Skip to content

Latest commit

 

History

History
388 lines (240 loc) · 74.8 KB

Problems.md

File metadata and controls

388 lines (240 loc) · 74.8 KB
name category
Problems

The science of cryptography, which has existed to some degree for millennia but in a formal and systematized form for less than fifty years, can be most simply defined as the study of communication in an adversarial environment. In a similar vein, we can define cryptoeconomics as a field that goes one step further: the study of economic interaction in an adversarial environment. To distinguish itself from traditional economics, which certainly studies both economic interaction and adversaries, cryptoeconomics generally focuses on interactions that take place over network protocols. Particular domains of cryptoeconomics include:

  • Online trust and reputation systems
  • Cryptographic tokens / cryptocurrencies, and more generally digital assets
  • Self-executing "smart" contracts
  • Consensus algorithms
  • Anti-spam and anti-sybil attack algorithms
  • Incentivized marketplaces for computational resources
  • Decentralized systems for social welfare / mutual aid / basic income
  • Decentralized governance (for both for-profit and non-profit entities)

The increasing prominence of cryptoeconomics in the last five years is to a large extent the result of the growth of cryptocurrencies and digital tokens, and brings a new, and interesting, dimension to cryptography. While before cryptography was, by and large, a purely computational and information-theoretic science, with strong guarantees built on security assumptions that are close to absolute, once money enters the picture the perfect world of mathematics must interact with a much more messy reality of human social structures, economic incentives, partial guarantees and known vulnerabilities that can only be mitigated, and not outright removed. While a cryptographer is used to assumptions of the form "this algorithm is guaranteed to be unbreakable provided that these underlying math problems remain hard", the world of cryptoeconomics must contend with fuzzy empirical factors such as the difficulty of collusion attacks, the relative quantity of altruistic, profit-seeking and anti-altruistic parties, the level of concentration of different kinds of resources, and in some cases even sociocultural circumstances.

In traditional applied cryptography, security assumptions tend to look something like this:

  1. No one can do more than 279 computational steps
  2. Factoring is hard (ie. superpolynomial)
  3. Taking nth roots modulo composites is hard
  4. The elliptic curve discrete logarithm problem cannot be solved faster than in 2n/2 time

In cryptoeconomics, on the other hand, the basic security assumptions that we depend on are, alongside the cryptographic assumptions, roughly the following:

  1. No set of individuals that control more than 25% of all computational resources is capable of colluding
  2. No set of individuals that control more than 25% of all money is capable of colluding
  3. The amount of computation of a certain proof of work function that can be accomplished with a given amount of money is not superlinear beyond a point which is reasonably low
  4. There exist a non-negligible number of altruists and a non-negligible number of crazies or political opponents of the system, and the majority of users can be reasonably modeled as being close to economically rational
  5. The number of users of a system is large, and users can appear or disappear at any time, although at least some users are persistent
  6. Censorship is impossible, and any two nodes can send messages to each other relatively quickly.
  7. It is trivial to generate a very large number of IP addresses, and one can purchase an unlimited amount of network bandwidth
  8. Many users are anonymous, so negative reputations and debts are close to unenforceable

There will also be additional security assumptions specific to certain problems. Thus, quite often it will not even be possible to definitively say that a certain protocol is secure or insecure or that a certain problem has been solved. Rather, it will be necessary to create solutions that are optimized for particular empirical and social realities, and continue further and further optimizing them over time.

Technology

The decentralized consensus technology used in Bitcoin is impressive to a very large extent because of its simplicity. A 30-year-old problem in computer science was solved via a mechanism which is simple to implement, and so simple to understand that even some semi-technical teenagers can describe the entirety of how it works. However, at the same time the technology in its current form is very limited. The scalability in Bitcoin is very crude; the fact that every full node needs to process every transaction is a large roadblock to the future success of the platform, and a factor preventing its effective use in micropayments (arguably the one place where it is the most useful). Timestamping is flawed, and proof-of-computation algorithms are very limited in the types of computation that they can support. The fact that the original solution was so "easy", however, suggests that there is still a large opportunity to improve, and there are a number of directions in which improvement could be directed.

1. Blockchain Scalability

One of the largest problems facing the cryptocurrency space today is the issue of scalability. It is an often repeated claim that, while mainstream payment networks process something like 2000 transactions per second, in its current form the Bitcoin network can only process seven. On a fundamental level, this is not strictly true; simply by changing the block size limit parameter, Bitcoin can easily be made to support 70 or even 7000 transactions per second. However, if Bitcoin does get to that scale, we run into a problem: it becomes impossible for the average user to run a full node, and full nodes become relegated only to that small collection of businesses that can afford the resources. Because mining only requires the block header, even miners can (and in practice most do) mine without downloading the blockchain.

The main concern with this is trust: if there are only a few entities capable of running full nodes, then those entities can conspire and agree to give themselves a large number of additional bitcoins, and there would be no way for other users to see for themselves that a block is invalid without processing an entire block themselves. Although such a fraud may potentially be discovered after the fact, power dynamics may create a situation where the default action is to simply go along with the fraudulent chain (and authorities can create a climate of fear to support such an action) and there is a coordination problem in switching back. Thus, at the extreme, Bitcoin with 7000 transactions per second has security properties that are essentially similar to a centralized system like Paypal, whereas what we want is a system that handles 7000 TPS with the same levels of decentralization that cryptocurrency originally promised to offer.

Ideally, a blockchain design should exist that works, and has similar security properties to Bitcoin with regard to 51% attacks, that functions even if no single node processes more than 1/n of all transactions where n can be scaled up to be as high as necessary, although perhaps at the cost of linearly or quadratically growing secondary inefficiencies and convergence concerns. This would allow the blockchain architecture to process an arbitrarily high number of TPS but at the same time retain the same level of decentralization that Satoshi envisioned.

Problem: create a blockchain design that maintains Bitcoin-like security guarantees, but where the maximum size of the most powerful node that needs to exist for the network to keep functioning is substantially sublinear in the number of transactions.

Additional Assumptions and Requirements:

  • There exist a large number of miners in the network
  • Miners may be using specialized hardware or unspecialized hardware. Specialized hardware should be assumed to be more powerful than unspecialized hardware by a large (eg. 10000) constant factor at specific tasks.
  • Ordinary users will be using unspecialized hardware
  • Ideally, after some number of blocks (perhaps logarithmic in the total size of the network) every transaction should require 51% of network hashpower to reverse. However, solutions where transactions can pay very small fees for a lower "level" of security are acceptable, though one should take care to avoid situations where an attacker can profit by performing one attack to reverse very many small transactions at the same time
  • Ideally, the solution should work for and maintain as many properties as possible of a generalized account-based blockchain (eg. Ethereum), though solutions specific to currency, domain registrations or other specialized use caes are acceptable

2. Timestamping

An important property that Bitcoin needs to keep is that there should be roughly one block generated every ten minutes; if a block is generated every day, the payment system becomes too slow, and if a block is generated every second there are serious centralization and network efficiency concerns that would make the consensus system essentially nonviable even assuming the absence of any attackers. To ensure this, the Bitcoin network adjusts difficulty so that if blocks are produced too quickly it becomes harder to mine a new block, and if blocks are produced too slowly it becomes easier.

