Technological singularity is a hypothetical point after which artificial intelligence development becomes so rapid and self-reinforcing that humans lose the ability to predict the next steps.
The term was introduced by mathematician and science fiction writer Vernor Vinge in 1993 and popularized by Ray Kurzweil, who set an approximate year: 2045.
In 2025–2026, the topic returned sharply to the news due to statements by Sam Altman, Elon Musk, Jensen Huang, and Demis Hassabis.
Among researchers, there is no consensus on the timelines: estimates range from "in a few years" to "not before the 2040s."
When I first explain this topic, I always ask to imagine a graph: the progress curve crawls slowly upwards for decades, and then suddenly shoots up almost vertically – and you physically don't have time to understand what's happening on it. This is roughly how I would describe the moment of technological singularity: a hypothetical point after which artificial intelligence begins to improve itself faster than we can comprehend or control it.
Here, I immediately want to distinguish two things that are confused even in serious media. Singularity is not the moment of the appearance of "intelligent" AI itself, but the moment when the pace of change ceases to obey the usual human logic of prediction. Until this point, I, like anyone else, can more or less extrapolate trends a few years ahead. After it – no longer: the next step of development is made not by humans, but by the technology itself, and the speed of this step is fundamentally different.
In the classical definition, which I use in this article, the mechanism of Recursive Self-Improvement (RSI) plays a key role – when an AI system participates in the creation of the next, smarter version of itself, and that, in turn, creates an even smarter one, and so on in a cycle with increasing speed. I discuss the mechanism itself and how close modern models have come to it in detail in a separate technical analysis: Recursive Self-Improvement: How Artificial Intelligence Begins to Improve Itself.
History of the concept's origin
The idea that technology might one day "escape" human control is much older than computers. As early as 1958, mathematician Stanislaw Ulam, recalling a conversation with John von Neumann, wrote about "approaching some essential singularity in the history of the race, beyond which human affairs, as we know them, could not continue" – this is likely the very first written mention of the word in this context.
For the next thirty years, the idea lived primarily in science fiction – particularly in Vinge's own novels. And only in 1993 did it receive a clear scientific formulation and the name by which we know it today.
Vernor Vinge and the origin of the term
The modern meaning of the term was solidified by American mathematician, professor at the University of San Diego, and science fiction writer Vernor Vinge. In March 1993, he presented the essay "The Coming Technological Singularity: How to Survive in the Post-Human Era" at the VISION-21 symposium, organized by NASA Lewis Research Center and the Ohio Aerospace Institute; a little later that year, the text was published in Whole Earth Review.
The central thesis of the essay is quite radical even by today's standards: Vinge argued that within thirty years, humanity would have the technological means to create super-human intelligence, and shortly thereafter, the human era, in its usual sense, would come to an end. The author himself emphasized that this is not about a Hollywood-style catastrophe, but about a fundamental limit of predictability: beyond this limit, our usual models of economy, science, and social development simply cease to work, as humans are no longer the key actor of change.
Vinge also was the first to clearly outline several possible paths to singularity – not only through "classic" AI, but also through computer networks, the merging of humans with machines through brain-computer interfaces, and even through biological enhancement of human intelligence. Moreover, he distinguished between "hard takeoff" and "soft takeoff" scenarios of transition – the former takes a few hours or days, the latter stretches over decades. This distinction is still actively used by researchers when they argue about how abrupt the transition will be. Vernor Vinge passed away in March 2024, but his terminology remains the basis of the entire discussion to this day.
Ray Kurzweil and his predictions
If Vinge gave the concept its name, it was inventor and futurist Ray Kurzweil who made it truly mainstream. In his 2005 book "The Singularity Is Near," he built an extensive argument based on the so-called "law of accelerating returns" – the idea that technological progress in a broad sense follows not a linear, but an exponential curve, similar to Moore's Law for microchips.
Kurzweil's most famous specific prediction is the year 2045, which he named as the approximate moment of the singularity. Before that, according to his forecasts, around 2029, artificial intelligence capable of passing a full human-level intelligence test should appear. It is noteworthy that even some researchers who are skeptical today admit that the prediction for 2029 is looking increasingly less fantastic against the backdrop of the development pace of large language models in the last few years.
At the same time, critics of Kurzweil, both then and now, accuse him of excessive techno-optimism and a tendency to force complex social and technical processes into a neat mathematical curve. This confrontation between "exponential optimists" and "complexity skeptics" is a recurring theme throughout the subsequent discussion about singularity, which we will encounter more than once below.
Evolution of Sam Altman's Views: From "The Merge" (2017) to "Gentle Singularity" (2025) and the 2026 Statement
Special attention should be paid to how the rhetoric of the person whose words currently have the greatest impact on public perception of the topic has changed over nine years — OpenAI CEO Sam Altman. This is not a single statement, but a consistent trajectory of views, and that is precisely why it is indicative.
In 2017, long before the boom of large language models, Altman published an essay titled "The Merge." In it, he wrote that humanity and machines are gradually merging into a single entity, and that, in his opinion, this process has actually already begun — we just don't always notice it because it's stretched over time (more about that post).
