Virtual Assistants

It is estimated that IDC the growth rate of investment in the creation of digital assistants is among the highest in IT-. industries Moreover, they reached this level after the companies "tried" the possibilities artificial intelligence() AI in the processes of customer service and to automate internal and external routine operations.

The existence of such systems became possible with the development of a very important and extensive direction in, machine learning which is called natural language processing (Natural Language Processing,), NLP - says the head Ilya Pomerantsev of the ML department of the company. - The Globus IT work virtual of assistants is provided by the "three" main technologies: (speech recognition Speech-to-Text), which is used in assistants that are able to interact with a person, by voice text analysis and (speech synthesis Text-to-Speech). "

The developers have learned to better convey emotions in speech synthesis, virtual characters spoke in different voices, telephone secretaries defended users from spam calls, he says in an article in Techinsider magazine (February 2022).

At the same time, the basic technologies of the bot are two main problems: text classification and named entity recognition (NER), says Ilya Pomerantsev: Шаблон:Quote 'And only with a high-quality solution to these problems, it is worth adding subsequent improvements to the chatbot.

Bot creation tools

The demand for digital assistants gives rise to an offer: today on the Russian market there are several types of systems designed to create smart virtual assistants: ready-made bots, bot designers, dialog platforms, frameworks.

Using a finished bot is the fastest way to implement a virtual assistant in your project, which is suitable for small and narrowly focused tasks and is not designed for customization. Such bots are able to conduct polls, moderate chats, place orders, write down for a consultation or give users the necessary information on demand, but a ready-made bot is always aimed at solving problems in a specific business segment.

Just AI has created a marketplace for ready-made bots Solution Store, where you can find an assistant for narrow tasks: housing and communal services, retail, e-commerce, fintech and others.

The bot designer is a set of out-of-the-box tools that allow you to create and customize bots according to your own scenarios without technical skills and developers. For example, the universal Aimylogic designer of Just AI includes 30 ready-made integrations and channels: popular chats and instant messengers, CRM and analytics tools, mailing services, etc. Users can not only place their bots in many channels, but also create complex scripts for the chatbot, launch mass calls, automate work with incoming phone calls and even create new skills for voice assistants. For example, you can teach a bot to send images and videos, call and write on a schedule, run mailings, etc.

The Aimylogic constructor contains a Natural Language Understanding (NLU) block that allows you to teach the bot new meanings and phrases. Aimylogic bots can communicate in different languages: Russian, English, Kazakh, Portuguese, Spanish.

The designer of JINNEE, developed by ISS, is based on the basic set of tools, Andrey Kulyashov, director of business development at ISS:

• a proofreading module that corrects errors common in the natural language;
• a tokenizer that forms large text blocks that are understandable to the machine;
• lexical analysis module responsible for determining significant sequences in the text;
• a morphological module that normalizes words that highlight lemmas, as well as is responsible for eliminating non-characters of words;
• a parsing module that establishes links between words and groups;
• a semantic analysis module that identifies entities and intentions;
• NER (Named Entity Recognition) module, which is responsible for recognizing named entities in the text.

This set of tools is capable of solving business problems in any industry. At the same time, the bot designer supports a wide range of functionality: in your personal account, you can write any scenarios for processing essential requests and maintaining dialogs. " Thus, the virtual assistant of the HR department takes on a wide range of personnel tasks: from selection of candidates and primary questionnaire to exit interviews and collecting reviews about the company. In addition to recruiting and onboarding, JINNEE bots are engaged in automation of technical support, processing of incoming requests, document management, work with subcontractors, sales, etc. One of the JINNEE chat bots is integrated into the all-Russian system of services for employees and employers Онлайнинспекция.рф - there he answers questions on labor law. Next in line are new tasks: providing pre-configured, ready-made solutions for retail, banks, insurance companies, etc., which are planned to be implemented during the current year.

Thanks to its self-learning ability, the JINNEE NLU module accurately understands the context of the questions asked, ISS says. The point is that JINNEE recognizes entities and their context using an NLU module by searching the library and keywords. It is known that the company reminds that many of the current bots, having answered the first question in the conversation thread, immediately "forget" the context, and even if the next question is related to the previous one, the dialogue begins anew, in fact, from scratch.

Шаблон:Quote 'This most often annoys users, - notes Andrey Kulyashov. - Our task is to implement context-keeping functionality in JINNEE so that the bot can build a long sequential conversation, and does not perceive each subsequent replica as a new request.

