Blockchain Provides The Perfect Balance To The Potentially Unbridled Power Of AI

It’s only a matter of time before we create an AI that surpasses human intelligence. Estimates for this event known as singularity range from anywhere between ten and fifty years from now. What happens next is anyone’s guess, but there are two prevailing theories – (i) humans are fast-tracked for more technological innovation, and (ii) extinction at the hands of a sentient super AI who is indifferent to our survival.

There is more than just one category of AI –

There are a few distinctions when talking about advanced AI recognized as follows –

1. Weak AI – On the lower end of the intelligence spectrum, there’s weak or narrow software that is useful within a fairly limited range of tasks. Alexa, Siri, and all the other personal assistants are the most obvious examples, but we are actually surrounded by narrow AI. Your Netflix caters to your taste in movies and TV, Amazon suggests what products you like based on your browsing history and previous purchases. Machine learning algorithms represent another form of narrow AI. Any program that can adapt based on user interaction can be lumped into this category. This type of AI is rarely considered a threat. Take Siri for example. There is no real intelligence there, no mind, nor self-awareness. It will answer some questions for you but is useless outside of that.
2. AGI – Moving up one level, we have artificial general intelligence known as AGI. This type of AI is able to apply its intelligence to a much wider range of applications and can successfully perform any intellectual action a human can. This level of AI is generally the goal for projects aiming to develop advanced intelligence, and academics- call it strong AI.
3. Artificial superintelligence – Next, we come to the version of AI that could potentially spell the end of mankind – artificial superintelligence. A superintelligence AI would be smarter and more capable than all of the human minds put together – infinite knowledge, the ability to perform multiple tasks at once, instant answers to questions we haven’t even thought of yet. The biggest question we have thought about AI is whether or not it will be friendly. How would an entity a million times more intelligent than humans treat us? We tend to disregard the danger because we assume AI will be human-like, but if it saw humans as a threat to its own existence, it could take any step to protect itself. How can we control an entity that has access to all our knowledge and more, can control our computer systems, can create and destroy at will using our already existing infrastructure, or worse, by building its own that is far beyond human understanding? A sentient software based AI could travel instantly between electronic devices and inhabit many of them at once, making it an immense challenge to contain such a situation were it to go out of hand. Unless of course AI is combined with another breakthrough tech innovation – blockchain.
   

How can blockchain address the drawbacks of AI?

Although the integration of AI and blockchain is in its nascent stage with software development companies such as HashCash Consultants beginning to introduce combined services, distributed ledger technology AI and blockchain are based on polar concepts. One centralizes and analyzes vast amounts of data, the other is inherently decentralized. However, it can be argued that blockchain can mitigate some of the problems that AI faces, such as,

Blockchain and the Black Box Problem

The neurons in the neural network are interconnected through the layers. Each neuron has a threshold called a “weight” where it makes a decision. The network could take in a massive amount of data for training.

Unfortunately, this does not always give positive results. To counter this, there is a process called back propagation where the correct result is introduced to the neural network when it produces incorrect output. This information is sent back through the network to train it a little better. In many of these cases the accuracy is enough, but when dealing with more critical decisions like driving autonomous cars, doing medical diagnosis, or pushing the frontiers of science to understand why a discovery was made, it becomes necessary to know what the AI is thinking, not what the results are. Since there are all these layers it just engages in such complex decision-making that it is essentially, a black box. One does not know they think.

However, all is not lost in this black box problem. Immutable and transparent records of all the decisions taken by AI can be attained by a blockchain. Studying the process of how an AI system arrived at a particular conclusion makes it possible for human operators to detect instances where their involvement becomes mandatory and where AI fails. Blockchain, in other words, with make it easier for engineers to cope with the outcomes and enhance the predictability of AI systems.

Blockchain and Data Reliability

A neural network’s efficiency depends on the data that it learns from, and the most reliable data is expected to satisfy multiple requirements pertaining to quality and quantity. These requirements are difficult to meet in numerous cases at the same time. If a network is trained on an inadequate amount of data, its output will have low accuracy, and if it is trained on unreliable data, the output will be biased.

Blockchain solves the issue of low accuracy since it can provide a massive amount of data, and checks bias since blockchain validates all the data by default. This way, blockchain reduces the scope of mistakes to occur by improving the quantity and quality of data.

Blockchain and Overfitting

This issue is closely related to the problem with data reliability in AI. When excessive data is fed to a neural network, it begins to detect random and insignificant patterns leading to incorrect conclusions. Consequently, the AI system performs better on historical data than when it comes to making predictions on new data. Overfitting, in mathematical terms, indicates that AI tends to ignore off-the-curve data points.

The blockchain space has recently proposed rectification of the overfitting problem in a form of a token intended as an incentive for data scientists. The goal is to encourage them to create models that can work more efficiently on new data.

So, what happens when AI becomes more powerful?

Honestly, AI could be designed as a benevolent force whose only task is to improve human life. No one really knows what would happen if we reach this singularity, except that it would change the course of human history.