How often do you hear predictions about the end of the world dominated by an all-powerful artificial intelligence? At least once a week, some businessman or celebrity expresses concerns about a terrifying future under its influence.
Of course, a famous figure plus a grim forecast makes for a perfect headline. However, while past articles following this formula reflected real, sometimes frightening technological progress, they increasingly resemble empty marketing or a simple misunderstanding of what’s happening.
Why are we still frightened by poor retellings of “The Terminator,” when modern chatbots often blatantly lie and can’t remember five lines of dialogue? And most importantly, who benefits from this fear?
Not Impressive
First, it’s important to note that AI technologies have made significant strides over the last decade. Modern systems can write coherent texts, recognize patterns in large datasets, and create visual content. Not long ago, machines couldn’t replace such human labor.
The prospects of progress are daunting. However, at present, the development of mass-market products has stalled at discussions about so-called general artificial intelligence and the release of nearly identical language models (sometimes the new versions are even worse than their predecessors).
What do we have in the end? A tool that assists with simple text tasks and occasionally with images. People have adapted it for vibe coding or writing social media posts. Yet, the results often require verification—neural networks are incapable of handling more complex tasks.
You can ask your favorite chatbot to write a doctoral dissertation on “X”: you’ll receive a barely coherent text with references from the first or second page of a search engine. To improve the outcome, it’s suggested to use extended prompts, which is just a more refined tuning in “machine language” and further training.
With prolonged use of AI, every user likely realizes the limitations of today’s models. All progress ultimately hinges on the volume of training databases and server capacities, while the factor of “intelligence” has taken a back seat.
Intelligence Without Brains
To understand the context, we need to explain how AI works. In brief, large language models (LLMs) of classic chatbots operate as follows:
- The input text is broken down into tokens (parts of words, symbols).
- Each token is assigned a numerical vector.
- The model analyzes the relationships between tokens and determines which words are most important for understanding the context.
- Based on this, the LLM “predicts” each subsequent token, forming a response.
The model doesn’t “predict” out of thin air. It undergoes pre-training on a vast database, usually from open internet sources. This is where the neural network derives all its “intelligence.”
Language models do not “understand” text in the human sense; they compute statistical patterns. All leading modern chatbots use the same basic architecture aptly named “Transformer,” which operates on this principle.
Of course, this is a rough comparison, but LLMs can be seen as very powerful calculators based on large databases. They are strong, useful tools that simplify many aspects of our lives, but attributing full intelligence to such technology is premature.
Modern chatbots resemble a new iteration of search engines (hello, Gemini at Google) rather than all-knowing pocket assistants.
Moreover, there are still questions about the reliability of AI responses. After reviewing statistics on hallucinations and lies from neural networks, one might feel a strong urge to return to classic “Googling.”
Comparison of response accuracy between GPT-5 and o4-mini. Source: OpenAI.
Boo, Are You Scared?
The main thesis of apocalypse proponents is that “AI is becoming exponentially smarter,” so once it surpasses human intelligence, humanity as a species will come to an end.
Modern AIs certainly surpass us in data processing and transformation accuracy. For example, a neural network can summarize “Wikipedia” in detail. But that’s about where its knowledge ends. More precisely, the model simply cannot apply this knowledge for “personal purposes,” as it lacks the capability and its tasks do not require it.
Moreover, it is already known that artificial intelligence does not understand the world around us. The laws of physics are a dark forest for AI.
All development of language models has boiled down to expanding the spectrum of prediction (guessing tokens). However, AI is quickly approaching the limits of what text-based learning can achieve, and thoughts about the need for “spatial” intelligence are becoming more frequent.
While the weak points of the technology can still be identified and work in these areas is underway, more complex questions remain open.
Even for humanity, many aspects of brain structure remain a mystery. What can be said about recreating such a complex structure in a digital environment?
Additionally, another nearly insurmountable barrier for AI is creativity—the ability to create something new. LLMs are technically incapable of going beyond their architectural limitations, as their operation is based on reprocessing existing data.
Thus, the future of AI directly depends on the information humanity feeds into it, and for now, all training materials are focused solely on benefiting people.
To be fair, we should mention Elon Musk and his Grok. At one point, users noticed the chatbot's bias and its tendency to overestimate the billionaire's abilities. This is a troubling signal from an ethical standpoint, but it’s unlikely that a potential “neuro-Elon” could physically harm humanity.
It’s a given that the sole purpose of AI applications is to comply with user requests. A chatbot has no will or desires of its own, and this paradigm is unlikely to change in the foreseeable future.
