Artificial intelligence

We live on the very thresh­old of the era of AI or arti­fi­cial intel­li­gence, but the con­tours of the future are already vis­i­ble. About what AI is and how it devel­ops — in the mate­r­i­al of Kom­so­mol­skaya Prav­da
Artificial intelligence
Arti­fi­cial intel­li­gence. Pho­to: pix­abay

For the first time, sci­en­tists start­ed talk­ing about arti­fi­cial intel­li­gence (AI) in the ear­ly 40s of the last cen­tu­ry. In Eng­lish, it is called Arti­fi­cial intel­li­gence and the abbre­vi­a­tion AI is used. Today, we increas­ing­ly meet this phrase in every­day life. Kom­so­mol­skaya Prav­da set out to explain in sim­ple terms what arti­fi­cial intel­li­gence is, what is the essence and tech­nol­o­gy of this sys­tem.

What is artificial intelligence

“You can’t see or feel arti­fi­cial intel­li­gence. It is a pro­gram that takes a sequence of data and ana­lyzes it. The result can be face recog­ni­tion, text recog­ni­tion, text gen­er­a­tion, and much more, to which the char­ac­ter­is­tic “tem­plate approach” is applic­a­ble, — says tech­ni­cal direc­tor of Chrono.tech Mikhail Savchenko.

- Arti­fi­cial intel­li­gence in the broad­est sense — a set of meth­ods and mod­els that receive data and draw a cer­tain con­clu­sion based on them. This may be assign­ing a pho­to to a cer­tain class, group­ing texts on sim­i­lar top­ics, or pre­dict­ing the exchange rate. All mod­els are unit­ed by the extrac­tion of knowl­edge from data. Cal­cu­la­tion of a sin­gle func­tion that can con­tain mil­lions and even bil­lions of para­me­ters, says the soft­ware Ilya Makarov, direc­tor of the MADE Big Data Acad­e­my at Mail.ru.



In essence, AI is a pro­gram, like some noto­ri­ous text edi­tor. But it dif­fers in that it can be trained.

- The pro­gram itself is a set of code and a knowl­edge base that it received in the learn­ing process. Both are just a set of data on the hard dri­ve of some com­put­er, notes Evge­ny Kar­pov, AI prod­uct man­ag­er at Zyfra.

Experts notice that the con­cept is often spec­u­lat­ed, espe­cial­ly in the mar­ket­ing of var­i­ous prod­ucts.

- We dis­tin­guish a more intel­li­gent per­son by behav­ior, and the abil­i­ty to find the right solu­tions for more com­plex prob­lems. Sim­i­lar­ly, an arti­fi­cial intel­li­gence sys­tem is dis­tin­guished not by its spe­cial appear­ance, but by the behav­ior and tasks that it is able to solve. Devel­op­ers, of course, can “see” arti­fi­cial intel­li­gence in the text of the algo­rithm, explains Alex­ey Dyachenko, Deputy Gen­er­al Direc­tor of Open Tech­nolo­gies.

The essence of artificial intelligence

Like any oth­er com­put­er pro­gram, AI works like this: it takes data, ana­lyzes it, and pro­duces a result. But the essence of arti­fi­cial intel­li­gence lies in its fea­tures.

“The struc­ture dis­tin­guish­es it from a reg­u­lar pro­gram: an arti­fi­cial neur­al net­work, which, after being cre­at­ed by a pro­gram­mer, goes through a train­ing pro­ce­dure in order to make the results of data analy­sis as accu­rate as pos­si­ble,” says Alex­ey Dyachenko.

If we sim­ply explain the essence of AI, then the algo­rithm has two modes: work and train­ing.

- For exam­ple, we have an AI that deter­mines whether a per­son wears a hel­met at the fac­to­ry or not. To make it work, we show 10,000 pho­tos of peo­ple in hel­mets and say that they are peo­ple in hel­mets, then we show 10,000 peo­ple with­out hel­mets and say that they are with­out hel­mets. After that, we trans­fer it to the oper­at­ing mode and show a new pho­to. And our AI can already deter­mine whether a per­son is wear­ing a hel­met or not. Thanks to this, we can sig­nal a safe­ty vio­la­tion and save work­ers from dan­ger­ous injuries,” notes Evge­ny Kar­pov.

But this exam­ple does not exhaust the essence of arti­fi­cial intel­li­gence. One of the most impor­tant com­po­nents of tech­nol­o­gy is self-learn­ing.

