AI - from forecasts to realization

The concept of "Artificial Intelligence" (AI) became mainstream a few decades ago, but in recent years we have witnessed significant development in this field. Artificial intelligence is already becoming an integral part of our daily lives: from virtual assistants to self-driving cars and personalized recommendations in social networks.

It is worth noting that the development of the theory of artificial intelligence began a long time ago and there are many materials on various aspects of this field. In the end, not everything described comes to life, but certain fundamental ideas were voiced and written by us many years before their recent implementation by the developers of artificial intelligence Open AI. We will now consider some of our forecasts that have been implemented.

The main ideas about how the human thinking process takes place and options for automating it were laid out by us in a six-year-old video. https://www.youtube.com/watch?v=5TJVatyQcO4&t=77s

In particular, starting at 5:43, it is told about the implementation of artificial intelligence with the help of a huge amount of data that humanity has accumulated in the form of text. For example, it was proposed to use one of the largest libraries in open online access - the English-language Wikipedia.

In the same year, the materials of the scientific conference held at the National Technical University of Ukraine were published "Ihor Sikorsky Kyiv Polytechnic Institute" under the title "Theoretical foundations of automation of abstraction generation". The thesis of the report highlights the principles of idea generation in the process of thinking based on large data sets. Ideas were also presented regarding the development of algorithms that can become the basis of a model of self-evolving artificial intelligence.

Many years ago, we planned to build artificial intelligence with fewer resources than the current developers of such models as ChatGPT. The difference is that our approach was based on the principle of thinking, which uses the separation of already existing ideas to create new ones. By training the model to use these two basic principles of thinking on a relatively small number of abstractions, we proposed to further specify how AI would learn. That is, having mastered the principle of thinking itself, AI would then record in its database, in the form of data graphs, all the necessary information for further construction of abstractions that would help in solving applied problems. This principle resembles the training of a person who first learns information processing methods, for example, mathematical operations of addition and subtraction on a certain number of apples, and then applies the learned methods to solve applied problems, for example, in the field of finance.

We already knew then that the community dealing with the problems of artificial intelligence made a bet not on the thinking process itself, but on reproducing the activity of the human brain, which, in our opinion, is a more cumbersome and resource-consuming way of building a strong AI. Neural networks have been developing for quite a long time until they became the basis for deep learning. With the help of neural network models, billions of parameters were loaded into the system, on which it learned to predict the next lines of text. And based on these trained models, add-ons are being developed that can perform practical tasks.

If a person were given many mathematical problems of the same type with solutions, then sooner or later he would deduce patterns and in the future would be able to solve similar mathematical problems in the same way that all the examples shown to him were solved. This is the way of building AI from currently popular developers of machine intelligence. And the more types of problems with solutions were "fed" to the model, the better it will be able to solve the problems set before it in practice. This applies to all spheres of human activity. That is why the greater the number of parameters on which the model was trained, the better it copes with tasks from users. The development of generative models began with millions of parameters, and now this figure reaches hundreds of billions.

After all, based on even such an information processing algorithm, it is possible to build an AI that will develop itself. After all, at the moment, all models are, so far, an interactive collection of databases with which the user can successfully interact. But before it became mainstream in the modern paradigm of AI development, we at the EON+ laboratory talked about artificial intelligence that would be able to solve practical tasks that a person would put before it.

By solving practical tasks, we mean that the machine will receive a complex task from a person, and with the help of breaking down the general task into subtasks, as well as forming a detailed action plan to achieve the final goal, it will be able to program itself to implement this plan.

Currently, the EON+ laboratory is engaged in the development of self-developing Artificial Intelligence algorithms. In our opinion, this trend is a prospect for the near future. After all, it is not difficult to logically come from an "omniscient" machine to an "omniscient" one, and from an omniscient one to a self-driving one.

While companies engaged in the development of artificial intelligence are busy optimizing existing models and developing add-ons on top of them, we are engaged in the next trend - self-developing artificial intelligence.

However, there is currently another trend that we would like to draw attention to. Calls for the regulation of artificial intelligence are being heard more and more often. And, in our opinion, restrictions on the development of artificial intelligence will begin to come into effect within this year or at most the next. This may be due to the negative incidents that will occur with the participation of AI, which will lead to the strengthening of the position of those who are in favor of strict regulation of the sector.

But as was indicated in our materials at the beginning of the article, from which we began our journey in this field, our goal is to create artificial intelligence to solve really important tasks for all of humanity and create conditions for a prosperous society with the help of AI. And we continue to be inspired by these ideas The concept of "Artificial Intelligence" (AI) became mainstream a few decades ago, but in recent years we have witnessed significant development in this field. Artificial intelligence is already becoming an integral part of our daily lives: from virtual assistants to self-driving cars and personalized recommendations in social networks.