Read on Twitter
The problem with free speech is that truth doesn't win in the short term (see: QAnon)
The problem with restricted speech is that truth can't even win in the long-term if whoever has censorship power decides to suppress it - and in general everyone wants to suppress something. https://twitter.com/emilybazelon/status/1316438330901442561
The problem with restricted speech is that truth can't even win in the long-term if whoever has censorship power decides to suppress it - and in general everyone wants to suppress something. https://twitter.com/emilybazelon/status/1316438330901442561
The best way to restrict speech is to place limitations on speaking and sharing that are as close to entirely procedural as possible, with no entity able to exert causal influence on what gets said.
An automated classifier might be the best approach, perhaps with some cryptoeconomic aspects.
For example, you might be allowed to say anything you like, and then some classifier rates it to see whether it seems to be engaging in certain kinds of speech that are undesirable.
For example, you might be allowed to say anything you like, and then some classifier rates it to see whether it seems to be engaging in certain kinds of speech that are undesirable.
You might post a tweet about some false or crazy theory, pay a small fee to submit it to the network, and get refunded if the classifier marks it as OK. A stake for speech.
If the classifier flags it, some other algorithm might assess how bad it is, and then charge you for it.
If the classifier flags it, some other algorithm might assess how bad it is, and then charge you for it.
You can pay to post it anyway (or maybe not even pay - just do something marginally inconvenient like a CAPTCHA), and perhaps it gets some penalty on its virality.
The biggest difficulty with this right now is how you get a classifier to work in a transparent way that doesn't also allow people to "hack" the classification function.
Also, even contemporary AI struggles to understand reasoning, so it wouldn't be a classifier for truth.
Also, even contemporary AI struggles to understand reasoning, so it wouldn't be a classifier for truth.
You could have a classifier for emotion and you could perhaps automatically identify trends, and that might be useful enough on its own.
Another way to look at it is a "classifier for truth" will never be useful. If you have one, speech itself becomes unnecessary!
Another way to look at it is a "classifier for truth" will never be useful. If you have one, speech itself becomes unnecessary!
Why? Because if you have a machine that can classify truth and falsity, you can just get the machine to run civilization - you have a superintelligent AI.
So the question is what we can do with existing weak/narrow AI methods and economic/behavioral incentives.
So the question is what we can do with existing weak/narrow AI methods and economic/behavioral incentives.
