It would have to have the requirement that your computational problem is easily verified, but it sounds feasible as a separate kind of transaction. Gavin Anderson wrote a bit about the possible types of different "smart contract" transactions that are possible. Solving a given computational problem sounds like something that would fall into that realm.

I looked into smart contracts a bit and they are very fascinating, but much of it went over my head. They tend to be very tiny though, such as escrow, multiple signatures, delayed processing, etc. As execution and state external to the blockchain...

A computational exchange could be established (within the blockchain?) in which users and 'miners' negotiate prices for CPU or GPU cycles, temporary storage during computation, input/output bandwidth, and even long term encrypted/replicated storage.

As a user, I could create or use an existing program (perhaps compiled ECMAScript or a minimal turing-complete language such as brainfuck) to run within a specific amount of RAM, and receive periodic state and proof-of-work dumps. As long as the program were single-threaded (or the user acted as the mult-thread control center) then the user could start, stop, restart, and switch processing between any number of miner nodes after any periodic state dump.

I imagine most of this is already well researched in the field of grid computing and darknets. What may not already exist is a mechanism to monetize processes in a fully decentralized environment. Crypto-money like bitcoin would be necessary without a central mediator such as IBM or Amazon. Why not just use Amazon? Perhaps you'd prefer anonymity. Perhaps your computational units are many but infrequent or tiny. Perhaps a commoditized service will have greater liquidity and competition would drive prices down and reliability up. Perhaps you need a resource with no single point of failure. Services could be funded directly by the users who find them useful (through donations or usage tax).

If such a computational network and exchange existed, then it would be very easy for a company to accept existing fiat money from customers, including credit card transactions, to purchase computational time denominated in bitcoin. Similarly a miner could negotiate contracts and accept cash from any company offering the commoditized service of computational power.

If commodification drove prices down and reliability up, then anonymous, secure, distributed computing funded by crypto-currency would truly establish the vision of 'the cloud'.

Very true indeed.  I took an interest in mining for the first time just a few days ago and started researching what was available in FPGA-land.  'winning' a few BTC now and then is very much a secondary interest of mine (no pun intended.)

I have an internal debate about whether it makes more sense to go for some big-iron GPU machines to use periodically in times of crisis.  Maybe if some demoralized miners are giving there gear away I would think about it, but Moores law and supply-chain factors tend to argue against this investment.

You should read one of the inspirations for the orignal Bitcoin paper:
W. Dai, "b-money," http://www.weidai.com/bmoney.txt, 1998.

If the work has value, by itself, that is a problem for establishing "Proof of work". It would essentially allow participants to sell the same work twice. If not carefully implemented, it also opens up the possibility of "banking" up work in isolation from the network in order to mount an attack on the network.

Thanks for the b-money link. Fascinating read; It's obvious from which ideas Satoshi's design evolved.


Yeah, 100% verification or repetitive guess and check problems such as bitcoin's (hash+nonce < target) are not typically of any practical value. But what I am envisioning are deterministic computational units. Any one of them could be recreated and verified by the 'buyer'. He can try to outsource verification to other parties if he could be sure there was no malicious collaboration.

While computers could be anonymous, their identifiers (addresses) could have reputation, and buyers and sellers rate each other, whose values would be individually weighted by a web of trust. A good rating held by a node at the center of a dense web of trust could in itself could be factored into a miner/computer's exchange price.

I've never designed a VM and know even less about distributed computing, but the model I have in my head are finite-state-machines (with state) or instruction/data stacks. The buyer sends the computer/miner a stack, buys a specific number of cycles and then receives the modified stack at the end whether or not the program completed. The buyer can then send the stack to another computer and buy some more cycles. At any point, the buyer could rerun the stack to see if he got the exact same state. Perhaps the buyer verifies 10% randomly and then 1% later if he trusts the computer.

Ultimately, it comes down to trust. Perhaps I can not trust a random miner on the other side of the earth, but I can trust a pool who has verified innumerable computations from hundreds of its 'sub contractors'.

well, why not just pay for miners to do folding proteins?

I'm on the fence about these arguments. If the bitcoin experiment were to catch on, then we will all claim its success as proof that a physical commodity were not necessary. Failure would seem to validate our dissenters, despite fiat paper providing prior proof. Neither scenario actually proves anything, just as a few generations experimenting with communism, anarchy, capitalism, or any other poli-economical ideology.

