Artezio Presents its Vision on Development of Machine Learning Technologies
Posted on Jun 8, 2018
Speaking to the conference attendees, Pavel Adylin raised the issue of understanding the term "artificial intelligence".
“The situation with the term “AI” is similar to how the term “digital transformation” was understood. Many people talk about artificial intelligence without understanding clearly what this term means. The problem is that when talking about artificial intelligence, the press uses the abbreviation “AI”, rather than “machine learning” as it was originally. The problem with the perception of terms occurred due to the analysts’ fault who replaced the “boring” term “machine learning” with a more beautiful one “AI”, says Artezio CEO.
“Today the development of AI goes in three directions: image recognition, natural language processing, and machine learning that includes everything that didn’t fall into the first two groups. At the same time, artificial intelligence develops cyclically. A problem is set that at first seems impossible: for example, training the system to recognize faces, and this problem is referred to AI. The technology develops gradually and finally, a breakthrough is made. The achievement causes a real stir in the society, and then it becomes commonplace,” he adds.
The expert is confident that in future software products will be created taking into account the parameters of a model and the weights of neural networks. The time of Software 2.0 will come.
“The term Software 2.0, first introduced by Andrej Karpathy, is closely related to the concept of AI. Software 1.0 means that instructions to the computer are written in one of the programming languages, whereas Software 2.0 – the parameters of a model or the weights of neural networks are programmed. At the same time, the stack of Software 2.0 doesn’t replace or compete with Software 1.0. It applies the developments made within Software 1.0, and the program code itself is created using the classical development stack. But more and more solutions will contain the elements of a new stack,” stresses Pavel Adylin.
Artezio is involved not only in the practical implementation of AI projects but also conducts research and educational work in this field. The company has its own AI lab that aims at finding and building solutions for the practical application of AI.
According to Pavel Adylin, “Together with the Nizhny Novgorod branch of the Higher School of Economics, we have developed an educational course for our employees. Its goal is to train engineers who will develop and implement applications based on AI. Initially, in 2015, it was mainly based on big data and technologies for big data storing and processing, but over the next three years, the course has been almost completely reworked. It demonstrates the speed with which this direction is evolving.”
As for the implementation of AI projects for specific customers, Artezio expert points out that “it is always better and more appropriate to offer the latest technologies within the context of the customer’s business.”
“If you take into account a particular business process, then the conversation becomes more productive. In terms of methodology, it makes sense to adhere to existing cross-industry standards developed for Data Mining. And in terms of the project estimation, first, analysts work on Time & Material, they create an app prototype, and then after the scope of work is clarified, the Fix Price model is applied. The Success Fee model, while appealing, in terms of profitability, contains a lot of risks due to a small number of precedents for the implementation of similar projects and opportunities for interpreting results,” adds Pavel Adylin.