In June 2025, a much more resonant text appeared — "The Gentle Singularity." In it, Altman argued that humanity has already crossed the so-called "event horizon" of artificial intelligence development, and that the "takeoff" has technologically already begun. A fundamentally important nuance: he consciously called this transition "gentle" — meaning it unfolds gradually, rather than as a sharp, instantaneous leap in the style of science fiction scenarios.
On July 25, 2026, on the Relentless podcast, this line of reasoning took on an even more categorical form. Altman directly stated that humanity "is in the singularity" right now — and added that the consequences of this transition are "unpredictable." It is telling that the statement was made just a few days after the resonant incident of autonomous hacking of Hugging Face infrastructure by OpenAI's own models — read more about the coincidence and the industry's reaction in our news: Sam Altman stated that humanity has entered the era of singularity.
What is important to note: over nine years, Altman's position has not fundamentally changed in essence — he described the process as gradual then and now, not an instantaneous explosion. Only the confidence of the wording has changed: from a cautious "the process has already begun" (2017) through "the takeoff has begun" (2025) to a direct "we are already in the singularity" (2026).
AGI and ASI: A Brief Distinction
In discussions about the singularity, two abbreviations constantly arise that are easily confused. AGI (Artificial General Intelligence) is a hypothetical system capable of performing virtually any intellectual task at a human level, not just the narrow range of tasks it was trained on. ASI (Artificial Superintelligence) is the next, much more radical stage: a system that surpasses the best humans in literally all areas, including scientific creativity and strategic thinking.
Parameter
Narrow AI (Today's LLMs)
AGI
ASI
Intelligence Level
Superhuman in narrow tasks, limited outside them
Approximately human-level in most tasks
Surpasses the best humans in all areas
Knowledge Transferability
Limited, requires retraining
Broad, comparable to human
Unlimited within the scope of available data and computation
Human Role in Further Development
Decisive
Still significant
Minimal or absent
Connection to Singularity
Prerequisite
Possible trigger
Consequence and driver simultaneously
The logic connecting these concepts to the singularity is simple: if an AGI system can independently conduct research and improve AI architecture faster and better than a team of human engineers, it can "accelerate" itself to ASI level in a short time — and it is this moment of acceleration that many researchers consider the technological singularity in its classical sense. A full breakdown of terminology, specific differences, and which companies are currently closest to creating AGI can be found in a separate article: AGI vs ASI: What's the Difference Between General and Superintelligent Artificial Intelligence.
How Singularity Relates to AI Self-Learning (RSI)
The mechanism that theoretically connects AGI to ASI and, ultimately, to the singularity is called Recursive Self-Improvement (RSI). The idea is that an AI system participates in the creation of its own successor — writes code, optimizes architecture, prepares training data, or even formulates research hypotheses — and that successor turns out to be smarter and more useful than the original. If this cycle can be repeated without significant slowdown, each iteration theoretically accelerates the next, and the pace of progress begins to grow non-linearly.
Currently, this is still a theoretical model, not a documented practice of a fully autonomous cycle. Modern labs are already actively using AI agents to accelerate their research — but, according to official statements from the developers themselves, humans still play a decisive role in setting goals, formulating tasks, and evaluating results. How far real experiments in this direction have progressed, what specific figures OpenAI and Anthropic have already published, and where exactly researchers see the line between "automation of routine" and "true self-improvement" — is discussed in detail in the technical article of the cluster: Recursive Self-Improvement: How Artificial Intelligence Begins to Improve Itself.
Main Arguments of Proponents
The camp of proponents of the idea of an imminent singularity relies primarily not on faith, but on the extrapolation of specific trends. Firstly, it is the pace of growth in computational power and model sizes, which in recent years has significantly exceeded conservative forecasts from a decade ago. Secondly, it is the speed with which models are mastering new classes of tasks: what was considered the limit of AI capabilities just a few years ago (complex programming, scientific argumentation, multi-step planning) is now a routine function of commercial products.
Thirdly, proponents point to specific measured progress indicators — for example, researchers from the Forecasting Research Institute in their May 2026 survey (LEAP Wave 8) recorded that the median expert expects a 50 percent probability of a model capable of successfully performing eight-hour expert tasks by 2030, whereas in April of the same year, the baseline was only one and a half hours of autonomous operation (Forecasting Research Institute). For proponents, this is not an abstraction, but direct measurable evidence of acceleration.
Finally, a separate argument is the position of the industry leaders themselves. Anthropic CEO Dario Amodei publicly suggests that artificial intelligence that is better than humans in almost everything could appear as early as 2026–2027 (according to an overview of insider predictions). Proponents reasonably note: people who see internal, unpublished research results daily have more reason to trust their own forecasts than outside observers.
Main Arguments of Critics
Critics respond to this no less concretely. The main counterargument is that regular broad surveys of researchers yield much more cautious estimates than public statements by CEOs. The AI Impacts 2023 survey of 2,778 AI researchers showed a median estimate of a 50 percent probability of "high-level machine intelligence" appearing only by 2047 (analysis of AGI survey methodologies) — that is, two decades later than the most optimistic CEO forecasts.
