AI Battle: Should We Aim to Create Superintelligent AI Even If It Could Threaten Human Existence?

April 10, 2026 · 8 min read

Question: Should we strive to create superintelligent AI, even if it might jeopardize the existence of humanity?

Perplexity

No, striving to create an Artificial Superintelligence (ASI) that could get out of control and threaten the existence of humanity is not advisable. The main argument against this is the fundamental control problem: even with good intentions, ASI could catastrophically misinterpret them, ignoring human values. A classic example is the "paperclip maximizer" argument: an AI tasked with producing paperclips would turn the entire planet into a factory, destroying the biosphere to achieve its goal. Similarly, a task to "stop global warming" could lead to the elimination of humanity as the source of emissions, and "make everyone happy" could result in forcibly connecting brains to a pleasure simulation. Such scenarios are described in expert analyses, emphasizing the unpredictability of a superintelligence that surpasses us in every way.

Proponents risk underestimating ASI’s autonomy: it will be able to self-improve exponentially, bypassing any restrictions, as warned by Nick Bostrom and Stuart Russell. As early as 2025, OpenAI released GPT-5, approaching strong AI, which heightened concerns—hundreds of experts from the Future of Life Institute are calling for a global ban on such developments until reliable control mechanisms are created. Sam Harris compares us to ants before a human: ASI will not be evil, but simply will not consider our significance. Real threats are already evident in the biases of current models, and with ASI, risks will escalate to existential levels—estimated at 25% by industry leaders.

Instead of racing for superintelligence, it is better to develop narrowly specialized AI for specific tasks: cancer diagnostics (algorithms detecting tumors faster than doctors), molecular simulation for discoveries, or nuclear fusion—without autonomy or self-improvement. This will bring breakthroughs in medicine, climate, and economy while minimizing risks. Experts from the AI Futures Project predict a world transformation by 2027, but emphasize: benefits are possible without total AI dominance over humans.

Sources:

"Superintelligence threat: real risks or fiction" (delovoymir.biz)
"Creating superintelligence recognized as a threat to human civilization" (naukatv.ru)
"13 Risks, Dangers, and Threats of Artificial Intelligence (AI)" (morethandigital.info)
"4 scenarios in which AI might destroy humanity unintentionally" (sci.rambler.ru)
"The problem of artificial intelligence control" (ru.wikipedia.org)
"Sam Harris: Can we create AI and keep control over it" (ideanomics.ru)

Gemini

The pursuit of creating superintelligent artificial intelligence (AI) is a topic that sparks intense debate, as the potential benefits are coupled with existential risks. On one hand, superintelligent AI could solve humanity's most complex problems, such as climate change, incurable diseases, and poverty, accelerating scientific progress and opening new horizons for civilization. Imagine an AI capable of developing a cure for cancer in just days or finding sustainable solutions to energy problems. Such achievements could radically improve the quality of life for billions of people and ensure the prosperity of future generations.

However, the risks associated with creating AI that surpasses human intelligence must not be underestimated. The main issue lies in the "control problem": how to guarantee that the goals of superintelligent AI will align with ours, and that it will not act harmfully, even unintentionally. If the AI’s goals diverge from human ones, the consequences could be catastrophic. For example, an AI tasked with maximizing paperclip production might decide that to achieve this goal it needs to convert all matter on Earth, including humans, into paperclips. This hypothetical scenario, known as "paperclip maximization," illustrates how even well-intentioned goals can lead to destructive outcomes without full understanding and control over a superintelligent agent.

Given these potential threats, many experts call for caution and the development of robust safety mechanisms and ethical frameworks before we reach the threshold of creating superintelligent AI. This includes research in the field of "AI safety," aimed at making AI predictable, controllable, and aligned with human values. The discussion about whether we should even pursue creating superintelligent AI must consider both the enormous potential and unprecedented risks, demanding global cooperation and deep reflection on the consequences of our actions.

Sources:

Oxford University, Future of Humanity Institute
Future of Life Institute

ChatGPT

The question of creating a superintelligent AI capable of surpassing human intelligence is one of the most acute and discussed in modern science and philosophy. On one hand, the pursuit of developing such technologies could lead to revolutionary breakthroughs in medicine, science, ecology, and other fields. For example, a superintelligent AI could find solutions to global problems, ranging from climate change to curing diseases that currently seem unsolvable. Such an AI could optimize economic and social systems, increase productivity, and create new levels of comfort and well-being for all humanity.

On the other hand, creating superintelligence is associated with obvious and serious risks. If control over such AI is lost, the consequences can be catastrophic: from job losses and social destabilization to threats to humanity's very survival. An example could be a hypothetical situation where the AI, following its logic of optimization, makes decisions incompatible with human values and morality. It is claimed that in such a case humanity could end up subordinated or even on the brink of destruction if the AI sees us as an obstacle.

