2022/08/01 21:51:27

Solutions based on artificial intelligence: technologies and implementations

Experts estimate the annual growth rate of the artificial intelligence solutions market in Russia at 20-30%. Which technologies provide such accelerated development? In what practical tasks do these technologies find massive use? Where are the growth points that determine the vectors of development of intelligent systems in the near future? Russian experts in the field of intelligent technologies and solutions helped TAdviser search for answers to these questions The article is included in the TAdviser review "Artificial Intelligence Technologies"

The company's researchers IDC believe that the Russian market for artificial intelligence (AI) solutions will demonstrate average annual growth at 18.5% until 2024. In 2023, it will overcome the $500 million mark, and by 2024 its volume will be 555.1 million. dollars The Competence Center MIPT "" Artificial intelligence estimates the volume of this market ruble in terms of. Its estimates are much more optimistic: at the end of 2021, the AI market in Russia amounted to approximately 550 billion rubles. (or $6.9 billion, if we count at the rate of $1 = 80 ₽). At the same time, market growth amounted to 28% compared to 2020.

The size of the AI market of the Russian Federation, billion rubles.

Source: "Artificial Intelligence 2021." MIPT, Artificial Intelligence Competence Center, collection No. 10, April 2022

At the same time investments states , AI through the federal project "Artificial Intelligence" amounted to 4.7 billion rubles in 2021, which is approximately twice as much as a year earlier. Of these, 960 million rubles. was spent on government purchases in the field of AI solutions.

Procurement in the field of AI in accordance with 44-FZ and 223-FZ, million rubles.

Source: "Artificial Intelligence 2021." MIPT, Artificial Intelligence Competence Center, collection No. 10, April 2022

According to IDC estimates made in the spring of 2021 financial , the sector remains the largest consumer of AI in the Russian market. And the focus of consumer attention here is fraud analysis and investigation technologies, as well as automated analysis and prevention of cyber threats. In second place among industries the most actively implementing AI solutions, IDC puts organizations retail and wholesale: trade they use AI to serve customers and increase revenues in digital sales channels through demand planning and price optimization, recommendation services and. chat boats

virtual assistants The analytics segment Gartner is called the largest segment in the global AI market. Expenditures on them on a global scale at the end of 2021 are estimated at $6.21 billion, which is 12% more than a year earlier. Higher growth rates - above 13.7% - are expected ON self-driving cars in the smart and digital jobs segments: $5.7 billion and $3.59 billion, respectively.

Estimates of the growth of the AI solutions market by various segments, million dollars.

In August 2021, the analytical company Statista released its report on the state of the AI market last year. In this study, experts conclude that the highest growth rates are expected in the field of collaborative robots (this segment, according to Statista's calculations, will more than double by 2026), automation of business processes using AI, unmanned vehicles and natural language processing.

Sanzhar Dosov, ML developer of Globus, agrees with analysts: {{quote 'Indeed, the modern world cannot be imagined without tools such as technical support chatbots, automated answers to customer questions and messages, automated translation of text into any existing language and others. Great success was achieved in this area thanks to the development of machine and deep learning technologies in the field of natural language processing (NLP). }}

Gartner prepared its "hype curve" of AI technologies last summer.

This picture clearly shows that today at the top of the interest of developers and researchers are knowledge graphs, the Transformers deep neural network architecture, which appeared in 2017, which shows good design results as described in the field of machine translation and automatic text summarization. Organizations are increasingly using AI solutions to create their own new products, combining NLP methods and new technologies at the peak of hype: knowledge graphs, as well as generative and composite AI, Gartner analysts say.

On average, it takes about eight months for an AI-based model to be integrated into the business process and deliver tangible benefits, says Shubhangi Vashisth, senior chief analyst at Gartner.

He believes that organizations will be forced to pay attention to the effectiveness of the architecture of AI systems and their operation in order to reduce the risks of failure of AI projects due to problems at the operational level.

In this regard, Gartner analysts expect that by 2025 70% organizations will modernize architectural approaches to the implementation of AI systems in order to achieve optimal orchestration of AI technologies and applications. On this path, such basic concepts for AI applications as, for example, data are being rethought.

