The law is for Facebook etc. Not tiny sites.

These are expensive AI solutions

Right so no matter how many times and ways I ask you are not gong to do it.

You will keep repeating there are many easy ways they can do something but never once spell it out.

All because you took a position you have no answer for and in forum land you just don't want to go back on it. Nice.

As if you don't know my prior question is equally applicable to SM, as I have said that many times, just tell exactly what advice you would give Sm Mods or anyone. What can they do to stop me from posting stuff that will get them in trouble with the law if I am intent to get around any AI filters?

But this aint going anywhere as clearly you have no answer.

I answered your question. It was a very simple answer but I'm not sure you undersood it. It has no legal impact on mods so that's a complete red herring

I will ignore the silly. Let's try not to that

If this gobbledygook was your answer I have already told you why it would not work.

There is simply no way, IT IS NOT POSSIBLE, to have AI currently pick out what you or I might say in advance of it learning it.

All the machine learning will be done on the back side. What you are referring to is done on the back side and not the front side like letting an AI drive down streets and try to identify obstacles. The way it learns is when you correct the mistakes enough that it then can identify that issue the next time.

If I wanted to write the word ni**er but replace with 'Explosive People' as they used to do on the old MMA forums to avoid the filters it will not know to block that until it learns after the fact.

Then the person switches to other words. It is game of chasing from behind and would not prevent anyone getting past the filters on the front side who wanted to.

Do you understand that or deny it. That a person with intent will easily get around it thus posting things the sight can then get sued and charged for.

It was very basic AI stuff.

There is no doubt that people, will circumvent the system at times (accidentally and on purpose). No-one thinks otherwise. You seem fixated on this point but it's not the be all and end all. Someone like you propose will quickly get flagged in a way that raises the bar for their posts to get through

From the white paper
https://www.gov.uk/government/consul...nment-response

It's no different from the fact that failure to remove content doesn't neccesary lead to fines/etc. Nothing at all changes in that regard because it's about preventing posts.

This is regulation 101 stuff

I am saying it will change nothing meaningful of what is actually posted and serve no function similar to what an Editor does. An editor on his/her game can and will catch anything and everything in most instances in advance of its publication. That simply is not possible (yet) for SM and is no where near ready.

People who want to get around it will and easily so. You can take down what they say after the fact and ban them after the fact but that is what they can do now.

You are simply wrong in the functionality that you think it will get.

You are simply wrong. Going beyond the declarations of simple victory. I'll put the problem into three groups

1) dumb users and happy to comply users.
2) determined to break the rules intelligent users
3) Sophisticated professional fraudsters/etc and a few determined very high ability types


1) is a vast number of posters that will have a huge proportion of offending posts eradicated
2) is a relatively small number of people who will get themselves into trouble and in a losing battle against advancing AI/ improved processes. They will remain a diminishing problem for long time with residure likely to persist
3) is an arms race

It wont be close to perfect both in missing offending posts and incorrectly blocking ok posts. So that would be a win for you in the argument if anyone had ever said it would be.

No i am not and that is why I would bet and you would not. This is pure rhetoric for you.

And I am not saying AI, Machine Learning could not do some of the more straight forward parts of the mod job in the way current word filters can block words and even more than that.

What I am saying is that so much will still get thru that the law will be unmanageable for Big Tech.


Take a newspaper today, get rid of all Editors, and let all users submit and publish their own content but still hold the NewsPaper to the same standard as an editor and they quickly shut down. Their is no AI saviour that will save them from the wave of lawsuits about to hit as large numbers of people do what i just here and do things like write ni**er, instead of ****** to get around the filter, AND DO, until the AI learns and adapts.

Humans are infinitely adaptable and can use completely indirect language where a reader can still clearly know what the writer meant. The readers can see right thru it and understand the slander or lie being told. I can write a post now about you, Chez, being a pedophile and harming kids that uses neither of those words and instead uses 'puppies' for kids and 'plays with' for harm. Words you would be able to sue a publisher for missing and printing the article but an AI filter would not catch.

Agreed. AI moderation is miles off where it needs to be, which we know full well from the horrific stories we hear from overworked and understaffed moderation teams working for big media companies. These people basically swim in the worst dredge humanity has to offer day in and out, and still the efforts often come much too late and leave many victims behind.

And that is for the stuff where the debate isn't even controversial. The kind of content that anyone except die-hard free speech advocates or extremists reject should have a place online. We're talking absolute human horror.

Let us not forget that this is the situation for the various "lingua franca", big languages with enormous spread. For smaller languages, which after all make up vast parts of communication world-wide, the AIs are even more lacking and human moderation often non-existent. The classic example is Myanmar, where Facebook moderation completely failed to a stop a large scale social media campaign and widespread fake news that contributed to the genocide of the Rohingya people.

aI only moderation is way off but AI is already used as part of moderation an it continues to develop. There's still a big role for humans that will increase with the new legislation. The beef with cuepee is the idea that anyhting like every post will require being read by human and the measure of what sucess is. Will it be perfect - not close. Is perfection even a real thing in this context - No..

