Translation. Region: Russian Federation –
Source: Moscow Government – Moscow Government –
An important disclaimer is at the bottom of this article.
Moscow is a global leader in implementing artificial intelligence (AI) technologies in urban services. Chatbots and voice assistants, recommendation services, computer vision systems in transportation and public services, medical assistants, a "Digital Teacher" service for schoolchildren, and many other solutions are currently working to make life in the city easier, more comfortable, and safer. In the capital Department of Information Technology They explained how Moscow creates and trains AI models for a wide range of areas of life.
"Following Sergei Sobyanin's instructions, Moscow is expanding the implementation of artificial intelligence technologies across all areas of city life. The city creates new AI solutions annually, and for this purpose, it operates its own platform for developing, training, and operating AI models. Residents also assist developers by testing AI services on real user requests. This increases both the effectiveness of these services and their popularity among city residents. Today, over 90 percent of Muscovites support the implementation of AI in at least one city project," the department's press service reported.
First, the data
Training AI requires massive amounts of data. For example, a traffic violation recording system requires numerous photographs of road markings, parked cars, or traffic lights, while a chatbot requires hundreds of thousands of real-world conversations with an operator. Therefore, the first step is to collect the data into datasets. First, all personal information unnecessary for training the AI is removed. Next, the dataset is formatted so that the AI can process the data. Typically, it is collected in a structured table with clearly labeled rows and columns. Datasets can contain various data types, such as images, text, audio files, numerical values, and other formats.
When working with photographs, medical images, scanned copies of documents, or conversations, the data collected for training is traditionally pre-labeled. For example, when working with documents, document management specialists mark stamps, signatures, watermarks, and other areas containing important elements. When processing X-ray images, a radiologist manually highlights tumors, fractures, inflammation, and other areas of concern. Typically, these are double-checked by another specialist. The more accurate the labeling, the better the AI model will perform its tasks. Therefore, this work is carried out by highly qualified specialists.
In Moscow, some city datasets involved in the development of AI-based solutions are available on the page ay.mos.ruSome help chatbots answer questions about public works or housing and utilities more accurately, others make performance or concert recommendations as personalized as possible, and still others improve the content and navigation of Moscow's main online portal. Developers can apply and gain access to these datasets to create their own solutions for the city and its residents.
Time to choose a model
An AI model is a program that learns from collected data. The model chosen depends on the task being solved. There are many ways to build AI models, but the most common is machine learning. This is the process of creating specialized algorithms that enable programs to learn autonomously from collected data. A key difference from standard algorithms is that a computer learns to solve problems by analyzing numerous examples rather than following precise instructions: it automatically finds patterns in the data and applies these patterns to new information.
Another area of machine learning is computer vision—the ability of a machine to understand and interpret photographs, videos, and other visual images. Today, this technology is widely used by Moscow doctors. For example, it helps analyze and describe imaging studies, such as fluorography, mammography, X-rays, CT scans, and MRIs. Algorithms identify signs of potential pathologies in images, enabling accurate diagnosis and treatment. Moscow radiologists already have over 60 AI services at their disposal, which are used in over 40 clinical areas.
Machine learning also underlies decision support systems, natural language processing, speech synthesis and recognition, and generative AI.
AI also needs to be taught
At this stage, the model is trained using the collected data: it analyzes it and identifies patterns. For example, Moscow uses AI to verify treasury documents. To ensure the program can automatically determine whether they are filled out correctly, the AI model was trained on a large number of different scanned copies of contracts, acceptance certificates, and invoices, both correctly and incorrectly filled out. The model "views" these images and learns to distinguish them. As a result, the service's accuracy exceeds 90 percent, which speeds up document processing and significantly reduces the labor intensity of the process.
To teach the AI to identify shortcomings in city cleaning, such as overflowing trash containers, ice on rooftops, or uncleared snow, the model was trained on relevant photographs, both with and without the defects. Currently, the system receives over 70,000 screenshots daily from the city's CCTV cameras and automatically analyzes them with up to 90 percent accuracy. It then reports any shortcomings to the Center for Automated Recording of Administrative Violations. There, specialists double-check the information and forward it to the responsible city services. Thanks to the neural network, they quickly identify areas where work needs to be done, making the city even more comfortable for millions of residents.
Last year, the Moscow Electronic School (MES) launched the "Digital Teacher" service for mathematics. It analyzes students' knowledge, identifies poorly understood topics, and offers assignments to fill gaps. It is powered by an AI model that, among other things, boasts high-speed natural language processing, is capable of recognizing each student's learning characteristics, and updating recommendations based on new data. To develop these "superpowers," it was trained using a large number of educational materials of varying difficulty levels, sample curriculums and educational pathways, data on test results, academic performance, and more. The "Digital Teacher" for mathematics is now one of MES's most popular services—since its launch, 850,000 students, their parents, and teachers have used it. This year, it was expanded to include materials on English and Probability and Statistics.
Stage Four: Testing
Next, the model is tested using new data not presented to it during training. This helps determine how well it can absorb new information and solve problems. If the model doesn't perform well, the developers continue training it or modify the algorithms. This process can be repeated several times until the desired results are achieved.
The city often invites Muscovites to try out new AI services. This allows the algorithms to be trained on a large volume of real user queries. As a result, the resulting services become even faster, more accurate, and more useful.
Today, AI solutions are integrated into more than 120 city projects in healthcare, education, transportation, public amenities, security, and digital government. Learn more about how the capital is using artificial intelligence atproject page.
Support for the development and implementation of artificial intelligence technologies is in line with the objectives of the national project "Data Economy and Digital Transformation of the State"More information about Russia's national projects and the capital's contribution can be found atspecial page.
Get the latest breaking news quickly on official Moscow messaging channels. MAX And Telegram.
Please note: This information is raw content obtained directly from the source. It represents an accurate account of the source's assertions and does not necessarily reflect the position of MIL-OSI or its clients.