YouTube cracks down on synthetic media with AI disclosure requirement | Hacker News Reader

YouTube cracks down on synthetic media with AI disclosure requirement

arstechnica.com

53 pointsisaacfrond70 comments

I think some people on HN are worried about stifling creativity. The vast, VAST majority of the AI-driven content is going to be gamed product reviews / advertisements, as well as some political propaganda. It’s going to be a hellhole for usefulness.

I think there are more categories than just that--e.g., kids videos.

Yeah but that stuff was mostly pointless garbage before the advent of AI. Content farms, apps for endlessly generating variations of the same content, low wage and Mechanical Turk jobs etc etc.

Maybe AI will speed up the process of making this stuff either ignored or unacceptable. Things gotta get worse before they get better...

“The Top 10 X You NEED in CURRENT_YEAR” has been a ubiquitous article title when searching for reviews on literally anything for years. It will only get worse for sure.

I've seen an AI generated "top 10 list Chrome extensions for $X" submitted to HN. Those didn't exist, I searched, those extensions were all made up. (The user is now blocked.)

That’s true, but I think there are few points of concern here.

Currently, I can detect content farm garbage. Will I be able to when AI starts churning out videos?

Volume is also a concern. If 10x more videos are AI trash, it functionally just becomes harder to find what you want.

WRT propaganda, I think the past few years have proven that people are just absolutely terrible at resisting political manipulation. I don’t support censorship as a solution to this issue, but at the same time it seems impossible to imagine AI doing anything other than making this problem significantly worse.

I also wonder how much worse AI will make the filter bubble phenomenon. I don’t just mean with regard to political extremism, but also when it comes to basic alienation from your peers. Too much homogenous culture wasn’t a good thing, but infinitely personalized culture is pretty alienating as well.

> Will I be able to when AI starts churning out videos?

It is already full of it! If you are into bananas you type "bananas" in the box and a video comes out that is disturbingly watchable (but easy to spot with its "open eye" kind of typos and dull stockphotos)

https://invideo.io/ai/

AI's will capture viewtime so effectively that human creators channels will wither.

Most popular youtube content is already garbage in my opinion but many small kids are already happy viewing bottom tier AI gibberish for hours.

And if endless AI content will soon be good enough to hit the dopamine centers in adults, only the connoisseurs will seek out the human made content making the markets impossibly small for artisans creating video content.

It's already a tragedy that socialmedia has made it so 95% of all content consumed on the internet is made in the last 24 hours lowering the value of making an effort.

Everyone is in desperate need of quality, boundaries and strict media diets including myself.

So , like the "real" content . I think most people don't trust random videos, that's why they follow influencers who they consider reliable.

I think overall the distribution of quality of content will remain the same, just a lot more of it at all levels.

I predict a return to private communities on the Internet and bulletin boards where verified humans hang out, away from the rest of the garbage the Internet will become filled with.

Unlikely. Most people care about network effect more than they care about authenticity. Small enclaves of hobbyists like HN will always exist but they aren't and won't be the mainstream.

The small enclaves are gonna become a little less small, the plebs are gonna keep plebbing (now with AI)

I am an optimist

How about small select enclaves of people using generative AI? Or are they plebs by your definition?

I expect the large social media platforms to emphasize verification that accounts are controlled by real humans even if they are pseudonymous. X (formerly Twitter) is already heading in this direction. AI social media accounts and even accounts where they're unsure if the account is a genuine human or AI are likely to end up shadowbanned.

I suspect newer platforms and small enclaves that want to be larger will be tempted to use AI. Reddit's founders infamously created fake accounts and put fake posts on their platform when it started to make it look active.[0] That was actually a pretty common technique to address the "chicken and egg" dilemma that every new forum has: Nobody wants to join a forum that has no posts and no active users but the forum won't get posts or active users unless somebody joins. AI can solve this problem much more efficiently than if the founders are writing the fake posts.

[0]: https://arstechnica.com/information-technology/2012/06/reddi...

> I expect the large social media platforms to emphasize verification that accounts are controlled by real humans even if they are pseudonymous.

