Bitcoins coolest trick is its voting mechanism, it incentivises playing according to the rules and creates trust. We should reinvent the principle for smart Machine Learning Algorithms. This will allow us to introduce ‘facts’ about the real world into our Systems
If you have two major buzz words like Blockchains and Artificial Intelligence in an industry you can be certain, someone will try to mash them up — the Marketing sound is just too good. But why? My answer is: it’s all about ‘facts’: It is a great idea to marry Blockchains with Smart Machine Learning Algorithms, because smart Algorithms could allow Blockchain ‘the trust engine’ to reach out into the real world to establish facts: A neural network can detect a car, many neural networks with the right incentives will establish the fact there really is a car. Then we record the fact into the Blockchain and our smart Contracts rewards a false parking ticket. This is the idea, I will explain.
If you know what Decentralized Oracle’s are, you will find here a suggestion to address the question Ethereum Oracle’s raise: how to get data and facts into smart contracts. Most smart Contracts and with them platforms offering them are useless without facts.
Don’t expect a design or code or a prototype, this is just a pamphlet and it says: The ‘real deal’ of Bitcoin is the Voting Mechanism that converges towards the truth because we incentivise the network accordingly. We better have a 3-rd , 4-th and n-thlook at Satoshis trick .
So what’s ‘really’ that Bitcoin trick?
Have a look here, if you are not familiar with the Bitcoin Proof of Work, this is main idea. It’s not that Blockchain doesn’t matter, Turing Completeness helps for some use cases, but if you just replace that Mechanism with another Consensus Mechanism, you need to be careful what implications this has for the System. The Consensus Mechanism is cannot stand alone and is not the only invention of Bitcoin, but without it and it’s incentive scheme nothing would work.
This is what it does in a nutshell: Miners take a fingerprint, a hash of a Block with a set of Bitcoin Transactions. The result has to meet certain conditions (to be below a difficulty target, adjusted by the amount of 0s in front of the number), you add a random number, a nonce to the set of Transactions you collected, generate the hash and you probably failed to meet the difficulty target. You pick a different nonce, take the hash, check — until you made it. It is very hard to get there (brute force, try again and again) but it is very easy to prove you have the right nonce to add to the Block to hit the difficulty target. You won, pfff. It’s a mouthful, hope I got it right. You better watch that video and read a bit about hash functions, if that doesn’t sound like english to you.
If you do not win, you lost your electricity costs, because to run those chips to make those hashs, to mine you needed that electricity. If you do not play by the Rules, you try double spending or you are sloppy with how you generate your Blocks, you will definitely lose your money. Other nodes will just reject your Block. So, you are in for a race and if you behave you will have a statistically definable chance to win, if you win you create a Bitcoin and send it to your own address.
One way to look at the Proof of Work Mechanism is to see it as a prediction market. In a prediction market you bet on a future event. Since you put money on the table you are careful what you predict. These markets often turn out to deliver quite accurate predictions. The Bitcoin network offers a specialized prediction market in asking: what will be the longest chain? If you are a Miner, only if you operate on the longest chain, then you will have a chance to win a reward. Contrary to a prediction market, where you are passive and can really just bet, as a Bitcoin Miner you can influence, what the longest chain will be. If you win the race, your Block will lead the pack and you get to reward yourself some bitcoin.
So this is not about ‘Blockchain is the Bitcoin invention (the Banks)’, ‘Turing Complete Blockchains are Bitcoin spiritual heir (ethereum)’ or ‘You can choose whatever consensus mechanism you like (hyperledger)’. These strategy all have their merits, but what I focus on here is the mass voting mechanism, where voters can win and loose and will go with the mass, because only there they can win. The best strategy to be with the mass is to play by the rules.
Blockchains and Neural Networks for the facts
Now, for the combination of Blockchains and Machine Learning.
The Bitcoin Blockchain tells without any room for doubt, if transaction A happened before or after transaction B, between which addresses the transactions took place and which amounts were ‘sent’. They are actually not really ‘sent’ but just written into the public Blockchain. In the Bitcoin Blockchain this time ordered vector is necessary to avoid the ‘double spending’ problem, which simply means you can spend every bitcoin only once like we spend cash money only once.
Publishing and distributing the Data set public is a means to an end. If there was another design to avoid double spending, which does not use a central clearing house, that might (pure speculation!) well have been Satoshis choice.