An interesting function of JINNEE is a map of the spread of intention recognition in a request, which is maintained in your personal account. The company explains: a person turns to the bot, he answers, and how accurately he does this is displayed on the map. Thus, it becomes clear whether the module copes well with this type of appeal: does it give the desired answer right away? Does he need clarification? Or does he not understand what it is about at all? You can see this clearly on a visualized map: if the distance between the bot's answer and the necessary answer is large enough, then it is in this area that you need to work, teach the bot new vocabulary, constructions. Thus, it is possible to continuously analyze the bot and improve it.

Another improvement tool is feedback: if you invite the user to evaluate the effectiveness of the bot after each communication session, then you can get enough data to identify problem areas.

An important aspect of the bot is the ability to interact with several information systems at the same time, receiving the necessary data from them. To this end, JINNEE has received an advanced API, making it easy to integrate with any CRM system or corporate portal.

Not all of our customers want to use the NLU engine. Often they need a script bot that can be easily customized for their tasks in the designer. JINNEE allows you to solve, among other things, such problems, - notes Andrey Kulyashov, ISS.

Scenario bots allow you to quickly solve linear business problems: "catch" a client at the time of communication, get his contact and answer a popular question, for example, whether red smartphones remained in the store.

The VoiceBox platform, developed by MTT, allows you to assemble a robot depending on the specifics and needs of a particular organization. The robot is assembled from ready-made functional units: incoming and outgoing calls, speech recognition and synthesis, interactive menu, logical processing, forwarding, integration with an external database. A voice robot can be integrated with most CRM systems, such as Bitrix 24 or amoCRM, which allows you to quickly generate reports and improve the quality of service.

In May, a voice assistant for applicants created on the platform was launched at the Institute of International Economic Relations (IMES). The voice robot calls those who left the application and finds out whether it is relevant for him to receive an education, as well as whether he needs advice from a specialist in the selection committee. In real time, the digital assistant sends information to the university database and forms a report, helping members of the admissions committee to prioritize further work with applicants.

The big plus of the VoiceBox voice system is the speed of deployment and flexibility of scaling, "comments Ivan Artemyev, MTT Product Director.

Depending on the needs of the organization, the system can automate dialogues with both dozens and thousands of subscribers.

Dialog platforms

Dialog platforms include a set of all services and solutions for prototyping, development, testing, deployment, quality control and subsequent support of the dialog solution throughout its life cycle. They are designed for IT professionals to work with them: platforms usually provide the opportunity to develop scripts, including in programming languages. It is important that they have advanced functionality for managing many branches of scenarios and provide the ability to deploy conversational enterprise-level solutions.

In some platforms, NLU allows you to use only the simplest patterns, in others - to deeply teach complex models, - says Kirill Petrov. - If you do not have enough built-in service capabilities, it's great if the platform has an API to connect other NLUs to your project.

Platform vendors claim that they make it possible to create bot assistants in various subject areas and skills for them with extensive logic and cheat chat ("chatter").

JAICP

JAICP is the development of Just AI - a platform with a built-in NLU service for the development of powerful AI bots: chatbots, virtual operators, voice assistants and skills for them. The vendor positions JAICP as a tool for creating complex conversational solutions: smart chatbots, voice games, skills for smart speakers and voice assistants (for example, Alice).

JAICP is integrated with CAILA's own NLU core, which allows bots to understand natural speech. Managing dialog logs makes it possible to learn how to retrain NLU directly from the JAICP interface, as well as load training samples in order to train NLU to better recognize client intentions at the start of the project.

Dialog OS

A professional platform for creating intelligent voice and text robots developed by Nanosemantics. Uses the Rules + ML hybrid model, combines Machine Learning and fuzzy search with rules.

This means that the technologies used do not depend on the required natural language. At least Eskimo: if there is material for developing a knowledge base, then the dialogue engine will work with the resulting base, "says Stanislav Ashmanov, General Director of Nanosemantics.

Of course, the bot's knowledge base will have to be developed anyway. For example, prepare datacets for training a neural network or explicitly describe models of potential incoming replicas ("rules"). This work on the DialogOS platform can be performed by a person who is not IT qualified.

The main problem with combining ML and rule-based is that these systems produce results in completely different dimensions. Rule-based counts the number of word matches in templates, taking into account coefficients that depend on the context of the dialogue and increase or decrease the weight of the candidate template. And ML produces probabilities from 0.0 to 1.0 for each node. Our system leads these indicators to one scale according to a special formula and ranks candidates.