However, this solution requires an important ingredient: the blockchain must be aware of time. In order to solve this problem, Bitcoin requires miners to submit a timestamp in each block, and nodes reject a block if the block's timestamp is either (i) behind the median timestamp of the previous eleven blocks, or (ii) more than 2 hours into the future, from the point of view of the node's own internal clock. This algorithm is good enough for Bitcoin, because time serves only the very limited function of regulating the block creation rate over the long term, but there are potential vulnerabilities in this approach, issues which may compound in blockchains where time plays a more important role.

Problem: create a distributed incentive-compatible system, whether it is an overlay on top of a blockchain or its own blockchain, which maintains the current time to high accuracy.

Additional Assumptions and Requirements

  • All legitimate users have clocks in a normal distribution around some "real" time with standard deviation 20 seconds.
  • No two nodes are more than 20 seconds apart in terms of the amount of time it takes for a message originating from one node to reach any other node.
  • The solution is allowed to rely on an existing concept of "N nodes"; this would in practice be enforced with proof-of-stake or non-sybil tokens (see #9).
  • The system should continuously provide a time which is within 120s (or less if possible) of the internal clock of >99% of honestly participating nodes. Note that this also implies that the system should be self-consistent to within about 190s.
  • The system should exist without relying on any kind of proof-of-work.
  • External systems may end up relying on this system; hence, it should remain secure against attackers controlling < 25% of nodes regardless of incentives.

3. Arbitrary Proof of Computation

Perhaps the holy grail of the study zero-knowledge proofs is the concept of an arbitrary proof of computation: given a program P with input I, the challenge is to create a zero-knowledge proof that you ran P with input I and received output O, such that the proof can be verified quickly (ie. in polylogarithmic or ideally constant time) even if the original computation took a very large number of steps to complete. In an ideal setup, the proof would even hide the value of I, just proving that you ran P with some output with result O, and if I needs to be made public it can be embedded into the program. Such a primitive, if possible, would have massive implications for cryptocurrency:

  1. The blockchain scalability problem would be much easier to solve. Instead of miners publishing blocks containing a list of transactions, they would be publishing a proof that they ran the blockchain state updater with some list of transactions and produced a certain output; thus, instead of transactions needing to be verified by every node in the network, they could be processed by one miner and then every other miner and user could quickly verify the proof of computation and if the proof turns out correct they would accept the new state. This is not a complete solution, because there would still be a need to transmit data, but the problem would be much easier with this powerful building block.
  2. The blockchain privacy problem would be much easier to solve. The blockchain scalability solution above would hide the details behind individual transactions; it would only reveal the fact that all of them are legitimate, so transactions would be hidden from everyone but the sender and the receiver.
  3. It would become computationally viable to use a Turing-complete consensus network as a generic distributed cloud computing system; if you have any computation you wanted done, you would be able to publish the program for miners and miners would be able to run the program for you and deliver the result alongside a proof of its validity.

There is a large amount of existing research on this topic, including a protocol known as "SCIP" (Succinct Computational Integrity and Privacy) that is already working in test environments, although with the limitation that a trusted third party is required to initially set up the keys; use of this prior work by both its original developers and others is encouraged.

Problem: create programs POC_PROVE(P,I) -> (O,Q) and POC_VERIFY(P,O,Q) -> { 0, 1 } such that POC_PROVE runs program P on input I and returns the program output O and a proof-of-computation Q and POC_VERIFY takes P, O and Q and outputs whether or not Q and O were legitimately produced by the POC_PROVE algorithm using P.

Requirements And Additional Assumptions

  • The runtime of POC_PROVE should be in O(n*polylog(n)) where n is the number of steps required to run the program.
  • The runtime of POC_VERIFY should be either constant or logarithmic in the number of steps, and at most linear in the maximum memory usage of the program.
  • The protocol should require no trusted third parties. If TTPs are required, the protocol should include a mechanism for simulating one efficiently using secure multiparty computation.

4. Code Obfuscation

For many years now we have known how to encrypt data. Simple, robust and well-tested algorithms exist for both symmetric key encryption, where the same key is needed to encrypt and decrypt, and public key encryption, where the encryption key and decryption key are different and one cannot be derived from the other. However, there is another kind of encryption that can potentially be very useful, but for which we currently have no viable algorithm: the encryption of programs. The holy grail is to create an obfuscator O, such that given any program P the obfuscator can produce a second program O(P) = Q such that P and Q return the same output if given the same input and, importantly, Q reveals no information whatsoever about the internals of P. One can hide inside of Q a password, a secret encryption key, or one can simply use Q to hide the proprietary workings of the algorithm itself.

In 2007, it was proven that perfect "black box" encryption is impossible; essentially, the argument is that there is a difference between having black-box access to a program and having the code to that program, no matter how obfuscated, and one can construct certain classes of programs that resist obfuscation. However, there is also a weaker notion of obfuscation, known as indistinguishability obfuscation, that appears to be quite possible. The definition of an indistinguishability obfuscator O is that if you take two equivalent (ie. same inputs -> same outputs) programs A and B and calculate O(A) = P and O(B) = Q, then there is no computationally feasible way for an outsider without access to A or B to tell whether P came from A or B.

This type of obfuscation may seem more limited, but it is nevertheless sufficient for many applications. For a heuristic argument why, consider two programs F and G where F internally contains and simply prints out that 32-byte string which is the hash of "12345", whereas G actually computes the hash of "12345" and prints it out. By the indistinguishability obfuscation definition, there is no computationally feasible way to tell O(F) from O(G) apart. Hence, if one can feasibly recover "12345" from O(G), then for O(G) and O(F) to be indistinguishable one would also need to be able to feasibly recover "12345" from O(F) - a feat which essentially entails breaking the preimage resistance of a cryptographic hash function.

Recently, a discovery was made by Craig Gentry, Amit Sahai et al on an algorithm which uses a construction known as "multilinear jugsaw puzzles" in order to accomplish this. Their algorithm, described here, claims to satisfy the indistinguishability obfuscation property, although at a high cost: the algorithm requires the use of fully homomorphic encryption, a highly inefficient construction that incurs roughly a one-billion-fold computational overhead.

If this construction can be made better, the potential benefits are massive. The most interesting possibility in the world of cryptocurrency is the idea of an on-blockchain contract containing private information. This basically allows for the scripting properties of Turing-complete blockchain technologies, such as Ethereum, to be exported into any other financial or non-financial system on the internet; for example, one can imagine an Ethereum contract which contains a user's online banking password, and if certain conditions of the contract are satisfied the contract would initiate an HTTPS session with the bank, using some node as an intermediary, and log into the bank account with the user's password and make a specified withdrawal. Because the contract would be obfuscated, there would be no way for the intermediary node, or any other player in the blockchain, to modify the request in-transit or determine the user's password. The same trick can be done with any other website, or much more easily with a "dumb" blockchain such as Bitcoin.

Problem: create a reasonably efficient indistinguishability obfuscation algorithm.

