Will Blockchain Unlock the Next Generation of AI?

It’s conquered Go and cracked the labyrinthine code of human language but, on a global scale, the technology of artificial intelligence is still maturing. To get to the next generation of AI, we need to equip it with accountability and democratize it with scale. AI that is able to show us exactly how it makes all decisions, will allow us to understand its reasoning. AI that is lightweight, yet powerful enough to efficiently apply to applications that are distributed across multiple nodes.

Once we get a good grip on these two, we can really teach AI to be able to handle conflicting inputs, the way humans do, and let it loose over infinite knowledge, like the internet. These are huge challenges but, in the blockchain database system, we might find the key to their unpicking. Before we get into how, let’s revise a few ABCs:

• Artificial Intelligence: In its current state of art, AI refers to computer programs that are trained and used to predict outcomes. AI loves data – the more data it is trained with, the better it is at making predictions.
• Blockchain: Blockchain is a distributed digital ledger that can manage almost any type of transaction. In essence, a blockchain-enabled database stores an unchangeable history of how any variable got its current value.
• Cognitive Computing: Cognitive Computing is the genus of artificial intelligence specifically written to work with unstructured datasets, like natural language. It makes predictions by emulating the process of human thought.

Putting it together

Okay, let’s put these together. The job of any kind of AI is to identify the relationships between input data and output data. For example, it can predict which shows you will like by looking at your viewing history, or when are you likely to buy season tickets, by looking at call transcripts. Understanding of these relationships develops gradually with each set of training data, until the AI is capable of making accurate predictions.

But, often, training data contains conflicting inputs, which can make life difficult for the AI. Consider an AI that has been trained to understand the predictive relationship between the clauses in these sentences:

“The sky is grey, so it is about to rain.”
“The sky is grey, so a smog is about to set in.”

Both are logical, but apply to two completely different scenarios. How is a machine to determine what happens when the sky is grey? There has been some progress with concepts like Cognitive Calibration, but, in general, if you ask most of today’s AI what happens when the sky goes grey, it will conclude that there is a 50% chance of rain, and a 50% chance of smog. Which, clearly, doesn’t come anywhere near to capturing the true contexts at play.

If the training process had been stored in a blockchain, the machine would be able to make a much more intelligent decision. It could dig back to the training inputs that brought it to its current dilemma; it could see that the first sentence was uttered in San Francisco, and the second by a someone far away in smog-prone Delhi. By recording the history of training, we’re making it incredibly easy for the AI to find new dependencies. It can remember that San Francisco associates to rain, and Delhi to smog when sky is grey; couple that knowledge with an understanding that the agent running the query is currently in the Bay Area; and provide a more accurate answer (high probability of rain). In short, blockchain helps AI make sense of conflicting inputs. By so doing, it represents a significant step towards machines that solve problems like humans.

The question still remains, why use blockchain? Even if this idea is worth pursuing, why can’t we use a simple register of transactions?

This is where magic begins. Being a secure, decentralized database, a blockchain-trained AI can scale almost infinitely. It can collect and train on information from any source in its database, and grow exponentially.

Blockchain also solves the accountability problem, providing a relatively simple, highly secure means of registering and understanding the combination of inputs that led to an AI’s action. When we combine blockchain with cognitive computing in particular, we reach a very powerful solution.

Of brains and blockchains

In Coseer’s approach to cognitive computing (called tactical cognitive computing), the system begins by assuming all possible answers are equally probable. Then, a mix of constraints are added from other known information. This tells the system that some answers are more probable than others, allowing it to narrow down its search. Eventually, one answer wins out – this, in turn, becomes new information the system can use for finding other answers.

While there are multiple iterations of this process, the number of steps dwarfs compared to traditional AI algorithms like Deep Learning. Simplest of Deep Learning algos, which run on neural networks, have trillions of steps, which make storage of transactions that much harder. This is why tactical cognitive computing fits so well with blockchain technology. It is tangible each step of the way. It can cope with the real world as it is. No human has to sit around for months or even years labeling that information first.

As we have written before, these strengths of cognitive computing make us ambitious. Coseer’s ultimate ambition is to create a machine brain; a program that processes information using the same paradigms used by human cognition. Assimilating the technologies of blockchain and cognitive computing will be critical – and we’ve already begun writing the software, Cobrain.

Building upon blockchain, we can teach AI how to better handle conflicting data. Then we can scale a single intelligence across multiple contexts securely. This promises to unleash a new wave of collaborative and amazingly powerful AI.

For each of our clients, across a multitude of contexts and locations, Cobrain is going to be solving a matrix of business problems; from difficult single tasks to entire workflows and functions. When all this data is merged into one central intelligence for each client, there is bound to be an array of conflicting learnings, on a scale thousands of times more complex than the ‘sky is grey’ example above.

At that point it is crucial that a history of transactions agreed upon by various nodes exists. Only a radically decentralized blockchain database is capable of creating an accessible, reliable ledger on this scale.

By offering solutions to the questions of accountability and scale, blockchain is certain to accelerate AI’s predictive power. Imagine these advances not just on the scale of one organization, but that of one planet: cognitive computing running on a blockchain database could, theoretically, train on the entire internet.  

The prospect is exciting and, thanks to blockchain, within our grasp.