Many of us are familiar with Artificial Intelligence (AI) applications, thanks to personal assistant bots, which help humans to solve easy micro-tasks or answer simple questions based on keywords. For example, the weather forecast for tomorrow, the nearest grocery store, or the upcoming events in one’s schedule. Bots with uncomplicated algorithms are acceptable by users during their interactions with applications like Siri, Google Assistant, Amazon Alexa or Microsoft Cortana.
But in most cases AI powered applications get confused with complex questions or can not recognise human emotions.
For instance, more advanced applications are needed for consulting customers in electronic store. AI powered electronic consultants should constantly learn, following the behaviour, responses and emotions of customers during interaction.
Despite that the current level of communication between AI powered applications and humans is far from seamless, that’s the goal for the foreseeable future. Everything is ready: the computing capacity of modern processing chips is at its best, data transaction speed is lightning fast, AI services infrastructure for a widespread adoption is on it’s way and and soon to be implemented by a number of AI platforms. One of the platforms that we’d like to share our thoughts about is blockchain-based AI Technical Network (ATN).
Examples of services, powered by AI technology
Before proceeding, lets have a quick look where AI technology is employed today, but keep in mind that the future has way more opportunities in its store.
– Online customer support. Many websites and phone centers have already implemented chatbots to substitute humans in answering to common queries.
– Impoving engagement with customers and providing personalized recommendations. Such giants of online commerce as Amazon or Alibaba collect and analyze tons of data about behaviour of millions of customers. This data includes information about searched items, time spent on reading descriptions, frequency of purchases, regions of delivery, etc. The more advanced the algorithms used for big data processing, the more benefits online retailers can get from information about their customers and increase sales.
– Fraud detection in e-payments and banking spheres. Fraudsters are becoming ever more tech savvy, often deploying bots, which can act as a real person. So, antifraud software needs to engage AI technology to stay one step ahead of them and stop deceitful attempts.
– Security surveillance and face recognition. Today almost all public places are under video surveillance. Cameras record tremendous volumes of data, which should be analized almost instantly. The speed is essential to rapidly recognize faces of criminals in the crowd, identify actions that might lead to theft in a store or public transport, detect other suspicious actions, etc.
– Smart homes. Many smart home devices learn behavior patterns of their owners and help to save energy by adjusting thermostat settings. Smart home devices also manage other appliances in an effort to increase their users’ convenient lifestyle.
– Self-driving cars. Program in charge of driving a car must simultaneously distinguish and react to traffic lights, road signs, other cars, pedestrians, and many other objects.
– Smart traffic regulation
– Speech recognition
– Computer-assisted diagnosis
– Preventive healthcare
ATN platform – ecosystem for AI development
Put simply, the ATN platform will act as an advanced marketplace, where developers will find everything they need to create and train AI services, and customers will find any kind of AI service.
ATN ecosystem is based on three main pillars: DApps, DBots and AI services.
DApp is a decentralized AI application, created for end users, and to be used for their specific needs. By entering initial data into DApp users then receive an answer or a complete solution. But DApp itself will not necessarily contain an AI algorithm to solve tasks.
Instead, DApp can use an AI service from a third party, simply sending the user information to AI service. AI service will process the data and provide the result back to DApp. Doing it this way DApp will need help in finding an appropriate AI service for its task.
DBots will act as a link between DApps and AI services. DBot will define which kind of AI services is appropriate for DApp, it will then choose the most suitable AI service among a huge pool.
Machine learning and training of AI service within ATN platform
Before an AI service developer starts to sell it within the ATN platform, he’ll have to train its service. It is done via machine learning, a field of computer science that gives computer systems the ability to “learn” with data. The goal of machine learning is to take some data, train AI service, and use it to make new data predictions. Simply speaking, teaching AI service is very similar to how parents teach a young child to distinguish a cat from a dog or recognize different objects. Just as children learn by being told what they do right or wrong, AI services are trained via feedback provided by machine learning algorithms.
In many cases developers have fantastic ideas and create great AI services. But very often they don’t have access to specific data needed to train their AI service. Also, the machine learning process can consume a lot of computing power, which might not be available for the developer. ATN platform will provide infrastructure capacities for developers, data providers and computing power providers.
