Comment by api | Hacker News Reader

Comment by api | Hacker News Reader

I've thought this for many years, but I've never come up with so succinct and perfect a way of expressing it. Anonymity prevents free speech by making it easy to just drown any legitimate speech or valuable information in a torrent of bullshit.

This was always somewhat true due to the prevalence of human trolls and propagandists, but the ability to generate content automatically makes the problem orders of magnitude worse.

In the world of deep fakes and stuff like GPT-3 (which can be used for what amounts to deep fakes for text), any information without clear provenance is worse than worthless. It's a form of pollution whose result is to confuse, misdirect, misinform/disinform, and indoctrinate, and it's impossible to use it for anything legitimate since it's completely unverifiable and could easily be fake.

Right now it's possible to detect deep fakes, but it probably won't be long until the technology is good enough that this is very hard.

Then there's the fact that a lot of people won't even check. People just believe claims in stupid memes and forward them.

Example: a photo of a supposed Denver International Airport mural depicting kids in masks with different flags on them was going around a while back. The claim was that this showed some kind of foreknowledge of COVID-19 going all the way back to the 1990s. Turns out the photo was of a recent painting that was not from DIA. It was totally fake. I've still seen it making the rounds.

I've been predicting that deep fakes will make an appearance this October in the Presidential election. I think there's a really good chance. It'll be at the last minute so nobody has time to perform any analysis or fact check anything.

> In the world of deep fakes and stuff like GPT-3 (which can be used for what amounts to deep fakes for text)

Exactly what is GPT-3 doing in the example? You can already write whatever you want, better than GPT-3 can. You can attribute it however you like. People have been attributing undesirable positions to their political opponents and personal enemies for thousands of years. Where does GPT-3 come into it?

Enabling automated "could be real" spam/noise

Writing propaganda is labor intensive. GPT-3 could do it well enough and almost for free. Think automated troll farms and spam that can be generated in unlimited quantity and that is close enough to real text that it can't be filtered.

Basically it does for spam what the machine gun did for open field warfare.

Writing propaganda is labor intensive if you care what it says. But if you care what it says, you won't be trying to have it generated automatically.

There is a big difference between some labor and zero labor. That's the whole reason spam is a problem.

Meme warfare is not a new idea, but appears growing ever more sophisticated. Implication being, the value of attention and critical thinking is increasing exponentially as people struggle to navigate the information explosion online and to discern what really matters. Curious how people will evolve.

Can't expand the ideea right now too much, but think of Satoshi, anonynous, but consistent and the possibility to prove provenance.

Satoshi is pseudonymous, not anonymous. Public key cryptography (like that used with Bitcoin) allows the former but not the latter.

You're right, but in the context here, there isn't any significant difference. A pseudonymous machine account that is indistinguishable to a human carries the same type of risks to online discourse as an anonymous account.

Code is a very different thing. It's possible to run code and objectively see whether it does what it says or not.

Political speech and whistleblower type stuff is very different. Without a really deep investigation it's impossible to verify anything.

Let's say someone dumps an anonymous video showing Donald Trump or Joe Biden molesting a child. How would you verify that?

For a CS analogy: would you trust anonymous elliptic curves provided with no background or provenance? How can you prove they're not somehow weak in a way known only to the generator? This is why some people are suspicious of the NIST elliptic curves, though in that case they've been around for so long and used to protect so many high value targets they're believed to be probably safe. They've had what amounts to a huge bounty on them for quite a while.

But there already exist authentic videos of Joe Biden sniffing children without their consent. Doesn't seem to harm him any.

I guess we'll see if this is even an issue. Lying by making shit up is not as effective as lying by omission, and the mainstream press are _experts_ at the latter.

Reply on news.ycombinator.com