«The work of AI is no different from how the brain functions».

The story of a programmer with 30 years of experience

Vladimir Chernitsky, Faceter co-founder and CTO, talks about how the work of the programmer has changed since the late 1980s and whether neural networks will soon be able to surpass human capabilities.

I started programming almost 30 years ago

I started programming at school in Chita, almost 30 years ago. We had the first computer class in the city. In most Soviet schools, teachers explained computer science on the blackboard, and then students completed assignments in notebooks. We were lucky — we had real Yamaha computers.

I joined an informatics section, and at first, we mainly played video games there. Then a teacher appeared, who began to explain the basics of programming: Basic, cycles and arrays.

We read the «Modelist» magazine in which pieces of ready-made programs were published: an article with a short explanation and two or three pages of code.

There was a game named Sacoban; we tried to recreate it on our computers. We wrote a simple game together with a friend and showed it as an example of what can be done on computers to younger students for inspiration.

Programming became my main hobby, and I asked my parents to buy a home computer.

When we received feedback from Australia, we were shocked

In 1994, I graduated from school and wanted to enter the Tomsk Institute of Automated Control Systems and Radio Electronics. But my parents did not dare let me travel to another city.

So, I stayed in Chita. There was not a single IT-faculty, and I entered where there was strong mathematics. It was an energy faculty. A year later, I found a job in the information-analytical agency of the Chita region. At first, I just created the layouts of different magazines and books on computers in Photoshop and Corel, and set up the network.

In 1995, we received the Internet. In that time, the Runet area was completely empty, and the most popular search engine was Altavista.

Our department received an order to make the first site of Chita. I learned to write scripts, and made the first guestbook on Perl. I remember that someone from Australia wrote there, and it was a shock.

Then I started to work at the «Chita.net» company. It was the first Internet provider in the city. I developed a billing system to account for customer traffic. It was a beginning of deeper programming growth — C ++ and other stuff. I learned it through teaching myself — I could already find professional forums, publications and books on the Internet.

In the early 2000s, I received $300 per month

I was strongly influenced by an acquaintance with my colleague. He was a hardcore programmer, already at the time, in the early 2000s. He talked about data structures and trees, and developed software for foreign companies on outsourcing.

My colleague was looking for clients, and we earned serious money, although for Western companies, our work cost a penny. I took $ 6–8 per hour. In a month, I received $300, and it was 2–3 times more than my official salary.

We made a website for one customer from Europe — for his clan in the MMORPG game. Also, we developed a site for the Australian state office. As a result, my partner left for Australia.

I stopped enjoying my work

I was 30 years old when realized that I was tired of everything. At that time, I already had a family, an apartment, but at same time I felt a dissatisfaction, I wanted to grow. With my wife, we began to think of where to move.

We chose Novosibirsk. At that time, I was already a leading programmer in «Chita.net». When we moved, it turned out that I wasn’t the big fish that I thought I was back in Chita before. I sent my resume to four places, and I was invited for an interview, but I was refused everywhere.

I realized that my knowledge was still superficial. As a result, I went to Azoft company for an average position. After only a year of work there, I probably learned as much as during my previous career.

I began to learn different technologies, and projects became more serious. I wrote front-ends in PHP, as well as back-ends in Java. In that «before iOs- and Android-era», where JavaME and SymbianOS dominated, I developed mobile applications.

It was 2008 when iOS appeared, and we wrote a test application, and customers came to us. I headed the mobile development unit. The department grew to six people. We wrote under iOS, and a little later, Android appeared, which we quickly mastered.

We recreated the technique of drawing on the water in the application

Once a client came, he wanted to recreate Ebru, the Turkish art of drawing by colors on the water, in his mobile application. The customer wanted to create a software emulation of this technique. The activity of R&D company started with this project, and my new era began.

We built a physical and mathematical model, and then went to Akademgorodok to talk with professors to understand the hydrodynamics of the process.

As a result, we managed to recreate it quite accurately right on the GPU — on the iPad. Of course, the model was simplified, but the customer was satisfied.

About neural networks

In this project, it became clear that we needed to create an R&D department in the company. I headed it, and we started to do different non-standard projects. In the same way, we came to neural networks that just began to rise.

Neural networks are «ancient» technology, but it received a new life in 2012 with Alexnet. Jan Lekun also wrote a study on this subject in the 80's.

