Show HN: Pledge Your Human-Made Content | Hacker News Reader

Show HN: Pledge Your Human-Made Content

nonbot.org

39 pointsjonodonovan41 comments

AI generated content is on the rise. How can you be sure what you're reading is 100% human?

Like with the photography competition ending up with a winner who just generated an image via AI, you're going to get the same bad apples validating their AI-generated crap via this as there's no validation going on in reality.

They're also incentivizing them with backlinks and validation that it is legit content "because a third-party said so".

Sorry, I don't understand. How does this work? I haven't kept up with SOTA papers, but I thought it was currently not possible to detect AI generated content with good confidence?

How do you prevent someone who is generating AI content from applying the non-bot badge? If you can't, then what's the point?

Because it's virtue signaling. People are seen to be doing something. It's a LARP.

Use the non-bot badge to feed bots their own effluvia?

You have caught up. It doesn’t work. They appear to not even be trying here and instead just asking people to “pledge” it’s human made. I’ll ask ChatGPT to write my pledge

I do not like that this exists, this shouldn't need to exist, AI or ML should not write articles, but they should just be a companion who helps you put things into words if you struggle with it.

I guess it is too much to ask for in the clickbait-y, monetization per-view driven internet.

I agree fully with you! Unfortunately, law enforcement should not need to exist also...And yet, the incentives and behaviors of so many a$$holes in the world requires at least some mechanism to curb the bad stuff. (Not saying this not-bot effort is effective, but just saying that sometimes in life we need something even if not perfect...in order to keep some things in check). /sigh

I don't think it is correct to compare my above opinion with law enforcement, it doesn't need to be "enforced", may be use it as a nice guideline for publishing stuff.

If someone did write things purely with some algorithm, just annotate that it was not written by a human, so that people who don't want to read made-up/generated articles can just ignore it and move on.

Doing it the reverse way does not make any sense to me, it is the same as treating all articles not having "written by human" label as generated content. -- in the "law enforcement" perspective, this is "guilty until proven innocent".

What can and should be enforced is licence checks, like reusing things not in public domain should follow the license when publishing (looking at you Copilot).

> ...may be use it as a nice guideline for publishing stuff.

Totally agree that guidelines/guidance is ideal rather than zooming all the way to the enforcement side of the spectrum. I think i must have been thinking of "community policing" in that there is positive guidance and engagement well before actual enforcement is conducted...and your comment reminded me of this...so fully agree with you there. :-)

> ...If someone did write things purely with some algorithm, just annotate that it was not written by a human, so that people who don't want to read made-up/generated articles can just ignore it and move on...

I've heard others state something similar in this context for AI-authored content (or not AI-authoered content as whatever the case might be), and I'm entirely in favor of this. Its conceptually similar - at least in my mind - to "organic" or "fair trade" (or similar) that exists for products, which i think is necessary and valuable to society.

If you've ever searched for recipes online and had to read through ten pages of drivel before getting to the actual recipe at the bottom, you may have realized that just because articles are human-written doesn't make the article worthy of existing. SEO made sure of that long ago.

Yes, and something bad happened does not make another bad thing good.

Where did I say or imply that it was good? What the AI-content-apocalypse Chicken Littles seem to constantly miss is that the incentive structure of the public web for the last 20+ years has rewarded trash content, so the public web is an SEO-driven garbage heap. The cost to produce trash content is already effectively zero, and AI will neither reverse nor repair that fact unless the underlying incentive structure changes.

I said that because SEO (or your entire reply) has nothing to do with either the OP or my post said, your reply sounded like an argument, sorry if I misinterpreted.

But please don't change the topic to something irrelevant.

Should we also pledge not to use word processors or spell check?

AI and ChatGPT is a tool. When used appropriately, it can be used to produce good content. For me, I have used it to make my content more engaging and to do things like avoid using the same passive verbs over and over again as I sometimes have a tendency to do. At worst, it produces shallow or wrong or misleading writing, but people also do that.

It will not replace humans providing insightful writing. I think it is getting a lot of negative attention in the press because the advertising, click driven press is predicated on producing lots of words with minimal insight.

I'm not anti-ai.

I think there's a ton of value in 100% human made art, and I think AI art is still valid art - but I think it's still useful to differentiate the two. It's like how unprocessed photography is judged differently from photoshopped art.

Also useful for collecting training data.

However everything already was - and is - ML assisted at this point. Any photo from a smartphone, autocorrect from MS word, most photo editing, etc. How are those counted? Does github copilot count? Not to mention non-ML but still automated work with CI's, compilers, templates, etc. Does intellij's autocomplete count? One could argue templates/stencils/stamps predate computers.

I think there's also a difference between 100% AI made and AI assisted work that needs to be differentiated.

What does 100% human mean?

