Incentivized Machine Learning through Crypto-Economic Principles
By now I’m sure you have heard the phrases Proof-of-Work and Crypto-Economics. In this paper, I’ll cover how P-o-W secures an individual network for thousands of participants running full nodes. I’ll also showcase the flaws in using a nonce to iterate hashes and how this method can be improved. These powerful ideas are fundamental in understanding the benefits that Crypto-Economics has to offer. Without mining there is no Bitcoin. We rely heavily on miners to secure payments and add new blocks to the longest chain.
As of now Bitcoin works like a 777 slot machine. Every hashed block submitted is analogous analogous to a spin on the slots. 99.9999% of the spins are useless and result in a hash well above the target hash. A target Hash is what Bitcoin uses to determine whether or not the next block will be added to the longest chain. As more miners join, this target gets lower and lower to maintain a certain level of difficulty.
At the present moment, these hashes contribute nothing more than verifying transactions. I ask of you, would one not rather see fit that this Immense computing power go towards a more useful alternative? Perhaps a region of computer science where computational power is currently lacking? Well, there are hopeful signs in Bitcoin Core’s codebase that point to such a reality. Since every guess at a new block requires an iteration of the nonce, it is possible to reverse engineer this iteration process to solve a new puzzle.
We must first imagine how a blockchain comes to consensus on each new block. A host computer running Bitcoin Core signals the Bitcoin network once a new hash is made. This hash is looking to meet some specific requirement. In our case it’s that a hash is less than 000000000000010000. This number is what we call the target hash. If our host computer reaches an iteration that is below such a hash, a prize is dispensed in the form of 2.5BTC being added to that user’s address. However, this method of iteration is inefficient and costly. Bitcoin alone consumes as much energy as the entire nation of Czech Republic. But what if there was a way to using this hashing in a positive light. A field of research deeply lacking computational resources… enter machine learning.
What if we could reverse engineer the hashing functionality of cryptocurrency to solve an image recognition game instead of reaching a target hash? Computer scientists would then be able to train their models on a decentralized network, as opposed to using time on a national supercomputer. It’s a riveting idea and one that holds much potential in the crypto-economic community. Yet, there remains a downside. To replace the consensus algorithm of a running cryptocurrency is analogous to performing spinal surgery on a patient. However, the upside is worth the challenge. Incentivizing miners to perform computational tasks on machine learning data opens up the door to a whole world of possibilities. Even the ones yet to be discovered.