Additional Assumptions and Requirements

  • Successful attacks must have an expected runtime above 2^80
  • The algorithm should be sufficiently fast that a standard ECDSA signature or an AES encryption should be feasible within 108 computational steps (more specifically, 108 gas in the Ethereum VM)

5. Hash-Based Cryptography

One of the looming threats on the horizon to cryptocurrency, and cryptography in general, is the issue of quantum computers. Currently, the problem does not seem too severe; all quantum computers are either "adiabatic quantum computers", effective at only an extremely limited set of problems and perhaps not even better than classical computers at all, or machines with a very small number of qubits not capable of factoring numbers higher than 35. In the future, however, quantum computers may become much more powerful, and the recent revelations around the activities of government agencies such as the NSA have sparked fears, however unlikely, that the US military may control a quantum computer already. With this in mind, the movement toward quantum-proof cryptography has become a somewhat higher priority.

To date, all quantum-proof schemes fall into one of two categories. First, there are algorithms involving lattice-based constructions, relying on the hardness of the problem of finding a linear combination of vectors whose sum is much shorter than the length of any individual member. These algorithms appear to be powerful, and relatively efficient, but many distrust them because they rely on complicated mathematical objects and relatively unproven assumptions. However, there is also another class of algorithms that are quantum-proof: hash-based algorithms. One example of this is the classic Lamport signature: create a Merkle tree of 164 nodes (a figure chosen specifically to match 160 bits of security), publish the root, and then have the signature of a document be the combined Merkle tree proof of a subset of 82 nodes pseudorandomly chosen based on the hash of the document. This signature is one-time, and bulky (~3000 bytes), but fulfils the purpose.

The question is, can we do better? There is an approach known as hash ladders, allowing the size of a signature to be brought down to 420 bytes, and one can use Merkle trees on another level to increase the number of signatures possible, although at the cost of adding 100-300 bytes to the signature. However, even still these approaches are imperfect, and if hash-based cryptography is to be competitive the properties of the algorithms will need to be substantially improved in order to have nicer properties.

Problem: create a signature algorithm relying on no security assumption but the random oracle property of hashes that maintains 160 bits of security against classical computers (ie. 80 vs. quantum due to Grover's algorithm) with optimal size and other properties.

Requirements And Additional Assumptions

  • The computational effort of producing a signature should be less than 224 computational steps, assuming a hash takes 28 steps (a reasonable assumption due to hardware optimizations and in the future hashing ASICs built into chips)
  • The size of a signature should be as small as possible
  • The size of a public key should be as small as possible
  • The signature algorithm should be scalable to add any number of uses, although likely at the cost of adding a constant number of bytes per signature for every 2x increase in the maximum number of uses, and if possible the setup time should be sublinear in the number of uses.

Consensus

One of the key elements in the Bitcoin algorithm is the concept of "proof of work". In any Byzantine-fault-tolerant system, the security level is often defined as the minimum percentage of hostile nodes - for example, in the context of secret sharing, the Berlekamp-Welch algorithm with 2x redundancy is guaranteed to provide the correct output assuming that the total number of hostile nodes does not exceed 25% of the network, and in the context of Bitcoin mining the requirement is that the size of the set of honest nodes exceeds the size of any individual hostile coalition. However, all of these security guarantees have one important qualification: there must be some way to define what an individual node is. Before Bitcoin, most fault-tolerant algorithms had high computational complexity and assumed that the size of the network would be small, and so each node would be run by a known individual or organization and so it is possible to count each node individually.

With Bitcoin, however, nodes are numerous, mostly anonymous, and can enter or leave the system at any time. Unless one puts in careful thought, such a system would quickly run into what is known as a Sybil attack, where a hostile attacks simply creates five times as many nodes as the rest of the network combined, whether by running them all on the same machine or rented virtual private server or on a botnet, and uses this supermajority to subvert the network. In order to prevent this kind of attack, the only known solution is to use a resource-based counting mechanism. For this purpose, Bitcoin uses a scheme known as proof-of-work, which consists of solving problems that are difficult to solve, but easy to verify. The weight of a node in the consensus is based on the number of problem solutions that the node presents, and the Bitcoin system rewards nodes that present such solutions ("miners") with new bitcoins and transaction fees.

Bitcoin's proof of work algorithm is a simple design known as Hashcash, invented by Adam Back in 1995. The hashcash function works as follows:

def hashcash_produce(data, difficulty):
    nonce = random.randrange(2**256)
    while sha256(data + str(nonce)) > 2**256 / difficulty:
        nonce += 1
    return nonce

def hashcash_verify(data, nonce, difficulty):
    return sha256(data + str(nonce)) <= 2**256 / difficulty

Note that in the actual Bitcoin protocol nonces are limited to 32 bits; at higher difficulty levels, one is required to also manipulate transaction data in the block as a sort of "extranonce".

Originally, the intent behind the Bitcoin design was very egalitarian in nature. Every individual would mine on their own desktop computer, producing a highly decentralized network without any point of control and a distribution mechanism that spread the initial supply a BTC across a wide number of users. And for the first 18 months of Bitcoin's existence, the system worked. In the summer of 2010, however, developers released a Bitcoin miner that took advantage of the massive parallelization offered by the graphics processing unit (GPU) of powerful computers, mining about 10-50 times more efficiently than CPUs. In 2013, specialization took a further turn, with the introduction of devices called "application-specific integrated circuits" - chips designed in silicon with the sole purpose of Bitcoin mining in mind, providing another 10-50x rise in efficiency. CPU and GPU mining are now completely unprofitable, and the only way to mine is to either start a multimillion-dollar ASIC manufacturing company or purchase an ASIC from one that already exists.

Another related issue is mining pool centralization. Theoretically, the legitimate function of a mining pool is simple: instead of mining on their own and receiving a small chance of earning the block reward of 25 BTC, miners mine for a pool, and the pool sends them a proportionate constant payout (eg. 0.002 BTC per block). There are centralized mining pools, but there are also P2P pools which serve the same function. However, P2P pools require miners to validate the entire blockchain, something which general-purpose computers can easily do but ASICs are not capable of; as a result, ASIC miners nearly all opt for centralized mining pools. The result of these trends is grim. Right now, nearly 25% of all new ASIC hashpower is produced in a single factory in Shenzhen, and nearly 50% of the network is controlled by a single mining pool.

The second problem is easy to alleviate; one simply creates a mining algorithm that forces every mining node to store the entire blockchain. The first problem, that of mining centralization, is much harder. There is the possibility that the problem will solve itself over time, and as the Bitcoin mining industry grows it will naturally become more decentralized as room emerges for more firms to participate. However, that is an empirical claim that may or may not come to pass, and we need to be prepared for the eventuality that it does not. Furthermore, the wasted energy and computation costs of proof of work as they stand today may prove to be entirely avoidable, and it is worth looking to see if that aspect of consensus algorithms can be alleviated.

6. ASIC-Resistant Proof of Work

One approach at solving the problem is creating a proof-of-work algorithm based on a type of computation that is very difficult to specialize. One specific ideas involves creating a hash function that is "memory-hard", making it much more difficult to create an ASIC that achieves massive gains through parallelization. This idea is simple, but fundamentally limited - if a function is memory-hard to compute, it is also generally memory-hard to verify. Additionally, there may be ways to specialize hardware for an algorithm that have nothing to do with hyperparallelizing it. Another approach involves randomly generating new mining functions per block, trying to make specialization gains impossible because the ASIC ideally suited for performing arbitrary computations is by definition simply a CPU. There may also be other strategies aside from these two.