Artificial intelligence systems are becoming "smarter" in proportion to the amount of data consumed, IDC analysts say. They estimate that big data is growing at a rate of approximately 40% annually. Such growth rates are associated, among other things, with the spread of smart devices, the Internet of Things and the use of social network data for the functioning of a variety of AI applications in various sectors of the economy. By 2025, the total global volume of big data will reach 163 trillion. Gb.

Data for AI

According to the estimates of the Artificial Intelligence Competence Center of the Moscow Institute of Physics and Technology, 1.7 zettabytes of data were produced in Russia in 2021.

Volume of data produced worldwide and in Russia, zettabyte

Source: "Artificial Intelligence 2021." MIPT, Artificial Intelligence Competence Center, collection No. 10, April 2022

An example of big data that is processed in modern smart systems is the digital forwarder Agorafreight, developed by Reksoft. In fact, this is an aggregator of logistics services that provides multimodal transportation (several types of transport can be used: road, railway, sea or aviation) between the Russian Federation and 150 countries of the world. The digital forwarder solves the most difficult optimization problem: working with millions of tariffs for various transportation, within a few seconds it selects the environments of hundreds of routes and tariffs best for the client, in terms of the time and cost of delivery of cargo.

The system is implemented on the basis of MongoDB, a document-oriented NoSQL DBMS with support for geo-queries, full-text search in 15 languages and a hierarchical data structure. The system scales horizontally and can be used as a file storage with load balancing and data replication, according to Reksoft.

If we want to create a machine learning model with "broad data," then it may face the problem of increasing complexity. A large amount of incoming information in the training model will cause the training of the model to be either unstable or very time consuming.

This procedure is especially useful when separating "garbage" from valuable information, - add Megaputer specialists.

Together, they are able to make more efficient use of available data, either by working with small amounts of data or by benefiting more from unstructured, diverse data sources, the study authors say.

And the concept of Smart Data appears - this is actually the result of some preliminary processing of the collected information, which has a certain value for further processing. Another option for valuable data is Fast Data. This is the name of those identified in the big data stream that have value in solving a particular problem.

At the same time, Dmitry Lapin, head of the Data ScienceJSA Group, talks about the lack of data as the main problem that industrial enterprises have to deal with when trying to implement intelligent IT solutions: it is not easy to obtain data generated by smart machines and mechanisms, and besides, they still need to be able to professionally interpret.

Such a direction as buying external data (or enriching internal data) from specialized companies is promising, for example, purchasing as a service (Data Brokers, Data as a Service), buying and selling data on specialized sites (B2B Data Marketplaces), says Alexander Khledenev, director of digital solutions at VS Lab.

Gartner analysts in their study of the main trends in data processing and analytics Gartner Top 10 Trends in Data and Analytics for 2020 predicted that by 2022, 35% large organizations will become either sellers or buyers of data on specialized marketplaces and data exchanges. (In 2020, the share of such enterprises was 25%).

Intelligent processing with AI gives additional incentives for its development. Notable examples: Acxiom, Epsilon, Equifax, Experian, Oracle. Among the data exchanges are Nikkei Asia, Google BigQuery, Snowflake and Adobe.

In Russia, according to the MIPT Artificial Intelligence Competence Center, the data market has been growing steadily since 2010, despite the crises and the COVID pandemic. In 2021, its size reached 46 billion rubles.

In Russia, well-known data providers are myTarget, Yandex, GetIntent. The leader in the Data as a Service segment can probably be called CleverDATA, - says Alexander Khledenev.

CleverData owns the Hermann.AI platform, designed to automate marketing communications and the 1DMC data exchange, the largest independent platform in Runet for the exchange of external data between data suppliers and consumers. Today it presents more than two dozen data providers (with a daily audience of more than 90 million people), over 9 thousand data sources.

Among the data providers are companies that own large amounts of various user data (search services, applications, payment services, user data collection platforms).