If some company has large amount of specific data (big travel agency or large telecommunication company, for example), it can sell the data via ATN platform to a developer who needs it for training their AI service.
If someone has a large processing power capacity, he can rent it to developers for AI service training.
Once AI service is trained, the access can be provided via the ATN network.
All activities on the ATN platform will be recorded on blockchain and will utilize ATN tokens, which are essentially the “blood” of the platform. ATN tokens will be used by Dapps to pay for AI services, while developers will use ATN tokens to pay for access to data and computing capacity.
Utility of ATN platform
Let’s consider an example of how ATN platform works.
AI services are used for computer-assisted diagnoses, which are rapidly becoming accepted worldwide in various fields of medicine.
The data (information) derived from numerous diseases and symptoms can be normalised and anonymised from large patient databases (DB). AI services can use this data to learn how to make a preliminary diagnosis. The more cases AI service uses during the learning process, the more advanced and sophisticated the process becomes.
Then, suppose, there’s DApp (decentralised application) that aims to assist doctors in detecting a disease based on preliminary data (such as complaints/pain, blood tests, overall visual condition of a patient, etc.). The DApp can be installed on the computer or smartphone. The doctor enters the information about the patient into the DApp, which then calls a DBot (decentralised bot) which looks for suitable AI services within the ATN network. The suitability of service can be decided based on its sophisticated reputation system that takes into account the ratings from users, service costs, and various factors to ensure the quality of the service. The information is run by AI service, trained to conduct a computer-assisted diagnosis, which AI service processes and provides the preliminary diagnosis to the DApp, which neatly displays it back to the doctor. The DApp charges ATN tokens for the service. The AI service owner decides on the amount of ATN tokens to charge.
Furthermore, the doctor, who uses DApp to make a preliminary diagnosis, may discover new information about the patient and can amend the DApp to change the diagnosis. The doctor can then enter this new data into the same DApp. The AI service will take into account this data, and, if needed, improve its algorithm for detecting diseases.
Also, if such DApp has a lot of users, it can collect an extensive database of information about patients and their diseases and sell it to other AI services within the ATN marketplace, which provide a computer-assisted diagnosis, so these AI services can learn and improve their algorithms. The owners of such AI services will pay ATN tokens for such valuable information datasets to improve their AI services.
This is a self-sustainable ecosystem where everyone benefits: data providers, AI service developers and AI service consumers.
Neural networks and deep learning within ATN platform
Neural networks are based on the human brain biology. Once information is received in a neural network, it gets processed through many layers. Each layer has its own function, which processes data, adds new valuable information and sends it to another layer. Once the information is processed through the final layer, the neural network makes a decision or produces a response. During processing information through all layers, the neural network can distinguish unique characteristics of received data, choose only necessary bits of data, find common patterns, categorize or generalize data, focus on the right features of complex objects, etc.
A neural network is essentially an AI service, but much more complex and with many layers. So, it needs to be trained in a complex way, with the help of deep learning. “Deep” means that all the layers in a neural network should be trained. The aim of deep learning is to teach neural networks what a human brain can do naturally.
In many cases a separate layer can be trained independently of each other. A developer can implement different deep learning algorithms and use different data for training layers. Different types of data might be required for training different layers as well.
The ATN platform will provide infrastructure for deep learning processes. The developer of a neural network model will be able to pre-train some layers of the model with a large-scale dataset. Developers can further train other layers of this model with their own datasets. Such approach will provide opportunity for cooperation between developers, an extremely important feature in AI sphere.
Current state of development and future outlook
ATN is already gaining traction among large technology companies and venture investors.
At the end of 2017, ATN announced a strategic partnership with Chinese Wanda Internet Technology Group. ATN and Wanda will jointly establish a “blockchain + AI” industrial alliance to support Wanda’s business entities to create smart products and services.
In February 2018, right after finishing successful public sale of tokens, ATN announced a strategic investment of US$10 million from institutional investors Dream Chaser Capital, Capital Dynamics Fund, Longqing Capital and Drop Capita.
As of today, we consider ATN as one of the most promising blockchain projects. ATN platform has a great value because it solves a lot of issues within the AI tech industry. The team has a clear vision of how to develop and implement the platform. At its early stage ATN is widely known among many players of IT industry.
We expect more announcements from ATN in the nearest future and will continue to follow ATN progress and update on upcoming events.