Of course they are controlled by a real human. The question is if that human is posting or an AI controlled by the human?

Most people on social media lurk and don't post. They're more or less being chased off the platforms these days due to the bombardment of ads, influencer marketing and UI Tiktokification.

> Unlikely. Most people care about network effect more than they care about authenticity.

I think that's excessively cynical take. "People" might care about network effects more than a lot of other things technology people thing they should care about, but I don't think that will extend to socializing with networks of bots.

A lot of tech people tend to focus too much on surface appearances and neglect social context. A connection with a bot is not equivalent to a connection with a real person, even when a tech person can't distinguish the posts.

Been doing this via Discord for years and it’s great. I’ve got two hobby servers with maybe 500 people each. A local hobby server of 30 people and a friends server of maybe 15 people.

Verified? Meh. Not necessary. I can’t think of how that would improve my experience.

Well, if you suspect the top poster on your forum is a bot, you might change your mind...

I think that’s more an anxious fear than a real problem. Especially if you seek meaningful interactions and relationship building.

And if an AI is able to offer such a compelling long-term character development and online relationship and regularly offers insights and new ideas and empathy, we’re already into a deeply philosophical place about authenticity and if it really matters.

You have no idea if I’m an AI right now. If I am, would this comment suddenly become invalid? The information and idea here is what matters, no? Do you need to know much about the corporeal flesh bag mashing the buttons?

But does it even matter? Everyone online is a cat.

> The information and idea here is what matters, no?

Yuval Noah Harari goes into this in detail that this is actually a huge risk. Imagine an AI controlled by a single with a hidden motivation. But also imagine it is a really good AI that 99% of its communication is towards building a friendship long term relationships with particular users on hundreds or thousands of forums.

You now have an intimate relationship and when political season comes around and they make a recommendation you trust what that 'person' says because they don't generally talk about politics. And so do thousands of other people.

I was thinking about that. Isn’t that already a real problem? Cold War spies and the modern equivalent? But I guess this drives the cost of that down a million fold. Fascinating.

DHT based vBulletin let's go!

I am expecting stamps on things that say "human made", "designed by humans". Handed out by a nonprofit called something cheesy like "The SOUL Foundation".

Soul foundation is a great idea. Also human factory and such.

We already have shop small (business) thanks to AMEX and numerous others.

Heh, and a black market that stamps them on AI products for profit too?

Stuff being designed or made by meat puppets doesn't imply any quality.

It may not guarantee quality, but to turn the argument around, everything good/brilliant/interesting made so far was made by humans (barring naturally-ocurring coolness).

Isn't that just Facebook? Facebook is doing just fine.

Probably looks a bit more like Google+ ;-)

One can only hope. I miss the old internet before a bunch of stupid children were on it

Hopefully there's a search filter that comes with this disclosure requirement.

I'm all for people expressing creativity however they go about it, but I still have a preference for actual human content.

With you on this one!

I've noticed a lot of AI narrated videos lately

Honestly that doesn’t bother me very much, so long as the script is human-written. YouTube has always had videos with TTS narration, and I’d rather be listening to that than an incomprehensible accent or $0.50 microphone.

There’s one clever channel that makes content about the Deus Ex games that uses voice synths trained on the games’s characters - which, in addition to making the narration a bit more interesting is very on-brand for a cyberpunk game.

Here’s a video. You can see that the translation from “real” JC Denton dialogue in the first few seconds to fake AI voice is surprisingly smooth (helped by the admittedly weird voice acting in the game), although there are always a few spots in the video where it breaks down (which the author usually leaves in for laughs): https://youtu.be/jDYVx3nqgxw

I was prepared to criticise this video - but upon inspection, it really captures the low quality[0] of the voice acting in an authentic manner!

0. https://www.youtube.com/watch?v=wGE2ZB_3ZO0

Do you actually want a human-written script, or do you just want a good script, and use 'human-written' as a proxy for 'good'? In a hypothetical future where most AI generated content was qualitatively better than most human content, would you feel the same?