The DialogOS platform does not require an array of data before working on a virtual assistant. All data can be built from scratch when working in the DialogOS platform on a specific assistant, the company says.

Of course, it is better if the developer of the assistant has logs of real conversations with clients, - notes Stanislav Ashmanov. - But if they are not, he can quickly "out of his head" sketch out 10-15 examples of remarks on a given topic in the platform and on this minimum set already get a trained neural network and a bot that can communicate with clients.

The platform supports the functionality of working with user intentions. They are described as rules and are accompanied by examples of specific phrases related to these intentions. It is also possible to view the communication history of users with the virtual assistant and mark up intentions.

The platform also has tools for debugging training. For example, a table of potential intent conflicts, as well as the ability to see information about the set of candidate intents for a particular replica, NER candidates and their weights.

Erudite

The Erudite AI platform for creating dialog robots and controlling their work was created by Naumen. A comfortable and natural dialogue with the robot is realized thanks to the NLU engine and machine learning, the company says.

Domain pre-training has been implemented on Russian-language data bodies to improve the quality of understanding of the natural language in projects of various topics. The preprocessing of replicas in the dialogue is implemented using tokenization based on the BPE algorithm. Machine learning is used to correct spelling. It is possible to train the robot on customer data to develop skills in working with narrow-industry topics.

To understand the meaning of text and replicas, encoder-decoder models are used, to classify intents (intentions) - the BERT algorithm. Bidirectional LTSM (bidirectional LTSM) and CNN (convolutional network) with attention block are used to extract facts from client requests.

The Erudite platform implements a hybrid approach to learning, combining machine learning and rule-based methods, which provides complete control over dialog robots and the ability to make adjustments to their work in a short time.

Greater transparency of machine learning can be achieved through the use of LIME and LRP methods, which allows you to visualize the operation of the ML model, make the logic and actions of the robot accessible for interpretation and understanding, as well as assess the influence of certain training examples on its behavior.

The platform contains ready-to-use dialog modules for quick start and scaling, pre-configured work with common types of facts and intents (date and time, consent and denial, operator request, city and country) is implemented.

There is also a built-in social chat module for recognizing replicas that are not related to the subject of the consultation, and a mechanism for smoothly returning the client to the main subject of the dialogue. Ready-made script fragments can be reused without reconfiguring.

Robots on the Naumen Erudite platform can solve various problems:

• processing of incoming calls and chats (product and service consultant, reception of meter readings, HR support, prompter for Call center operators);
• outgoing calls (tracking of orders and objects, balance check, technical support);
• replacement IVR in the contact center (search for addresses and objects, execution of applications and documents, taxi order).

Bot training involves not only data preparation and training in working with facts and Intentas, but also automatic combining of historical dialogs into clusters by keywords, marking up training dialogs and compiling a hierarchical tree of topics for classifying queries, as well as training the robot to recognize and interpret intents and facts in replicas of the interlocutor. Regular bot training is supported on real "combat" data from worked out dialogues.

In May, the intellectual chat bot Naumen earned in the contact center of the Federal Treasury of Russia in the chat of a personal account on the state automated system portal "Management." The chatbot provides round-the-clock portal advice and authorization issues, and helps you apply for technical support. In addition, the chatbot answers frequently asked user questions, for example, about the extension of the EDS, filling out and submitting reports, the capabilities of the state automated system "Management" and the peculiarities of working with the portal..

CraftTalk

The CraftTalk platform, developed by the company of the same name, is an all-in-one solution for providing the service through chat with automation based on the knowledge base and artificial intelligence.

The platform supports developed functionality: omnicanality communication with customers is possible through chats on the site, in, and messengers social networks on. e-mail At the same time, as they say in the company, the platform is fully ready for use in: contact center organizing queues for processing requests, working in high load mode with optimization of operator downtime, operator and supervisor workstation with detailed statistics of operators and artificial intelligence on more than 50 indicators.

A separate product - Knowledge Base 2.0 - serves as an omnichannel source of knowledge not only for people, but also for chat bots: it contains ready-made answers for chat bots and information for machine learning, helps to create complex scenarios. They can be developed in a visual editor by ordinary contact center employees without programming, even if the scripts contain complex logic and integration with other services.

This will help the contact center survive even a 10-fold increase in traffic without attracting additional operators, according to CraftTalk experts.

