Anyhow, I feel it’s time to take other hypotheses into consideration, such as the role AI can play in cryptocurrency’s worldwide adoption. I agree this is not an easy task as most of us are not technical people and our understanding of how technology works and interacts with users is, in fact, limited.
Nonetheless, my goal today is to analyse some AI tools which can assist cryptocurrency projects in deploying better products, with more intuitive interfaces. We will discuss four types of tools which can be combined with natural-language processing (both with known and unknown goals), with a relatively simple aim: to allow any agent to better interact with blockchains.
Lastly, we appreciate the importance of data visualization as a key ingredient for users to understand blockchain data better.
Like having an epiphany, it now seems obvious that the influence AI-enabled tools, such as oracles or sovereigns like bots and virtual assistants, can have in cryptocurrency usability. So many issues arise from bad user-interfaces and smart contract code bugs, I feel implementing autonomous preventive measures around smart-contracts auditing or creating interactive user-focused bots could definitely lead to a serious improvement in adoption. How can users benefit at the end? Let’s find out below!
–This article shouldn’t be taken as financial advisement as it represents my personal opinion and views. I have savings invested in cryptocurrency so take whatever I write with a grain of salt. Do not invest what you cannot afford to lose and always read as much as possible about a project before investing. Never forget: with great power, comes great responsibility. Being your own bank means you’re always responsible for your own money—
If you want to understand how AI learns and improves its knowledge, you should first read about artificial neural networks. It’s the latest and most up-to-date research on machine learning.
Neural networks (and natural language processing algorithms) allow machines to learn much like we humans do. Particularly, they are inspired by the behavior of neurons and the electrical signals they convey between input (such as from the eyes or nerve endings in the hand), processing, and output from the brain (such as reacting to light, touch, or heat). In a sense, neural networks take a heuristic approach to mechanistic learning, as they promote failure as a method of gaining knowledge. Key concepts to study, if you wish to widen your understanding of neural networks, are Game-theory, utility maximization, and optimization.
I won’t describe all types of neural networks, for obvious reasons, although I do want to mention the most important ones, which can be used in natural language processing (NLP). NLP can then be used by AI-enabled devices and algorithms to learn how to translate from language to language, text-to-speech, image-to-text or voice-to-text, by using common machine learning techniques — such as the Monte Carlo simulation and the application of random walks.
The ultimate goal of neural networks is to endow machines the power to learn human logic by trial and error.
In MLP networks, every node in a layer connects to each node in the following layer making the network fully connected. For example, multilayer perception natural language processing (NLP) applications are speech recognition and machine translation.
Convolutional layers use a convolution operation to the input passing the result to the next layer. This means that convolutional neural networks show outstanding results in image and speech applications, such as translators and bots.
In a simple way, LSTM networks have some internal contextual state cells that act as long-term or short-term memory cells. The output of the LSTM network is modulated by the state of these cells. This is a very important property when we need the prediction of the neural network to depend on the historical context of inputs, rather than only on the very last input.
Now that you have a grasp on which type of learning processes can be implemented, let’s explain how these networks communicate with the average Joe.
Usually, when I think of assistants, my mind travels to the typical 90’s installation wizards, which were a great tool to help users installing software. Do you still remember those? Of course, nowadays, we simply called them interfaces. Only now with the advancements in AI new types of smart-assistants are being developed and introduced in different products. We can find some examples by looking at Telegram and some websites which use communication bots — a very basic type of oracle or genie (depending on its properties).
There are four main types of assistants we can choose from, that grant their users the power to communicate easily with a given AI.
Depending on the assistant, there are higher or lower chances of the interacting agent making bad decisions — which can be any decision that goes against the end user’s goal. The strength of each AI’s predictions’ success is related, as we’ve seen before, to the data-set and machine learning algorithms. One thing worth mentioning, before diving deep into each AI assistant type, is the extreme importance of comprehending the strengths and weaknesses of assisted AI interaction.