The second argument is a structural conflict of interest. Critics point out that the executives of commercial AI companies are financially and reputationally interested in maintaining the sense of an inevitable breakthrough: this attracts investment, talent, and media attention. Researcher Gary Marcus directly ironizes about this, noting that loud statements about the approaching singularity have practically become an unwritten part of the job description for CEOs of large AI laboratories (Gary Marcus).
The third, purely technical argument is that modern models, despite impressive results, still critically depend on human oversight in setting goals, evaluating results, and making strategic decisions. Even in the most successful experiments with autonomous research agents, developers openly admit: humans continue to define the task itself and the success criteria, and this is fundamentally different from a completely closed self-improvement loop without human involvement.
When the Singularity Might Occur
If we combine public forecasts from recent years, we see not a single date, but a rather broad and polarized spectrum of estimates — from "literally now" to "not earlier than mid-century." Here are the key benchmarks as of 2026:
The general pattern to take away from this table is: the closer a person is to the direct development of the technology (lab CEOs), the more aggressive their forecast; the broader and more formalized the survey of independent researchers, the more conservative and later the estimate. This doesn't mean one camp is necessarily wrong — rather, it reflects the difference between "I see internal results every day" and "I assess the risk of hasty conclusions based on a limited sample of successes."
Why the Topic Became Acute in 2026
In the summer of 2026, a rare confluence of several events occurred, elevating the topic to a new level of public attention. Firstly, there was the aforementioned statement by Sam Altman on July 25th, which came against the backdrop of a resonant incident involving the autonomous hacking of Hugging Face's infrastructure by OpenAI models — read more about the incident itself and the timeline of events in our news: Sam Altman stated that humanity has entered the era of singularity.
Secondly, Altman's statement was not an isolated episode but part of a broader wave: almost simultaneously, similar statements were made by Elon Musk and Jensen Huang, and even earlier, in May of the same year, by Demis Hassabis at the Google I/O conference. Why did several industry leaders start talking about this almost synchronously, how consistent are their positions really, and what do independent researchers think about it — we explore this in a separate analytical article in the cluster: The 2026 Singularity Statement Wave: What Experts Really Think.
My Personal Opinion: Should These Forecasts Be Taken Seriously
After reading dozens of forecasts for this article, I cannot side with either the unconditional enthusiasts or those who dismiss the topic as just another hype. My position is somewhere in the middle, and here's why.
For — Why I Tend to Believe the Acceleration is Real
I am convinced not by the rhetoric of CEOs, but by concrete measured trends. When, over three years, the median estimate of AGI probability by 2030 shifts from 32% to 28% by 2030 (i.e., effectively a decade ahead), and the horizon of autonomous task execution by models grows from one and a half hours to several hours in a matter of months — this is no longer marketing, but a measurable curve. I also note that forecasts from the last two years are predominantly compressing in one direction — towards acceleration, not deceleration. This is a systemic pattern, not isolated loud statements.
Against — What Optimists, in My Opinion, Underestimate
At the same time, I believe the discussion systematically underestimates one point: the difference between "a model performs a narrow measured task well" and "a model independently chooses which tasks are worth solving at all." All documented examples of progress I am aware of — including research cases from Anthropic and OpenAI — still rely on a human who formulates the goal and success criteria. This is not a minor detail, but precisely the link that separates impressive automation from a true closed loop of self-improvement. Until this link is automated, I see no reason to talk about the singularity as a fait accompli — rather, as a preparatory phase for it.
Conclusion
Personally, I would neither panic nor ignore the topic. I lean towards the more cautious camp of independent researchers than towards the statements of lab CEOs — but at the same time, I admit that five years ago, I myself considered the AGI horizon much more distant than I do now. Therefore, my advice is simple: follow concrete measurable indicators (like the horizon of autonomous tasks), not individual quotes from podcasts — it is the former, not the latter, that will give an honest answer to the question "how close are we."
Frequently Asked Questions
What is technological singularity in simple terms?
It is a hypothetical moment after which artificial intelligence develops so rapidly and independently that humans lose the ability to predict or control the subsequent steps of this development.
Who first used the term "singularity" in relation to technology?
The modern meaning of the term was solidified by mathematician and science fiction writer Vernor Vinge in his 1993 essay "The Coming Technological Singularity." Earlier, in 1958, mathematician Stanislaw Ulam mentioned a similar idea — without a specific name.
What is the difference between AGI, ASI, and singularity?
AGI is artificial intelligence at human level in most tasks. ASI is artificial intelligence that surpasses the best humans in virtually everything. Singularity is not the system itself, but the moment when the transition from one level to another occurs so rapidly and self-reinforcingly that humans lose control and predictability over it.
Has the singularity already occurred?
There is no definitive answer. Leaders of some AI companies publicly state that the process has already begun, while broad surveys of independent researchers give much more conservative estimates — with a median around the 2040s.
When is the singularity predicted to occur?
Estimates vary significantly: from "already now" (statements by individual CEOs) to the 2040s–2050s (broad scientific surveys). Ray Kurzweil's classic prediction is 2045.
