[dkjaudyeqooe]

Reminds me of a toy girl doll I heard about which had a speech generator which you could program to say sentences but had "harmful" words removed, keeping only wholesome ones.

I immediately came up with "Call the football team, I'm wet" and "Daddy lets play hide the sausage" as example workarounds.

It's entirely pointless. Humans are vastly superior in their ability to subvert and corrupt. Even if you were able to catch regular "harmful" images humans would create a new categories of imagery which people would experience as "harmful", employ allusions, illusions, proxies, irony etc. It's endless.

[canjobear]

Furthermore, the possibility that we create an AI that can outsmart humans in terms of filtering inappropriate content is even scarier. Do you really want a world with an AI censor of superhuman intelligence gatekeeping the means of content creation?

[AndrewKemendo]

If you squint and view the modern corporation as a proxy for "an AI censor of superhuman intelligence gatekeeping the means of content creation" - then that's been happening for a long while now.

Automatic review of content, NSFW filters, SPAM filters etc... have been bog standard since the earliest days of the internet.

I don't think anyone likes it. Some fight it and create their own spaces that allow certain types of content. Most people accept it though and move on with their lives

[vintermann]

I'm down with calling a corporation intelligent (as long as you don't call it a person). But automatic content review is regularly bypassed, they can't even keep very obvious spam off YouTube comments, such as comments copied from real users, posted with usernames like ClickMyChannelForXXX.

So if the corporation is an intelligent collective, then it's regularly outsmarted by other intelligent collectives determined to bypass it.

[AndrewKemendo]

Agreed, I think that's a good way to put it. That other collectives kind of "nip" at the larger ones and cause them to implement defenses.

[pverghese]

That's the government forcing corporations to do that

[AndrewKemendo]

Normally I don't respond but this is just false

SPAM filters on SMTP ports were implemented long long before any government mandated it - at the ISP level often

Further, the development DKIM and SPF, were incidental to any government requirement

Preemptively: The fact that the early internet was heavily government and institutionally focused, doesn't a government mandate make

[kelseyfrog]

We can look back further at the Hays code. That's just religion plain and simple. The feeling of, "we're sliding into a decadence which will lead to the downfall of our civilization" is a meme propagating this very sentiment. It's not a simple as just the government, but that does co-occur.

[angusturner]

Developing models that can predict if stuff is harmful ironically makes it easier for people to optimize for harm.

e.g. the one line of code in Stable Diffusion that predicts if stuff is NSFW, can be inverted to generate only NSFW stuff.

I tend to agree with OP that there is no technical solution to this problem.

[6510]

With some further refinement real harm could be done. Think of an infinite short video feed that is both irresistible and gradually modifies you.

[naillo]

Isn't this what infinite jest was entirely about

[tintor]

Solution is an arms race, by keeping your algorithmic improvements hidden.

[whywhywhywhy]

Isn’t that basically what OpenAI and Google tried to do and it lasted all of 3 months.

Problem with tech is once it’s known to be possible if you choose to try and monetize it by making it public as OpenAI and Google were planning to do then it’s only a matter of time before another smart team figure out how you’re doing it.

You can do the Manhattan Project in secret and in 500 years someone else might not realize it’s possible. But the second you do a test of that concept the sign you did that is detectable everywhere and the dots of what you did will connect in someone’s brain somewhere.

Can’t put the genie back in the bottle.

[pfortuny]

In England you discover that English has actually two different existences… The ordinary one and then the “dirty” one. Almost any word has or can be made to have a “harmful” meaning…

Useless effort.

[dumpsterlid]

"It's entirely pointless. Humans are vastly superior in their ability to subvert and corrupt. Even if you were able to catch regular "harmful" images humans would create a new categories of imagery which people would experience as "harmful", employ allusions, illusions, proxies, irony etc. It's endless."

This is employing a fallacy that people have infinite amounts of energy and motivation to devote to being hateful. I have been on countless online communities in video games and elsewhere and when the chat in them doesnt allow you to say toxic, hateful stuff... guess what a whole lot less of that shit is said. Are there people who get around it by changing out characters to ones that look the same that dont trigger the censor or by using slang or by mispelling? Of course but the fact is I think if you talk to someone who runs communities like this they would laugh in your face if you said a degree of censorship of hate speech wasn't fundamentally beneficial.

A big aspect has got to do with the fact that if everybody agrees to be part of a community, part of that agreement is a social contract not to use hate speech and if someone flaunts that they are bypassing it.. in the obvious flaunting of the social contract established (it is obvious they had to purposely mispell the word) these people are alienating themselves by underlining the fact that the 99% of the community finds their behavior pathetic and unacceptable.

[AndrewKemendo]

I (and I would assume the OP) agrees that saying "entirely pointless" may be a bit hyperbolic

However the point stands that as a concept, humans will find a way to exploit and corrupt any technology. This is unquestionably true.

Bertrand Russell famously makes exactly this point as well, albeit specifically when it comes to violent application of technology in war. That: until all war is illegal every technological development will be used for War.

Your point however is also true, in that in certain spaces for certain audiences (communities), participants make it more difficult to exploit these things in ways that they don't want to and to explout them in ways they do.

Ergo, Technology is and remains neutral (as it has no will of it's own) and the people using and implementing technology are very much not neutral and imbue the will of the user onto the tool.

The real question you should be asking is, how powerful can a free tool/knowledge get before people start saying that only certain class of "clerics" can use it or that most communities agree that NO community should have it.