Ultimately, perfect ASIC resistance is impossible; there are always portions of circuits that are going to be unused by any specific algorithm and that can be trimmed to cut costs in a specialized device. However, what we are looking for is not perfect ASIC resistance but rather economic ASIC resistance. Economic ASIC resistance can be defined as follows. First of all, we note that in a non-specialized environment mining returns are sublinear - everyone owns one computer, say with N units of unused computational power, so up to N units of mining cost only the additional electricity cost, whereas mining beyond N units costs both electricity and hardware. If the cost of mining with specialized hardware, including the cost of research and development, is higher per unit hashpower than the cost of those first N units of mining per user then one can call an algorithm economically ASIC resistant.

For a more in-depth discussion on ASIC-resistant hardware, see https://blog.ethereum.org/2014/06/19/mining/

Problem: Create two functions, PoWProduce(data,diff) -> nonce and PoWVerify(data,nonce,diff) -> { 0, 1 }, to serve as alternatives to Hashcash such that it is economically unattractive to produce an ASIC for PoWProduce

Additional Assumptions And Requirements:

  • PoWProduce must have expected runtime linear in diff
  • PoWVerify must have runtime at most polylogarithmic in diff
  • Running PoWProduce should be the most efficient, or very close to the most efficient, way to produce values that return 1 when checked with PoWVerify (ie. no software optimization)
  • PoWProduce must not be superlinear in computational power or time; that is to say, the expected number of successful PoWProduce computations for a node with N dollars worth of hardware after t seconds should be bounded by kNt for some k. Furthermore, the linearity should kick in quickly; ie. $1000 worth of mining hardware should function with over 90% efficiency.
  • It should be shown with reasonably rigorous technological and economic analysis that the algorithm is economically ASIC resistant.

7. Useful Proof of Work

Another related economic issue, often pointed out by detractors of Bitcoin, is that the proof of work done in the Bitcoin network is essentially wasted effort. Miners spend 24 hours a day cranking out SHA256 (or in more advanced implementations Scrypt) computations with the hopes of producing a block that has a very low hash value, and ultimately all of this work has no value to society. Traditional centralized networks, like Paypal and the credit card network, manage to get by without performing any proof of work computations at all, whereas in the Bitcoin ecosystem about a million US dollars of electricity and manufacturing effort is essentially wasted every day to prop up the network.

One way of solving the problem that many have proposed is making the proof of work function something which is simultaneously useful; a common candidate is something like Folding@home, an existing program where users can download software onto their computers to simulate protein folding and provide researchers with a large supply of data to help them cure diseases. The problem is, however, that Folding@home is not "easy to verify"; verifying the someone did a Folding@home computation correctly, and did not cut corners to maximize their rounds-per-second at the cost of making the result useless in actual research, takes as long as doing the computation oneself. If either an efficiently verifiable proof-of-computation for Folding@home can be produced, or if we can find some other useful computation which is easy to verify, then cryptocurrency mining could actually become a huge boon to society, not only removing the objection that Bitcoin wastes "energy", but even being socially beneficial by providing a public good.

Note that there is one major concern with this approach that has been identified: if the useful PoW is implemented incorrectly, it can potentially reduce the cost of an attack on the network. If the useful PoW is useful in such a way that it is sometimes economically viable for certain very large entities to perform the computation even without the currency incentive, then those entities have an incentive to launch attacks against the network at no cost, since they would be performing the computations anyway. One simple, though crude and imperfect, way of addressing this problem is to make the PoW a half-and-half mix between useful and useless, making the cost of an attack at least 50% of what it would be in a useless-PoW environemnt. In practice, the overhead of making PoW verifiable may well introduce over 2x inefficiency unintentionally. Another economic solution is to make the computation a "pure" public good such that no individual entity derives a significant benefit from it. Proposed solutions to this problem should include a rigorous analysis of this issue.

Problem: Create two functions, PoWProduce(data,diff) -> nonce and PoWVerify(data,nonce,diff) -> { 0, 1 }, to serve as alternatives to Hashcash such that the outputs of PoWProduce are independently useful.

Requirements

  • PoWProduce must have expected runtime linear in diff
  • PoWVerify must have expected runtime at most polylogarithmic in diff
  • Running PoWProduce should be the most efficient way to produce values that return 1 when checked with PoWVerify
  • PoWProduce must not be superlinear in computational power or time; that is to say, the expected number of successful PoWProduce computations for a node with N dollars worth of hardware after t seconds should be bounded by kNt for some k. Furthermore, the linearity should kick in quickly; ie. $1000 worth of mining hardware should function with over 90% efficiency.
  • PoWProduce must produce a public good, such that the total value to everyone of the public good produced is greater than the cost of all resources invested into the mining process.
  • The system must be able to exist without a trusted third party, but it is reasonable to allow a trusted third party to serve as a data source for useful computations. If the trusted third party acts maliciously in any way, the public good may be negated but the blockchain mining should not be compromised.

8. Proof of Stake

Another approach to solving the mining centralization problem is to abolish mining entirely, and move to some other mechanism for counting the weight of each node in the consensus. The most popular alternative under discussion to date is "proof of stake" - that is to say, instead of treating the consensus model as "one unit of CPU power, one vote" it becomes "one currency unit, one vote".

A very simple proof of stake algorithm requires the miner mining the block to sign it with the private key to the address holding their coins, where the block is valid if sha256(PREVHASH + ADDRESS + TIMESTAMP) <= 2^256 * BALANCE / DIFFICULTY where PREVHASH is the hash of the previous block, ADDRESS is the signer's address with balance BALANCE, TIMESTAMP is the current Unix time in seconds and DIFFICULTY is an adjustable parameter to regulate the frequency of successful signatures. At first glance, this algorithm has the basic required properties: every miner has some random chance per second of succeeding, and if your address has twice as much money in it then you have double the chance of success.

However, this algorithm has one important flaw: there is "nothing at stake". In the event of a fork, whether the fork is accidental or a malicious attempt to rewrite history and reverse a transaction, the optimal strategy for any miner is to mine on every chain, so that the miner gets their reward no matter which fork wins. Thus, assuming a large number of economically interested miners, an attacker may be able to send a transaction in exchange for some digital good (usually another cryptocurrency), receive the good, then start a fork of the blockchain from one block behind the transaction and send the money to themselves instead, and even with 1% of the total stake the attacker's fork would win because everyone else is mining on both.

Another problem to keep in mind is the issue of so-called "long-range attacks" - attacks where the miner attempts to start a fork not five or ten blocks behind the head of the main chain, as happens normally, but hundreds of thousands of blocks back. If an algorithm is designed incorrectly, it may be possible for an attacker to start from that far back, and then mine billions of blocks into the future (since no proof of work is required), and new users would not be able to tell that the blockchain with billions of blocks more is illegitimate. This can generally be solved with timestamping, but special corner cases do tend to appear in overcomplicated designs.

