Hello again and welcome to part 2 of The Merger of AI and Blockchain: Two Great Hypes that Hype Together. Check it out if you haven't done so already.
Blockchain Can Get the Data AI Needs
A blockchain is a decentralized network hosting a distributed ledger -- literally, it is a list of data written and hosted in a very particular, non-centralized way. This means that a blockchain could host data that is then used by a machine learning algorithm to develop an AI, and there are a number of advantages to developing AI that way.
For one, blockchains are highly democratized and secured, meaning that blockchains can allow for public contribution of data while still making the intentional insertion of bad data far less likely. This has been a big problem for AI -- just ask Microsoft. With blockchain it would still be possible for users to slip in bad data, but that data would be visible to the community and thus able to be vetted for accuracy. Smart contracts can administer the collection of data, allowing only certain forms of data from certain sources in certain volumes, greatly speeding the collection process. And since the blockchain should be permanent and transparent, it means the dataset remains available for the development (in parlance: “training”) of a new AI that needs to develop the same ability.
The other issue is that in a blockchain environment, users can be incentivized to provide the data needed for any particular AI project, even if it doesn’t spark significant user excitement all on its own. Cryptocurrencies allow blockchain platforms to financially reward users who contribute data of a particular type and in a particular form. This makes it far easier to collect large, curated datasets that require significant amounts of human work -- blockchain makes it possible for digital economies to spring up and provide sustainable support the evolution of AI.
The most prominent example right now is probably DeepBrain Chain, which uses blockchain to secure and incentivize the posting of both neural network/machine learning code and properly annotated datasets for machine learning. By using blockchain, DeepBrain Chain can track access to posted material, ensuring payment, and provide further capital gains through increasing token value. It could be a solution to the most pressing problem currently facing AI: how to get all the data needed to train them.
On the other hand, sometimes the difficulty of collecting information arises from the fact that the information is highly sensitive. Patients generally don’t want money in return for their medical info, but they do fear the improper use or protection of their data. With a blockchain platform, anonymized data can be kept totally secure, so only very specific recipients get to see it. More importantly, information kept on-chain can be more easily controlled, so a permissioned reader can’t re-transmit it to non-permissioned parties, even by mistake. That makes it far, far easier to actually create medical industry AIs, since blockchain can allow practical collection of datasets.
But AI needs more than data -- it also needs an incredible amount of computing power, and blockchain can incentivize the sharing of computer time just as easily as it can the sharing of data. Tatau is just one startup aimed at using blockchain to run a decentralized network of computers available to rent for machine learning jobs, while others like Subutai and Golem allow sharing of computing resources for virtually any purpose. Got a great idea, and all the curated data you need to implement it via an artificial neural network (ANN)? You still need the computer time needed to put all that data to use training a neural network. Cloud computing providers are turning to blockchain to make sure that it’s not just large tech companies that have access to the computing power needed to make world-changing machine learning products.
So, that’s the needs of AI, as seen from the perspective of blockchain tech. On the other side of the equation, we have the needs of blockchain, and the ability of AI technology to meet them. Blockchain has data needs too, this time for quality data rather than quantity, and AI is by far the most promising source of quality data anywhere in the world today.
AI Can Provide The Insight Blockchain Needs
Blockchains exist behind a tightly controlled informational border. Inside the border, we have the blockchain’s version of reality, called consensus; outside the border, we have the real world. The advantage of the blockchain reality is that it is reliable -- the truth is the truth, and every node hosting the blockchain understands the exact same version of it. In the real world, we have competing standards for truth, both between multiple sources and across multiple periods of time.
In order to function properly, every node in a blockchain needs to agree about the value of X, so if the value of X changes by the hour, how are we supposed to keep everything in agreement? What happens if, looking out into our big chaotic world, one blockchain-hosting computer sees a different value for X than another, even for an instant? What happens when the blockchain has competing standards for truth? Problems, that’s what happens.
But in order to be useful, a blockchain does have to be able to take in some level of information from the world at large -- to do that, it needs what’s called a “blockchain oracle.” Oracles are simple blockchain smart contracts that have unique authority to report new information into the blockchain, making them incredibly important to a whole range of applications. An oracle is required for any application that requires a blockchain to be capable of incorporating new values for data on an ongoing basis, for instance, a service that takes an action when a particular stock reaches a particular price. The price fluctuates throughout the day, but the blockchain requires new values to enter in an orderly fashion -- that’s where blockchain oracles come in.
This all dovetails very nicely with the strengths of AI: gobbling up large volumes of information, sifting through a diversity of forms and values, and finding some hidden pattern it holds. Neural networks can boil down a whole internet’s-worth of social data to a single multivariate statement that can be easily incorporated into a blockchain’s internal reality.
