You cannot be sovereign if you do not control your medium of exchange, if your enemies can insert themselves into every transaction and tax it or block it at will.

Cryptocurrencies do not solve this problem. You are simply building thicker walls around your doomed fortress. It might take your enemies more time to break through, but they know exactly where you are, they will find you, and they will hurt you or your loved ones until you give up your private keys, your signal messages, and the balance assigned to you in the ledger of record. The defensive approach to warfare hasn't worked since Napoleon at latest, if it ever did. The only way to win against an opponent with a monopoly on physical violence is to be invisible.

They cannot know where to find you, or even that there is a "you" to find. Instead of having a balance in a ledger, a fact that can be tracked and taxed and discovered in court, perhaps your "reputation" is represented by a 4000-dimensional vector. When we transact - say I help fix something in your house, or lend you my car - our reputations both move around in this vector space in an opaque and somewhat random way. Eventually, the system guarantees, everyone will be made whole, all debts repaid, all good actions rewarded. But at any given point, all bets are off. How much did I pay you for that car rental? How much goodwill do I have accumulated? How influential am I in this social network? Here's a few thousand floating point numbers, good luck figuring it out.

Illegibility is the point. The physics of this medium of exchange must be such that it's cheap to do a binary spot check, like "can I trust this person to rent my house", but intractably costly to figure out anyone's "aggregate" level of wealth or influence.

I don't know how to implement this yet, but a partial sketch might help illustrate the point.

Instead of having some global notion of karma or "account balance", you could have a relation between pairs of network participants trust(A,B) that returns the level of trust or accumulated credit that A has in B (thanks to their history together, but also the history of transactions B has with people who A already trust). This relation can only be computed with A's secret key, but additionally it is /costly/ to compute - like a hard credit check, computing it reduces B's trust in A. For A to even understand /their own/ overall level of wealth or influence, they'd have to evaluate trust() thousands of times, burning their reputation in the process and making the whole process futile. Having a gun to their head makes no difference.

Information flow through this trust network is viscous and heisenbergian; by the time a central node accumulates enough information to intervene, the situation has completely changed.

This is in fact closer to the human default way of doing business - most transactions are informal, illegible favors embedded in a social network to be paid back at some unspecified future time - but uploaded and upgraded to 21st century economic scale. We will wander through our cyberpoleis giving and receiving favors, never worrying about money or taxes or our daily bread, but trusting that our computational karmic god will give us our due.

Does anything like this exist already? Who wants to help me build it?

a key amplifier of centralized power is perfect knowledge of system state, as peasants resisting the census have known for millenia. this proposal tries to heavily tax such concentrations of knowledge

I like the idea of building a solution to this problem, but I don't understand OP's proposed sketch.

What is the purpose of making the reputation vector high-dimensional? Is it just to make the reputation illegible and difficult to compute, like a kind of steganography in vector space?

If so, may I suggest zero-knowledge proofs (ZKPs) as a better approach? ZKPs let you convey the proof of a predicate (e.g., "my reputation is greater than k along particular dimensions") without revealing the data that would ordinarily be required to prove that predicate.

>>1955
I think OP's sketch has to be taken as a vague dream more than a proposal. But it's a good one. Why 4000 dimensions not zk proof? I take it to be coming from the intuition that reputation and credit should actually be fairly high dimensional, context-specific, tied to particular social circumstance. Do these people trust you with this resource in this way in this social context? That's a high dimensional question. The zk proofs would be good, but even if we couldn't figure that out, part of the proposal here is just for the reputation ledger to be complex and opaque enough that it's onerous for bureaucracies to make legal claims or surveillance against it.

>>1955
OP here, I think you're right that something like ZKP is required, but high-dimensionality, social-network-situated reputations are needed in addition to the usual zero-knowledge guarantees. It's critical to illegibility that even you yourself not know fully what your reputation is, which ZKP doesn't give.

Again, I really like this general idea. That's why I'm motivated to press a bit on the technical details. Please take this pressing as entirely friendly.

>>1957
> ... reputation and credit should actually be fairly high dimensional, context-specific, tied to particular social circumstance. Do these people trust you with this resource in this way in this social context? That's a high dimensional question.

Are you sure this couldn't be well-represented with ~5 dimensions? Especially if the vector is computed in relation to a social graph?

> ...part of the proposal here is just for the reputation ledger to be complex and opaque enough that it's onerous for bureaucracies to make legal claims or surveillance against it.

That's what I was describing as "steganography in vector space." I'm just not convinced that this is the best way to achieve the desired protection.

>>1963
> ...It's critical to illegibility that even you yourself not know fully what your reputation is ...

