Contents
Artificial intelligence (AI) is a general-purpose technology similar to electricity and the internet. However, it is much more powerful, as it is claimed to automate cognitive functions and perform independent actions. Scenarios for AI’s impact on the economy are polar, from an ‘era of abundance’ to a ‘machine uprising’.
The U.S. Bureau of Economic Analysis is already trying to integrate (link in Russian) the effects of AI in the system of national accounts, and Deutsche Bank experts claim that almost all of the growth in fixed capital investment in the U.S. since 2019 has been driven by AI-related investments, whereas in other sectors, this growth is largely non-existent. AI-related companies are reaching new heights in the stock market. Open AI and Anthropic, companies developing models, have announced plans to go public, each valuing themselves at approximately $1 trillion, a sum exceeding the GDP of most countries in the world.
On one hand, the development of AI presents the economy with opportunities for productivity growth, but it also poses risks of inequality, bubbles, and cyber threats. ‘For us, as a regulator, it is very important to understand what to expect in the future to take the right measures, possibly right now,’ said Elizaveta Danilova, Director of the Financial Stability Department of the Bank of Russia and moderator of the AI session at the Financial Congress of the Bank of Russia. Econs provides excerpts from this discussion.
Opportunities: ‘We Are at a Very Early Stage’
Maksim Bolotskikh, partner, Yakov and Partners:
– According to our estimates (link in Russian), the expected economic effect of AI in Russia could amount to approximately ₽8 trillion to ₽13 trillion per year, which corresponds to up to 5.5% of the forecasted GDP. For comparison: this effect is more than twice the net profit of the entire banking sector in 2025. (₽3.5 trillion, link in Russian). If we look at the individual industries that are most effectively using AI, we see that, according to company estimates, AI brings them about 20–22% EBITDA. This is significant.
There are two components of this positive effect. The first is revenue growth as such, the creation of new businesses and areas of business. The second is cost reduction. Here lies the biggest problem: everyone likes that they can start working 25–35% faster, but the downside is the reduction of part of the staff, and managers need to decide if they are ready for this.
Alexander Kraynov, Director of AI Technology Development, Yandex:
– When talking about the contribution AI will make to the economy, I always ask a simple question: can anyone quantify the contribution of personal computers to GDP or to any organisation’s business? We know for sure that nothing would work without them. But calculating their contribution is very challenging.
The story with AI is similar. Currently, ‘AI’ refers to large language models and, recently, AI agents. But AI is an umbrella concept under which a huge number of technologies live. When someone says they don’t use AI, it’s not true because we use AI daily, including generative AI, without which, for example, modern search cannot work. If we remove AI from our lives, many processes will simply stop.
There are also certain extra ‘complications’. Everything related to AI is calculated based on sales, but AI technology sales do little to reflect its significance. AI is largely a product for internal consumption. It is produced internally and consumed internally. This does not mean that there is no market. Of course there is, and it is growing strongly. But when it comes to the impact of AI, this is just the tip of the iceberg. The lower part involves very significant and substantial internal production and internal consumption. We are at a very early stage, and what AI will mean for us, we will understand in about five years when it unfolds to the scales we should expect from it.
Qiao Liu, Professor of Finance, Guanghua School of Management, Peking University:
– Productivity is the main driver of economic growth, and if you look at what is happening in China, it faces the same problems as the rest of the world. Productivity growth in China is gradually declining, and many believe that AI has great potential to boost productivity and efficiency. But at the moment, it is quite difficult to conclude what exactly AI will turn out to be for us, whether it will meet these expectations. A number of estimates are quite pessimistic: for example, in 2024, an MIT study showed that over ten years, productivity growth due to AI will be small, only about 0.6%. It seems that so far, productivity growth from AI is only visible on paper.
