What to do once you admit that decentralization never seems to work (2018) • Hacker News Board

Re: What to do once you admit that decentralization never seems to work (2018)

#51

Post by Jasper_ » Sat, Aug 31, 2019, 3:31 PM UTC

zrm wrote: post #47

humanrebar wrote: post #39

Earlier quoted context omitted.

Fraud is probably a harder problem than discoverability. You can always exchange (maybe through an automated pairing) IPv6 addresses if nothing else. Supporting chargebacks, eliminating spam, blocking phishing, etc. if both harder and more important to a functioning culture.

Fraud and spam are two separate problems. One of the best solutions for spam for a new system is to impose a proof of work requirement on sending the first message between any pair of users, or any (new username, hosting service) pairing for federated services. For any normal pairing that happens only once, but for a spammer they have to pay the cost for every message or else you can trivially block every message fro…

Spammers have large networks of computers they aren't paying for (botnets), so they aren't paying the cost of sending messages. Proof of work really doesn't solve spam anymore.

Re: What to do once you admit that decentralization never seems to work (2018)

#52

Post by superkuh » Sat, Aug 31, 2019, 3:34 PM UTC

oblio wrote: post #18

msla wrote: post #13

Earlier quoted context omitted.

The Web and email are both decentralized. The Internet is decentralized to the extent there's no single "Internet company" which can kick you off. Various companies try very hard to make you forget these things.

Almost no new B2C web company can work without Google.

Companies are not the web.

Re: What to do once you admit that decentralization never seems to work (2018)

#53

Post by mikeash » Sat, Aug 31, 2019, 3:39 PM UTC

nyolfen wrote: post #19

nwah1 wrote: post #15

Decentralization and devolution are often confused. For instance, in the US Civil War, the splintering could be described as devolution. But it certainly wasn't decentralizing power to the slaves. Which side was the side of "decentralization?" It's almost a meaningless question, unless being very well defined. Obviously, the Pareto Principle is usually at work, even if protocols are decentralized by design. Email is…

>For instance, in the US Civil War, the splintering could be described as devolution. But it certainly wasn't decentralizing power to the slaves. it was decentralizing in the sense that secession was meant to grant political authority to local elites. 'devolution' just sounds like a moral judgment, which while fair is not very useful for these purposes.

In what way does it sound like a moral judgement?

Re: What to do once you admit that decentralization never seems to work (2018)

#54

Post by vageli » Sat, Aug 31, 2019, 3:39 PM UTC

sweeneyrod wrote: post #40

nwah1 wrote: post #20

Earlier quoted context omitted.

Most markets exhibit oligopolistic competition, and email is no exception. Google controls the most, yes, but they aren't the only ones to consider. The number of competitors who are significant players are only a handful. The top 5 providers likely account for the overwhelming majority of all email. Email being an open protocol allows interoperability and lowers barriers to entry, and all of that is good, but power…

> The top 5 providers likely account for the overwhelming majority of all email. I don't think this is true. At least, it wasn't true in 2016 according to this ( https://email-verify.my-addr.com/list-of-most-popular-email-... ), these figures ( https://emailclientmarketshare.com ) for email clients in 2019 don't suggest a massive change, and anecdotally it seems unlikely -- a lot of people use addresses from universi…

> > The top 5 providers likely account for the overwhelming majority of all email. I don't think this is true. At least, it wasn't true in 2016 according to this (https://email-verify.my-addr.com/list-of-most-popular-email-...), these figures (https://emailclientmarketshare.com) for email clients in 2019 don't suggest a massive change, and anecdotally it seems unlikely -- a lot of people use addresses from university or work.

> Certainly it is more centralised than say the automotive industry. But it seems much more similar to that than operating systems, browsers, social networks etc..

But those addresses from university or work are increasingly just gsuite deployments.

Re: What to do once you admit that decentralization never seems to work (2018)

#55

Post by o09rdk » Sat, Aug 31, 2019, 3:42 PM UTC

eternalban wrote: post #43

> You lose out on the discoverability that centralized hubs give you. Obviously you can have decentralized services that address this important 'presence' issue. IMO the issue is that we have conflated the 'functional goal' with 'architecture'. The 'goal', for me at least, is public information systems that protect civil rights and resist the rise of tyranny. That goal, after thinking about this space for the past 25…

I was going to say something similar, or at least related.

Centralization processes can occur on a decentralized architecture, but I don't think that means the decentralized architecture has failed. I think people need to distinguish decentralization as a network system state, and decentralization as an affordance. Centralized structures on a decentralized protocol can reorganize themselves, but if you build the system as centralized, it's much harder, if not impossible, barring some implicit decentralized aspect of the architecture.

The question isn't so much "do decentralized systems end up as centralized," it is "can a disrupted decentralized system reorganize itself?"

It's obviously relevant to your point in that the reason for decentralized architectures, as you say, is to maintain communication in the presence of threats, especially to free exchange of information. It's not necessarily to maintain a given network structure state.

I couldn't care less if a handful of servers become dominant on Mastodon or IPFS or whatever, if it's easy to change which handful of servers they are should they start to become compromised in some way.

Re: What to do once you admit that decentralization never seems to work (2018)

#56

Post by nyolfen » Sat, Aug 31, 2019, 3:48 PM UTC

mikeash wrote: post #53

nyolfen wrote: post #19

Earlier quoted context omitted.