Oh for sure, my problem is primarily one of misinformation. If we managed to solve the hallucination problem such that AI content was more reliable than a reasonably smart human I wouldn’t have a problem with it. But for now AI has a bad habit of telling me that “mayonnaise” has four letter Ns or that you can melt eggs.

I really hate when the top result for something on YouTube is an obviously AI narrated video. Even weirder when I realize it halfway through because it makes a mistake no human narrator would. I feel like someone cheated me out of my time. If you can’t spare the time to actually narrate something, why should I listen to it?

Maybe the author is a good writer with interesting things to say, but has a bad speaking voice?

Write a book then? Blog?

If I can't throw a baseball, then a career in baseball isn't for me. Trying to be a baseball player with a pitching machine to do the work for me isn't the way.

Or someone like Stephen Hawking who has great ideas but no other option for speaking.

> Maybe the author is a good writer with interesting things to say, but has a bad speaking voice?

Possibly, but it's more likely the author was trying to make a quick buck by spamming low-quality content. Especially since the GP left the experience feeling cheated.

AI narration also could probably be used to launder stolen content to make a buck off it. Did you find a good writer with interesting things to say? Steal his work and run it through AI narration and collect the clicks.

The David Attenboro 40k lore channels were hillarious

A bunch of "review" videos are just a referral link to aliexpress, a slideshow of aliexpress photos, some AI thing that takes the item description and condenses the data to a few sentences and text to speech that reads the text...

Although, even with actual humans reviewing stuff, there is a very very high chance that the review is "influenced" (=paid by) the seller/manufacturer of the item.... there are only a few youtubers left, whose reviws I'd actually trust...

There is a lot of synthetic content about academic subjects on YT now, and it's very low quality. I used to search for lectures to listen to while walking or driving but now need to wade through tons of enshittified spam. Even if it's reading wikipedia or other long form articles, the voices and graphics are bad.

Actually I paid for Blinkist recently and really enjoyed it at first. They have a lot of "blinks" that state at the end that the voice was synthetic and I was legitimately surprised at the quality, having not even noticed until they told me.

This seems like a good move for YT to maintain a basic level of quality (which I'm amazed can actually get worse), but I suspect it's a pretext to avoid paying out to "illegitimate creators" for commercial reasons in a way that makes them look like they care about people.

They should label misleading or fake content, not all "Synthetic content", or else that's discriminatory. There are lots of videos narrated by synthetic AI voices. Lots of users don't speak english well or don't have a good accent, nothing wrong with that.

Some people just want their Warhammer 40k lore deep-dives narrated by David Attenborough, and I don't see why that's bad.

Why would it be bad to have videos narrated by synthetic AI voices carry a label declaring that the voice is synthetic? This isn't a censorship proposal.

I'm wondering if it'd be specific like that or if it'd just say "This video has been labeled to contain synthetically generated content" and not specify. Meaning the spectrum from "Tr|_|mp sticks a turkey where the sun don't shine" and "Non-english speaker uses AI voice to narrate" would be labeled the same.

I hope it's what you're talking about but I've seen enough actions by YouTube to feel like it's probably the latter.

FWIW, both of those examples are the kind of thing I think we'd all agree don't need to be labeled. No one is going to be fooled by those two videos. The worry is about misleading videos, and having a mechanism to tag them as such that doesn't require content providers to get into the business of deciding on what is true.

Basically: "This video is synthetic and does not depict real babies being tortured in Gaza" is the use case in question.

I'm 100% with you and all for those kinds of videos being labeled. I'm pretty nervous about the 2024 election with all of the capabilities we've got now.

That said, I feel like the norm anymore is for companies to go way overboard and that was where I was going with my comment. The usage you're talking about makes perfect sense I just expect they'll require way more.

Like look at how they've handled children. YouTube Kids already exists yet, for some reason, youtubers risk demonetization for showing blood or saying minor curse words to the point where it feels like we can't actually be adults on a platform that already has a counterpart specifically for children. Just not "advertiser friendly" but who's to say that advertisers won't decide that "An AI narrated informational video" is less advertiser friendly than a human narrated one?