- In addi­tion to the ini­tial­ly laid down, the algo­rithm is able to receive and process new infor­ma­tion in order to make more com­plex and effec­tive deci­sions based on it in the future. In addi­tion, learn­ing occurs with each new com­plet­ed task. Arti­fi­cial intel­li­gence improves on its own expe­ri­ence. This is called machine learn­ing, explains Direc­tor for Inno­v­a­tive Solu­tions at Oberon Evge­ny Ovcharov.



- AI has advan­tages — it does not get tired and is able to process a huge amount of infor­ma­tion in a short time. It dif­fers from human think­ing in that it does not yet have emo­tions and does not depend on them. Where­as the think­ing of man is always sub­ject to them, says IT spe­cial­ist Alexan­der Sere­bryakov.

Development of artificial intelligence technologies

How it all started

If you dig very deep, you need to go back to the 17th cen­tu­ry, when Wil­helm Schickard cre­at­ed the first mechan­i­cal cal­cu­lat­ing machine — the cal­cu­la­tor. And the philoso­pher and math­e­mati­cian Descartes pre­sent­ed ani­mals as a think­ing mech­a­nism and thus set the task of human civ­i­liza­tion to cre­ate its own intel­li­gent mech­a­nism.

First ideas

The con­cept of an arti­fi­cial neur­al net­work was pro­posed in 1943 by Amer­i­can sci­en­tists — the founder of cyber­net­ics McCol­loch and math­e­mati­cian Pitts. A year lat­er, their col­league from Prince­ton (USA), von Neu­mann, will offer a vari­ant of the archi­tec­ture that formed the basis of all mod­ern com­put­ers. In 1950, math­e­mati­cian Alan Tur­ing pub­lished his famous test to detect strong AI. When a per­son can­not under­stand that he is talk­ing to a machine, but thinks that anoth­er per­son is stand­ing behind the screen, then true arti­fi­cial intel­li­gence has been born.


In the 60s of the XX cen­tu­ry, sci­en­tif­ic uni­ver­si­ties of world pow­ers are active­ly work­ing on the devel­op­ment of tech­nol­o­gy. For exam­ple, in the USSR it was the cyber­net­ics Pospelov and Tsetlin. And already in the next decade, an explana­to­ry dic­tio­nary, ref­er­ence books and volu­mi­nous sci­en­tif­ic works devot­ed to the top­ic of arti­fi­cial intel­li­gence as an inde­pen­dent phe­nom­e­non were pub­lished. In the West, in 1965, they cre­ate a robot, Eliza, who could speak Eng­lish. Today we would call it a chat­bot. Four years lat­er, the android She­ki appears, who also knew how to move.

Interview with Eliza.  Photo: wikipedia.org
Inter­view with Eliza. Pho­to: wikipedia.org

Breakthrough in the development of artificial intelligence in the 80–90s

In the 80s, a qual­i­ta­tive­ly dif­fer­ent era in the devel­op­ment of AI began. There are even more advanced robot con­sul­tants that could solve math­e­mat­i­cal and some every­day prob­lems, learned even bet­ter to keep up a con­ver­sa­tion. At the end of the decade, they devel­op the Deep Thought pro­gram, which beats Grand­mas­ter Larsen. Then the chess play­er Gar­ry Kas­parov will say that he does not believe in the capa­bil­i­ties of the machine. They say if they can beat the best chess play­ers, then they will com­pose the best music and write the best books. The grand­mas­ter announced that on behalf of the human race he would fight in a chess duel with AI. Such an oppor­tu­ni­ty will be pre­sent­ed to him in 1996 and 1997. Kas­parov won the first match and lost the sec­ond.


Start­ing from the “zero” they begin to intro­duce the devel­op­ment of a smart home. Arti­fi­cial intel­li­gence is find­ing appli­ca­tions in the grow­ing PC gam­ing indus­try. There are a num­ber of com­plaints against him, they say he still acts accord­ing to pat­terns, and is not guid­ed by the sit­u­a­tion. But over time, devel­op­ers elim­i­nate this prob­lem, and the machine grad­u­al­ly learns by itself.

AI is active­ly begin­ning to be used in urban ser­vices. The same def­i­n­i­tion of faces through cam­eras. And with the mass char­ac­ter of smart­phones and the abil­i­ty to eas­i­ly install appli­ca­tions on each mobile phone, arti­fi­cial intel­li­gence is some­how being intro­duced into pro­grams.

To under­stand the scale of the seri­ous­ness of the atti­tude of top gov­ern­ment offi­cials to the top­ic of AI, just look at the decree of the Pres­i­dent of Rus­sia, which refers to the cre­ation of a nation­al strat­e­gy for the devel­op­ment of arti­fi­cial intel­li­gence in the coun­try. The plan is up to 2030.

