There are at least a few kinds of bad:
Spam, google bombing, and related activities. These are noise, generally.
Misinformation is slippery. Often this gets conflated with differences of opinion. That is happening a lot right now as moderation is politicized and weaponized. More than we think is debatable and should be debated rather than legislated or canonized into an orthodoxy, flirting with facism.
Clearly criminal speech, kiddie pr0n, inciting violence, etc. These are not noise and can be linked to either real harm as a matter of the production of the speech (kiddie pr0n), or can be linked to the very likely prospect of harm. Material harm, is an important distinction segway to:
Offensive material.
Being offended is as harmful as we all think it is. Here me out, please:
To a person of deep religious conviction, some speech can offend them just as deeply. They may struggle to differentiate it from criminal speech, and in some parts of the world this is resolved by making the speech criminal anyway. Blasphemy.
That same speech might be laughable to some who are not religious, or who perhaps hold faith of a different order, sect.
Notably, we have yet to get around to the intent of the speaker.
Say the intent was nefarious! That intent would hit the mark sometimes, and other times it would not.
Say the intent was benign. Same outcome!
With me so far?
Before I continue, perhaps it makes sense to match tools up with speech.
For the noise, rule based, AI type systems can help. People can appeal, and the burden here is modest. Could be well distributed with reasonable outcomes more than not. Potentially a lot more.
Misinformation is a very hard problem, and one we need to work more on. People are required. AI, rule based schemes are blunt instruments at best. Total mess right now.
For the criminal speech, people are needed, and the law is invoked, or should be. The burden here is high, and may not be so well distributed, despite the cost paid by those people involved.
Offensive material overlaps with misinformation, in that rule based, and AI systems are only marginally effective, and people are required.
Now, back to why I wrote this:
Barring criminal speech, how the recipient responds is just as important as the moderation system is!
I said we are as offended as we think we are above, and here is what I mean by that:
Say a clown calls you an ass, or says your god is a false god, or the like. Could be pretty offensive stuff, right?
But when we assign a weighting of the words, just how much weight do the words of a clown carry? Not much!
And people have options. One response to the above may be to laugh as what is arguably laughable.
Another may be to ask questions to clarify intent.
Yet another option is to express righteous indignation.
Trolling, along with misinformation share something in common, and that is they tend to work best when many people respond with either righteous indignation (trolling), or passionate affirmation and or concern. (Misinformation)
Notably, how people respond has a major league impact on both the potency and effectiveness of the speech. How we respond also has a similar impact on how much of a problem the speech can be too.
There are feedback loops here that can amplify speech better left with out resonance.
A quick look at trolling can yield insight too:
The cost of trolling is low and the rewards can be super high! A good troll can cast an entire community into grave angst and do so for almost nothing, for example.
However, that same troll may come to regret they ever even thought of trying it in a different community, say one where most of its members are inoculated against trolling. How?
They understand their options. Righteous indignation is the least desirable response because it is easily amplified and is a very high reward for the troll.
Laughing them off the stage can work well.
But there is more!
I did this with a community as it was very effective:
Assign a cost to speakers who cost more than their contributions deliver value! Also, do not silence them. Daylight on the whole process can be enlightening for all involved as well as open the door for all possible options to happen.
People showed up to troll, stayed for the high value conversation and friends they ended up with.
Others left and were reluctant to try again.
The basic mechanism was to require posts conform to one or more rules to be visible. That's it.
Example costs:
No 4 letter words allowed.
Contribution must contain, "I like [something harmless]"
Contribution may not contain the letter "e".
And they have to get it right first time, and edits are evaluated each edit. Any failure renders the contribution hidden.
Both of these did not limit expression. They did impose a cost, sometimes high (no letter "e"), sometimes subtle (no four letter words)...
But what they did do was start a conversation about cost, intent, and