trust someone who uses AI to generate paragraphs from bullet points, your web-of-trust will likely have people that have a lax approach to AI, and what it means for content to be human-created. With all that said, I think we can update the spec to say something like: > The presence of this file indicates that the content of your site was generated primarily by a human: you. Any use of AI is assistive only, and is described in accompanying documentation (consider publishing an /ai slashpage with your policy).