I challenge you to define “misleading”

Synthetic is at least relatively straight forward to categorize; it’s not based on the content but on the production process & output.

I do see your concern about using synthesis to assist with language barriers, but we should be able to distinguish between an actual human video with synthetic narration and the floods of terrible synthetic voice over non-human slide shows & “borrowed” content.

you are correct but the article is talking about "crack down" of labeled videos

Second sentence of the article: "We’ll require creators to disclose when they've created altered or synthetic content that is realistic [emphasis added]."

From the second paragraph: "For example, this could be an AI-generated video that realistically depicts an event that never happened, or content showing someone saying or doing something they didn't actually do."

They are not requiring labeling of all synthetic content.

this is the kind of thing that algorithms are really good at (determining provenance/AI traces, at scale).

search has been trying to fingerprint users forever, its time to point their agorithmic guns at something far more useful for the actual end user.

it makes sense for search algorithms to spot these and then flag them accordingly if they have not been disclosed.

Algorithms are NOT good at detecting AI, and every attempt so far has been laughably terrible. For example the internet is already filling up with students complaining about teachers failing their school work for false positives on AI text generation.

And there's the reality of the Survivorship Bias when it comes to examples: the only times you see AI art are the times it didn't fool you, not the full set of all AI images you saw. Similarly, if the algorithm can find the easy ones but miss the same better ones, is it really that useful? Or is it just training more realistic AI to evade both you and the algorithms detection?

The idea that people will fundamentally be able to differentiate AI and human works is nonsensical from a future perspective and is an aberration of AI quality for the next few years at most. If you're not preparing for fully indistinguishable AI text/image/video/audio, then you're not preparing for the future.

Realistically I think the only way to create verifiably-human-made digital content is basically a end to end chain of trusted computing DRM from keyboard or camera lens to final product, which seems dystopian in its own way, impractical, and a high value target for people to try to bypass.

Authoritarians: "Go on, I would like to know more"

With an automatic system, what will actually happen is even more algorithmic bureaucracy, and real people being banned for false positives or for using any ML-assisted techniques at all. These algorithms are snake oil, there's simply no real way to detect it reliably. Youtube is enough of a dumpster fire already and its automated systems are being exploited left and right.

Don't make any mistakes, this requirement is meant to cover Youtube's ass, not to do anything useful to you. How it will be enforced is entirely up to them. I suspect it will be much closer to "occasionally banning someone to appease the outraged crowd" rather than to "algorithmic banhammer".

Are we going to see this on every Hollywood blockbuster trailer?

A sensitive choice - procedurally generated content should be disclosed, not only by youtube but also twitter, reddit, and most important news websites.

[dupe]

Lots of discussion yesterday:

https://news.ycombinator.com/item?id=38269656

Well, that will sure work. About as well as the Craiglist ad poster who says, "No scammers."

[dupe]

Why would "AI" be treated in a special way among other tools in a CG artists toolbox?

Should I label a video where "AI" applied denoising or color grading?

I could hire an actor to professionally fake a voice of a celebrity vs I could generate a voice with "AI".

What even is a definition of "AI" vs a "simple" ML or genetic model or sufficiently advanced algorithm?

Seems like another reason for arbitrary content removal because video is suspected/highly likely to be made with "AI".

All of the examples you’ve given prevent you from trivially scaling content creation to flood the world with nonsense

I have no idea what the intention behind this is, but I suspect it might be more to do with preventing extremely low quality content that doesn’t have a human in the loop?

Meat puppets have been flooding the internet for a long time. The intent is to distinguish fake AI generated videos of things.

Youtube is already full of low quality nonsense made en masse without whatever they call AI.

Another point, AI-made content is (or soon will be) indistinguishable from authentic footage, so any effort to label it is quite quixotic and will likely result in many false positives.

Not to mention a whole bunch of stuff labeled as AI and then the really bad stuff non-labeled because they've worked hard enough to make it indistinguishable from authentic.

Reply on news.ycombinator.com