According to surveys conducted by the Higher School of Economics (HSE), 85% of Russian scientists already employ artificial intelligence (AI) in their work. However, the adoption of innovative technologies in practice does not automatically make those who use them innovators. Scientists are becoming consumers of third-party developments, most often foreign, which help automate routine processes or mask a lack of foreign language skills. This is convenient, but insufficient for achieving a scientific breakthrough. The surveys have yet to reveal any significant impact of AI technologies on the ultimate outcomes of scientific research in Russia
| Универсальные ИИ-сервисы, ИИ-чат-боты / General-purpose AI services, AI chatbots | |
| ИИ для машинного перевода / AI for machine translation | |
| ИИ-поисковики (кроме умного поиска в Яндекс, Google) / AI-powered search engines (excluding smart search in Yandex and Google) | |
| ИИ для генерации изображений / AI image generation | |
| ИИ для распознавания речи, преобразования ее в текст / AI speech recognition and speech-to-text | |
| ИИ для генерации программного кода / AI code generation | |
| ИИ для генерации речи, преобразования текста в речь / AI text-to-speech and speech generation | |
| Другие сервисы / Other services |
Share of Russian scientists using different types of generative AI, as a percentage of surveyed researchers who reported using neural networks. Source: Higher School of Economics (HSE).
Few sectors of the Russian economy can match the level of AI adoption seen in the country’s scientific community
That is according to an online survey conducted in May by the Institute for Statistical Studies and Economics of Knowledge at HSE among approximately 1,200 researchers working at universities and research institutes
The survey found that around 85% of Russian scientists already use AI technologies in their work. Generative AI models, capable of producing text, images, audio and video after being trained on large datasets, have become widely adopted within the research community.
General-purpose AI services and chatbots are the most popular tools. Nearly three-quarters of researchers who use generative AI rely on them. The overwhelming majority use foreign platforms (88%), while around half also use domestic services (51%). The total exceeds 100% because many respondents use multiple platforms.
Age differences are relatively modest. About 90% of respondents under the age of 44 use AI. Among researchers aged 45 to 69, adoption ranges from 83% to 86%. Even among scientists aged 70 and older, 71% reported using AI.
Adoption also varies by discipline. Social sciences lead with an AI penetration rate of 89%. Engineering, medical and natural sciences also report high levels of use, ranging from 84% to 87%. Usage is somewhat lower in the humanities and agricultural sciences, at around 76-77%.
For most Russian scientists, AI is a recent addition to their workflow. Half of AI users began using these tools one or two years ago, while almost 30% adopted them within the past year.
Nevertheless, generative AI has already become a routine part of research. More than two-thirds of users rely on it at least once a week, while over one-third use AI every day.
The survey asked respondents about the various tasks encountered during different stages of the research process.
The most common uses were exploratory information gathering, literature searches and bibliography preparation, and translating Russian texts into foreign languages. More than half of researchers using neural networks employ AI for each of these tasks.
AI services are also widely used to improve written texts and translate materials from foreign languages into Russian, with just under half of respondents citing these applications.
In addition, one-third of AI users rely on these tools to brainstorm research ideas, discuss those ideas with AI systems on something approaching equal terms, prepare presentations and teaching materials, design examination questions, write academic papers and edit manuscripts.
One in five researchers uses AI to generate or refine programming code for data analysis or visualisation.
Yet widespread adoption of innovative technologies does not automatically make everyone using them an innovator.
Researchers’ assessments of AI’s impact on the quality of their scientific output proved mixed. ‘Forty-two per cent reported a positive effect, while 47% saw no impact,’ HSE experts concluded. ‘This suggests that generative AI remains primarily a supporting tool that improves the efficiency of certain stages of research without significantly affecting the final outcomes of scientific work.’
Even without expecting sensational discoveries or scientific breakthroughs, it is notable that AI has not significantly increased publication output in either Russian or international academic journals.
Despite AI’s assistance with literature reviews, drafting and editing manuscripts, around 60% of users reported no noticeable change in their publication activity.
The survey therefore suggests that AI mainly helps Russian scientists automate routine tasks, compensate, one might infer, for limited foreign-language proficiency, and reduce bureaucratic paperwork. These benefits are convenient and useful to an extent, but insufficient to produce genuine scientific breakthroughs. By the same logic, researchers could once have boasted about their proficiency with internet search engines or online translation services.
Still, the headline figure of 85% AI adoption will undoubtedly look impressive in reports highlighting the digital transformation of Russian science.
The methodology also deserves consideration. The study was conducted as an online survey and relies on respondents’ self-assessments. Scientists who actively use AI may simply have been more inclined to participate, noted Binyatov Murad Bakhtiyar ogly, head of a research programme at the Presidential Academy.
Experts interviewed by Nezavisimaya Gazeta cautioned against equating widespread adoption with scientific effectiveness. ‘The real impact should be measured by reductions in research time, improvements in publication quality, and the number of patents and technologies successfully commercialised,’ said Yaroslav Seliverstov, a leading AI expert at University 2035.