The first problem is that, most times, agents cannot predict with absolute certainty whether the AI is following the proposed programmed logic to achieve the best outcome possible; in other words, the AI’s objective might change depending on its preferences, hence, applying the correct rewards and incentives game-like mechanics is so crucial for any AI’s success.
The second problem is that no AI developer will ever be able to create enough fail-safes to prevent dumb users from making mistakes. The reason being, expertise comes from knowledge and practice, meaning, if a certain AI evolves above a certain threshold of human understanding, even the AI creator won’t possess enough skill to understand its goals and logic.
Right, now that those problems are out of the way (not really, but at least we know they exist), let’s focus in discussing each assistant type.
1. Oracles: question & answer system (read-only)
Oracles are the first line of interface between user and machine. Usually, oracles serve the purpose of giving quick answers or instructions to users about the mechanics of a certain program. Oracles can be domain-limited, so each answer is mathematically based; output-restricted oracles come next, and these represent systems which give a yes/no/NA type of answers and, finally, there are oracles which can have fail-safes like not answering questions that go into a disaster criteria list. Telegram bots (found on most ICO projects) usually come in this category, as users are able to interact with specific groups by requesting the bot information about a topic related to said project.
2. Genies: command executing system (read-write)
Genies are the second layer of tools and assistants, as these possess a certain degree of perceived intelligence by users. Usually, this assistant performs actions on a system based on the user logic — which can be through natural language (again, look at Telegram bots). In most cases users communicate with the genie through instructions (commands) or questions/answers, and the system will perform actions based on the users desired outcome. Fail-safes can be created to prevent agents from promoting bad outcomes, although this system is prone to human error, as usually, it’s considerably more open-ended than oracles
3. Sovereigns: open-ended autonomous operation (with known goals or unknown goals)
Sovereigns can usually be connected to natural language techniques, as these open-ended systems are created with limited instructions to achieve certain goals. The AI is supposed to learn using models, such as the Monte Carlo (random walks) and find the most efficient path by trial and error.
4. AI-Tools: a system not designed to exhibit goal-directed behavior
Tools represent open-ended search processes. Creating an AI-enabled tool would mean its behavior would be mechanistic and not agent-based like, which can be a problem as open search boxes might create unexpected results. Tools allow users to define parameters and context, and they are commonly associated with business applications like IBM Watson or Google Big Query.
Can you imagine now what we will be capable of, in just a few years, when AI fully develops easy-to-use interfaces? The immense power users will have over data analytics like predictive behavior and preventive monitoring will be exquisite. Fewer errors, better performance.
Now that you understand the basics of neural networks and their purpose, as well as the different types of assistants available to help users interacting with machines (IoT devices, websites, platforms, protocols, etc.), can you try to guess the missing piece?
If you thought of data visualization, you stand absolutely correct.
The final frontier between user and AI is how data can be expressed; by adding visuals like graphics, infographics, images, and charts, it becomes incredibly easier for humans to retain information. That’s why it is so important that new AI-enabled visualization tools can be implemented alongside bots and such NLP systems. It’s the quickest and most effective way of transmitting information between machines and humans.
Platforms like Blockchain, DappRadar, Ethplorer, and Coin360 are doing an amazing job in creating data visuals we can use to understand the cryptocurrency market better; however, I feel more options are needed. If oracles or sovereigns can be taught to interact with data visuals, it will be easier to teach complex information to all cryptocurrency users.
There are already a lot of interesting blockchain projects out there with lots of different use cases; consider the list below to be the AI projects which caught my personal interest. Of course, there are a lot more cryptocurrency projects out there, many of whom I may not know yet. Honorable mentions go to Golem, DeepBrainChain, and SingularityNET.
Share your opinions and thoughts down below. Don’t forget to give it a like!
Disclaimer: The views expressed in the article are solely those of the author and do not represent those of, nor should they be attributed to, CCN.
Featured Image from Shutterstock