At the end of May, CraftTalk announced the creation of an online chat for the Finservy personal finance platform, created by the Moscow Exchange. Online chat is intended for consultations of individuals throughout the range of services "Finservug."

Sergey Budnik, Product Director of the Finuslugi platform of the Moscow Exchange, explains the choice of product: {{quote 'Today our main channel of online communication with users is various instant messengers. Therefore, we chose the CraftTalk chat platform for implementation, which helped organize omnichannel communication in chats not only on the portal, but also in the two most popular instant messengers - Telegram and WhatsApp. }}

Through the open API, the platform is integrated with CRM, personal account, website and other systems of the Moscow Exchange. Voluminous work has also been done to configure complex analytics, says Sergey Budnik. It helps in the work of the operator and groups of experts on specific topics and gives an overall picture of the effectiveness of the contact center to its supervisors and managers.

Developed support for omnichannel has become the basis for choosing the CraftTalk platform to create a chatbot for client communication of Ingosstrakh. Today, according to a study by SDI 360), Ingosstrakh is the only large insurance company in Russia that provides customers with support in all major instant messengers. Ingosstrakh uses scripted chatbots while receiving a client's appeal, and its algorithm is a tree with a large set of scripted branches.

In addition to clarifying questions, the chatbot can additionally request documents necessary for further consultation.

Open-source frameworks

This allows you to flexibly work with classes and data input and output formats, change code for project goals, and customize responses automatically, depending on the class.

Code written using the framework must be placed in the runtime on its own. To do this, you can use your own servers in the cloud or circuit, or use a platform that will take on all the tasks of hosting, scaling and balancing. The same applies to NLU models. If the virtual assistant script uses natural language understanding in its work, such NLU models also need to be placed in the runtime and, accordingly, scale and balance the load.

RASA

Open source toolkit built in Python, which the project developers call the third generation chat bot: it not only walks the state graph, but is able to save and use the context of the previous dialogue.

For example, defining intent in RASA algorithms will work well for any language, as well as any specific words that you specify in the training examples. When implemented through pre-trained vector representations like GloVe or word2vec, localization of the bot and its use in highly specialized areas will bring a lot of headache.

An important part of this framework is RASA Stories, examples of real bot dialogs formatted in Intention-Reaction format. Based on these stories, a recurrent neural network (LSTM) is trained, which compares the previous message history into the required action. This allows you not to set dialog graphs rigidly, and also not to determine all possible states and transitions between them, Parallels says: with a sufficient number of examples, the network will adequately predict the next state for the transition, regardless of the presence of a specific example.

Deep Pavlov

Developed at the Neural Systems and Deep Learning Laboratory, the DeepPavlov Project MIPT is a library for virtual assistant creation and text analysis built on TensorFlow and Keras. It contains a set of components for rapid prototyping of dialog systems that allow you to automate communication processes in various areas of activity. The platform provides a complete cycle of development of dialog agents designed to automate communication processes.

The advantage of DeepPavlov in comparison with the RASA library is the flexibility in the configuration of dialog agents, as well as a set of pre-trained models for NLP tasks of the Russian language.

The DeepPavlov library contains a set of trained neural network models for text analysis (ML/DL/Rule-based), components of dialog systems and pipes, a library for creating and testing dialog models, application development and integration tools (instant messengers, support services software, etc.). It declares support for 53 languages.

The models are packaged in easy-to-deploy containers hosted on Nvidia NGC and Docker Hub. We tried to describe the specifics of working with the code in the documentation as much as possible.

Jovo

The Jovo framework is built on TypeScript. It allows you to create voice skills that work on different devices and platforms, including Amazon Alexa, Google Assistant, mobile phones, Raspberry Pi, etc.

BotPress

BotPress is an open source conversational AI platform built on TypeScript. Allows you to create projects that automate communications and workflows in companies. BotPress has convenient features such as advanced permissions and secure storage of personal data. The framework is aimed mainly at developing bots, not voice solutions. There is no support for various languages.

JAICF

Kotlin follows the concept of context programming, so it is best suited for creating conversational solutions, where the context of the dialogue is the main idea and value, the expert emphasizes.

The JAICF framework provides free access to all professional tools necessary to develop a full-fledged conversational environment: a ready-made dialog machine, libraries, NLU engines (built-in NLU CAILA service and external engines, for example, Dialogflow or RASA), analytics and storage systems, data ready-made integrations with messengers and ecosystems of voice assistants Alice, Alexa, Google Assistant Facebook Messenger, etc Slack Telegram.

