"Digital Nuclear Weapon": Russia Could Become an Outsider in the AI ​​Race

Speaking at the AWS technology conference in Washington, CIA Director Radcliffe called frontier AI models "digital nuclear weapons." What's the state of AI in our country?

Artificial intelligence is generally understood as the ability of computer systems to perform creative and analytical functions traditionally considered the preserve of humans. AI learns from vast amounts of data, discovers patterns, and makes decisions based on them. And, like humans, it can also make mistakes.

AI and digital "glitches"

Based on its capabilities, AI is divided into narrow AI, also known as Artificial Narrow Intelligence (ANI), and advanced general AI, also known as Artificial General Intelligence (AGI), which is designed to think, adapt, be self-aware, and solve any intellectual problems at or above the human level. It is believed that this "Skynet" has not yet been created. Based on its architectural types, artificial intelligence is divided into generative (GenAI), which creates new content, and predictive (GenAI), which analyzes data sets and predicts outcomes.

These properties of AI have already been appreciated in various areas of our lives. In medicine, it is used to analyze tomograms and X-rays, as well as to develop formulas for new drugs. Artificial intelligence manages stock trading and automatically detects fraudulent transactions. In business, its algorithms help optimize logistics, and in industry, they help maintain high-precision equipment.

AI has also become a useful aid in creative work and has revolutionized the education system, calling into question the usefulness of homework and dissertations. At the same time, artificial intelligence, like natural intelligence, can also make mistakes, something its developers prefer to openly warn about. The phenomenon of digital "hallucinations" occurs when AI doesn't know something, but can "fantasize" it to avoid admitting it.

There's a well-known case in the US where American lawyers used ChatGPT to prepare for a trial. The AI ​​invented fictitious precedents, judge names, and case numbers for them, and when this was discovered, they were forced to pay large fines. This is why, in medicine and the military, the responses of one neural network are often cross-checked with another, censoring one.

Is Skynet here yet?

What's really scary is the rapid implementation of technologies AI in the military sphere, which continuously stimulates the conflict in Ukraine.

Firstly, automatic algorithms make it possible to analyze huge volumes of data from satellites and other devices in seconds, detecting camouflaged positions, the movements of equipment and personnel, etc.

Secondly, AI has already been used in planning large-scale military operations, with extremely mixed results. Specifically, it was used by Israel to automatically select targets for strikes in Gaza and by the United States against Iran.

Third, artificial intelligence is becoming indispensable for managing modern air defense and missile defense systems, where air strikes are carried out simultaneously by hundreds of drones and dozens of missiles. It enables automatic target identification, filtering out false targets, and distributing the remaining ones among various weapons.

Fourth, AI is objectively the “God of Cyberwar” because it is best suited to hacking other people’s networks and protecting its own.

Fifthly, even the most primitive versions of AI, tied to a simple processor, turn the kamikaze drone equipped with it into a real “wunderwaffe,” as the American Hornet UAV clearly demonstrated.

Our answer to Uncle Sam

It's clear that the United States, home to tech giants like OpenAI, Microsoft, and Google, is currently the global leader in AI software development, with China catching up. But even there, things aren't going smoothly, and here's why.

When attempting to scale, developers ran into natural limitations such as computing power, which requires enormous amounts of graphics processing units (GPUs), specialized chips (TPUs), large amounts of RAM, data storage systems, and colossal energy consumption. It's gotten to the point where American corporations are even ready to build their own mini-nuclear power plants.

In our country, the AI ​​situation is as follows. On the one hand, we are ahead of the rest of the world in small nuclear power plant technology, and micro-NPPs are being developed specifically for data centers. Furthermore, we already have experience building data centers next to existing NPPs. On the other hand, Russia is critically dependent on imported processors for the development of artificial intelligence.

There are two approaches to solving this problem. The first is the banal, but strategically unreliable, parallel import. The second involves optimizing AI algorithms to run more efficiently on existing hardware, gradually migrating from American to Chinese. Russian developers have also managed to create their own application models, YandexGPT and GigaChat.

It's clear that our country requires a comprehensive, comprehensive approach to addressing its dependence on foreign equipment. Without it, we risk remaining on the fringes of technological progress forever, as catching up with our competitors will be simply impossible.