‘A high level of engagement with a technological trend does not guarantee meaningful scientific or technological results. If “using AI” primarily means translating text, formatting papers or preparing presentations, then this represents improved operational efficiency rather than a technological breakthrough,’ said Irina Mezheneva, lead engineering analyst at the AI laboratory of Gazinformservice.
According to her, such figures may look impressive in official reports but fail to answer the key question: are researchers creating new methods, models, scientific schools and globally competitive applied technologies? ‘The true value of artificial intelligence in science lies not in the number of users but in the quality of the problems it helps solve,’ said Andrei Kondratyev, Director of CDO Global.
‘International discussion about AI in science is increasingly moving in this direction. The focus is shifting away from user numbers toward identifying research questions that were previously impossible to address but have become solvable thanks to artificial intelligence,’ Binyatov Murad Bakhtiyar ogly added.
Regardless of the specific research applications or the effectiveness with which Russian scientists use AI, concerns surrounding the technology continue to grow.
One obvious risk is hallucinations. ‘AI can generate inaccurate information or entirely fictitious facts. Every result it produces therefore needs to be verified,’ warned Sergei Karpovich, Deputy Head of the T1 IT holding company.
A recent controversy in the Russian publishing industry illustrates the point. A newly published popular science book on the mythology created by an American author was found to contain clear traces of AI-generated content, including factual errors, fabricated quotations, references to non-existent studies and fictitious translators. The publisher acknowledged the problem.
Collecting information and testing hypotheses constitute a substantial and labour-intensive part of scientific research. ‘AI makes it possible to process large datasets more quickly, compare sources, test hypotheses and identify promising research directions,’ said Lyudmila Bogatyreva, Head of Digital Solutions at the Polylog communications agency.
‘But there is another side to the story. The same technologies can accelerate not only high-quality research but also pseudo-research. They make it easy to generate text, simulated analysis or conclusions that lack a solid evidential basis,’ she warned.
Experts also point to the danger of cognitive bias alongside hallucinations. As AI systems interact repeatedly with the same user, they begin adapting to that user’s preferences. Researchers’ own behaviour may also change.
‘Generative AI can reinforce a scientist’s existing views, creating the illusion that they are more convincing than they really are. It may also encourage researchers to accept ready-made interpretations more readily instead of seeking alternative explanations,’ said Maria Lopukhina, Head of AI Studio 1331.
Other risks are legal and technical, particularly when foreign AI platforms are used.
According to Kondratyev, providers of consumer-level subscription plans, whether free or low-cost, generally reserve the right to use uploaded data to further train their models. Russian copyright law offers no protection in such cases.
Meanwhile, plans guaranteeing that uploaded data will not be used for training are unavailable to Russian legal entities because of sanctions. ‘That means Russian scientists are effectively limited precisely to plans where the risks are greatest,’ he explained.
In other words, when a researcher uploads a draft scientific paper, experimental dataset or methodological description to a public version of ChatGPT, they are effectively transferring unpublished intellectual property to a foreign company.
Ideally, that confidential information becomes lost within anonymised training data. However, there is also the possibility that it could later be extracted using carefully designed prompts. In some cases, such material may even include state secrets or commercially valuable know-how
‘For organisations working on classified subjects or commercially sensitive technologies, this creates risks of losing priority over intellectual property. Without appropriate monitoring tools, organisations often discover such leaks only after the fact,’ said Sergei Shcherbakov, Chief Technology Officer at Stakhanovets.
Kondratyev also noted that copyright-related vulnerabilities exist when using domestic AI platforms.
Dependence on foreign digital infrastructure presents another risk: access can be restricted at any moment for a variety of reasons, warned Denis Sugaipov, a researcher at the Gaidar Institute’s Laboratory for Mathematical Modelling of Economic Processes.
‘Reliance on external services that lack domestic alternatives creates vulnerabilities for the continuity of scientific research.’ Seliverstov agreed. ‘Heavy dependence on foreign platforms makes research sensitive to changes in access conditions, pricing and the policies of technology providers.’ For that reason, he argued, the use of generative AI should be combined with internal data protection rules, mandatory expert verification of AI-generated results and the development of specialised domestic solutions.
Such criticism does not imply that the technology should be abandoned. ‘We have acquired a new tool for working with large volumes of information. It automatically filters, consolidates and organises data, effectively producing summaries,’ Pavel Terelyansky, Deputy Head of Digital Transformation at the Plekhanov Russian University of Economics, told Nezavisimaya Gazeta.
‘The previous generation of tools consisted of automated searches across digitised collections, namely search engines. Who still spends their days sitting in dusty libraries? Hardly anyone. Search engines now deliver information with a single click, and that has become an entirely normal and indispensable part of scientific work. The next stage is using automatically generated summaries. That stage has already arrived,’ Terelyansky said. ‘Before long, it will be impossible to work without tools like these.’