Unit testing minimizes errors by automatically testing dialog scripts. JAICF uses the frameworks of the usual unit testing and Kotlin capabilities for convenient and concise DSL. An example of the advanced use of frameworks to create a virtual assistant is an intelligent chatbot that QSOFT has launched on the website of one of the leading automotive brands. The customer decided to create a universal assistant with artificial intelligence, which is able to provide information about the car in a convenient format, answer frequently asked questions, replace the function of searching for information on the site and even help issue an application for the purchase of a car.

The project was carried out in several stages using Kotlin, JAICF, RASA NLU, Python. During the project, the implementation team worked out more than 100 scenarios for navigating the site. The dialogue architecture is based on more than 100 intentions, each of which is a separate branch of the dialogue scenario, according to QSOFT. The intelligent chatbot has been trained on a base of more than 3,000 training phrases and dialogues.

The chatbot learns to determine the user's intention, context and respond to the interlocutor's statements. Depending on the situation, the chatbot can provide the optimal scenario to the client and issue more than 200 reactions.

Towards thematic universalization

Such an assistant should be able to communicate as an ordinary person, not limited to a pre-prepared dictionary or a set of sentences, the expert explains.

But the conversation that virtual assistants and chatbots are trying to imitate as a form of human communication is an extremely complex discipline, "says Alexander Khledenev, director of digital solutions at VS Lab. - In addition to the fact that it is fundamentally based on the general cognitive abilities of our brain, which AI is not yet clear when it will reproduce, it has a huge number of forms and aspects that need to be taken into account for full-fledged imitation. These are, first of all, aspects of verbal and non-verbal forms of communication that are included in the conversation - context, emotional coloring (tone), gestures, facial expression, etc.

Ilya Pomerantsev notes that from a technological point of view, most of the problems that impede the universality of assistants arise at the stage of speech recognition and analysis of the resulting text.

The severity of the problem of correct speech recognition was reduced by using an approach based on the recognition of elementary parts (the so-called tokens), followed by composing a meaningful sentence from tokens. Moreover, one that does not require huge sets of source data and at the same time gives an acceptable recognition quality. This, for example, is the Byte Pair Encoding (BPE) algorithm.

But after we deal with a clear understanding of the current context, the next step is needed. It will be associated with understanding the dialogue at a higher level, at which you will have to model the interlocutor, plan the dialogue, draw up dialogue scenarios for the future, "says Ilya Pomerantsev.

It is necessary to achieve a deeper understanding of the dialogue, which means that it is necessary to move away from solving private problems that are dealt with by question-response systems, the expert emphasizes.

The presence of this data, it seems, suggests that they have the opportunity to develop an assistant capable of supporting a conversation close to human. But it is worth talking longer with such systems and it becomes clear that in situations that they were not trained in advance, they will give general answers, and in some cases even say that they do not understand the interlocutor, - notes Ilya Pomerantsev.

Our current experience of communications with chat bots and assistants is that in mass order we "communicate" not with Alexa or Alice at all, but with representatives of their first generation, the so-called rule-based or transactional bots, - notes Alexander Khledenev. - We can understand this when we are asked to press one of the buttons, answer with standard phrases in text or voice. Actions on the developed scripts and rules are "sewn" into the logic of such bots. Their "intelligence" consists in recognizing our query (by voice or text), identifying keywords in it, specifying intentions, and then executing the program or finding an answer in the knowledge base.

Projects with training such bots can take up to six months, but they are able to perform only the simplest operations characteristic of a Call Center or support service, notes Alexander Khledenev, although they can be beneficial even at this level.

Such products are trained on historical data from real communications, more easily scalable to other scenarios and capable of additional training based on feedback, - emphasizes Alexander Khledenev.

When this generation of assistants becomes widespread, they will definitely stop annoying users with their own "stupidity," the expert believes.

"Annoying" is still a rather soft definition. People often literally hate voice bots without even trying their capabilities, - Stanislav Ashmanov laments. - But in general, yes, the main complaints about virtual assistants are a lack of understanding of the context of the conversation, insufficient personalization of communication, lack of emotional coloring, "voice like a robot." We are at a stage of technology development where these flaws still exist. But the ways to resolve them are already understandable. "

Areas of improvement of virtual assistants

Therefore, the question: what should I do with access? - this is a question about access to "1C," even if the mention of "1C" sounded several remarks earlier, - explains Stanislav Ashmanov.