In other words, a blockchain is only as good as the accuracy and usefulness of the data it hosts. Right now, only neural networks can do the complex analysis necessary to provide that accurate and useful data -- either the biological neural networks of human data entry specialists or software ANNs of modern AI. Using AIs rather than human employees obviously leads to much faster and cheaper data entry, allowing the blockchain to be truly reactive to the world around them. A non-AI-powered blockchain oracle can watch for a particular integer approaching a particular value -- but an AI-powered blockchain oracle can watch some large array of disorganized human behavior and deduce when that integer has reached that threshold value.
That’s a far, far more powerful ability to have, and one that will be absolutely crucial if we’re ever going to see prophesied future-tech like blockchain-powered smart cities that react intelligently to the behavior of the city’s inhabitants. If we’re ever going to enjoy truly next-gen blockchain supply chain management, it will only come from an ability to react to changing conditions in real-time and push out the implications of those changes as efficiently as possible. For that, we need AI.
But AI can also do its analysis on the contents of a blockchain, rather than delivering its analysis into those same contents. This allows AI to analyze the transactions on a ledger and watch for evidence of malicious activity. Empowered to halt transactions that look suspicious, and we have an on-chain neural network cyber-security guard. Pretty dang cool.
But, let’s be honest. Nobody is interested in the idea of blockchain AI because of supply chains. The fact is that blockchain is a uniquely powerful platform for hosting and executing code -- and AI is a uniquely powerful form of code that can be hosted and executed. It seems like a match made in heaven: an AI actually runs and learns on a blockchain platform and, in a very real way, a decentralized brain.
The Blockchain As Machine Learning Platform
There’s a pretty obvious overlap in terminology between blockchain and AI, and that’s the word “node.” A blockchain “node” is a computer, one helping to run the blockchain, while an AI node is a simulated neuron, a computing unit within the larger neural network. They perform very different functions in very different ways, but they are networks of independent computing units -- could we marry the two concepts? Could we have an AI in which each neuron is a computer, and each synaptic connection between those neurons is a smart contract-secured token transaction? Could the passing of an input from one neuron to another in a neural network be handled as a token transfer?
All this stuff is extremely preliminary. There is simply no such thing as a truly blockchain-based AI at present -- but it could be just around the corner.
Google recently unveiled a concept called Federated Learning, which allows disparate machines (even mobile phones) to do machine learning on data even while that data is distributed between all of the machines doing the learning. In other words, phones can collect data on their users then usefully contribute to a machine learning process derived from that data, without needing to actually transmit that data to any external party. That helps enormously with both privacy and energy consumption, but it also means that there’s generally no need to centralize machine learning data in a single monolithic dataset. Now, a decentralized collection of devices can collaborate toward training an AI. Sounds like a job for blockchain!
Machine learning techniques are just now becoming versatile enough to apply in a context as unique as blockchain, but they’re getting there. It’s hard to overstate the potential of combining the strengths of these two new technological paradigms -- so much so that a decentralized AI is a very good (or, bad) candidate to become the sort of rampant, world-destroying AI that has obsessed science fiction since the concept first arose. A conventional computer program is still limited by its nature as a digital entity, as tied to the survival of a particular device as a human being is to the survival of their physical body -- but an intelligent neural network running on a decentralized platform like a blockchain could be incredibly difficult to shut down. And if you want just a little more nightmare-fuel, imagine if IP-anonymizing technology like deep web “onion routing” could be applied to the traffic between nodes in a blockchain -- it would be very, very difficult to completely stop such a thing without shutting down the internet itself.
What Blockchain-AI synergy means for business
A lot of the more out-there applications of AI in the blockchain space that will probably remain in the realm of fantasy until the neuromorphic hardware revolution begins. This is the oncoming wave of computers that are physically structured like a neural network, built to embody the types of organization that conventional computers simulate in an ANN. They are orders of magnitude more energy efficient at executing neural network code than a digital computer, meaning that machine learning can finally fulfill its entire potential.
Both the blockchain and AI revolutions have been enormously important to entrepreneurs already, allowing more more and more complex services to be rolled out by smaller and smaller challengers. Now, AI’s enabling abilities make many blockchain applications possible for the very first time -- and blockchain is making AI profitable in all-new ways. There’s still ample room for new entrants to a space as replete with money as this one, new combinations of technological strengths that have not yet been dreamed.
Make sure to keep an eye out for investment opportunities with companies or individuals poised to take advantage of this trend, but also watch out for over-use of nonsense jargon. If you understand the very basic mechanics of how AI and blockchain function, then their interactions should be fairly simple to imagine -- and the potential of some such interactions to change the landscape of the tech industry forever.