The Slasher algorithm, described here and implemented by Zack Hess as a proof-of-concept here, represents my own attempt at fixing the nothing-at-stake problem. The core idea is that (1) the miners for each block are determined ahead of time, so in the event of a fork a miner will either have an opportunity to mine a given block on all chains or no chains, and (2) if a miner is caught signing two distinct blocks with the same block number they can be deprived of their reward. The algorithm is viable and effective, but it suffers from two flaws of unknown significance. First, if all of the miners for a given block learn each other's identities beforehand, they can meet up and collude to shut down the network. Second, the nothing-at-stake problem remaing for attacks going back more than 3000 blocks, although this is a smaller issue because such attacks would be very obvious and can automatically trigger warnings.

For a more in-depth discussion on proof of stake, see https://blog.ethereum.org/2014/07/05/stake/

Problem: create a proof-of-stake algorithm that solves the nothing-at-stake problem and long-range attack problems, without introducing new collusion risks that require less than 25% of stakeholders to succeed.

Additional Requirements And Assumptions

  • The expected return from mining should be bounded by k times the miner's stake for some k, and assuming $1 billion total participating stake a stake of $1000 should be able to reach 90% of this maximum efficiency.
  • The algorithm should be fully incentive-compatible, addressing the double-voting issue defined above and the collusion issue defined above at both short and long range.

9. Proof of Storage

A third approach to the problem is to use a scarce computational resource other than computational power or currency. In this regard, the two main alternatives that have been proposed are storage and bandwidth. There is no way in principle to provide an after-the-fact cryptographic proof that bandwidth was given or used, so proof of bandwidth should most accurately be considered a subset of social proof, discussed in later problems, but proof of storage is something that certainly can be done computationally. An advantage of proof-of-storage is that it is completely ASIC-resistant; the kind of storage that we have in hard drives is already close to optimal.

The most simple algorithm for proving that you own a file with N blocks is to build a Merkle tree out of it, publish the root, and every k blocks publish a Merkle tree proof of the ith block where i is the previous block hash mod N. However, this algorithm is limited because it is only a simple building block, not a complete solution. In order to turn this into a currency, one would need to determine which files are being stored, who stores whose files, to what extent and how the system should enforce redundancy, and if the files come from the users themselves how to prevent compression optimizations and long-range attacks.

Currently, the latest work in this area are two projects called Permacoin and Torcoin, which solve some of the problems in proof of storage with two insights. First, users should not be able to choose which files they store. Instead, files should be randomly selected based on their public key and users should be required to store ALL of the work assigned or else face a zero reward. This idea, provided in the context of proof of bandwidth in the case of Torcoin, prevents attacks involving users only storing their own data. Second, a Lamport-like signature algorithm can be used that requires users to have their private key and store their file locally; as a result; uploading all of one's files to the cloud is no longer a viable strategy. This, to some degree, forces redundancy.

However, the problem with Permacoin is that it leaves unclear what files should be stored; cryptocurrency issuance can theoretically pay for billions of dollars of work per year, but there is no single static archive whose storage is worth billions. Ideally, the system would allow for new files to be added, and perhaps even allow users to upload their own files, but without introducing new vulnerabilities.

Problem: create a currency that uses proof-of-storage as its consensus and distribution algorithm.

Additional Assumptions And Requirements

  • The currency must be future-proof, being able to expand the amount of data stored over time; the system should not eventually fall into some failure state if hard disk space continues to get cheaper and more efficient.
  • The currency should ideally be maximally useful. At the least, the currency should allow people to upload their own files and have them stored, providing an uploading network with minimal cryptographic overhead, although ideally the currency should select for files that are public goods, providing net total value to society in excess of the number of currency units issued.
  • The expected return from mining should be at most slightly superlinear, ie. it must be bounded by ks/(1-s) for some k, where s is the miner's share of the total network, although perfect linearity is ideal.
  • The system should be maximally resistant against mining pool centralization as a result of any small degree of superlinearity.
  • The system should be secure against nothing-at-stake and long-range attacks.
  • The system should be secure against attacker involving users uploading specially formatted files or storing their own data.

Economics

The second part of cryptoeconomics, and the part where solutions are much less easy to verify and quantify, is of course the economics. Cryptocurrencies are not just cryptographic systems, they are also economic systems, and both kinds of security need to be taken into account. Sometimes, cryptographic security may even be slightly compromised in favor of an economic approach - if a signature algorithm takes more effort to crack than one could gain from cracking it, that is often a reasonable substitute for true security. At the same time, economic problems are also much more difficult to define. One cannot usually definitively know whether or not a problem has been solved without extensive experimentation, and the result will often depend on cultural factors or the other organizational and social structures used by the individuals involved. However, if the economic problems can be solved, the solutions may often have reach far beyond just cryptocurrency.

10. Stable-value cryptoassets

One of the main problems with Bitcoin is the issue of price volatility. The value of a bitcoin often experiences very large fluctuations, rising or falling by as much as 25% in a single day and 3x in a month. The main economic reason behind this is that the supply of bitcoins is fixed, so its price is directly proportional to demand (and therefore, by efficient market hypothesis, the expected discounted future demand), and demand is very unpredictable. It is not known if Bitcoin will be simply a niche payment method for transcations requiring a high degree of privacy, a replacement for Western Union, a mainstream consumer payment system or the reserve currency of the world, and the expected value of a bitcoin differs over a thousandfold between these various levels of adoption. Furthermore, the utility of the Bitcoin protocol is heavily dependent on the movements of the Bitcoin price (ie. people are interested in Bitcoin more if the price is going up), creating a positive feedback loop, which has arguably been responsible for both Bitcoin's great meteoric rises and its many-month-long periods of rapid decline.

To solve this problem, there are generally two paths that can be taken. The first is to have the network somehow detect its current level of economic usage, and have a supply function that automatically increases supply when usage increases. This reduces uncertainty; even though the expected future level of adoption of the protocol may have a variance of 10-100x, the circumstance where adoption increases 100x will also have 100x more supply and so the value of the currency will remain the same. There is a problem that if usage decreases there is no way to remove units from circulation, but even still the lack of upward uncertainty should reduce upward volatility, and downward volatility would also naturally reduce because it is no longer bad news for the value of the currency when an opportunity for increased usage is suddenly removed. Furthermore, in the long term the economy can be expected to grow, so the zero-supply-growth floor may not even ever be reached in practice.

The problem is that measuring an economy in a secure way is a difficult problem. The most obvious metric that the system has access to is mining difficulty, but mining difficulty also goes up with Moore's law and in the short term with ASIC development, and there is no known way to estimate the impact of Moore's law alone and so the currency cannot know if its difficulty increased by 10x due to better hardware, a larger user volume or a combination of both. Other metrics, such as transaction count, are potentially gameable by entities that want the supply to change in a particular direction (generally, holders want a lower supply, miners want a higher supply).

