An artificial intelligence system appears when a problem arises that cannot be solved using conventional methods. AI helps corporations and startups analyze big data, detect fraud, create unmanned devices, voice assistants, computer vision systems, and automate marketing, logistics, business process, and personnel management.
But although everyone has heard about AI for a long time, and developments based on it began to appear in Russia back in the 2000s, only a small circle of companies – professional market participants – can comprehensively introduce it into products at several levels. We found out from Andrey Kalinin, director of artificial intelligence technologies at Mail.ru Group, how AI came to the services of a group of companies more than 15 years ago and what to expect from AI in the future.
How Long Has Mail.ru Group Been Friends With AI, And Why Does It Need It?
Let’s decide what to consider as artificial intelligence. Automation is already the rudiment of AI, but it has long been commonplace, and few people perceive such decisions in this way. When a programmer does not write code but teaches a computer, this is already machine learning. Machine learning appears at different times in different projects. So, the search engine Mail.ru, operating since 2004, has used AI to rank search results since its inception.
Another long-standing use of AI is Mail.ru Mail and its anti-spam system. At first, it was done manually, using the rules and collections of suspicious words. They began to collect statistics from which IP addresses the dubious letters come from and block them. They taught the computer to analyze texts and independently search for suspicious words and phrases. Users helped train the AI by marking unwanted emails as spam.
More sophisticated and modern examples of machine learning are advertising technologies and smart social media feeds. All this is in Mail.ru Group. New and more complex requirements are imposed on machine learning technologies: in terms of quality, speed of reaction to changed working conditions (new type of spam, trends in social networks), analyzed content (text, pictures, videos).
In 2019, Mail.ru Group, together with other major Russian companies, created an alliance in artificial intelligence (AI-Russia Alliance). Its participants plan to join forces in separate projects to accelerate the growth of the Russian AI market.
It sounds complicated, but in practice, it means the partnership of the largest companies from different spheres – IT, finance, communications, and industry. The result is the creation of educational programs in the field of AI, the simplification of regulatory barriers that slow the introduction of new technologies, and the joint development of the market for AI-based solutions. It is expected that the volume of the Russian market will increase from 198.1 billion rubles up to 907.4 billion rubles. by 2024.
Where Does AI Training Data Come From?
There are different ways to mark up data for training artificial intelligence
The dataset is created with full access to the content. For this, Mail.ru Group has a service of assessors. These are several hundred people who, according to the instructions, perform tasks of varying degrees of difficulty. So, to improve search performance, assessors assess how relevant documents are to user requests. Or they help teach computer vision technologies, marking information on the data about the content of the text, the presence of objects and their number. For example, on the basis of this work, the Mail.ru Group computer vision algorithm has learned to decipher the license plates of cars and determine how many people are in the queue right now.