Much more is being invested in developing such systems precisely because pressure is being put on these companies to moderate better. They can no longer get away with their preferred level of moderation. the diea tehre wont be significant improvement is bizarre, A key difference of this legisdlation is the change of focus to prepost moderation. Thats's a big problem for humans but only different for AI in limited respects.

The language problem is of course very real. UK legislation will focus on English and other languages common in the UK unsuprisingly but it wont stop there.

Did they even try seriously to stop it? The claim is that they amplified it along with much hate speech. Did they use humans to amplify it? Do humans look at every post and decide which ones to push to other users? Some of the improvments are likely to be trivial to implement. And much is language independent - patterns of posts and volume/speed of responses for example.

This is tue as even we here suffered from. One common mistake which sems to be being applied to AI in this conversation is comparing it to some non-existent, impossible standard. The measure of these new laws is nothign to with perfection. Its whether it leads to significant improvements. Or even as part of that process as this is not the end or even the beginning of the end. There is no end.

The reality is always going to very messy, way from perfect and a mix of AI/humans (untol maybe one day we dont need any humans for anything). As it is now but with a slightly different focus and the addition of real corporate pressure on these companies.

That is not the real beef.

The real beef is what SM, or a forum like this, facing that mandate would be able to really do to change what reached their site, before moderation catches it.

If the current standard of after the fact, reactive moderation, which is the bulk of it, is not acceptable in a world with new laws and instead like a Newspaper editor the SM sites are expected to pre approve/edit every posti FIRST, they simply will not be able to do that, and nothing close to that.

The AI is not sufficient to come close to stopping it.

Only human moderation (1 mod for every 3 posters with a time delay) would suffice.

That is our beef. You think AI is 'good enough' that it could work well enough. I am saying it is not. Especially if bad actors then target a site like FB deliberately to get around the filters to purposely get them in trouble with the law.

The reason why we all still get these captcha's to access various articles or websites



is because the Big AI data collection companies are paying sites to require them so humans can identify the objects in the pictures and the AI (big data) can learn from us. it is reactive learning from hindsight analysis STILL.

Having AI capturing human language in a way that could intuit sarcasm, innuendo or out right replacement of words in a pig Latin type way, is far more easy to teach a 6 year old child than an AI currently. Having an AI solve 'driving' and be able to detect a road obstacle is far easier then solving language.

And solving language is one of the holy grails of AI and something that would be necessary to have AI pre read and effectively moderate content to pick out INTENT to prevent publication prior.

it is an older article but still relevant.

the below is the type of thing I heard constant reference to from the AI 'experts'. The issue being the complexity of human language. You can teach AI how to deal with questions asked a very certain way and have a great conversation with it. But if you vary or just allow others to address it conversationally it will be lost and not understand what is being said or meant.

Some recent attempts to get around that with 'robot's' at Trade Shows replying to human speech is to pre program sets of 'canned' opened ended replies to certain key words the AI might pick up but not fully understand. In other words the AI is faking it.

If you can't follow the actual meaning of a conversation and understand what is being said or meant your ability to moderate is small to none. You are then left to key word moderation like this sight uses for words like **** but a site like this struggles with you saying F*ck. Filter beaten. Easy as that. Innuendo... forget it. No chance.

Lwet's not confuse AI assisted modding with AI's ability to understand natural langugae when it comes to be able to make a significant impact. Also wow IQ has jumped from 25.5 to 47.38 and is now approaching a 6 year old. That's damn impressive. Nor far to go for humanity then.

I think you are still hung up on this rather than on the actual issue of making a significnat impact in preventing harmful posts. I'm not going to indulge the silly bet thingy but it can help clarify the discussion to consider how itt would be framed. The question I would ask an agreed on expert is:

"What impact could AI make on preventing harmful content being posted on facebook over the next 5 to 10 years"?

We would then agree on someone like T_D to determine if the answer amounted to 'not significant' when I would lose the bet. You would be nuts to take that bet but I dont think you would agree that question because you're talking about something else.

This question makes any bet and almost the entire discussion almost worthless...

"What impact could AI make on preventing harmful content being posted on facebook over the next 5 to 10 years"?

There is vast money pursuing various areas of AI breakthrough's and this area of being able to engage in 'language' is considered one of, if not the biggest and toughest holy grails to crack.

If you asked 100 top AI professional scientists/theorists you would get 100 differing answers but no one would be certain. There are many who believe this area will never be cracked without AI acquiring some level of near sentient ability, which some AI scientists can evolve but other feels cannot.

It is however irrelevant to the discussion I am having with you here. If you are saying' TODAY WE GOT NOTHING, but i am betting that within 5-10 years the AI tech will evolve to handle this.' than we are talking right past one another.

I have repeated constantly that 'these new laws, if imposed TODAY, will be inoperable for Big Tech or forums like this one. The Tech is just NOT there yet right now' and thus that is the threat that the far right KNOWS they are wielding, that since the tech is not available, then FB and other such platforms would have to shut down. So the threat is 'open up to us or we will wield this threat of making you an editor, to shut you down'.

