''Google developed an algorithm similar to Yandex’s Korolyov already in 2013''
Yandex has updated its search engine and presented the new ranking algorithm Korolyov. The presentation was held with fanfare, almost all Russian media wrote about this, and some experts considered this step of the Russian Internet companies to be revolutionary. The main difference from previous version is that the new algorithm assesses requests not by key words but meaning, what is more — with high accuracy due to the use of learning neural networks. Realnoe Vremya had a conversation with Vice President of the Russian-Tunisian business council and an IT entrepreneur Elbrus Latypov, who using a plain language explained how Korolyov will work, how soon the users will notice the difference in the results given to them, whether Yandex is a pioneer and how many people are now training the neural network of the Russian company.
''It is a good PR move, more directed to attract a market share in Russia''
Elbrus, how do you assess Korolyov? Is it true that this algorithm can be called revolutionary?
As for revolutionary, it is rather a PR move. In this case, Yandex adopts experience from American partners, with an intention to present something ''new''. In fact, if to take Google, they developed a similar algorithm called Hummingbird already back in 2013. Google that time behaved in a more modest way – they developed the algorithm and presented it exclusively in the professional field, not to the wide public.
Are there other companies in addition to Google and Yandex that use neural network?
Neural network is a buzzword today. Machine learning, neural network, artificial intelligence are a kind of trend, it is not the future but the present. In general, the information volume on the Internet is such that the old algorithms that were developed by Google and Yandex are already not suitable. Linear analysis of requests does not work anymore. There is a huge amount of information, which intersects with other information, and people who use serch engines do not get appropriate results.
''If to take Google, they developed a similar algorithm called Hummingbird already back in 2013. Google that time behaved in a more modest way – they developed the algorithm and presented it exclusively in the professional field, not to the wide public.'' Photo: siteclinic.ru
Hummingbird, Korolyov, Palekh – such system is effective when a person tries to find, for example, a name of a movie or a song, but they have only a part of its description. In that case the linear search doesn't work. Here it should work like a human brain – to analyse the picture as a whole and to give an answer not linear but semantic. The human brain operates with meanings, not some particular linear indicators.
For example, you are looking for the film Armageddon but forgot its name. You write in the search: ''People fly to the meteor and save the Earth''. Linear search will give you some separate links on each word or phrase. It will give you a meteor, the Earth, some topical images, but not the film itself. The new algorithm search meaning of all the words together, it unites them and give the only correct answer that the human brain would find.
If we go back to Google and Yandex, then after doing an elementary analysis we get the same result. Google and Yandex have almost the same algorithms. We do not know them, we do not see them – it is a secret, but in the end we have almost the same result. None of the systems has some serious advantage.
Naturally, Yandex work more correctly with Russian requests. However, Google has also advanced in recent years in this field and gives correct answers. By the way, Google has a higher relevancy than Yandex does.
That's why it is difficult to tell something about revolutionness here. It is a good presentation, good PR move, but more directed to attract a market share on Russian market.
''Yandex has attracted more than a million of volunteers who teach the neural network by personal experience. Each person is given the task: to search something and choose more correct answers. It is the way how the neural network learns.'' Photo: inorehovo.ru
The neural network is in the process of learning now. What do you think, how long it will take to show a tangible result to the public?
The neural network is an algorithm that learns with participation of real people. Today, according to the information I have, Yandex has attracted more than a million of volunteers who teach the neural network by personal experience. Each person is given the task: to search something and choose more correct answers. It is the way how the neural network learns. The effectiveness of the neural network will more depend on algorithms that would be able to process the information. In fact, it is a kind of threshold and likeness to artificial intelligence that will analyse by its own, but for now it needs learning from real people.
The problem is that the human brain works much more powerfully that any technology. All the existing neural networks are very primitive in comparison to out brain. Today it is difficult to tell how long it will take in order the technology approached human capabilities, to the analysis and perception of the same level.
''By and large, neural networks and artificial intelligence are in embryo state''
Do you agree with those who tell that the quality of the issued content will increase due to Korolyov? When will the user see it?
I think now it will be difficult to notice. This huge work will be noticeable in a few years. For now it is impossible to see it with the ''naked eye''. By and large, the neural networks and artificial intelligence are being in embryo state, and a huge work awaits us. I thing, for now, even 10% of the work have not been done yet.
You probably have heard about a research, the result of which show that women usually formulate their request more detailed, while men use key words. Will the algorithm unify this moment, how do you think?
You have touched an interesting moment. We will return to the uniqueness of the human brain, that it functions differently of men and women. Men think more specifically, women – more abstract. It is a huge work for artificial intelligence as well as for learning of neural networks – to differentiate between male and female.
''The developers of AI and neural networks have a huge task – not only differentiate the wishes of the users but also recognise their gender. The task is very interesting.'' Photo: iab.ru
Even if we take a similar request of a woman and of a man, the necessary result for each may be very different, although they both will use a similar set words. They want different result, you see? It is a different story. That is why the developers of AI and neural networks have a huge task – not only differentiate the wishes of the users but also recognise their gender. The task is very interesting.
Is it true that website promotion will be more complicated because of Koroyov? What will happen to the methods of SEO copywriting?
I think that SEO won't change. Now there is a confrontation – there are those who promote websites and there search engines themselves. They try to optimise search and to give their websites in the top, but in many cases it is not completely fair. In their turn, the search engines fight with this. One should also bear in mind that they both make money on it.
Perhaps, we will see a different mean of promoting. Maybe, in the result, in order to optimise the search and to make it fair for all players, particularly neural networks and AI will be involved. I don't think that the rules of the game will change significantly, rather new mechanisms will be used because it is a business first of all.