Where, why and how AI, cryptocurrency and blockchain come together

The future is here, and it's a doozy.

Cryptocurrency, distributed ledger technology and artificial intelligence (AI) have a lot of overlap.

First, cryptocurrency mining and machine learning cycles are extremely similar. Crypto has already delivered a way to directly use computing cycles as a currency, which helps open the door to much quicker development of machine learning systems. There are already signs that the future will feature a global machine learning cloud computer network where machine learning warehouses, similar to bitcoin mining warehouses, crop up to educate future generations of robots on demand.

Beyond that, you can also see how AI firms are using cryptocurrency and blockchain technology, and how developments in different areas can be used across multiple industries.

Just look at image recognition

CloudSight is an image recognition AI firm that's using the bitcoin network to decentralise its machine learning and let its products better serve customers. It's currently geared towards relatively straightforward applications such as reliably and automatically captioning images on online marketplaces, but is eyeing cryptocurrency technology to get to the next level.

To that end, it recently announced its support for bitcoin lightning network payments, unlocking the ability to handle digital microtransactions to speed up development and open new use cases.

"Bitcoin's utility extends far beyond its monetary value. The secure, decentralised technology enables two parties, including AI, to engage in a trustless transaction," it said.

"These features are critical for autonomous entities to communicate with each other, enabling information exchange in a new AI-based economy. Should apps, autonomous AI, or any other application need to gather visual knowledge, CloudSight will now be able to serve and exchange with these queries through this new monetary medium."

Assuming the bitcoin lightning network ever gets off the ground, the ability to instantly, privately and securely process microtransactions, including pure data transactions as well as purely monetary transactions, is valuable.

Decentralised eyeballs on the blockchain

It all comes back to computing power. Image recognition is undeniably valuable for robots, but building those eyes into a robot involves building in the cameras, the computing power and systems to work out what those images are as well as data storage for the library of images a robot could conceivably recognise.

Decentralised computing lets individual robots outsource their image recognition, basically just sending over a picture of what they see for off-site processing and recognition, and then getting back the answer as well as a set of appropriate instructions for how to respond to it.

This could be done in a fashion with existing cloud computing systems, but blockchain technology can introduce some key elements.

The first is a secure network to transfer information, where it can be mathematically proven to be tamper-proof and secure, allowing it to be used for a much wider range of applications.

The second is a system that can simultaneously transfer monetary value along the same network. The off-site image recognition robots aren't going to work for free after all, and it's useful to have this on the same network.

Separate payment systems (your robot identified x objects in the last month, your bill comes to y dollars) are unwieldy, inefficient and expensive, and prevent these systems from being widely used on-demand.

The system fabric

The pieces for creating these kinds of systems are all vaguely there, but it's hard to find the right fabric to hold these systems together.

Finding that fabric is one of the main end-games of that "Internet of Things" thing that's so in vogue right now and for the foreseeable future. AI is one key piece of the puzzle, connected data-collecting devices are another and the network architecture of this vast system is a third.

It's increasingly clear that distributed ledger technology is the best material for this fabric.

With the correct decentralised architecture, one can set aside, or at least automate, many of the concerns around data privacy. But just as importantly, cryptocurrency in the form of digitised units of monetary value is key, and distributed ledger technology is the only possible way of making it happen.

This is because it's the only possible way of turning data into something finite, non-replicable and individually identifiable. In other words, it's the only way of directly putting a monetary value on data itself. For example, it might seem like you're transferring money electronically when you swipe your card or make an electronic transfer, but you're actually just creating a record of payment, not moving value itself.

Both are functionally similar to the end user, who just wants to buy their muffin and coffee and move on with their life, but in practical terms they're completely different, and you can't have that fully autonomous universal mega hyper-future without it.

Why crypto is key

Cryptocurrency is a token that can be worth things. You already have tokens that are worth a certain amount of computer rendering power, or that can be directly redeemed for electricity, or mobile phone data, or spins of a roulette wheel, or computer network services, or a representation of the ownership of a share of an asset, or commercially valuable data, or seconds of human attention on advertising or a crowd-sourced prediction.

All of those examples are already live and functional, and can to a certain extent be exchanged for those types of resources as intended.

All of them are cryptocurrencies, and all of them have monetary value, whether it's denominated in dollars, cents or satoshis.more info button

Monetary value is a shared language for indicating the value of very different things and for allowing the seamless exchange of different things. This is what it has always done, and whether you're exchanging a certain amount of labour for dollars, or a certain amount of data analysis for a certain number of joules of energy, the principle is the same.

You don't actually have to exchange these things via monetary units like dollars. It's traditionally been a convenient option, and it saves you the trouble of needing to find a buyer for whatever you just bartered up, but practically the dollar can just sit on the sidelines as a means of denominating the value of each side of the equation and making sure both parties are enjoying a fair trade, even if it doesn't actually do anything.

Consider the previous example of Bob the driverless car, who notably lacks the pockets to carry cash and probably isn't eligible for a bank account. And consider the many other robots that will be dependent on external machine learning and image recognition centres, and other things not yet imagined, to enjoy full functionality. There's no reason to futz with some intermediary unit of monetary value just so the banks don't feel left out, and it's certainly not feasible to have some human third party sign off on a robot's every thought.

Why not just let Bob pay with units of electricity instead? He needs electricity, the third parties need electricity, it's win-win.

Wireless electricity transfer will probably be making an appearance in this hypothetical-ish future, probably in some way that lets Bob charge up without ever pulling into a charging station. The transfer will just be unit data that can be redeemed for a set value of electricity – or a cryptocurrency as it's called these days.

Bob the driverless car isn't just a chauffeur, and his commercial interests don't stop at being an energy broker. He's also a very keen data-collector, and people are willing to pay for his insights.