And then others not on the far right, become unwitting allies of the far right in assuming this can be done right now (and not 5-10 years from now) and make the same call.

There is of course also the question that even if AI can reduce specific types content, to what extent are social media companies interested in letting them do just that.

Big data and social media algorithms operate in near unregulated space, their specific mechanisms pretty much a complete secret and run a business that to the outside observer seems to thrive on controversy and echo chambers.

If a foreign nation-state did to our citizens what social media companies and big data do, we'd fuel the bombers. I'm not saying that as a call to aggression, I'm saying to put what is going on into perspective.

Totally agree

But maybe you've been grunching as the need for regulation has long been my point and the recent flurry of posts are because I posted about the upcoming UK law that includes requring SM (the big ones, not 2+2) to prevent harmful posts being made.

Most of the following argument is I suspect mosrly a mismatch between what we consider it working. I'm looking for a significant reduction as per the bet I outlined bit am not offering. I dont really think that QP disagree with that buy has a much higher and more difficult standard in mind.

Correct.

Our disagreement is over the bolded.

My view that a law today, with real teeth and threat of escalating fines and potential imprisonment of company exec's if any type of Editorial vigor approximate to the standard that Newspaper editors or CNN has, would mean the death of the formats of most SM and chat forums like this.

These forums simply cannot 'prevent' what will be posted in any meaningful way with current tech (AI) and only human power (moderators) could do the job. Even with the law passed today, the only compliance SM would get is the voluntary kind by people who will simply follow new TOS, but everyone intent in getting around, will and anything they write that breaks the law will then become the legal liability for the site.

I would bet that if FB and Twitter did not immediate reinstate Trump they would be deliberately barraged with posts purposefully breaking those laws by his Derps, as quick as they could pull them down, until they just shut the site in futile response.

Ok if you dont think it can and will prevent a significant number then we disagree

I suspect that Just preventing straightforward reposts of material identified as harmful after it was posted will be a very significant reduction in the number of harmful posts. No human will be required to read any of them.

I'm not offering the bet I described but it was maybe useful for clarifcation. Would you really take it if you could?

Edit: to be clear this is not a trap. You can say yes and there's still no bet. there will be no real bet.

Lets be clear.

If 2+2 did not have a technology filter for the word ****, years ago, I am sure countless examples of the word being used could be found.

But as soon as they put in the filter people just switch to f*ck, because we are crafty like that.


So if the law said 'if you allow the word f*ck or the intent of the word f*ck to reach your audience and it is not stopped first, then here are you penalties both in fines and in legal personal liability' oh and btw 'intent matters' meaning even if the person is using 'duck' but everyone can easily see and substitute what is MEANT... and you must eliminate that too'...


Then what i am saying is this forum would be done. There is no way back then or now to prevent people getting around it.


You say 'but look at all the instances of the word **** being removed... that is an improvement and we "prevented significant numbers" and i am saying that is a very wrong way to consider it as it does not speak to a company, in the way the lawyers would want, to protect them from the legal jeopardy they would be exposed to.

What you are speaking to as a 'success' and implementation response to the new legislation would do almost nothing to protect the company from its user base still transgressing the law and thus the companies absorbing liability.

All they can they do is just resort to hindsight adding new words to the filters and banning people (which is what is done now), which gives them zero additional legal protection from the new law.

2+2 or SM would have two options only. Shut down the forums or accept the punishments. They could not stop the acts and "preventing significant numbers" of one action when the people can just switch instantly from '****', to 'F*ck', and then 'duck', and then 'Spaghetti' and everyone reading can see they are using it as code for **** to get around it.

I don't see there being anything to bet on there if I understand your question correctly.


If is like saying 'would you bet on mankind discovering significant evidence of intelligent alien life in the next 50 years?

And thinking anyone could mediate a bet when all they would be doing is offering an opinions in a sea of opinions on each side and you are just coin flipping and hoping you ask the guy that aligns with your side.


What I would bet on, is that if i asked a World leader in the field, such as the UofA IA head, he would say 'we don't have that answer yet. It is a break through billions of dollars have been chasing for some time now and no one can be sure if we will make or not. Some believe we certainly can and will and others believe that without quantum supremacy first being solved and then some level of AI sentience being achieved, language will just be too nebulous to solve'.

Why would I bet on that? Because I have sat in on numerous AI theoretical discussions at the Uni and online and that is the active debate now.

There is a belief that fully autonomous Auto's is a minor question to solve compared to AI solving language.

It's not close to a coinflip

Would you take the hypothetical bet if we asked 6 experts and I needed 4 of them to take my side? How about 5? 6?

I am not sue what you are trying to get at or if you think you caught me in some thing clever.

I am fine in taking a bet but the wording of the bet will be an issue as you are not asking the right question.

I'm not trying to catch you. I'm trying to clarify my position.

I dont expect you to take that bet even hypothetically as I dont think you disagree with my position. Which is that a significant reduction in harmful posts is both possible (and likely if the upcoming law passes) without anything close to human modding of every post.

Hopefully the wording of the 'bet' clarifies this.