The first notable project in the field of neural networks was called Neocognitron. Japanese scientists in the 80’s dissected cats, studied how the visual cortex of the brain is arranged. Even then, the technology of pattern recognition appeared, and the machine was taught to «learn» the numbers.

Later, Neocognitron finalized and created convolutional neural networks. In 2010, it became clear that these algorithms can be run even on ordinary home computers, and a new life of neural networks began.

I studied the «neurons» from and to

In 2013, a customer came from Spain. He wanted to develop an application for the recognition of data from credit cards for bank customers. It turned out that the best way to achieve this is to use convolutional neural networks.

We created the application, and the customer was very satisfied. It took about a year. And then there was a new client who wanted the same application — Robert Pothier from South Africa.

At that time, I worked for 7 years in Novosibirsk, I continued to develop as a professional, but again I had a strong desire to change something. Robert wanted to create a development office in Moscow and offered me a position in leading it.

I took my family and just went to Moscow. In the project Pay.cards, I got the opportunity to study «neurons» from and to, squeeze out a maximum of them. We managed to create an excellent product.

Sometimes neural networks surpass human capabilities

In 2014, an unfortunate accident occurred, which eventually led us to the Faceter. Robert was robbed on the street in Johannesburg, moreover, under several cameras of video surveillance. He did not suffer, although the attackers threatened him with a knife.

It affected him so much that he started to think how to make video surveillance more effective.

In this way, we came up with the Faceter software, which actually replaces the guard or the policeman on duty, and оbserves everything happening continuously without days off and smoke breaks. It is a vivid example of how artificial intelligence can replace some human abilities.

Even more, neural networks in some cases surpass our capabilities — a system like ours can store data on millions of people and recognize them in the video stream in real time. Of course, one person is incapable of doing this. We want to make sure that the «neurons» could recognize not only faces, but emotions as well, and also analyze the chain of events and diagnose the situation. For example, if a person with an aggressive expression takes a sharp metal object out of his pocket, the system must turn on the alarm mode.

Neuron systems will change everything in the world in the coming decades

I believe that, in the next 20–50 years, this technology will change everything in the world, and fundamentally. I am inspired by the opportunity to develop systems that will be part of this process; I feel I am on the front line of the fourth technological revolution.

Our system uses computer vision. It has already been tested in real conditions. We achieved recognition accuracy of 99.78% for the LFW test and 79.46% for MegaFace.

We completed three PoC projects:

• Face recognition is used for loyalty programs in famous South African pizzerias Debonairs.
• In the largest casino network, it is applied to identify those who are prohibited from visiting gambling establishments.
• At one plant, we helped to cope with the fact that employees, when they cannot take a shift, give their passes to untrained relatives or friends. The system compares everyone who passes the checkpoint to a database and reports when it sees an unfamiliar person.

AI can bring real benefits

We are working on advanced recognition technologies. Now the system recognizes people by face, and soon it will be able to identify silhouette, clothing, and track movements of a person from the camera to the camera. Then we will learn to recognize situations — the system will not analyze single frames, but chains of events, for a long period of time.

Now, systems that can detect individual events already exist. But those who can recognize the chain of events and diagnose situations has still not been developed.

Many companies lead such developments. But our main mission is to make the product for mass users.

We want to show that artificial intelligence not only can print photos in different colors or draw paintings in Van Gogh style, but also bring a real benefit to everyone.

The work of AI is not much different from the work of the brain

Of course, I share the concerns about the dangers of artificial intelligence. Progress is speedy, and we just cannot follow it. If you could recently say that AI is a primitive organism with limited abilities, now robots can do a backflip and understand the English speech better than most people.

In my opinion, the work of AI is no different from how the brain functions in principle. It is, evidently, a simplified model, but it reproduces what is happening in our head.

Now, AI is like a children’s toy car in comparison with a standard adult car. But the technology develops rapidly. Everything is radically different from the technologies of the first generation, since the appearance of which only 5 years have passed.

Evolution is intense. People try to reproduce the processes in the brain. Extensive development is due to the growth of equipment productivity: networks can be made deeper every year. The computing power of the neural network will be equal to the human brain in the next decade.

Humanity stands at the threshold of universal systems creation, and will be enough to indicate the rules, and they will find the best ways for solutions!

This article was originally published in Russian here