Hacker News Reader

And not just law abiding speech being deplatformed, but what will AI do with new speech it has never seen before? Will it understand its context, its references, its sarcasm, its wink, its intention? It's another way to trap us into only having conversations that have already been pre-approved by the past.

> In general, he wants the end (no "bad" speech) but doesn't provide a clear picture of the means to get there. This is common.

The MPAA example is a perfect and very clear example. Lots of publishers have gone through the same thing: regulate yourself or be regulated by the people.

As long the Internet is not subject to government packet filtering based upon content (ie, the US doesn't become China), there's no way to prevent the eventual outcome that someone can say bad things and it reach millions of people. If that person breaks the law, the legal system can hold them accountable. But if they are not breaking the law, it seems to me that in the limit such speech cannot somehow be stopped from being spread. It's not a new lesson. It's hundreds of years old, all that is different now is the speed of dissemination and scale of reach. The current centralized platforms are transitional technology. There's some kind of Nash equilibrium where they can exist in a regime where a certain level of legal speech is stifled, but as soon as that threshold is breached and a sufficient sub-network of people conclude their speech needs to go around any centralized provider, the incentives are there to do the work to make it possible -- and internet infrastructure which already exists would arguably be the hardest problem they'd need to solve.

> As long the Internet is not subject to government packet filtering based upon content (ie, the US doesn't become China), there's no way to prevent the eventual outcome that someone can say bad things and it reach millions of people.

Here's the cool thing about trade group regulation like MPAA: we have much less to worry about with respect to slippery slopes. If there's someone who's too small to be a member, then yes: they do get to distribute their content. This means that people can still host individually horrible pages and blogs that few care about. But managing legal liability / responsibility for all those horrible blogs would be a real challenge for wordpress or github, e.g. Writing legislation to thread those kinds of holes but still be upheld in courts might also be difficult. But that's the threat.

If the lion's share of deceptive "news" and bogus claims are spread via Youtube/Twitter/FB, just think of how valuable it would be for those three to come together and have a trade group like the MPAA that would set standards for them to audit content.

> But if they are not breaking the law, it seems to me that in the limit such speech cannot somehow be stopped from being spread.

It can't be stopped, but it can be severely limited. See: 8chan and Stormfront.

I agree that the centralization of internet infrastructure (like Cloudflare, who was involved to the sites you mention) is a corollary to my claim that government packet filtering as a potential regime which could result in limiting global speech.

I agree with pretty much all of your comment. The only thing I don't agree with is this:

> The irony will be that the net effect will be way worse than today: completely unmoderated, unregulated speech with global reach.

I certainly don't see unregulated speech on decentralised platforms with global reach as a bad thing. Indeed, I see it as the goal. And as to moderation (which can be a good thing), that can be applied by a third party of the user's choosing, without requiring regulation or policing i.e. everyone gets to say whatever they like and you can look at it raw or you can choose what subset you see determined by a moderator of your (not the government's, nor the hosting platform's) choice.

I actually agree with you -- my point was more about things being worse than today because, much like today, the future we are headed into will be fallen into unintentionally if we don't recognize now that we should be designing for what we want it to look like and come to a common understanding of what ethics that system should enforce.

For example, it seems likely these decentralized networks will have their designs (such as moderation) iterated upon in response to points of crisis, as opposed to being methodically designed around a collective set of ethics that serve as a foundational bedrock for decision-making. To do so will require "adult decisions" which will have distinct tradeoffs. Without that, the loudest or most powerful voices will end up driving the ultimate long term structure, and we will be left almost certainly with a suboptimal outcome.

You can see this happening today with Mastodon: there's a huge amount of contention around what is the proper ethics with regards to developing software that would allow connectivity agnostic to content, vs stipulating that clients could not be used by certain groups (despite the network itself being inherently decentralized.)

The fact that this kind of question wasn't answered from the beginning before the first line of code was written, but is being answered in response to a blatantly predictable 'crisis' in the form of the arrival of a node like Gab full of Nazis, shows that many participants building these networks have actually failed to come to a meaningful consensus about how they think things ought to work and what the optimum social system is to balance these competing concerns of decentralization and harm minimization. It's trivial to design a decentralized system that works as long as nobody comes along who has counter incentives to the system designers.

It seems like we need to have a wider conversation, along with education, about what the various end states of this can be, recognize that the future will not necessarily represent the present, and then design incentives and legal structures to guide towards the ones we want to actually see happen.

Currently, we are being almost entirely reactionary and have an incredible lack of imagination (and in turn, optimism.)

