Knowledge Fight Interactive Search Tool

All Episodes

Previous Next

Nov. 5, 2022 - Freedomain Radio - Stefan Molyneux

07:22

Social Media Moderation in 7 Minutes!

Large

Audio Only |

Scroll

Time	Text
	Hi everybody, this is Lepan Molyneux from Free Domain.
	So I have put my, I guess, fairly considerable entrepreneurial and technical and software skills as well as my understanding of economics and philosophy and morality together to tell you how social media moderation can work in an objective and effective and consistent way, which I think is what we're all kind of looking for.
	So how would this work?
	Okay, so a couple of basic principles if you're designing a system Which is supposed to enforce moral consistency and universality.
	Understand that people respond to incentives.
	If the right incentives aren't in place, people in general won't respond in a positive and productive way.
	Of course, another principle is that power corrupts.
	That's an inevitable process.
	Whenever people have power over each other, the tendency is for them to misuse that power for various reasons.
	Now, consistency has to be incentivized and crowdsourced.
	And what that means, of course, is because people respond to incentives and power corrupts, you have to find a way to automate, externalize, and crowdsource consistency in your social media moderation rules.
	So here's an example from a food court.
	We've probably all seen this when we're snacky in a mall where the food court restaurant or fast food place says, if we don't give you a receipt with your order, then your order is free, if you tell us straight away that is.
	Now, the reason they do that is if you pay cash to the cashier and the cashier pockets the money, then the restaurant loses money.
	Now, you could, of course, have a system of video cameras watching the cashier and so on, but that's a lot of labor.
	It's expensive. And, of course, the person who reviews the footage could get half the take from the stolen cost of the meal.
	So what you do is you outsource this to the customers.
	The customers are the ones who enforce the Not stealing cash and not ringing up the sale.
	So you incentivize the customer, right?
	So you understand that the power to handle money could corrupt.
	You might pocket it. People respond to incentives.
	So you outsource, you crowdsource the enforcement of ringing up the sale of the meal to the customers.
	And this, of course, works very well.
	It's very cheap, very effective.
	Okay, so our goal is to reduce, can't eliminate, reduce prejudice as much as possible.
	So the first place you need to look to eliminate prejudice is on the part of the moderators.
	Not because they have more prejudice than the general population, but they have the most power, and remember, power corrupts.
	So we really have to make sure that the moderators are not going to be corrupted by power as much as possible.
	So how are we going to eliminate prejudice?
	Well, there's a process.
	We start with a warning to the offensive account.
	And then what you do is you incentivize consistency.
	Remember, people respond to incentives.
	So we want to make sure that if we're applying a rule, that we apply it consistently, but it has to be incentivized.
	If it's not incentivized, you're just speaking into the wind.
	And then what you want to do is you want to abstract the prejudice.
	So if the prejudice is particular to a group, you want to remove that group and say, imagine this was to all groups.
	You could very easily automate that using computer code.
	And then you want to universalize the application of your rule that this is offensive.
	So let's take an example of that.
	Okay. So somebody posts something offensive, right?
	Okay, so they get a seven-day warning, and that warning is not you're banned for seven days.
	It's that your account will be restricted or eliminated in seven days, but there's a process you follow to try and save your account.
	So you incentivize finding similar posts, right?
	Because if you're going to say your post is offensive, then clearly you want to use this to find as many similar posts as possible.
	And you will incentivize that by giving people free advertising credits.
	Maybe it's a dollar of advertising credit for each post you find that is similar or something like that.
	And then what happens is the similar posts get automatically stripped of emotional content.
	So, if somebody says something horrible like, Group X is a cancer on the world, but Group X is a particular group, well, then you take that particular group, you remove the content of that particular group, and you replace it with Group X. And so, if you say, so-and-so, Elbonians, right? It's a made-up country.
	Elbonians are cockroaches, right?
	And then you would say, well, you replace Elbonians with Group X, right?
	And then you send that automatically to third-party moderators.
	And then you say, okay, this statement stripped of its emotional content because you want to make sure you're eliminating prejudice.
	And then you say, is this a violation of the moderation rules?
	Now, if they say yes, then everyone who has said something very similar to the original warning, right?
	Everyone gets the same warning.
	Now, in the long run, either everyone who says Group X is a cancer on the world or something like it, everyone is banned or the original ban is restored.
	Now, if the ban is reversed, then there's a seven-day ban for the original flagger because, again, we want to remember that power corrupts and the power to get people banned It's an awesome power in the modern world.
	So if someone posts something that is offensive and a violation of the moderation rules, you want to use that to incentivize everyone to find similar posts and then either ban everyone, because that's consistency, that's universality, or rescind your original warning.
	The incentives work.
	The fact that power corrupts works.
	Much of this process can be automated.
	And this is the best way to universalize this.
	Now, of course, what it does is it reduces significantly the power of the moderators.
	And that's what you want to do because power corrupts and you want to make sure that your process is as objective and moral and universal as possible.
	And this way, if somebody says, well, I want to flag this guy because he's really upsetting me.
	I don't like him. I don't like what he's saying.
	You don't know whether your friends will be caught up in the same net.
	Maybe your friends have said offensive things.
	Maybe you've said offensive things a while ago.
	So if you want to claim a rule as universal, you have to abstract the content and universalize it and incentivize finding similar posts so that you can clean up the site.
	So now this is a process.
	It's much cheaper than what's going on.
	You can't object to it unless you want to be prejudicial and fight for your side and oppose the other side and so on.
	Much cheaper. And boy, I could make a lot of money if I patented this, but I won't because I'm here for the good of mankind as a whole.
	Way cheaper, way more automated, way more universal.
	There's nothing to complain about unless you want prejudicial power over others.
	This, of course, is how it should work.
	It would make a fortune for people and it would give people a sense of peace and calm and serenity that the rules were being applied as universally, as humanly and automatically possible.

▲