We are creating a platform in a simple language, which will enable ordinary people who do not know how to program, who do not understand mathematics and data science, to benefit from the data that they have stored. We have an idea: to create a product that will enable an ordinary user without programming skills to receive solutions using AI.
Many would ask if GraphGrail Ai is seeking to empower people by offering them instruments for creating high-end software products.GraphGrail Ai is providing not just software that will analyze the data. Several cases are ready and are waiting for the launch of our platform.
A typical example is the developer of a computer game. The developer regularly releases updates for his game and starts receiving feedback in the information field, in social networks, in chat rooms and so on. It is very important for the developer to get negative feedback, which can help him improve his products. Let’s say if users say something negative about the payment system, for example, the payment system does not work or the game does not load, then the developer needs to react to it instantly, and if this is a game with a large number of downloads, say more than 10 million, it’s about 1000 comments a day and they need to somehow sort them into critical-not critical, positive-negative. Now the entire amount of work is based on the resources of an ordinary person, who, on the basis of their subjective view, make a decision about what is in front of them in the form of a response, which is positive or negative, and have formed it as a report to the management, having spent time on proofreading, on transferring and other processes. The platform helps to prepare data already available to the user so that all subsequent mentions of their game that will be obtained with each next update are sorted automatically and divided into positive / negative ones, and in turn broken into subgroups. For example, negative responses can be attributed to critical and not critical. Questions about payment and problems with the installation of the game are critical questions, they need to be reacted on immediately. And, suppose that some item in-game does not correspond to its description, it has more ammunition or armor for example, these are not very critical reviews, but they also need to be worked out. To avoid doing this manually, one can take the platform tools we provide. Let us say we need to get information, so we take a block of the parser and input information. Further, it is necessary to break the information into positive and negative segments, else a problem may arise. There are solutions on today’s market that provide such analytics, but they are primitive, without the use of data science algorithms and neural networks. They are like a regular search engine looking for words. If there is some positive or negative word next to these words, the feedback will be marked as negative or positive. Although, in fact, it may not be so. There is the following problem if the computer game is made for children and the children begin to leave negative comments about the game with mistakes or in slang, and children at 8–16 years old have specific errors, in principle they speak in a figurative sense. Current solutions cannot allow the developer to get all that benefit from using AI without such tools. Moreover, our platform allows users to break the feedback into positive and negative segments. In addition, if there is a need to break into additional subgroups, it is possible to teach the machine to understand where the positive, and where the negative values are, given slang and errors in words.
For instance, let us take a business that needs to receive messages or comments from posts in social networks. Decisions about the answer to each comment are accepted by a living person. However, a machine can learn to understand the essence of a question or mentions and sort all subsequent questions into the same categories, and using the tools of the platform the user can use the information received for your needs. The user needs to explain the neural network how to understand this or that question.
Those who accepts incoming messages and somehow answers them, including those who support the operations are among those that can be removed from the process. For example, support helps with some number of categories of questions. Questions about payment, about installation, about quality and the answers to such questions are usually similar, but they are posed in various forms.
Lawyers are another profession that can be optimized. A common lawyer fills out documents with respect to some persons and actions. The lawyer takes data about a person, an event, an incident and fills in this data into a certain template of documents. Such examples are already working. In America there is a simple bot-lawyer who fills out the documents. In America, it is very expensive to just fill out the documents correctly, especially the templates. Having received any text data, the machine can understand what this data refers to, or determine which template should be used for this situation. The work of a lawyer can be automated.
We were contacted by one potential platform user with the task of analyzing the number of videos from each category and on a large video hosting hub. The customer wanted to understand how many videos he had on the site and how much more or less than his competitors. Therefore, he needed to either upload some content into some category or update the content. Suppose the user has a lot of videos about fishing and there are absolutely no videos about mothers. The viewers interested in the category of motherhood will leave, because this host has little interesting content. In the current situation, the owner of the hosting should sort his videos into categories. We made a selection, set the data for the analysis of headings and the machine analyzing the data quickly sorted all the videos according to the given characteristics. There is no need to view thousands of videos in order to determine which category to put them into. They are sorted automatically.
Sales managers may be in a bad mood, or maybe they don’t feel motivated, because the cycle for getting answers from clients is very long and plus all the typical questions for a person who constantly responds to the same thing seem already understandable and primitive, resulting in short and dry answers. And the client may not have enough of this answer, due to the fact that he receives information for the first time.
Payments and their different formats are another issue to be tackled as the existence of a form for unifying payments would have improved the issue of transactions.
