Will AI Bring the End of the World? The Answer May Surprise You

Ask whether AI could end humanity, and a familiar scene appears: machines become autonomous, turn against us, launch nuclear weapons, enable biological catastrophe or shut down power, communications and water. Civilization ends before people can respond.

These risks cannot be dismissed. They deserve serious research. But I am more concerned about another route: AI may not need to decide to destroy humanity for humans to destabilize their own society through its use.

Wealth concentrates. More people lose secure livelihoods. Economic pressure becomes political anger, political antagonism becomes international conflict, and eventually someone lets a weapon settle the argument. Even if AI presses the final button, might the catastrophe still begin with our greed, laziness, selfishness and ignorance?

My judgment is that this human-driven pathway deserves more attention than a debate focused entirely on the moment AI rebels. I think it is more likely to erode civilization first. That is my assessment of risk, however, not a scientifically established ranking of apocalypse probabilities.

The planet can survive while people's worlds fall apart

“The end of the world” can mean human extinction or the collapse of civilization. For someone who loses an income, a home and a support network, a familiar world can also disintegrate while the planet continues turning. These are different scales of harm. We should keep them distinct while taking each seriously.

Suppose AI enables companies to accomplish more with fewer workers. That could be a considerable benefit: less repetitive labour, better health care, more accessible education and stronger public services. But if most gains flow to a small group of capital owners and platforms while workers, families and local communities bear the transition costs, higher productivity need not improve shared life.

This does not establish that AI will cause mass unemployment. The ILO's May 2025 research summary estimates that roughly one in four workers worldwide is in an occupation with some exposure to generative AI. It also concludes that, because human input remains necessary, most jobs are more likely to change than disappear. Occupational exposure is not a forecast of job losses.

The IMF's January 2024 analysis warns that productivity gains and increased capital returns could deepen income and wealth inequality. Social safety nets, retraining and policy choices can influence how the benefits are distributed.

These sources support concern about work and distribution. They do not prove that AI will collapse the economy. My question is more concrete: if businesses count only the wages they save, and governments celebrate only higher output, who helps displaced people regain a secure place in society? If that question remains unanswered, efficiency and insecurity may grow together.

Every step from economic strain to conflict involves choices

The pathway I fear runs like this: AI accelerates changes in some occupations, the distribution of gains fails to keep pace, and people lose their sense of security. Political leaders and interest groups turn anxiety into hostility. States turn competition into confrontation. Weapons are then asked to resolve what politics should have addressed.

This is a possible causal chain, not an inevitable sequence. Changes in work do not automatically produce poverty; poverty does not automatically produce war. Investment, public services, labour negotiations and diplomacy can interrupt the chain. Recessions and international conflicts also have many other causes. Their coincidence with AI adoption would not, by itself, establish causation.

But if AI does help accelerate instability, should the subsequent deaths be attributed to AI?

My answer is that AI can belong in a causal explanation without becoming the place where the investigation of responsibility ends. We should ask who chose deployment, who benefited, who expanded the system despite known risks and who could have reduced the harm but declined to act.

“Human responsibility” must not become collective blame that includes the victims. Someone who loses a job has different powers from someone who controls deployment, capital or public policy. The more authority a person or institution has, and the more capacity to foresee and prevent harm, the more scrutiny their decisions deserve. Saying that humans are inherently greedy cannot be allowed to dissolve specific responsibility.

Humanity possessed catastrophic power before AI

Long before modern AI, humanity created nuclear weapons capable of causing global catastrophe. If a major nuclear war devastated civilization, would we simply say that the bomb made a mistake? Or would we investigate who built and deployed the weapons, escalated the confrontation and ordered their use?

The weapons would be a technological condition of the harm. Human power structures, fear and decisions would also be part of the explanation. We cannot remove the technology from the chain of responsibility, but we cannot put the whole chain inside the technology either.

Nor would all the damage occur in the instant of a blast. A 2022 Rutgers-led study on nuclear war and food supplies modelled how soot-driven climate disruption could affect global food production. This is a scenario study with assumptions, illustrating indirect harm. It does not demonstrate that a single bomb would instantly kill every human being or physically destroy the Earth.

The analogy with AI is useful here: a technology's potential for harm depends partly on where we put it and what powers we give it. But the analogy has limits. AI can generate plans, select actions and execute tasks through connected tools. It can fail or behave in ways people did not anticipate. AI and nuclear weapons are different systems. Acknowledging human responsibility cannot replace research into AI safety.

The International AI Safety Report 2026 describes continuing expert disagreement about the likelihood and severity of loss-of-control scenarios. Human responsibility does not make autonomous loss of control impossible. Equally, the possibility of loss of control does not erase responsibility for deployment and authorization.