Comparing fiat paper (FP) and physical metal (PM) is interesting. Both have successfully been used as money and each has its advantages. FP is much more convenient while PM has 'historical staying power'. Bitcoin seems to be a more extreme case of FP. It is potentialy much more convenient, though its 'historical staying power' will always be questioned until of course history proves its power to stay.


I sympathise with this statement. I loath the brood of altchains. I think experimenting with new ideas is great, but I don't like the hack-y free-money environment. What it tells me though, is that altchains will never go away. Any (government/banking) institution could replicate bitcoin, throw in a few freebies for the creator, call it regulation and stability, and compete with bitcoin. I wouldn't bet gold on the winner.

It may very well be that I (or others) can not create a bitcoin like system in which the computation generates useful commodified work, can be stored (promise for future work), and for which the bitcoin units have an inherent value beyond their use as a money. If that is the case, then purchasing computational cycles, data storage, and bandwidth with bitcoin is functionally similar.

The differences (cycle=money vs money buying cycles) are quite profound though. Bitcoin as it is will forever be a speculative asset. It may have the best of FP and PM or the worst of both worlds. Its monetary stability may, like PM, act as an excellent store of value, or the innumerable altchains may inflate them all out of existence.

Suppose a single computational Cycle were our atomic monetary unit. A Cycle would have a direct cost related to technological innovation and the cost of energy, which I believe is the ultimate unit of value for living beings. There are many variables, but ultimately computation has three key aspects which remarkably parallel money:

RAM disk	: store of value
bandwidth	: medium of exchange
processing	: unit of account

As (if) energy becomes cheaper and technology more efficient, then the purchasing power of each Cycle would reduce (price inflation). On the other hand, the inflation would be very stable and predictable, currently roughly 50% annual inflation (Moore's Law and energy efficiency rates (Energy intensity calculated as units of energy per unit of GDP)).


Good question. That's the major whole in my thesis. But it's why this is a forum discussion and not a PDF manifesto.

Perhaps I have been arguing a contradiction in terms. A promise of future work is credit, not money. A money is not spent (consumed), it facilitates spending.

You may be on to something with Ripple. If it replaced its centralized aspects with a decentralized (bitcoin) ledger, I think Ripple would be more appealing. Thus far I haven't properly sunk my teeth into it. I think I'll give Ripple another go this weekend.

Thanks for the paper on delegated computation using homomorphic encryption. It's PRECISELY relevant to this discussion. I think I'll use their terminology from here on.

I can speak for neither Phillipsjk nor W. Dai, but I think "selling the proof of work twice" is short-hand for what researchers of computational delegation more simply called dishonesty:


In the case of bitcoin, you can imagine a miner sending their proof of failed work to multiple pools while keeping successful work for themselves. The pools have many ways to reduce these and similar risks, such as sending unique nonce prefixes, re-checking a tiny percentage of work, dividing work into pieces, etc.

I'm not particularly interested in generally 'more useful work' like protein folding and other involuntary donations. Sure, if bitcoin validation stabilized the climate as a side effect, that would be wonderful (though many here in Greenland may object). However, I am interested in arbitrary work. The promise to do MY work for which I'm willing trade quantifiable value. I want a universally valuable generic commodity (either good or service).

If a Satoshi represented the promise to compute a cycle in the future, then I can either 'cash it in' and get some processing power now, or just pass it around as credit notes to be redeemed far into the future.

Ripple handles credit while distributing/balancing the coincidence of simultaneous wants which plagues real-world barter. I can offer computational power, storage, but limited bandwidth and need tropical fruits. Someone in Africa can offer fruit and music and wants computation and to learn English. Someone in England can teach English, has bandwidth and wants music and storage. We can balance our wants and offers through Ripple and record the remaining credit balances and trust relationships in Bitcoin. ...or some such.

Computational cycles, bytes of storage, and bits of bandwidth are such tiny atomic units of value that for all practical purposes, as a service/commodity/money, they are infinitely divisible. They also lend themselves extraordinarily well to a distributed network - in fact they ARE the units of a distributed network.

I was. Thanks for pointing that out. I was feeling an overwhelming pressure to remain unbiased with respect to source.