Artificial intelligence control systems

Not all experts accept this for­mu­la­tion. Some believe that instead of con­trol sys­tems, it is more appro­pri­ate to speak of algo­rithms and arti­fi­cial intel­li­gence ser­vices. They are divid­ed into two types.

  • The spe­cial one is well able to solve a nar­row area of ​​tem­plate tasks. Each needs its own algo­rithm. This AI will nev­er be able to evolve into a full-fledged assis­tant or com­peti­tor to a per­son. In robot­ics, algo­rithms are only suit­able if the machine will per­form a nar­row task. For exam­ple, open and close the door.
  • Gen­er­al pur­pose (AGI Arti­fi­cial gen­er­al intel­li­gence) — in the future they will acquire cog­ni­tive (cog­ni­tive) abil­i­ties, emo­tions, the abil­i­ty to learn, make deci­sions for a wide range of areas, self-aware­ness, etc.



Who develops artificial intelligence


It would be strange if the tech­nol­o­gy giant who invent­ed the most pop­u­lar search engine and the most mas­sive mobile OS did not devel­op arti­fi­cial intel­li­gence. After all, they can imme­di­ate­ly imple­ment their devel­op­ments in prod­ucts, which means they will receive prof­it from this. Dif­fer­ent Google APIs — that is, tools for devel­op­ing pro­grams, are trained in speech recog­ni­tion, trans­la­tion from for­eign lan­guages, main­tain­ing a tem­plate con­ver­sa­tion, for exam­ple, in a call cen­ter. Of course, these are the most under­stand­able devel­op­ments for a sim­ple user. Although the com­pa­ny has hun­dreds of AI projects in one area or anoth­er. Most, of course, are relat­ed to online sales.


The com­pa­ny cre­at­ed a cog­ni­tive archi­tec­ture, in sim­ple terms, pieces of com­put­er code that are aimed at cre­at­ing arti­fi­cial intel­li­gence. The pro­gram is open, that is, it can be used by any devel­op­er. The com­pa­ny believes that by joint efforts the com­mu­ni­ty will quick­ly come to the cre­ation of a full-fledged AI that will be strong not only in its field, but in gen­er­al in all.

Chinese Institute of Artificial Brain

Sci­en­tists from the Celes­tial Empire have its lab­o­ra­to­ry to cre­ate a “brain-like” arti­fi­cial intel­li­gence. Offi­cial­ly, they are engaged in neur­al net­works capa­ble of rec­og­niz­ing objects, human mes­sages or cre­at­ing AI, which, in a sec­ond after an acci­dent, sig­nals an inci­dent to emer­gency ser­vices. But one of the most ambi­tious devel­op­ments of Asian sci­en­tists is the sys­tem of social trust.


The most pop­u­lar AI from an Amer­i­can cor­po­ra­tion is called Azure. His firm sells to cor­po­ra­tions that devel­op soft­ware. The neur­al net­work under­stands speech, is able to make pre­dic­tions and imi­tates oth­er human intel­lec­tu­al abil­i­ties. The com­pa­ny makes a spe­cial empha­sis on AI, which is trained to rec­og­nize errors in the code. And this means that soon AI will be taught to cre­ate pro­grams on its own.

How to con­nect phone to TV

The most trou­ble-free ways to con­nect your phone to your TV when you have a USB cable at hand

Russian players

— In Rus­sia, there are no nor­mal­ly func­tion­ing AI cen­ters of excel­lence inte­grat­ed with busi­ness. There are a num­ber of com­pa­nies that are suc­cess­ful­ly pro­mot­ing AI-based solu­tions in their busi­ness mod­els, such as Sber­bank and Mail.ru Group, Yan­dex and Kasper­sky Lab, Sibur Cor­po­ra­tion and Sev­er­stal. But the solu­tions are not unique in many ways. The rep­re­sen­ta­tion of Russ­ian sci­en­tists and AI devel­op­ers at lead­ing inter­na­tion­al con­fer­ences is extreme­ly small com­pared to sci­en­tists from the USA, Chi­na and Europe, Ilya Makarov believes.