Sergey Markov, Director of AI Technology Development, Sber:
– I really like to give an example related to the IT industry. In the late 1940s, there were only around a hundred programmers worldwide, who were doing things that seem rather strange by today’s standards: plugging connectors into switchboards, punching holes in punch cards. The work of a programmer has changed quite radically over 80 years. Thanks to technology, programmers today are roughly a hundred times more productive than programmers in the late 1940s. However, we do not find ourselves in a world where there is just one programmer doing the work of the other 99, but rather in a world where there are not 100, but rather 100 million programmers. Economists have long been familiar with this effect, known as the ‘Jevons paradox’. It is the phenomenon that increased efficiency in resource use paradoxically leads to higher consumption rather than lower.
The potential of any fundamental technologies from an economic point of view lies primarily not in replacing people in existing business processes. Technologies make it possible to create new products, services, and entire industries. For example, when the computer game industry appeared, it created a huge number of jobs not only for IT specialists but also for marketers, accountants, managers, and designers. The potential of AI technology lies precisely here.
There is a fly in the ointment, however. The new jobs that arise in the economy as a result of technological development often do not match the skills that people have and which become unneeded. This has happened many times, and society has faced waves of technological unemployment. Large corporations, which have the opportunity to retrain their employees, somehow cope with these problems. However, many companies in the market do not have such opportunities, and they are more likely to fire a person or not hire a new one to replace the one who left. In general, there will be no mass unemployment in the economy. However, certain social problems and questions about who should be responsible for retraining and social support for these people are acute and await answers.
Anantha Nageswaran, Chief Economic Advisor to the Ministry of Finance of India:
– India is in a unique demographic situation: we have a huge number of young people under 30, and 8 million jobs need to be created annually. The impact of AI in our context will be more pronounced, with 3 million people working in the IT sector, 2 million in global hubs, and an additional 10 million in related industries such as catering, hotels, and services. In total, 18 million people may be affected by the introduction of AI. The spread of AI is putting increasing pressure on the IT sector and salaries, so more and more people are opposing the use of AI.
However, in the near future, preference will still be given to people rather than AI. This is good news: the transition to the new technologies will not be abrupt, which will give time to adapt to the changes.
At the same time, of course, India will benefit from AI because we lack specialists in important areas, such as in education, medicine, and agriculture, especially in remote regions of the country. AI really can replace people there. Accordingly, for India, AI is both a blessing and a curse.
Ksenia Yudaeva, Executive Director for Russia at the IMF:
– When we look at the effects, we need to start with types of activities, with jobs. The effects of AI on jobs can be categorised into three distinct types:
- jobs that are unlikely to be affected,
- jobs where there will be a substitution effect,
- jobs with complementary effects, that is, where people remain in their places and become even more productive.
The main concern is not so much unemployment, as the substitution effect does not necessarily lead to unemployment, but rather the rise in inequality caused by emerging substituting technologies. For example, the problem with the Luddites was not that weavers completely lost their jobs, but that the new jobs were much lower paid and in worse conditions. A golden industrial era emerged 150 years later, marked by substantial technological progress and a profound societal shift, which ultimately led to the establishment of a social support system, resulting in enhanced living conditions for workers. The computer age also increased inequality: billionaires and trillionaires appeared, while a fairly large number of people lost not only in earnings but also in their social status.
This process may intensify with the advent of substituting technologies. The point is not that programmers will be replaced by vibe programmers. There may even be many more of these vibe programmers. But perhaps they will be lower paid and thus more profitable (link in Russian). In other words, in terms of overall employment, there may well be no effect. But in terms of rising inequality, the impact may be quite significant.
Risks: ‘a Tsunami Is Coming’
Qiao Liu:
– Firstly, if AI is a revolutionary technology, it should benefit society. Currently, AI is still far from this. More investment is needed in this area. A bubble is even necessary for the technology to become revolutionary.
Secondly, international coordination is necessary so that AI developers and AI models can follow certain fundamental principles applied worldwide. One of these principles is ‘Do no harm to humans’. The second principle is that the way AI solves problems should be understandable to humans. If there is no person who understands the processes behind AI decisions, we cannot use the decisions.