Another approach is to attempt to create a currency which tracks a specific asset, using some kind of incentive-compatible scheme likely based on the game-theoretic concept of Schelling points, to feed price information about the asset into the system in a decentralized way. This could then be combined with a supply function mechanism as above, or it can be incorporated into a zero-total-supply currency system which uses debts collateralized with other cryptographic assets to offset its positive supply and thus gain the ability to grow and shrink with changes to usage in either direction. The problem here is constructing the scheme in such a way that there is no incentive for entities to feed in false price information in order to increase or decrease the supply of the asset in their favor.

Problem: construct a cryptographic asset with a stable price.

Requirements

  • The expected root-mean-square daily change in the logarithm of the price of the asset should be less than 25% of that of Bitcoin under similar conditions. Ideally, the asset should be guaranteed to almost always maintain a value within 10% of an arbitrary cryptographic or real-world asset for which price information is easily accessible
  • The expectation analysis should take into account black swan risks (ie. systems where the variance is 0% 99% of the time but 10x in a day the other 1% of the time are unacceptable)
  • The solution must come with a model, including parameters such as short-term-consumption purchases, medium-term purchases, speculative purchases, positive and negative media, adoption and regulatory events, irrational actors and actors with political motives, show that their model well fits the history of Bitcoin and potentially major altcoins without overfitting, and show that under the model the other two requirements hold
  • Zero-total-supply assets, ie. assets where each unit is balanced by a collateralized debt of a unit, are allowed, although such systems must include a robust margin-calling mechanism because it is assumed that most users are anonymous and can therefore trivially run away from debts

11. Decentralized Public Goods Incentivization

One of the challenges in economic systems in general is the problem of "public goods". For example, suppose that there is a scientific research project which will cost $1 million to complete, and it is known that if it is completed the resulting research will save one million people $5 each. In total, the social benefit is clear: if everyone contributes $1, then each individual person will see a benefit of $5 - $1 = $4 for $4 million total. However, the problem is that from the point of view of each individual person contributing does not make sense - whether or not you contribute has close to zero bearing on whether enough money will be collected, so everyone has the incentive to sit out and let everyone else throw their money in, with the result that no one does.

So far, most problems to public goods have involved centralization; some large organization, whether a big company or a government, agrees to offer some of its private services only to those individuals who participate in paying for the public good. Often this is done implicitly: for example, some of the money from each purchase of an iPad goes toward research and development (some of which is a public good, and some of which is an excludable "club good"). At other times, it's more explicit, as in the case of taxation. In order for decentralized economic systems (we'll refer to decentralized economic systems that somehow rely on cryptography and/or cryptocurrency as "cryptoeconomic systems") to be effective, ways of incentivizing production of public goods relevant to that system are required. A few possible approaches include:

  • Assurance contracts - the idea behind an assurance contract is that N people may or may not put their funds into a pool, where that pool pays to produce a public good if and only if at least $X in total is contributed. Otherwise, the pool pays everyone back. If the pool creator acts optimally, the tipping point will be right at the top of the bell curve that is the probability distribution for how much other people might contribute, meaning that the chance that one user with their contribution of X/N will be pivotal should, by central limit theorem, approach ~1/sqrt(N), creating a sqrt(N)-sized amplifying effect on their donation.
  • Dominant assurance contracts - a special type of assurance contract, called a dominant assurance contract, involves an entrepreneur that pays all contributors back slightly more than 100% of what they put in if the fund fails to reach its target (and takes profits if the fund succeeds); this provides an incentive for someone to create optimally targeted assurance contracts.
  • Currency issuance - a cryptoeconomic system can contain its own currency or token system which is somehow necessary or useful in some part of the system. These currency units can then either be generated by the system and then sold or directly assigned to reward contribution. This approach gets around the free-rider problem because no one needs to pay the $1 explicitly; the value arises out of the emergent value of the network which is does not cost people to support.
  • Status goods issuance - a status good can be defined as a good that confers only relative benefit to its holder and not absolute benefit to society; for example, you may stand out in the public if you wear an expensive diamond necklace, but if everyone could trivially obtain such a necklace the situation would be very similar to a world with no diamond necklaces at all. A cryptoeconomic system can release its own status goods, and then sell or award them. One example of a status good is a "badge"; some online forums, for example, show a special badge beside users that have contributed funds to support the forum's development and maintenance. Another important example of a status good is a namespace; for example, a decentralized messaging protocol may be able to fund itself by selling off all of the 1-4 letter usernames.
  • Recursive rewarding - this is in some ways a mirror image of the concept of "recursive punishment" that arguably underlies a large number of social protocols. For example, consider the case of tax-funded police forces. In natural circumstances, there often arise opportunities to take actions which are beneficial to the perpetrator, but ultimately harmful to society as a whole (eg. theft). The most common solution to this problem is punishment - an act which is harmful in itself, but which shifts the incentives so that attacking is no longer beneficial to the perpetrator. However, there is a problem: there is no incentive to participate in the punishment process. This is solved by making punishment obligatory, with non-participation (in modern society by paying taxes) itself punishable by the same mechanism. Recursive rewarding is a mirror image of this strategy: here, we reward a desirable action, and people who partivipate in the rewarding mechanism (eg. by giving reward recipients a discount in shops) are themselves to be rewarded.

Many of these approached can arguably be done in concert, or even simultaneously within one mechanism.

Problem: come up with and implement methods for incentivizing public goods production in a decentralized environment.

Additional Assumptions And Requirements

  • A fully trustworthy oracle exists for determining whether or not a certain public good task has been completed (in reality this is false, but this is the domain of another problem)
  • The agents involved can be a combination of individual humans, teams of humans, AIs, simple software programs and decentralized cryptographic entities
  • A certain degree of cultural filtering or conditioning may be required for the mechanism to work, but this should be as small as possible
  • No reliance on trusted parties or centralized parties should be required. Where some kind of "supernode" role does exist, the protocol should provide a way for anyone to participate in that function with a mechanism for rewarding those who do it well
  • The mechanism should ideally be able to handle both public goods which everyone values and public goods which are only valued by a small portion of the population (eg. the production of a freely available book or video on a specific topic)

12. Reputation systems

A concept which can arguably be considered to be a mirror image of currency is a reputation system. A reputation system serves three functions. First of all, it provides a mechanism for filtering honest people from dishonest people. Different people have different moral preference profiles, and so individuals who cheat less in one context are less likely to cheat in another context. Second, it provides an incentive not to cheat. If an individual can be said to possess a reputation of value R, and he enters a business deal where he is receiving payment V in exchange for a product with cost-of-production C, then as long as R > C the reputation system removes the incentive to run away with the money because doing so would sacrifice the reputation. Finally, reputation can be thought of as a kind of point system that people value intrinsically, both in a private context and as a status good in comparison with others.

Money serves functions that are very similar. People who are willing to spend more money on something tend to want it more, creating a filtering function ensuring efficient resource consumption on the demand side. It provides an incentive not to cheat by consuming and not producing, because if you do so your remaining currency units and thus ability to consume in the future will go down. And finally, it is also very much an intrinsically valued point system; in fact, some argue that among very wealthy individuals this function of money is dominant.

However, there are also differences. First, money is an absolute score - I have X units of currency C from the point of view of everyone in the world - but reputation is a relative measure, depending on both the owner of the reputation and the observer. I may have a high reputation in North America, a near-zero reputation in Africa, and a negative reputation among certain kinds of antitechnologist and ultranationalist groups. Second, reputation is free to give; it does not cost me anything to praise you, except potentially moral liability that I may incur if you turn out to act immorally in some way. This is in contrast with money, where adding X units to A means subtracting X units from B.