Consider what happens when you look up the weather forecast.

The initial data collection is carried out by countless individual weather stations, and then routed through data centres which use this information to produce detailed forecasts for certain areas. The most powerful supercomputer in Australia is currently owned by the Bureau of Meteorology and is being used for this purpose. When you call up the report on your phone, it asks the supercomputer what the weather forecast for your area is.

There are two main parties involved. The first is the weather institutions such as the Bureau of Meteorology. They do all of the hard work and foot the bill, kindly picked up by taxpayers, for setting up and maintaining those individual weather stations, gathering and collating all the data and then using a system for translating that data into a forecast.

The second party is whoever runs that weather app. They pull that data and pipe it your way through the Internet and might pick up some advertiser dollars

This is a very oversimplified version of course, and it naturally varies by country, location, situation, existing commercial agreements and so on, but in broad strokes this is how it tends to work.

Consider the efficiencies that could be created by doing the following:

• Letting a wide range of devices record data to feed into this system, such as thermometers built into phones, air quality sensors in homes and hints from Bob the car. This allows the system to have a much wider variety of more detailed data.
• Being able to, as needed, access enough cloud computing power to perform high priority computations of this additional data. For example, it could be used to more accurately predict the path of a hurricane.
• Creating a system that can constantly learn from all its previous computations, and over time develop increasingly accurate weather forecasts based on the extra data available.
• Decentralising the computing power used to process all of it and assemble a coherent picture.

It's obviously easier said than done. So far, there isn't even any functional decentralised network for handling these kinds of problems, and different organisations are looking at different options.

CloudSight sees the bitcoin lightning network as a likely option, while Volkswagen, Bosch, Fujitsu and others are investigating the potential of the IOTA tangle.

What does a robot look like?

Not long ago, robots were seen as mindless, jangly, tin-man humanoids. More recently, they turned into tireless automated manufacturing systems, mostly performing highly specific repetitive physical tasks while remaining bolted to a factory floor. More recently still, they learned to move around. Now you can find humans and mobile robots co-working together at Amazon and elsewhere.

But those are just their robo-bodies, not their robo-souls.

The real action happens in the lines of code that keep them running as intended, and they don't have to be as closely connected to the physical world as humans are. With the physical realm turning into a kind of tip of the iceberg in the digital age, there's going to be no shortage of jobs for the ephemeral robots that are just lines of code, able to fly around anywhere in the world. Already, different car companies have created and hired their own ephemeral robot armies to live inside cars and perform small tasks like checking the engine temperature and preventing people from dying horribly.

These robots are more commonly called "software," and you've probably heard of them before. But add artificial intelligence and you're got yourself a high-risk party.

A dangerous recipe

Deep Patient is both a doctor and a patient at Mount Sinai Hospital in New York. It's a subject of a study in its own right, but is also a remarkable practitioner who's able to diagnose various cancers and even the onset of notoriously tricky-to-predict mental illnesses such as schizophrenia before anyone else can. But that jerk won't tell anyone how it does it, and no one can ask it.

That's because Deep Patient is an AI that applied deep machine learning techniques to the hospital's patient records. That loose cannon is damn good at what it does, but no one has any idea how it does it, how it works or what exactly the rationale for its prediction is. Even if it could save lives, which it clearly can, it's not possible to use it because there's no justification for its decisions, no one knows how it works and it's just a medical malpractice nightmare. Imagine a doctor arguing "the robot told me to do it" when asked why they killed a patient with an unprecedented medication change.

There's always the risk of robots turning the world into paperclips, but in the more immediate future it's more important to find a way of tracking the learning processes and creating systems to protect users, while making AI more useful in the real world. This will become especially important as AIs start worming their way closer into everyday life, and people naturally start trusting them more implicitly and leaning on them more.

But what happens when you download knock-off AI software, or your hired AI sells your business data? What if you don't know if two AIs are compatible with each other, or when your small business accounting AI starts diagnosing schizophrenia with unparalleled accuracy?

Once again, the solution might be on the blockchain. An immutable record is the perfect way of recording decision-making, and there will also be a clear need to bring disparate systems together.

"There's a common misconception that there will be general AI that is broadly useful on a huge range of tasks in the next few years," says Rob May, CEO of BotChain. "In reality, and arguably more helpful, the most notable progress on AI has been for narrow use cases. So, an expert on your company sales data is much more useful than a sales FAQ bot that pulls from publicly available data."

"Due to the level of expertise and the datasets available for each use case, there is going to be more than one clear winner in the AI space," he predicts.

"Bots and AI products are most useful when they are able to be proactive. Similar to human co-workers or service representatives, the most ideal situation is when someone can both anticipate what you need and deliver it to you at just the right time. With proactivity, and with AI behavior changes that come with machine learning over time, you must have both an identity of that bot and the inspectable record of the actions it's taken previously. Without that, there are huge implications for trust (and therefore, continued adoption) of AI."

May envisions a future where people can rent, hire, buy or sell software AI where needed. He sees a future where the robots can train on the job within predictable time frames and continue to add value to any task that involves data.

And remember, inspectable AI records like those produced by BotChain, and even the AIs themselves, can be transferred along the global Internet of Things like any other data and can carry monetary value in their own right the same way cryptocurrency can.

That super, ultra-hyper robot future isn't here yet, but with all the advances recently made in artificial intelligence, distributed ledgers and cryptocurrencies, we seem to be looking at the pieces of it.

All that's left to do now is put them together, and then rue the consequences of humanity's hubris.

Disclosure: At the time of writing, the author holds ETH, IOTA, ICX, VET, XLM, BTC and NANO.

Picture: Shutterstock