Not being obtuse, trying to figure out if the threshold is, like "ideas 100% my own, no auto spelling and grammar check, etc" or if it's more "I used GPT to workshop a page title but didn't copy/paste anything".

It's like "natural ingredients". Consumers seem to like it, and it doesn't really mean anything that you can't redefine to your benefit.

Looking at the linked page (which is claimed to be "human-made"), I'm sure that it was made with a significant automation assistance. A template engine had compiled the HTML (or someone was very sloppy with indentation and did a lot of copying and pasting), SVGs code was surely generated by a machine rather than coded by hand (and surely the fonts in those used a bunch of algorithms too), etc etc. The fact that they haven't used some particular kind of matrix multiplication feels irrelevant.

I wonder if it's better to just say "No language models were used to generate this text".

Yeah, I dunno. It's a sorites paradox. I'm annoyed when a search engine delivers me a site that's clearly GPT-generated rubbish (and greatly annoyed when it's not 100% clearly GPT-generated rubbish, but likely is), but it's unrealistic to say that something is 100% human also.

I'd love a way to generally assure me that there's a human behind the thing I'm reading, but I'm not even sure I can define exactly what I'm after when I say that.

I write my blog posts' text content 100% manually. However, I use AI image generators to produce decorative cover art for the posts. I guess my blog wouldn't qualify as 100% human...

Exactly what I was talking about at [0][1].

Happy to see it happening.

To be clear, I'm not anti-AI. It's just that if I'm going to do business with you I want to deal with the human, and this includes content.

Reading AI content for stuff I care about, feels like I'm being laughed off in the face. It feels fraudulent.

[0]: https://news.ycombinator.com/item?id=35531041

[1]: https://www.ivanmontilla.com/blog/certified-human-made

This is nonsense. You're falling for the toupée fallacy. You have no idea you're reading AI content when it's sufficiently good enough for you to not recognize it. Therefore, you view all AI generated content as bad.

I suspect this is just a way for mega corps to find actual artisanal content with which to train their AIs.

Ah this came to my mind as well. Although I didn’t consider it a serious thought. But then if one day we find out that this was the purpose (for a similar service), I won’t be surprised.

I'm a flesh automaton animated by neurotransmitters and I approve of this message!

Can some things, eg writing, actually ever be "verified"?

Because if it is impossible, sites like this further what we already have. A mix of AI and human, and the inability to - definitively - discern.

Most likely not https://arxiv.org/abs/2303.11156

tl;dr - it's highly unreliable to detect them and because of the risk of false-positives it's not likely a good idea to rely on it.

IMO the only way forward is a reputation based system.

Also moving forward identity verification might have to be cryptographic (IM NOT SAYING CRYPTOCURRENCY) because impersonating text/video/voice convincingly is becoming doable for home users.

Nice idea, but this is self-selecting groups of people that already would not be bad actors. Does this do anything except act as an "open letter" type pledge?

> AI generated content is on the rise. How can you be sure what you're reading is 100% human?

Why is that a concern? When you read a newspaper, how can you be sure what you're reading is 100% true? When you watch the news, how can you be sure what you're watching is 100% true?

What matters is not to have a "verification badge", what matters is to educate yourself to know what to take and what to discard regardless of its source of origin.

This endeavor is most akin to a performance art gesture—and not at a bad one.

Specifically because of the silence around every critical aspect of what this might mean, and why someone might pursue such a pledge, or what verification means, and with what aspiration.

All that is provided AFAICT is a "how"...

...and almost all that results is a wake of perplexity, punctuated by a froth of debate and publicity. Which I take to be the intended product.

Nice to see also "anti-AI" innovation. Something like this would be much needed also for newspapers.

Why? To both statements.

The first was mildly sensed humor. But the second one in my opinion is important for ensuring the integrity in journalism, i.e. a journalist should disclose his or her use of AI. You can count the sport news, financial summaries, etc. out.

I already only read the business section for laughs.

Why does it matter if an AI wrote the article, though? Does the nature of what's being reported change because of who or what is doing the reporting. You can still hold the organization accountable for any infactual or inaccurate statements. Having a human to hold accountable seems unnecessary and onerous.

An organization can not be accountable, unless a human is accountable.

Someone ordered the use of AI and someone gave it the prompts. Those someones have names that can be written on a subpoena or warrant.

We're talking about (real) journalism, so presumably there's going to be a byline. That person is responsible, as is the editor, whether AI is used or not.

Accountability is one thing; as actionfromafar notes below, a legal person may not be accountable in all cases. Transparency is another thing. The third thing relates to ethics; here the guidance is still in the making, but many have indeed stated that a journalist shouldn't put his or her name to a story generated by AI.

https://pressgazette.co.uk/publishers/digital-journalism/ai-...

Not sure what you find innovative in this? There's no actual verification going on. It's a circle jerk link farm with anti-AI virtue signaling.

Reminds me of https://notbyai.fyi

Reply on news.ycombinator.com