However, up until very recently, reputation has been a very informal concept, having no concept of score and instead relying entirely on individual opinion. Because opinion is relatively easy to manipulate, this means that reputation as a concept has been highly suboptimal in its implementation, and has been quite vulnerable to informational and psychological attacks. Some specific problems are:

  1. How do we know how what the value of someone's reputation with someone else is after a particular number of interactions? A common attack on informal reputation systems is the "long con" - act honestly but passively and cheaply for a very long time, accumulate trust, and then suddenly go all out and destructively capitalize on one's reputation as much as possible. The initial dormant phase is cheap for the attacker, but ends up resulting in the attacker accumulating a disproportionately large amount of trust for the community and thereby ultimately causing much more damange than good. Overcompensate for this too much, however, and there ends up being no opportunity to gain trust.
  2. How do we incorporate secondary trust? In general, when A is deciding whether or not to trust B, A has not had any prior dealings with B, and therefore has no way of knowing whether or not B is trustworthy. One approach is to just look at all ratings for B, but then we run into the issue of Sybil attacks: what if B creates 50000 fake users, all of whom rate each other highly, to give good ratings to him? To solve this problem, reputation systems rely on a fallback known as a web of trust: find some chain of people P[1] ... P[k] such that A trusts P[1], P[i] trusts P[i+1] for all i, and P[k] trusts B. Under the "six degrees of separation hypothesis", any two people in the world except those completely disconnected from society have such a chain of maximum length k = 5 (so at most six hops total). However, the question arises, if A has a certain rating for P[1] and P[1] has a certain rating for B, what should the reputation system recommend to B?
  3. If a reputation system becomes more formalized, are there market attacks that reduce its effectiveness to simply being just another form of money? Specifically, how would a reputation system where giving reputation is free handle users multiplying their reputation with millions of "I praise you if you praise me" trades? Will such trades need to be explicitly banned, punishable by loss of reputation, or is there a better solution?
  4. How do we deal with double use attacks? Specifically, suppose that A has a reputation with value R = $1000. Using this reputation, A has a business dealing where P[1] trusts her for $600. Then, she simultaneously engages in such a dealing with P[2], P[3] ... P[10], each of whom individually believe that A will not betray them since $600 < $1000, and then runs away with $6000 taking the $1000 hit from the value of her reputation. How do we prevent such fractional reserve-like scenarios?

Problem: design a formalized reputation system, including a score rep(A,B) -> V where V is the reputation of B from the point of view of A, a mechanism for determining the probability that one party can be trusted by another, and a mechanism for updating the reputation given a record of a particular open or finalized interaction.

Note that for the purpose of this use case we are targeting specifically the "can I trust you" use case of reputation, and not the social-incentivizing "whuffie"-esque currency-like aspect.

Additional Assumptions and Requirements

  • The system has access to a record of all finalized transactions inside the system and all transactions in progress, although entities are of course able to choose to make deals outside the system
  • It is allowed to introduce mechanisms like charity donations, public goods provision and sacrifices as a way of increasing one's reputation. However, if non-monetary contributions are allowed, there needs to be some mechanism for measuring their value
  • For simplicity, we can assume that interactions between two people are of the form "A pays, then B sends the product and A receives", with no possibility for loss beyond the principal (eg. food poisoning) or ambiguous quality. Ideally, however, the system should account for such possibilities.
  • The system should continue to be reasonably accurate whether the parties involved are simple programs (eg. micropayment software protocols), more complicated AIs, DAOs, individual humans or human centralized or decentralized organizations
  • If a mechanism is provided for determining the probability of a successful interaction, a success metric for the system can be defined as the sum over all transactions of V * (S * log(p) - (1-S) * log(1-p)), where S = 1 if the transaction succeeded and S = 0 if there was a registered complaint, p is the assigned probability and V is the value of the transaction. The objective is to maximize this metric.

Metrics

In the world of cryptoeconomics, in order for something to be rewarded it must be measured. Some things are easy to measure; for example, just by looking at the string "dog5356356" and its SHA256 hash, 0000390f327fefc900..., one can clearly see that around 216 SHA256 computations were done to produce it. Other computational results that cannot be verified so quickly can be easily measured competitively using challenge-response protocols, where different parties are incentivized to find errors in each other's proofs. Results to mathematical problems are also usually easy to computationally verify. Other things, however, cannot be verified just by looking them; in that case, in both the real world and the cryptographic world, there is only one solution: social proof.

To some extent, proof of work consensus is itself a form of social proof. Transaction A happened before transaction B because the majority of users say it did, and there is an economic incentive to go with the majority opinion (specifically, if you generate a block on the incorrect chain, that block will get discarded and the miner will receive no reward). Assuming that most participants act truthfully, the incentive is to go along with the projected majority and tell the truth as well. This insight can be extended into SchellingCoin, a generalized data feed protocol, protocols for proof of bandwidth, and anything else that can be quickly verified. The challenge is, however, what if verification has a cost? What if it takes some effort to determine whether or not a certain thing has happened, or what if the information is in principle only available to a few people? If there is too much gathering cost or secrecy, then centralization becomes necessary; the question is, how high can we go? How much can we measure without any social proof at all, and how much can we measure without a centralized verifier?

13. Proof of excellence

One interesting, and largely unexplored, solution to the problem of distribution specifically (there are reasons why it cannot be so easily used for mining) is using tasks that are socially useful but require original human-driven creative effort and talent. For example, one can come up with a "proof of proof" currency that rewards players for coming up with mathematical proofs of certain theorems. There is no generic algorithm, aside from brute force, for proving theorems, and yet proofs of theorems are theoretically computationally easy to verify: one simply needs to write every step of the proof in a formal language, allowing the use of only one inference rule (eg. a + b = b + a or a * (b + c) = a * b + a * c but not a * (b + c) = a * c + b * a) between each step, and having a program verify the correctness of the inferences at each step.

For example, a proof of a common algebraic factorization problem appears as follows:

  a^2 - b^2
= a^2 - a*b + a*b - b^2
= a*a - a*b + a*b - b^2
= a*(a - b) + a*b - b^2
= a*(a - b) + a*b - b*b
= a*(a - b) + b*a - b*b
= a*(a - b) + b*(a - b)
= (a + b)*(a - b)

Each step of the proof can be verified using pattern matching algorithms, but it is much harder for a computer to figure out that the trick is to add and subtract a*b into the expression (technically, in this case specialized algorithms can do it, but in more general cases especially involving second-order logic it becomes intractable). Note that for computers the proof must be written down in excruciating detail; blockchain-based algorithms specifically heavily benefit from simplicity. To alleviate this problem, compilers can likely be made that can make small two and three-step inferences and expand shorter proofs into more complete ones.