Popular questions and answers

Why are more com­pa­nies using arti­fi­cial intel­li­gence?
- Most ana­lysts pre­dict that the intro­duc­tion of AI will increase the effi­cien­cy of the com­pa­ny by tens of per­cent. In busi­ness, a huge num­ber of mis­takes hap­pen due to the human fac­tor. AI will increase the accu­ra­cy of oper­a­tional process­es, includ­ing when work­ing with doc­u­ments and when enter­ing data. There are already and are active­ly used neur­al net­works that rec­og­nize, for exam­ple, pass­port data and enter data into the cor­re­spond­ing cells of the sys­tem. Arti­fi­cial intel­li­gence is becom­ing a key ele­ment in increas­ing the com­pet­i­tive­ness of both com­pa­nies and entire states in the era of indus­tri­al rev­o­lu­tion 4.0, says Direc­tor of the IT Insti­tute of the Syn­er­gy Uni­ver­si­ty Stanislav Kosarev.
Where is arti­fi­cial intel­li­gence applied?
At the request of KP, ABBYY Rus­sia CEO Dmit­ry Shushkin told in what areas AI is used today.

in banks. Tech­nolo­gies allow open­ing accounts for peo­ple and com­pa­nies with lit­tle or no par­tic­i­pa­tion of employ­ees, includ­ing remote­ly. Based on infor­ma­tion about the bor­row­er, deter­mine the size and suit­able terms of the loan, ana­lyze news about the bank’s part­ners and iden­ti­fy risks, and effec­tive­ly trade on the stock exchange.

In retail. They help to under­stand the demo­graph­ic char­ac­ter­is­tics of the buy­er and offer the most suit­able prod­ucts. Con­trol the fill­ing of shelves, under­stand the rea­sons for queues in the store, bet­ter orga­nize deliv­ery and pur­chase from sup­pli­ers, sim­pli­fy the work of account­ing.

In tele­com. Pre­dicts the out­flow of sub­scribers, effec­tive­ly dis­trib­utes calls to call cen­ters.

In logis­tics. It helps to build routes for freight trans­porta­tion, con­trol fuel costs, extract and ana­lyze data from trans­port doc­u­ments.

Music ser­vices. They use machine learn­ing meth­ods to ana­lyze the user’s musi­cal tastes.

In med­i­cine. Cana­di­an com­pa­ny Blue­Dot uses AI to track the spread of infec­tious dis­eases. They warned about pneu­mo­nia in Wuhan a week before the announce­ment of the coro­n­avirus epi­dem­ic. AI trained by Sber­bank has learned to detect char­ac­ter­is­tic black­outs in the lungs caused by coro­n­avirus infec­tion. Sci­en­tists from the lead­ing Amer­i­can uni­ver­si­ty MIT used AI to cre­ate new antibi­otics.

How much does arti­fi­cial intel­li­gence cost?
- It is impos­si­ble to give a spe­cif­ic answer to this ques­tion: it all depends on the type of busi­ness, the scope of the tasks being solved. In short, it’s expen­sive. So far, only large busi­ness­es can afford it. As AI-relat­ed solu­tions devel­op, as new mod­els are trained and data is enriched, the cost of imple­men­ta­tion and use will decrease,” says Stanislav Kosarev.

There are many freely dis­trib­uted libraries that imple­ment arti­fi­cial intel­li­gence algo­rithms that a pro­gram­mer can sim­ply down­load, adapt to his data and use in his soft­ware prod­uct. There are also ready-made prod­ucts, for exam­ple, the Com­ma One project is an open source autopi­lot for a car that you can sim­ply down­load. Of course, there are also closed sys­tems. So, Google annu­al­ly spends bil­lions of dol­lars annu­al­ly on the devel­op­ment of arti­fi­cial intel­li­gence sys­tems, while, if the com­pa­ny pro­tects search algo­rithms, then some sec­ondary libraries are pub­lished in the pub­lic domain.

When will AI ful­ly enter our lives?
- If we are talk­ing about replac­ing a per­son with a formed and intel­li­gent per­son who can become a friend, then this should not be expect­ed in the fore­see­able future. Oth­er­wise, dai­ly life, ser­vices and com­mu­ni­ca­tion — AI has already pen­e­trat­ed tight­ly into these areas and reg­u­lar­ly helps us out. Algo­rithms work where we no longer notice them: from build­ing the opti­mal route by a nav­i­ga­tor to per­son­al rec­om­men­da­tions from a music ser­vice, says Evge­ny Ovcharov.
Can you rec­om­mend books, films, games that most plau­si­bly show arti­fi­cial intel­li­gence?
- All the Ter­mi­na­tor series, where the Skynet sys­tem, which has an inde­pen­dent life, acts as a source of dan­ger, and a per­son is already super­flu­ous in this world and is fight­ing for sal­va­tion. I also rec­om­mend the tril­o­gy “The Matrix”, “Zoey”, “Blade Run­ner. 2049”. The best movie in my opin­ion is Lucy. This is sci­ence fic­tion, where the main empha­sis is not on the pow­er of arti­fi­cial intel­li­gence, but on human intel­li­gence, on its unlim­it­ed abil­i­ties. This is like a sig­nal that human­i­ty should devel­op itself, and not machines that will replace them togeth­er and in every­thing,” says inno­v­a­tive tech­nol­o­gy expert Vio­let­ta Gali­cian.