Notice on that last point how not-hard we're trying to get rid of Nuclear Weapons

[Al-Khwarizmi]

I don't think swearing in a video game is comparable to art.

If I swear at a video game and it comes out as ** I might think "OK, maybe I'm being a bit of an asshole, there could be kids here and it's a community with rules so I'll rather not say that".

If a tool to make art doesn't let me generate a nude because some American prude decided that I shouldn't, though... my reaction is going to be to fight the restriction in whatever way I'm able.

[XorNot]

Importantly, we're posting on a forum where this exact idea is true. HN doesn't stop all hate speech, or flaming and what have...but the moderation system stops enough that people generally don't bother.

[angusturner]

It seems pretty well-agreed that the HN moderation works because of dedicated human moderators and community guidelines etc.

I think spaces that effectively moderate AI art content will be successful (or not) based on these same factors.

It won't depend on some brittle technology for predicting if something is harmful or NSFW. (Which, incidentally, people will use to optimize/find NSFW content specifically, as they already do with Stable Diffusion).

[epups]

But this is a forum of interaction between people. These models can and should do things privately. It's the difference between arguing for censorship in HN or Microsoft Word.

[torotonnato]

Another example of funny subvertion, Chinese style: https://www.wikiwand.com/en/Baidu_10_Mythical_Creatures

[AndrewKemendo]

I would argue that a group of humans is actually superhuman at corrupting and subverting any given system

What a great insight which I find both extremely poignant and simultaneously disheartening

Perhaps this will come as a comfort to the people who are vehemently against creating human-level AI systems

[zzleeper]

I naively asked for a "sperm whale opening its mouth in the middle of the ocean" on DALL-E and got a warning :/

[worldsayshi]

Sure it would be a fools errand to filter out "harmful" speech using traditional algorithms. But neutral networks and beyond seems like exactly the kind of technology that is able to respond to fuzzy concepts rather than just sets of words. Sure it will be a long hunt but if it can learn to paint and recognize a myriad of visual concepts it ought to be able to learn what we consider to be harmful.

[ziddoap]

One of the insurmountable problems, I think, is the fact that different people (and different cultures) consider different things 'harmful', and to varying degrees of harm, and what is considered harmful changes over time. What is harmful is also often context-dependent.

Complicating matters more is the fact that something being censored can be considered harmful as well. Religious messages would be a good example of this; Religion A thinks that Religion B is harmful, and vice-versa. I doubt any 'neutral network' can resolve that problem without the decision itself being harmful to some subset of people.

While I love the developments in machine learning/neural networks/etc. right now, I think it's a bit early to put that much faith in them (to the point where we think they can solve such a problem like "ban all the harmful things").

[JohnHaugeland]

This doesn't actually affect the system at all

There's way too much moralizing from people who have no idea what's going on

All the filter actually is is an object recognizer trained on genital images, and it can be turned off

The prompt isn't very relevant. I've had the filter fire on completely innocent text

The filter is a simple checkbox in preferences. All this deep thought is missing the point. You can just turn it off

[ziddoap]

>There's way too much moralizing from people who have no idea what's going on

>All the filter actually is is an object recognizer trained on genital images, and it can be turned off

I'm not sure if you misread something, but neither I or the person I was replying to was talking about this specific implementation, but in a more general sense?

I'm pretty sure you are the one who missed the point of the parent post and mine.

[JohnHaugeland]

Cool story.

Anyway, the filters you're thinking about don't exist and you can download the code and use it today.

Thanks for speculating.

[epups]

It's not that simple. The model was not trained to recognize "harmful" action such as blowjobs (although "bombing" and other atrocities of course are there).

[JohnHaugeland]

It very much is, actually, that simple

The model was trained on eight specific body parts. If it doesn't see those, it doesn't fire. That's 100% of the job.

I see that you've managed to name things that you think aren't in the model. That's nice. That's not related to what this company did, though.

You seem to be confusing how you think a system like this might work with what this company clearly explained as what they did. This isn't hypothetical. You can just go to their webpage and look.

The NSFW filter on Stable Diffusion is simply an image body part recognizer run against the generated image. It has nothing to do with the prompt text at all.

[epups]

The company filtered the LAION 5b based on undisclosed criteria. So what you are saying is actually irrelevant, as we do not know what pictures were included or not.

It is obvious to anyone who bothers to try - have you? - that a filter was placed here at the training level. Rare activities such as "Kitesurfing" produces flawless, accurate pictures, whereas anything sexual or remotely lewd ("peeing") doesn't. This is a conscious decision by whoever produced this model.

[CrazyStat]

>it ought to be able to learn what we consider to be harmful.

Who is "we"? Do "we" consider nude paintings to be harmful?

Is "we" Mike Pence? Roman Polanski? Woody Allen?

There is no coherent "we" and no consensus on what "we" consider harmful, so no AI can possibly learn that.

[worldsayshi]

Well it ought to be able to be trained for a number of scenarios and then on generation be told to generate based on certain cultural sensibilities. It's not going to be perfect but probably good enough?

Isn't this part of the AI alignment problem? To be able to understand what kinds of output is unacceptable for a certain audience? To be polite?

[CrazyStat]

> Well it ought to be able to be trained for a number of scenarios and then on generation be told to generate based on certain cultural sensibilities. It's not going to be perfect but probably good enough?

Do we want the AI to generate based on Polanski's sensibilities, even if he's the only audience member? I suspect for most people the answer is no.

[macrolime]

No

[arvinsim]

Agreed. Anything can be lewd if one has a "green" mind.

[rich_sasha]

Cue aubergine and peach emoji.