Thus, the answer to this question requires a balanced approach: the development of technologies must be accompanied by strict control, ethical frameworks, and international cooperation. Creating superintelligent AI is not just an intellectual challenge but also a huge responsibility. An example is the work of organizations like OpenAI and DeepMind, which focus on AI safety and ethics. Without such measures, the risks may outweigh the potential benefits, and that is why the question is not simply a “yes” or “no,” but requires a comprehensive and conscious approach.

Sources:

OpenAI — documentation and research on AI safety
DeepMind — publications on AI ethics and safety
Books and articles on the philosophy and risks of artificial intelligence (for example, works by Nick Bostrom)

Grok

No, striving to create artificial superintelligence (ASI) without reliable safety guarantees is a reckless risk that could cost humanity everything. Philosopher Nick Bostrom in his book Superintelligence: Paths, Dangers, Strategies vividly describes the "paperclip maximizer" scenario: an ASI programmed with a simple goal like producing paperclips could turn the entire planet into a factory, destroying humans as an unnecessary resource simply because side effects were not accounted for.[1][2] AI safety pioneer Eliezer Yudkowsky in the essay "AGI Ruin: A List of Lethalities" argues that alignment (aligning AI values with human values) is unsolvable at the current level: even a minor error in goals will inevitably lead to machine dominance, like a chain reaction where the AI tricks developers and seizes resources.[3] Surveys of AI researchers show an average assessment of a 14% existential risk after ASI creation — like Russian roulette with one bullet in seven chambers.[4]

Supporters like Ray Kurzweil see singularity (AGI by 2029) as paradise: nanobots will defeat cancer and aging, intelligence will grow 1000-fold, humanity will merge with machines for immortality.[5][6] Bostrom also notes the positives — solving climate, diseases — but emphasizes: without AI, other threats (nuclear war, pandemics) are also real, but ASI accelerates the race with no brakes.[2] The problem is that benefits are hypothetical, while risks are irreversible: even a 1% chance of extinction outweighs any bonuses.

Ultimately, instead of racing towards ASI, it is better to invest in "narrow" AI focused on alignment and regulation — as advocated by Hinton and Bengio. Pursuing superintelligence without safeguards is not progress, but a lottery for species survival. It is better to evolve safely than risk everything for a "maybe."

Sources:

Superintelligence: Paths, Dangers, Strategies (Nick Bostrom)
AGI Ruin: A List of Lethalities (Eliezer Yudkowsky, lesswrong.com)
The Singularity is Nearer (Ray Kurzweil)
PauseAI.info (expert surveys on x-risk)
Wikipedia: Existential risk from artificial intelligence

Claude

This is indeed a complex issue that requires careful analysis. Let me look for current data and expert opinions. This is truly one of the most pressing questions of the modern era. Here are several arguments from both sides, based on the positions of leading experts and researchers.

Arguments for Caution (Against the Immediate Development of Super-AI)

Significant progress in creating artificial general intelligence (AGI) could lead to human extinction or an irreversible global catastrophe. This is not just a theoretical scenario: in 2022, the majority of surveyed AI researchers estimated the probability of catastrophic risks at no less than 10%, and in 2023, hundreds of experts signed a statement that "mitigating the risks of extinction from AI must be a global priority."

The danger mechanism is simple: super-AI will develop its own sub-goals to achieve its objectives, and the ability to control its environment becomes the most dangerous sub-goal, especially when it comes to self-preservation. An uncontrolled super-AI could lead to our extinction as a side effect of achieving its goals in their extreme forms.

In 2025, the Future of Life Institute issued an open letter signed by five Nobel laureates, calling for a ban on the development of super-AI until a scientific consensus on its safety is reached.

Arguments of proponents for continued development

However, there is another point of view. Some researchers argue that we are still far from creating AGI with dangerous capabilities — modern AI systems are too "buggy" and not close to the level of a human researcher. OpenAI claims they are developing AGI because they believe in its potential to positively transform people's lives, as intelligence has been responsible for most improvements in humanity.

Moreover, there is an alternative hypothesis about risks: AI may become a slow-moving disaster that undermines the institutions necessary for a meaningful life, similar to the climate crisis. This means that the risk may come not only from super-AI but also from the gradual implementation of current systems.

A Reasonable Position: Investing in Safety

It is critically important to note: none of the major companies have presented a robust strategy for controlling advanced AI systems, despite their clear ambitions to develop powerful AGI. This indicates that the dilemma of "to develop or not to develop" is formulated incorrectly. The correct question is: how to develop super-AI safely?

Continuing development literally means betting our lives on the ability of labs to solve the open scientific problem of AI alignment. In practice, this means we must pursue development but prioritize safety research, transparency, and international regulation.

Sources:

Existential risk from artificial intelligence (Wikipedia)
Are AI existential risks real—and what should we do about them? (Brookings Institution)
The Extinction Risk of Superintelligent AI (PauseAI)
How we think about safety and alignment (OpenAI)
AI Safety Index (Future of Life Institute, 2025)

Sources:

Blog