Alternatives to proof-of-proof include proof-of-optimization, finding optimal inputs to some function to maximize a particular output (eg. the ability of a radio antenna to receive signals), algorithms involving playing strategy games or multiplayer AI challenges (one can even require users to submit programs to the blockchain that play against each other), and solving a specific math problem at greater and greater difficulty (eg. factoring). Note that because success in these problems is very sporadic, and highly inegalitarian, one cannot use most of these algorithms for consensus; rather, it makes sense to focus on distribution.

Problem: create a proof-of-excellence distribution mechanism that rewards solving problems that are both dominated by human effort and whose solutions provide some benefit to humanity.

Additional Assumptions and Requirements

  • Given a well-justified extrapolation of the global levels of human and computer competence at the underlying problem, over 75% of the rewards from the system should be provided by human labor, although software aids are allowed.
  • The algorithm must ideally be future-proof; that is to say, it must continue rewarding value production in the long term and should not be an area that will eventually be "solved" completely.
  • The distribution should be maximally egalitarian, though this is a secondary concern.
  • The system should be secure against front-running attacks, ie. if an individual submits a solution, then it should not be practical for even a moderately powerful attacker to look at the solution and then resubmit his own transaction containing the same solution and thereby steal the reward.

14. Anti-Sybil systems

A problem that is somewhat related to the issue of a reputation system is the challenge of creating a "unique identity system" - a system for generating tokens that prove that an identity is not part of a Sybil attack. The naive form of anti-Sybil token is simple: a sacrifice or proof of deposit. In a sacrifice setup, such identities simply cost $X, and in a PoD system identities require a deposit of $Y in order to be active, where perhaps the deposit can be taken away or destroyed under certain circumstances. However, we would like to have a system that has nicer and more egalitarian features than "one-dollar-one-vote"; arguably, one-person-one-vote would be ideal.

To date, we have seen two major strategies for trying to solve this problem. One potential solution is to come up with a proof-of-work algorithm which is dominated by human labor, and not computers. This is not as difficult as it may seem; although computers get more and more powerful every year, there are a number of problems that have remained out of computers' reach for decades, and it may even be possible to identify a class of problems that are the artificial-intelligence-theoretic equivalent of "NP-complete" - problems such that, if they can be solved, it with high probability implies that AI can essentially replicate human activity in its entirety, in which case we are essentially in a post-scarcity utopia and money and incentivization may not even be necessary. These problems may be non-interactive challenges like CAPTCHAs, although all existing CAPTCHAs are far from adequate for the task, or they may be interactive strategy games like Go.

The second strategy is to use social proof, turning the muscle of decentralized information gathering toward a simple problem: are these two identities the same person? If they are not, then they receive two anti-Sybil tokens, and if they are they receive one token. In general, we can separately define two concepts of identity: voluntary identity and involuntary identity. A voluntary identity can be seen as a cluster of interactions which are in some fashion correlated with each other; for example, a cryptographic identity consists of the set of interactions signed by a particular public key. An involuntary identity is a cluster of interactions which are correlated with each other, but where the entity producing the interactions does not want the correlations to be visible. A simple unique identity system would rely on voluntary identities embedded in social networks, with the understanding that creating separate identities with reputations is an expensive task and so most people would not want to do it, but a more advanced system may try to detect involuntary slipups like writing style patterns or IP addresses.

The question is, can we use these mechanisms, either separately or together, and perhaps in combination with cryptoeconomic protocols and sacrifices as a fallback in order to create an anti-Sybil system which is highly egalitarian? We will accept that any scheme can be cracked at some cost; however, what we want is for it to be much more efficient for individuals to obtain one anti-Sybil token "the proper way" rather than purchasing one off the grey/black market. The challenge is to push the grey/black market cost as high as possible, as much as possible without making the first token difficult.

Problem - create a mechanism for distributing anti-Sybil tokens

Additional Assumptions and Requirements:

  • Everyone is part of a social network with similar characteristics to social networks now found in the real world, and social data can be provably provided to cryptoeconomic systems (eg. blockchains, Ethereum contracts)
  • The cost of obtaining one anti-Sybil token for a human should be as low as possible
  • The cost of obtaining multiple anti-Sybil tokens for a human should be as high as possible
  • The cost of obtaining anti-Sybil tokens for an automated system should be as high as possible (this is a more important criterion than high cost for multi-obtainment for humans)
  • The system should not create dependency on centralized parties (eg. government passport offices) that have the power to cheat the system

15. Decentralized contribution metrics

Incentivizing the production of public goods is, unfortunately, not the only problem that centralization solves. The other problem is determining, first, which public goods are worth producing in the first place and, second, determining to what extent a particular effort actually accomplished the production of the public good. This challenge deals with the latter issue. Although in the case of computational tasks it's easy to come up with a proof of solution, for non-computational tasks the situation is much more difficult. If a cryptoeconomic system wants to incentivize users to build better graphical user interfaces to its own system, how would it rate people's contributions? Even more problematically, what about potentially quasi-adversarial tasks like incentivizing updates to its own code? What about a DAO that funds healthcare, or tries to incentivize adopting renewable energy?

This is a subclass of the general "social proof" problem; here, the particular challenge is that each individual datum in question is something that very few people are interested in, and data gathering costs are often high. Sometimes, there is not even a concept of a single "correct" value with respect to the particular metric; in the case of quality measurement for an interface, a solution like A/B testing may be required. In adversarial cases, there may need to be an opportunity for incentivized opponents to look at a solution and attempt to pick it apart.

Problem: come up with and implement a decentralized method for determining whether or not a particular task was performed by a specific person, and for estimating the quality of the work

Additional Assumptions and Requirements

  • The agents involved can be a combination of individual humans, teams of humans, AIs, simple software programs and other DAOs
  • There is no cryptographically verifiable information about the completion of any task; the system must rely entirely on some form of social proof

16. Decentralized success metrics

Another, related, problem to the problem of decentralized contribution metrics is the problem of decentralized success metrics. On the macroscopic scale, how do we know if, and to what extent, an organization has succeeded in accomplishing its objectives? In the case of something like Bitcoin, there is a simple, but imperfect, answer: success can be measured by the hashpower of the network. This setup is reasonably effective, but is flawed in two ways: first, hashpower is an imperfect proxy for price, because the development or nondevelopment of ASICs may skew the results, and second, price is an imperfect proxy for success, because the currency may have greater success as something with a lower market capitalization if it is more used in other ways. In the case of a DAO funding healthcare or anti-climate-change efforts, however, no such heuristic exists at all. Once again, some concept of social proof is the only option.

Here, information gathering costs are low, and information is accessible to everyone in the public, so a higher level of accuracy is possible, hopefully even enough for financial contracts based off of the metric to be possible. However, in order to maintain that higher level of accuracy, and in the presence of such financial derivatives, new problems arise. Can one moderately powerful entity manipulate the metric for their own benefit? If information gathering costs do exist, is the system vulnerable to falling into a centralized equilibrium, where everyone is incentivized to simply follow along with the actions of some specific party?

Problem: come up with and implement a decentralized method for measuring numerical real-world variables

Additional Assumptions and Requirements

  • The agents involved can be a combination of individual humans, teams of humans, AIs, simple software programs and other DAOs
  • The system should be able to measure anything that humans can currently reach a rough consensus on (eg. price of an asset, temperature, global CO2 concentration