Vasily Mazin has been striving to improve AI systems and create an AGI for 10 years. He graduated from the Faculty of Applied Mathematics of Kuban State University where he worked as a lecturer for 15 years. Mind Simulation is the end result of his scientific work. Now Vasily is a CRO of Mind Simulation Laboratory.
In his interview for AI Conference, the expert shared about the struggles of creating AI and explained his choice of hybrid systems.
A personal computer that I received for my 14th birthday from my parents dramatically influenced my choice of the future profession. It was a rather wacky Partner 01.01, like in the photo but red one. The computer didn’t have ROM, only a BASIC interpreter and an assembler. Every time I wanted to play a game on my PC, I had to write it first.
However, I liked programming, so I decided to enroll in the Applied Mathematics Faculty. Before the university, I was studying at the physics and mathematics lyceum famous for its brilliant information technologies and programming courses.
I started working when I was in the third year of university. I wrote inventory management, accounting, business and financial administration programs. Some of them are still available now.
When my programs became more complicated, aside from programming, I started to lead a development team. Then, we were focused on systems architecture and algorithms.
My latest project, not connected to the AI field, is program for fire risk assessment.
After graduating from Kuban State University, I became a lecturer at the Faculty of Applied Mathematics. Teaching used to be a good addition to the scientific activity. I had a lot of time after lectures to pursue science and do my personal projects. Besides, term papers and theses gave an opportunity to try out various ideas and approaches.
Later, I was flooded with paperwork, so it became difficult to do other projects. As a result, I left the university two years ago and dedicated all my time to Mind Simulation.
I became interested in AI and neural networks when I got my first PC. At school and university, I designed and programmed computer games (simulators of tanks and airplanes, strategies, role-playing games).
Coming up with algorithms of NPCs was the most interesting part of my work. I remember writing a tic-tac-toe game with an unlimited grid where you had to put 5 crosses or noughts in a row to win. I created the game from scratch since there were neither Internet connection nor professional liturature in Krasnodar at that time.
The program successfully bearen noobs and amateurs. However, it was defeated by an experienced player.
I became interested in neural networks in 2003 when they became a part of my research. Back in the days, everyone would say it was going to be a dead-end job. However to me it looked like neural networks have a big potential for solving specialized tasks and being a part of complex systems. This led me to the concept of deep learning networks but there was no opportunity to practice it.
I am currently on the pulse of events in this industry. I have tested almost all existing AI approaches.
During my university years, I was thinking about creating an AI-focused laboratory. In my free time, I worked with a group of students but I didn’t manage to launch a lab at the university. So, as soon as the opportunity arose, we created our own one.
At that moment, the popularity of chatbots and voice assistants increased. Later, we found out about Alice and analyzed it. Then, we looked into Alexa, Siri, Cortana, and others.
None of the abovementioned assistants could provide the essential thing – real communication. They call a taxi, show weather forecast, which is cool, but is that it? It is impossible to just casually chat with those assistants. We need AI not only to answer questions but also to be a friend and a partner who understands you, maintains a conversation, and helps.
Well-trained AI is the best solution for such cases. A single question came up: is this task worth dealing with? I brought to light my initial ideas of knowledge representation and hybrid models. We estimated our possibilities and resources, developed a detailed plan. That is how a new project was created. We are currently working only on it.
Even before the work process started, we already had general theoretical and partly algorithmic description of AI core architecture and working methods. However when we started working on it, some technical difficulties in implementing algorithms came up. Fortunately, brainstorming was the key to problem-solving.
We had a lot of difficulties with machine learning. Due to the lack of a critical mass of knowledge, essential for AI self-learning, we couldn’t automate the whole process on the first try. However, it still was a huge step.
It took me many years to learn how to use AI for problem-solving. During some experiments, we used two and more approaches simultaneously, which were successful. We started to combine several methods more often, which led to the creation of strong hybrids.
The audience shows a profound interest in AI products. Our presentation at Geek Picnic led to a 40-minute interview session though we were the last to speak.
It is very important to share knowledge and experience in science: you meet like-minded people, critics, expand your social circle. It helps to see your work in a different light.
Of course, we currently can not reveal all the details of our technology because it’s our competitive advantage. As time goes by, we will tell more and more information.
The ability to develop, discover new directions and sell products generates financial resources – that is why the commercial success of Mind Simulation is important for us.
We are planning to open an AI school at the MSL to help kids and teens with fulfilling their ambitions and getting knowledge. This idea can not be realized unless we raise enough money.
Before the commercial launch, we have to finish coding the basic core architecture. This project is divided into nine stages, we are currently finishing the fifth. The ninth will presumably be completed in February 2019. Some organizations have already shown their interest in using our core in the development of their products.