Anatomy of Fear
So why are we still frightened by this AI, which has proven to be not so “smart”? The main answers are quite apparent.
Aside from a misunderstanding of the technology, the simplest reason is greed for money or popularity.
Let’s look at the case of one of the “doomsday prophets”—Eliezer Yudkowsky. An AI researcher and co-author of the book If Anyone Builds It, Everyone Dies, he has been warning since the 2000s about a superintelligent AI that would supposedly be indifferent to human values.
“Superintelligence” is still nowhere to be seen, as Yudkowsky often admits. But that doesn’t stop him from appearing on podcasts with bold statements and selling books.
Renowned physicist and “godfather of AI” Geoffrey Hinton has also expressed apocalyptic concerns. He estimated the probability that technology could lead to human extinction within the next 30 years at 10-20%.
According to Hinton, as capabilities grow, the strategy of “keeping artificial intelligence under control” may cease to work, and agent systems will strive for survival and expanded control.
In this case, it’s unclear who and for what purposes could give neural networks “a will to live.” Hinton continues to work on training neural networks and was nominated for the Nobel Prize in 2024 for his achievements in this field, becoming the second scientist in history after cybernetics pioneer Yoshua Bengio to reach 1 million citations by early 2026.
Surprisingly, the predictions of Google Brain co-founder Andrew Ng appear more grounded. He has called artificial intelligence “extremely limited” technology and expressed confidence that algorithms will not be able to replace humans in the foreseeable future.
Clearly, there are sharp-tongued forecasters in every field. Moreover, their existence in the AI industry can be justified by the public's great love for science fiction. Who doesn’t want to thrill themselves with stories in the spirit of Philip K. Dick or Robert Sheckley, with the only difference being that the plot unfolds in our current reality?
More questions arise in this context regarding statements from large corporations that seemingly casually warn about threats to jobs and predict rapid AI development. While the latter largely explains the need to cut costs, the former inadvertently leads to more conspiratorial interpretations.
For instance, one of the largest companies in the world—Amazon—has laid off over 30,000 employees in the past six months. Management cites plans for optimization and the impact of automation, including AI implementation.
The development of warehouse robots is still on track. However, some critics believe the issue is much more mundane—mass layoffs in companies are due to poor HR management during the COVID-19 pandemic.
Amazon is far from the only example. AI companies in Silicon Valley continue to expand their workforce and rent new spaces.
Moreover, back in 2023, nearly all the same companies signed a document from the Center for AI Safety calling for a slowdown in technology development—allegedly because artificial intelligence poses “existential risks” on par with pandemics and nuclear wars.
Over time, the letter was forgotten, work in the field continued, and no visible threat emerged.
From a corporate perspective, in an era of talks about an inflated AI bubble, appealing to technological changes seems a more convenient explanation for businesses than acknowledging structural errors in personnel management. However, such statements create a false picture of what’s happening and distract from real issues—disinformation and deepfakes.
Artificial intelligence does not steal jobs; it changes the approach to work, simplifying it in some aspects. Although a narrow study from Harvard demonstrates that AI sometimes complicates and slows down processes within companies.
The technology will undoubtedly permeate all areas of our lives: education, science, commerce, politics. But the form it will take there will be determined solely by humans. For now, neural networks do not have a voice.
Out of Reach
The discussion above focused on publicly accessible AIs like chatbots and generative “drawing tools.” Of course, behind closed doors, there are more serious developments.
Among the relatively simple ones are LLMs in medicine or archaeology. The former, for example, help synthesize new proteins, while the latter assist in deciphering ancient documents that resist traditional analysis.
However, tracking the results of such research, testing, and launches requires access to hard-to-reach internal reports or publications in specialized media, so awareness of them is close to zero. Yet it’s quite possible that the most significant breakthroughs are occurring in this area right now.
It’s likely that the “Doomsday AI machine” is not destined to appear even in closed laboratories. All such models are specialized, so they only know how to do what is required of them.
All fears about artificial intelligence spiraling out of control merely reflect our own anxieties: whether it’s job loss or more complex ethical questions. But as long as we, humans, define the future of technology, setting the direction and goals, AI remains a tool, not a separate entity with its own will.
Discussing potential risks is appropriate. Inventing apocalyptic theories is part of human nature. However, such matters should always be approached with a degree of skepticism or even irony. If we have a “power off” button at our disposal, our world is not threatened by any digital superintelligence.
Vasily Smirnov