Convenience can become an excuse for unchecked authority

Having AI organize a calendar, search for information or draft a document differs fundamentally from allowing it to move funds, control critical infrastructure or determine who lives and dies. The latter delegation transfers power, not simply a larger workload.

By “laziness,” I do not mean using automation to save effort. I mean trying to outsource understanding, verification, refusal and responsibility along with the task. We can reduce routine work while retaining important judgment. Or we can progressively remove limits because a system has worked well before, until we no longer understand what it is doing.

Greed encourages us to pursue faster growth, lower costs and greater control while leaving other people's circumstances and long-term risks for later. Selfishness treats the gains as ours and the damage as society's. Ignorance mistakes fluency for truth, capability for wisdom and permission to act for the legitimacy to decide.

AI did not invent these weaknesses. It may allow them to operate faster, across a wider field and with less time to notice their consequences.

Weaponization follows the same concern. An AI need not want to destroy the world for people to connect increasingly difficult-to-understand systems to decisions about force. Fear of falling behind and the desire to defeat an opponent can provide the motivation. The ICRC's position on autonomous weapons treats the loss of human control and judgment as a serious concern, calling for limits and effective supervision, intervention and deactivation.

If AI eventually presses a catastrophic button, I still think our first questions must include: who made that action possible? Who created opportunities to bypass the limits? Who deployed the system despite inadequate oversight? An investigation focused only on the final second would miss the years of decisions that made it possible.

Self-cultivation must change our decisions

My conclusion is that humanity must improve itself to govern tools of such extraordinary power.

This does not mean becoming a saint before using AI. By self-cultivation, I mean examining our desires, admitting what we do not understand, practising restraint and including other people's lives and dignity in our decisions.

An individual can begin with one delegation: what do I want AI to do? Which decisions require my confirmation? If someone is harmed, will I acknowledge that I chose to use this system, instead of saying that AI told me to do it?

A company must look beyond efficiency to who bears transition costs, whether employees have opportunities to negotiate and learn, whether someone can intervene when the system fails, and whether performance incentives reward transferring risks to others.

Public institutions must translate reflection into practical arrangements: named responsibilities, limited permissions, review processes with genuine power to refuse, traceable records and channels through which affected people can challenge decisions. Good intentions matter, but safety cannot depend on everyone remaining virtuous.

Over the next year or two, I will watch three things: whether AI's productivity gains improve ordinary people's lives; whether human reviewers in consequential systems have time, information and veto power; and whether incident investigations identify responsibility for decisions and permissions. Sustained improvement would give me reason to revise my concern about social destabilization.

We can invest in technical safety, improve institutions and ask ourselves to act with less greed, less evasion and greater clarity. These efforts belong together.

I do not think AI must bring the end of the world. I believe it could help humanity reach another height of civilization. But realizing that possibility requires our judgment, restraint and willingness to bear responsibility to mature alongside our capabilities.

Choose one consequential task you are preparing to delegate to AI. Write down its permitted scope, the moment it must stop and the person who remains responsible. Self-cultivation can begin with how we exercise power today.

Frequently asked questions

Does M.K. believe AI cannot destroy humanity?

No. I do not rule out severe risks from direct attacks, misuse or autonomous loss of control. I give greater attention to social instability and unchecked delegation, while distinguishing that judgment from an established probability ranking.

Would AI-driven inequality inevitably lead to war?

No. Every link between changes in work, unequal distribution, political antagonism and war depends on institutions and human decisions. It is a possible pathway to prevent, not a predetermined outcome.

Why investigate human responsibility if AI acts on its own?

Design, deployment, permissions and oversight also shape consequences. That does not mean every action was directly ordered by a person. Investigators must still examine how models and systems failed and identify responsibility according to evidence and actual control.

Can self-reflection replace regulation and technical safety?

No. Reflection should motivate stronger institutions and safety measures, rather than require victims to trust powerful people's intentions. Personal restraint, organizational responsibility and verifiable safeguards need to develop together.

Sources and further reading

The sources below were revisited on October 14, 2026. The article distinguishes their findings from the author's assessment of risk.

• ILO: Generative AI and jobs: A 2025 update
• IMF: AI Will Transform the Global Economy, 2024
• International AI Safety Report 2026
• ICRC: Position on autonomous weapon systems, 2021
• Rutgers: Nuclear war, climate and global food scenarios, 2022
• What Should We Teach Children Before Teaching Them to Use AI?
• Could AI Become Public Infrastructure?
• Future City Governance: Does Better AI Management Make Us Freer? (Traditional Chinese)