Anantha Nageswaran:
– The use of AI is becoming widespread, but AI management may fall into the hands of those who can use it for harm rather than good. A cyber security breach could disrupt key economic systems – such as logistics, transactions, and money transfers – and paralyze the operations of any country. Ensuring cybersecurity will require enormous resources.
Maksim Bolotskikh:
– In terms of risks and the bubble associated with AI, frankly speaking, I am least worried about the Russian market. In the market, 86% of Russia’s largest companies use open-source models (open-source software). That is, if international models are suddenly taken offline or go bankrupt, we will not be cut off from the world. We will have models developed on our AI. A certain risk arises if the entire market depends on only one or two companies, such as the chip manufacturer Nvidia.Having a variety of competition and representation of different models, suppliers, and integrators is crucial in this scenario.
Alexander Kraynov:
– Concerns that AI will provide incredible opportunities for malefactors have been confirmed. The sad news is that the regulator cannot do anything in this case, at least not with bans. Before regulating something, it is necessary to understand to which extent it is possible. Imagine there is a tsunami coming. What should the regulator do – ban the tsunami? With cyber risks, it’s a similar story: it’s a natural process. It is quite predictable, but it is impossible to ban it. Barriers, security measures, and warning systems need to be built. It is important to realise that we will now have to invest much more in security. And to think about how to catch the perpetrators and deter them.
As for the risks of a bubble – there will, of course, be disappointments. It’s like a casino where half the players bet a huge amount on black and the other half bet a huge amount on red, and as a result, some will win it all, whilst others will lose everything. There is currently a great deal of investment in companies developing AI – will they all survive? No. Those who survive will come out on top.
Sergey Markov:
– We are currently facing a significant shortage of computing power because export restrictions have been imposed on us. Moreover, export restrictions on AI models are now emerging. Take the latest Anthropic model, for example. It turned out to be much worse than the one presented a month ago due to the restrictions imposed. There is a risk of dependence on open source. If Alibaba changes its strategy and the next version of Qwen is not released with open access, Russian companies will be left with nothing.
And in this situation, we are focusing on training our models from scratch, although this may seem dubious and very, very expensive. But the real product is not the models. The models turn into pumpkins in three to four months. The real product is the creation of a production pipeline capable of generating an AI model from scratch if needed. In my opinion, it is critically important for us that this production pipeline for creating our own models exists within our country. Otherwise, we will be in a very vulnerable position, just as happened with the electronics industry at one time – it is easy to destroy, but rebuilding from scratch is practically impossible. And I really don’t want this to happen with large language models.
Ksenia Yudaeva:
– It is impossible to say in advance whether there is a bubble or not. But most innovations, whether financial or technological, have been somehow associated with a bubble. The dot-com bubble was a rather mild crisis because the financing was mostly equity-based, not debt-based. There is now more debt financing, so the consequences for the economy could be more severe. I think it will happen sooner or later, but every problem can give rise to its own solution. The bursting of the bubble won’t halt further development – perhaps it will actually weed out the weaker players from the market.
As for cyber risks, regardless of whether we are talking about are our AI models or foreign ones, the risks will exist, and it is still unknown from whom more. I am probably the only person among the former and current employees of the central bank who has not received messages ‘from Elvira Sakhipzadovna’ [Nabiullina] asking for money. With the advent of AI, the scope for fraud using deepfakes has increased.
But this is a process evolving on both sides: both the risks and the analysis of the risks are developing. The regulator can and must set out requirements for the protection of information processed by AI, and for managing risks from third-party providers.
Here, the weapon of defence is the same as the weapon of attack: AI in the field of cybersecurity can help with AI in the field of cyberattacks. It can help identify vulnerabilities faster at the development stage, detect problems faster, and resolve them more quickly.
.png)