— There are few books in Russ­ian on AI. I would rec­om­mend the work of Sergey Nikolenko “Deep Learn­ing”. But to be in the sub­ject, you need to active­ly read arti­cles in Eng­lish from lead­ing con­fer­ences. Well, if you want to immerse your­self in the world of fan­ta­sy relat­ed to AI and its per­cep­tion, I advise you to check out the video games from the Mass Effect and Deus Ex series, as well as the films Arti­fi­cial Intel­li­gence, I, Robot, and oth­ers. For those who want to become a part of his­to­ry and cre­ate AI them­selves — the MADE Acad­e­my of Big Data, advis­es Ilya Makarov.

- The best option is a course from Yan­dex and MIPT on AI tech­nolo­gies. If you pass it, you will under­stand what AI is very well. Movies, books and games are full of fan­tasies that are very far from real­i­ty, notes Evge­ny Kar­pov.

Can AI pose a dan­ger to humans?
- A sig­nif­i­cant part of the worst fears about arti­fi­cial intel­li­gence is unfound­ed. Its devel­op­ment depends on the per­son and the cri­te­ria set by him for assess­ing the sur­round­ing world and events. That is, a sce­nario in which AI inde­pen­dent­ly “acquires” the desire to harm peo­ple is sim­ply impos­si­ble. By itself, the machine will not even under­stand what harm is and why it is nec­es­sary to inflict it, AI does not have such a com­po­nent as moti­va­tion and desire,” says Man­ag­ing Direc­tor of Applied Intel­li­gence Accen­ture Prac­tice in Rus­sia Lar­isa Malko­va.

Accord­ing to the expert, anoth­er cat­e­go­ry of fears — “robots and pro­grams will replace every­one” — is also great­ly over­es­ti­mat­ed. The labor mar­ket will change, but not dra­mat­i­cal­ly. The inter­ac­tion mod­el “Man + Machine” will become even wider.

What is the dif­fer­ence between arti­fi­cial intel­li­gence and neur­al net­works?
- AI solu­tions can work on dif­fer­ent algo­rithms, approx­i­mate­ly how a car can work on a gaso­line, diesel or elec­tric engine. One of these algo­rithms is neur­al net­works. Now they have become very pop­u­lar, explains Evge­ny Kar­pov.

- In a sep­a­rate cat­e­go­ry, we can sin­gle out the use of AI based on neur­al net­works for video image analy­sis. Such intel­li­gent video ana­lyt­ics is able to solve the most com­plex tasks. The num­ber of poten­tial appli­ca­tions of this tech­nol­o­gy for secu­ri­ty and smart video sur­veil­lance is tru­ly enor­mous. From the detec­tion by video ana­lyt­ics sys­tems of unusu­al or aggres­sive behav­ior aimed at harm­ing peo­ple, to the recog­ni­tion of indi­vid­u­als even in large crowds. And even the auto­mat­ic detec­tion of the noto­ri­ous “bomb in a suit­case” — the imple­men­ta­tion of bag­gage screen­ing with­out the need for X‑ray instal­la­tions, — adds DSSL Lead­ing Spe­cial­ist Mikhail Sharu­bin.

What is machine learn­ing?
“This is the prin­ci­ple of the evo­lu­tion of arti­fi­cial intel­li­gence, which is based on self-learn­ing algo­rithms,” says Vio­let­ta Gal­it­skaya. — The nec­es­sary infor­ma­tion and goals are loaded into the machine’s mem­o­ry. There are three types of machine learn­ing:

1) with the par­tic­i­pa­tion of a per­son when he cor­rects the goal;

2) with­out human inter­ven­tion. Robot­ic tech­nol­o­gy inde­pen­dent­ly, by com­par­ing pat­terns, is look­ing for a more opti­mal solu­tion, there­by hon­ing analy­sis skills;

3) mixed learn­ing process. The main fea­ture: the pro­cess­ing of a large amount of infor­ma­tion with the par­tic­i­pa­tion of neur­al net­works.