I don't understand why this is critical. It seems to me that it is sufficient for the reputation to be 1) sufficiently well-represented and 2) hidden apart from authorized use.

>>1948
> ... they will hurt you or your loved ones until you give up your private keys ...

OP's goal seems to be immunity from all attacks, including wrench attacks. I claim that this is literally impossible and therefore a bad goal.

To use a reputation representation, some computation against it must yield an answer, and at that point the answer is known to both parties to the transaction. So the import of the reputation becomes known upon use. This is the minimal possible degree of exposure for the system to remain useful. And this surface area can itself become an attack vector.

>>1968

really appreciate the fisking!

Agreed steganography in vector space is likely not the best way to achieve this practically, but felt like the easiest way to illustrate the concept.

> To use a reputation representation, some computation against it must yield an answer, and at that point the answer is known to both parties to the transaction.

Agreed, but the answer should be minimally informative - not "how good is my reputation overall" but "can I trust this specific person with this specific transaction right now, yes or no?". if these binary checks are sufficiently costly, even wrench attacks become impossible (as a way of forming a full picture of the target's reputation and influence); a wrench attack can still burn a person's reputation.

> Are you sure this couldn't be well-represented with ~5 dimensions? Especially if the vector is computed in relation to a social graph?

Yeah the relational formulation trust(A,B) is redundant with the high-dimensional vector space, I was thinking the vector space could be a decent implementation of trust(A,B), e.g. by placing people's vectors roughly in the vicinity of the people they have trust relationships with (or whatever steganographic equivalent)

>>1968
>I don't understand why this is critical. It seems to me that it is sufficient for the reputation to be 1) sufficiently well-represented and 2) hidden apart from authorized use.
There's something profound in having only a vague and unprovable notion of even your own "total" trust situation. It tracks reality in some ways and encourages open-ended positive-sum behavior if done right. On the other hand, the less you know the less you can plan. And I wonder if this isn't a flaw with the whole thing. The nice thing about money is we can know fairly well how much we can afford and can plan fairly definitely on having it if we don't spend or lose it. This abstracted reputation system fails to the extent that it reduces the power of the subjective picture available to the individual actor.

I think what we want is to maximize the power of the individual or medium-sized group's perspective (let's not bake in bourgeois individualist atomization too deeply) relative to bureaucratic monsters, even against wrench attacks. A toolbox approach of illegibility, deniability, encryption, and just plain obscurity of personal-local encoding seems best. What if I could hold some reputation in a highly legible (to me) form, or even a public form for prestige bragging, while holding other reputation in a maximally illegible black budget form?

>Yeah the relational formulation trust(A,B) is redundant with the high-dimensional vector space, I was thinking the vector space could be a decent implementation of trust(A,B), e.g. by placing people's vectors roughly in the vicinity of the people they have trust relationships with (or whatever steganographic equivalent)

Not sure if it is a useful intuition, but the fact that kernel methods makes vector spaces and graphs equivalent seems significant here. If you design the thing to be kernelized, you can abstract out what the underlying representation actually is, which may make it easier to get the desired steganographic/cryptographic or computational properties.

As for where to start on this, I would try to simplify the problem a bunch to where we can learn something. What if we're just trying to characterize how humans already track favors and reputations, and build a slick version of that that doesn't need any steganographic elements? What if we built a centralized ledger proof of concept for the underlying trust computation? How would we demonstrate the "repeated queries burn the underlying reality" technology in any application at all? What would a non-money decentralized trust accounting system look like without the steganographic elements?

There had been a thread on high software profit margins, which has since disappeared off sofiechan. The OP on hackernews might still be relevant (https://news.ycombinator.com/item?id=35904339):

> The popular reason pure software companies have great profit margins is for marginal investment in providing the product/service to more users, given software's replicability. That's the reason everyone likes to repeat.

> The true reason is that they price similarly to physical infrastructure companies, without the same recurring costs. The difference between physical infrastructure costs and total expenses which goes into reproducing the product/service for each new additional user, which software coys do not have deal with, is what makes up "high software margins".

> But pure software products/services shouldn't be priced like their counterparts with recurring costs. They have an abundance quality.

This would be an example of high knowledge of system state amplifying centralized power. The article also considers air and water as public goods with low profit margins, i.e. air is abundant but requires regulation to maintain quality, we've decided that water is essential enough to be a public good so service providers have to prioritize service quality over profit. The infrastructure supporting the medium of exchange is in itself a public good and has to be analyzed as such.

Suppose we wanted to restrict our attention to public goods, who gets credited when we consume the public good? How should this be funded out of the collective treasury? How would this medium of exchange interface with the various components of a public good provider (including the one providing the medium of exchange itself)?