> he feels that these companies should be trusted to regulate speech to a degree beyond what they already are doing

My extended interpretation: he feels that, if we are going to tolerate them, we should at least require them to dampen the worst of the ill effects they have on society. If, in doing so, they lose presence and that leads them to wither and die, too bad^W^Wthat's ok.

I've had an account on HN for more than 10 years. I'm sure we used to have the convention that you should not down vote comments just because you disagree with them. I don't see it in the FAQs anymore but I sorely wish we could go back to those days.

This is not the case. See dang's note https://news.ycombinator.com/item?id=16131314

The first link was posted by PG in 2008.

Characterizing the state limiting the actions a corporation may take, by regulation, as the state “trusting” the corporation is absurd.

Acting like we are talking about nebulously “bad” speech when the primary subject is 1) denial of the factual event of the Holocaust and 2) incitement towards a future Holocaust is deeply insensitive to the present and growing fear of that exact thing.

Pretending that any white supremacist or anti-semitic propaganda which comes from a politician must simply be “algorithmically similar to hate speech”, and could not conceivably be hate speech per se, is the pinnacle of politically motivated reasoning.

And somehow we put all of that together to arrive at the dramatic conclusion that... the status quo is fine with me, and nothing should change. It’s a nice speech, mind you.

I can’t possibly replace HN with another platform— where else am I going to find such passionately incoherent arguments from the radical middlebrow?

Algorithmic regulation of speech necessarily requires arguments around global effects, counterfactuals, and long term consequences. Focusing on a specific, extreme example to inductively derive a plan around making platform-wide, global scale, centralized enforcement of speech is something I expect of a child, not an adult. Anyone who has worked on machine learning understands this: if you are going to ignore false positives, it's trivial to make an algorithm with 100% accuracy. The challenge becomes what confusion matrix is acceptable. It's fundamentally impossible to avoid all Type I and Type II errors for anything but the most basic underlying function: I don't think anyone here suspects a system for labelling unacceptable speech is even close to such a problem.

In our justice system, we highly prefer to not send the innocent to jail. A similar level of tolerance should be articulated here. It is not being done, and the issue of false positives is largely glossed over by stirring up emotions around extreme examples.

This is similar to seeing people getting angry over when a suspected guilty person walks free in an individual case, winning due to proper legal reasons like insufficient evidence to prove guilt. We accept that outcome, not because we want the guilty to be free, but because the system we have designed has made tradeoffs where that outcome can happen in favor of minimizing the liklihood of the innocent from being imprisoned.

People should be forced to articulate what degree of non-hate speech they are willing to see silenced in favor of increasing the silencing of hate speech, and those who are happy with the status quo or who are interested in reduced regulation of speech should articulate the degree of acceptable hate speech that is let to continue so non-hate speech will not be inadvertently silenced.

> and could not conceivably be hate speech per se, is the pinnacle of politically motivated reasoning.

Nowhere did I state that political speech could not conceivably be hate speech. The inverse proposition, that political speech by politicians that isn't hate speech could be mis-labeled, was my claim. This isn't even a claim, it's self evident if we're talking about having a system that is labeling all instances of online speech as being either hate speech or not. Type I and Type II errors are unavoidable. Arguing the inverse shows you did not understand my logic.

My claim is that people who shrug their shoulders at the impact of false positives (like SBC) either underestimate the magnitude of them or are revealing fundamental beliefs about what world they think is acceptable to live in. (Namely, that these platforms should be willing to silence more people today than they are doing already, even if they are innocent, in the name of preventing others who are bad actors.) I'm sure there have been millions of people who have strongly argued in favor of a justice system where you must prove your innocence. A similar tack is being taken when you have people who wouldn't mind seeing people de-platformed who actually were not spreading hate, if it meant less dissemination of actual hate speech.

I don't blame celebrities for not understanding the nature of Type I and Type II errors, and the nature of unintended consequences, but I sure as heck expect more from commenters here.

And anyway, my point wasn't about what I think ought to be, what but I think will happen inevitably. Due to the Internet, all of these arguments are moot. Unless the government starts packet filtering, there will ultimately be no way to prevent non-illegal speech from spreading rapidly across the globe. I'd rather people focus on what we want the world to look like under a system where any voice can be heard by anyone, and how we can optimize that scenario, instead of focusing on the present-day system where a few centralized actors happen to have these artificial hammers they can use to clamp down the spread of speech. Leveraging those hammers for good or evil will not result in a good, long-term, sustainable outcome, since they are going to eventually evaporate and we will be back exactly where we started, if not in a worse scenario due to solving for the wrong problem.