>For instance, in the US Civil War, the splintering could be described as devolution. But it certainly wasn't decentralizing power to the slaves. it was decentralizing in the sense that secession was meant to grant political authority to local elites. 'devolution' just sounds like a moral judgment, which while fair is not very useful for these purposes.

In what way does it sound like a moral judgement?

devolution implies a return to a previous, lower level of sophistication. since the US was not previously split into two regional political structures, it's not an accurate descriptor, so it sounds more like an attack on the legitimacy of such an order. critiques of the legitimacy of the confederacy are typically moral in nature.

Re: What to do once you admit that decentralization never seems to work (2018)

#57

Post by WalterSear » Sat, Aug 31, 2019, 3:48 PM UTC

zelly wrote: post #34

Decentralization the way the crypto crowd wants can only happen if everyone has a server at home, some consumer-grade ARM box that you plug in and forget. It should be configured to work out of the box without the customer ever having to use the shell. • your own Nextcloud server to replace Dropbox, GCal • email server (replace Gmail) • torrent seedbox (replace Netflix, Spotify, etc.) • personal Bitcoin node (replace…

On the contrary, most of the crypto crowd wants people to decentralize on their platform.

Re: What to do once you admit that decentralization never seems to work (2018)

#58

Post by nostrademons » Sat, Aug 31, 2019, 3:54 PM UTC

vageli wrote: post #54

sweeneyrod wrote: post #40

Earlier quoted context omitted.

> The top 5 providers likely account for the overwhelming majority of all email. I don't think this is true. At least, it wasn't true in 2016 according to this ( https://email-verify.my-addr.com/list-of-most-popular-email-... ), these figures ( https://emailclientmarketshare.com ) for email clients in 2019 don't suggest a massive change, and anecdotally it seems unlikely -- a lot of people use addresses from universi…

> > The top 5 providers likely account for the overwhelming majority of all email. I don't think this is true. At least, it wasn't true in 2016 according to this ( https://email-verify.my-addr.com/list-of-most-popular-email-... ), these figures ( https://emailclientmarketshare.com ) for email clients in 2019 don't suggest a massive change, and anecdotally it seems unlikely -- a lot of people use addresses from univer…

A lot are still on Outlook. You might be surprised at the degree to which Microsoft still dominates enterprise IT.

Re: What to do once you admit that decentralization never seems to work (2018)

#59

Post by zelly » Sat, Aug 31, 2019, 3:55 PM UTC

humanrebar wrote: post #39

zelly wrote: post #34

Decentralization the way the crypto crowd wants can only happen if everyone has a server at home, some consumer-grade ARM box that you plug in and forget. It should be configured to work out of the box without the customer ever having to use the shell. • your own Nextcloud server to replace Dropbox, GCal • email server (replace Gmail) • torrent seedbox (replace Netflix, Spotify, etc.) • personal Bitcoin node (replace…

Fraud is probably a harder problem than discoverability. You can always exchange (maybe through an automated pairing) IPv6 addresses if nothing else. Supporting chargebacks, eliminating spam, blocking phishing, etc. if both harder and more important to a functioning culture.

Agreed. More broadly, decentralization is less equipped to counteract bad actors. You really are on your own, but isn't that what you asked for when you wanted to abandon Facebook/Google/government.

A decentralized network cannot be trusted by nature. Anyone can join and lie. Successful decentralized networks need some cryptographically-secure checksum to verify its inputs: BitTorrent uses SHA-1; Bitcoin uses Merkle trees. The network participants are just dumb mirrors to provide redundancy and compute. This solves the problem of the integrity of the information, but there is no such analogue for verifying the behavior of humans using the network. For example, you can download a valid BitTorrent file that is also a trojan (virus databases are centralized), or you could send Bitcoin to a Nigerian prince (third-party escrow layers are centralized). There is no algorithm to make sure what you're doing is ethical.

Imagine running reCAPTCHA as a decentralized service that runs on random people's machines. Impossible.

There is no way to flip a switch and change everything from centralized to decentralized without making major sacrifices. No one wants to go back to the dial-up era. Much of the lauded results of the internet are actually a result of its centralization.

Re: What to do once you admit that decentralization never seems to work (2018)

#60

Post by TheOtherHobbes » Sat, Aug 31, 2019, 3:55 PM UTC

zelly wrote: post #34

Decentralization the way the crypto crowd wants can only happen if everyone has a server at home, some consumer-grade ARM box that you plug in and forget. It should be configured to work out of the box without the customer ever having to use the shell. • your own Nextcloud server to replace Dropbox, GCal • email server (replace Gmail) • torrent seedbox (replace Netflix, Spotify, etc.) • personal Bitcoin node (replace…

This is how things should work, but IMO it leads to unsolvable problems.

I don't think discoverability is the issue. You could easily (for hard values of easily) run a distributed and cached search system - maybe with some form of content-aware DNS-a-like metaURLs for specific information.

The underlying issues are tougher.

If you don't have perfectly robust security the system is useless, because it can be abused in almost endless ways.

If you do have perfectly robust security you have perfect cover for a criminal Dark Net.

So you need some form of identity escrow which is robust enough to allow anonymous free speech but permeable enough to allow law-enforcement to take action to prevent crimes like child porn and human trafficking.

And even then, you'll still get people gaming the system in corrupt ways - for example by setting up astroturfing businesses, paying for followers, distributing fake news, and so on.

These are political and economic problems, and I don't think they can be solved with technology.