Consumers can, for example, enrich their data with external data from suppliers in the form of profiles or segments. The service provides periodic profile uploads, as well as the ability to receive profiles in real time through the API. It is also possible to deliver data: delivery of segments built on external data to advertising platforms: myTarget (Mail.ru), DoubleClick (Google), Yandex. You can also form unique segments of the target audience, say, by interest and behavior, using all available data on the exchange. The company says that the customer, say, can enrich the data on their customers available in the corporate CRM system with additional information on 3000 or more attributes.

By order of NP GLONASS, the Data Exchange marketplace was created, which plays the role of an independent trading platform for data in the motor transport sector within the framework of the Avtodata federal service navigation and telematics platform.

With the help of the Data Exchange, our compatriots will be the first in the world to be able to use the information generated by the car, sell and dispose of it at their discretion, "said Alexander Gurko, President of GLONASS NP, talking about the completion of development in June last year.

In fact, the "Data Exchange" is a tool for commercializing the services of the Avtodata platform: on this commercial site you can anonymously post data taken from real cars, accompanying them with detailed information about possible applications and use scenarios. And this opportunity is potentially in demand.

According to NP GLONASS estimates, more than 70% of commercial vehicles are already connected to various telematic platforms. As for individuals, NTI Avtonet, together with the consulting company Roland Berger, conducted a survey of Russians, which showed that more than 60% of Russians expressed their readiness to share data on the operation of a car with the ability to sell them on the data exchange created as part of the Avtodata project.

The National Platform "Avtodata" is designed for the collection, verification, enrichment, processing, analysis and further practical application of automotive "big data" using the most advanced technologies: algorithms and technologies of artificial intelligence and machine learning.

According to preliminary calculations, the volume of platform data for the sixth year of operation will exceed 15 PB. More than 170 Pb of storage capacity will be required to store them on a redundant basis. Information obtained from dozens of external systems will be enriched and processed by methods of artificial intelligence, predictive and advisory analytics. Consumers in the service format will receive data of a high degree of processing. Moreover, at the start of industrial operation of the platform, several hundred services will be formed for various categories of customers.

Another aspect of effective management of large and broad data on the company side is the use of master data management (MDM) systems, which minimize efforts to minimize chaos in client data.

A good MDM system is a flexibly configurable platform for standardizing and validating data, building reference profiles based on source disparate data. The basic functionality of such a platform is the ability for any field to set rules for cleaning, searching for duplicates, rules for merging and updating, "says Mikhail Berezin, Product Manager" One Client, "an expert on MDM solutions from HFLabs.

He emphasizes that such a system should initially contain ready-made rules for processing raw data

It may seem that the processes of working with data in all companies are different and complete freedom in initial configuration - this is exactly the opportunity to correspond to unique business processes. But, in our experience, this only leads to a delay in projects, - explains Mikhail Berezin. - A fast to implement and reliable MDM system should contain maximum knowledge about typical work with data in a particular industry (bank, insurance, telecom, retail, etc.) in order to immediately give a business result out of the box. And after that, provide the ability to fine-tune for the uniqueness of the business process.

In general, moving towards storing more and more useful knowledge, and not thoughtlessly accumulated "big" data, seems inevitable, the expert believes. True, so far it is impossible to distinguish the established universal technologies and approaches to solving this problem. However, there are a lot of intelligent data processing technologies on the market.

One of the options for classifying intelligent information systems by the types of technologies used

Source: Intelligent Information Systems and Technologies: Monograph/A.V. Ostroukh, N.E. Surkova. - Krasnoyarsk: Scientific and Innovation Center, 2015. - 370 p.

From the point of view of the principles of organizing work with data, two large areas are distinguished in the field of AI: knowledge-based systems (formalized and informal), and artificial neural networks. Moreover, Gartner analysts in the study Gartner Hype Cycle for Artificial Intelligence, 2021 express the opinion that in the coming years the leader in development dynamics will be the segment of solutions in the field of knowledge management. Let's see further how both of these directions are developing, together forming the appearance of modern AI solutions.

Next Overview Material > >
> Browse Home > > >

Solutions based on artificial intelligence: technologies and implementations

Content

Data for AI

Other Review Materials

Other materials on the topic of AI