This is already a relatively free dialogue, similar to the natural dialogue between people, says Stanislav Ashmanov.

If a person in the dialogue "heats up," shows irritation, then most often it is more correct to convey the dialogue to a person, and not try to solve the client's problem with a bot, - said Stanislav Ashmanov.

To make this part of the work easier for an assistant developer, the DialogOS platform provides a section that allows a technical specialist to enter any JavaScript or Python code into it, says Stanislav Ashmanov. - It is for the purpose of integration with the service of a particular company, with the settings and methods of accessing which the specialist is well acquainted with.

In the coming years, a new generation of personal virtual assistants may appear on the market, acting on behalf of a specific person - Bots who have information about their owners and perform routine operations for them will be able to save the client significant resources. At the same time, it should be noted that if such bots appear, it will be necessary to resolve issues related to ensuring the legitimacy of their actions on behalf of real people, their verification as true bots, the development of verification procedures, etc., - warns Leonid Perminov.

On the Way to Metaverse

The idea of ​ ​ deep personalization of smart virtual personalities leads straight into the space of metaverse. First, Meta introduced the metaverse, then Baidu talked about the digital world Xiang, inhabited by virtual characters.

Obviously, the new "digital counterparts" of people (Human Digital Twins, HDT will become quite complete copies of the personalities that gave rise to them, much more complete than those used in computer games, "says Timur Aitov, Deputy Chairman of the Commission on Digital Financial Technologies of the Chamber of Commerce and Industry of the Russian Federation on the pages of the Internet resource Finversia.

All this will be needed by the twin, first of all, to simulate the interaction of a pair of "twins" (robot-robot), instead of the usual communication format (robot-human). A new opportunity - interaction with others like yourself - today is important not for games, but for business, and therefore it will definitely appear in the concept of the Facebook metaverse, the ekpert emphasizes.

The list of tasks of such an autonomous robot may include creating consensus on controversial issues by enumerating existing options, receiving answers and comments from other doubles, discussing future agreements, etc., - suggests Timur Aitov.

A digital assistant can not only gain an extensive set of reference knowledge that allows him to replace his owner in situations of making typical decisions, but also gain new abilities, for example, knowledge of many foreign languages. And then get a high-paying job in several companies at the same time. Many more options can be offered for implementing the new habitat of human twins - the metaverse, relying on modern technological advances and numerous artificial intelligence options.

But here is an interesting question: will it be possible to "deceive" the double, attack him with numerous methods of social engineering, as is happening today? Will it be possible to find out his secrets? {{quote 'Most likely, it will be even easier to "hack" a double than to deceive an elderly, but still a person, because a person always has intuition, which is not easy for robots to "instill." And the double has only a "conscience," and even that is not real. So, everyone will decide for himself what can be entrusted to his double, - emphasizes Timur Aitov. }}

If the level of technological implementation of the metaverse is yet to be seen in the future, then at the practical level today there are examples of the creation of complex complex systems with a high intellectual level. Smart cities are an example.

Next Overview Material > >
> Browse Home > > >

Other Review Materials

• Data processing in deep neural networks: achievements and calls of the current moment
• AI: from data to knowledge
• Processing of natural language documents and texts
• Smart Process Processing
• Smart Voice Services
• Intelligent video analytics
• New Challenges for Information Security Tasks
• AI in analytics: What's beyond BI?
• Smart manufacturing
• Smart city
• Conclusion. Where does the world of practical AI implementations roll?

Other materials on the topic of AI

• The artificial intelligence market in Russia has reached a turning point. TAdviser 2020 Review
• Artificial intelligence (AI, Artificial intelligence, AI)
• Artificial Intelligence (Russian market)
• Artificial Intelligence (Global Market)
• Computer Vision: Technology, Market, Outlook
• Video Analytics Systems Video Analytics Systems and Projects Catalog
• National Strategy for the Development of Artificial Intelligence
• Machine Learning, Malicious Machine Learning, Data Labeling
• RPA - Robotic Process Automation
• Video analytics (machine vision)
• Machine intelligence
• Cognitive computing
• Data Science
• DataLake
• BigData
• Neuronets
• Chatbots
• Smart speakers Voice assistants
• Artificial intelligence in various fields: in banks, medicine, radiology, retail, military-industrial complex, production, education, Autopilot, transport, logistics, sports, media and literature, video (DeepFake, FakeApp), music
• Self-driving cars in the world
• Self-driving cars in Russia