But you can actually write anything into the Blockchain, it doesn’t have to be about Bitcoin’s Transactions. Satoshi him-/herself wrote a headline
‘The Times 03/Jan/2009 Chancellor on brink of second bailout for banks’
into the Genesis Block.
For example, you could put into the Blockchain ‘Brazil beat Germany in the 2014 world cup 7:1’. The Blockchain just records this into the Blockchain, it’s incorrect of course. A Blockchain doesn’t know anything about the world, its blind. It accepts whatever you put into it, even it is wrong. So that’s neat: it can be false but if it’s written into a Bitcoin Blockchain, you cannot rewind it. Not ideal, either rewind or make sure you get the facts right, because you keep them forever.
One use case of Machine Learning, specifically Neural Networks and Deep Learning is to simulate human senses and skills. Machines learn to ‘read’, in the sense of being able to summarize text or understand the sentiment of an author.
Machines also learn to categorize objects in pictures or movies. Give a Neural Network a cat picture and it tells you it is a cat, spectacular. Have a camera in a driving car and the Neural Network will detect objects like trees and persons and react accordingly, amazing.
With some training and a little phantasy let’s assume you could feed a Neural Network with an entire football game. It will detect, when a player tackled another one and it will understand and count the regular goals. Give it a camera and training and the Neural Network will ‘talk’ to you about the game. It cannot detect players, determine if there was a goal or a foul.
If the Neural Network detected a foul or a goal, it just puts it into the Blockchain.
‘Messi fouled minute 21.23 minutes, had dribbled around 5 Madrilenos. Barca crowd frantic for penalty.’
So, there is a really and capable neural network with a camera and it gives us this fact and we put it into the Blockchain, a fact.
But are they, how do we know this is true? Because we trust the Neural Network like we trust a referee, maybe a renowned organization guarantees for the validity of the Neural Network? There we go again, a Bank — a fact bank, but its the same ugly old middleman problem and we don’t want that. If we trust one Neural network, one referee we are just in the same situation as Bitcoin was as Satoshi out that headline into the Genesis Block. But then, why don’t we follow Satoshis footsteps if we have the same problem to solve?
Mass Vote with skin in the game documents facts
Imagine you are in a stadium with thousands of others watching a football game. During the break the speaker announces a phone number to send an SMS, into which you should send a ‘yes’ or ‘no’: was that tackle in the box a foul, so penalty? The games of the rules are: if you vote like the majority, you will participate in a lottery for a free ticket for the next game.
You could consider different strategies. If you were in a small stadium with locals, you probably are best off to check, with which team the audience sympathizes, because they will vote in favor of them no matter what actually happened. Maybe the sight of viewers could be a factor, You just know 80% of the spectators is seeing the field from very high up and they will not be able seeing that little push that you saw while standing right next to the sideline behind the fence.
But the larger the audience and in the overwhelming majority of the cases, you are best off to just vote as the facts present itself. The scene is repeated on the video screen and everyone can see it was a foul, so you vote foul and penalty and will have a chance to win.
This is a situation that can be analyzed with well known game theoretical concepts. Every participant chooses the strategy that is ideal for her or his benefit, even if the participants did not explicitly coordinate. If they had and had revealed their strategy, they would stick with the strategy. The strategy for our spectatiors is to use their best judgement to describe the facts.
Now take our little example to a more abstract level: our audience are actually some neural networks with a Camera. They are all over the stadium, someone let them off the leash.The question we have is: is it a foul yes or no? So, we don’t ask the Bitcoin Proof of Work question: give me that nonce that is below the current difficulty target and you get the right to write in a new Bitcoin and get a reward. We tell the neural network with its camera: Give me your answer, if this was a foul. If you vote as the majority of your fellow neural networks did, we will entitle you to participate in a lottery to win a price. The rewards must be high enough and the statistical distribution favorable enough to get you ‘do the work’, that is process that picture and tell me if it was a foul. For all networks participating, their best strategy is to go with the truth, because that is where you can expect the majority vote.
Important: Reuse Bitcoin’s voting Mechanism, create a prediction market participants can actively influence to their benefit with the behavior you want to see: tell the truth.
With ‘the truth’ I don’t refer to the relativity theory or a question like ‘ what is a just economic system’. I am talking simple facts, which we can already or hopefully soon automatically process with Artificial Intelligence.
Stay tuned for part 2, where I will say why I think this works, even though Cryptography and Machine Learning have really nothing to do with each other.