There are several algorithms that we can use in this situation. If everything is simplified, then this is a dictionary of synonyms, when a bot can match some words with others. And there is a more advanced tool — the algorithm for processing natural language, where words are translated into a vector and it will be difficult to comprehend their meaning for a machine and all similar words that carry the same meaning, have a similar vector. And when we have a phrase consisting of several vectors, that is, several words, then we understand in that direction whether we have a phrase. Without going deep into the mathematics of languages, we basically understand that the phrase is about the same context. If the probability is higher than 70%, then by default we accept that the phrase is roughly from the same category. But in our product, we envisaged the function of a prompter, so that the business that uses this chat-bot should not be afraid to give management of their sales to a chat-bot. As for the work of the chat bot, the chat-bot will send clarifying messages so that the owner can confirm or reject the request for classification of a particular question. And then, the possibility is excluded that the bot will do something wrong. For some time, the user responds to the bot in the format of training an ordinary salesperson-employee.
It is not appropriate to call this AI, but NLP, because in the 50’s even the calculator was called AI, all calculations were done manually. People were sitting in a certain sequence and processed the information manually and the calculator was artificial intelligence. The fact is that AI will soon become just another mathematical assistant and it does not really matter how we call it.
The task of processing a natural language is to understand the patterns and the program of the AI understands the essence of the conversation. There is a function of maintaining the context of a conversation. This is when we have a few messages in one message or we repeat the same thing several times. Let us say we sell medical clothes in an online store and the first message we receive is “Do you have a model with red sleeves?” We answer “Yes”. The customer asks the following question “And do you have size 42?”. The bot understands the chain that we are talking about and will take into account the previous information when answering the question posed. If we talk about the president he can be mentioned several times as President, Commander-in-Chief of the Russian Army, by name etc.
It all depends on the situation and on what tasks we decide that the AI will tackle. If you conduct dialogues, then most likely the dialogue ends with a specific point and next time you start talking about something else. The AI does not have an automatic function to take any action, if it was not assigned to it. It will never start a conversation on its own. The AI will respond to your messages. In order to activate it, we need to receive a request. An inquiry in the form of some kind of text.
When we conduct sales through neural networks and conduct a dialogue with the client on behalf of the chat-bot, we input the questions that we know the client may ask. We understand what the client wants to buy. We can ask a question, get an answer and classify every need. We can break the text into sentences, this process is called the tokenization function. We can tokenize words to give, for example, synonyms for each word if we want to strengthen the classification. We can vectorize the whole sentence, indicate to the machine that we will now classify a particular sentence or a specific word. The platform can understand the patterns. Another algorithm that we can discuss is that each person speaks and writes in their own style, and we can understand who wrote what, even identify a person by their style. We can digitize the patterns of Leo Tolstoy, we can have some kind of text on the neural network and it will alter the text in the style of Leo Tolstoy.
Firstly, the money is needed to attract the first users, as we use the reserve fund and the platform economy itself. We will motivate this foundation and those who will create algorithms, lead the markup, and upload their set data. The money is needed to load onto the platform already existing, assembled language blocks, libraries, and all sorts of algorithms. The greater part of the funds will be spent on entry onto the market, because the competitors are very powerful, they are not even competitors. Everyone has their own nuances, but to enter the market, you will need money. And since here, the very development of algorithms and integration into the platform is pure development work and it all costs money. Attraction of users, access to the international market, creation and promotion of tools that are needed by the whole community are all expenses. The bulk of the money will go to making the platform really safe and decentralized with all the bonuses that blockchain gives us.
The soft cap is 2 million dollars, and this money is enough to bring the platform to the market, realize everything that we conceived on the platform, in the road map, along with a full-fledged marketplace with tools. 20% will be spent on development, 30% for the purchase of algorithms and 50% will go on marketing. We have already collected about 550,000. We expect to take 3% of the market for processing natural languages in the end, and to date, this is about $200,000,000 per year. The volume of the market, which we are currently observing is 6 billion dollars.
For a platform, it is not important how you use it, what is important for the platform is that all actions are recorded in the blockchain, so that all users of the platform can see who everyone is, how they are working, what result they generate, how many orders they have performed, including all the comments and additional information. The GAI Token is used as the internal currency of the platform to this end. It is isometric.
The regularity of the data is the profit that cannot be gotten without using platform algorithms while processing data. The product is the specific patterns in the data that are collected. The patterns and their use.
Stay tuned to our news and updates and also take part in our TGE!
