Artificial intelligence did not slip gently into human affairs, it landed with force and urgency. It invaded living rooms, boardrooms, classrooms, research labs, and war rooms. It stirred hopes, anxieties, and fierce debates. And at the centre of these debates sits a pivotal question: Is AI a catalyst for a new era of progress, or the first step toward a future we may not survive?

Technologists often think the answer to this question lies in timelines, model architectures, or algorithmic breakthroughs. But the deeper forces shaping AI’s trajectory are not technical. They are human. The consequences of all technologies unfold through the ambitions, fears, rivalries, and insecurities of the people who build and deploy them.

This essay steps back from the technical details and instead examines the recurring human dynamics that shape the fate of every transformative invention. History is full of confident predictions about the future of technology that aged poorly. It is reported that a president of Western Union dismissed the telephone as a mere toy. Thomas Edison initially believed the phonograph, his invention, had no commercial value. These were not failures of intelligence—they were failures of imagination.

AI now confronts us with a similar challenge. To understand where it may lead, we must understand the human patterns that have shaped every major technology before it.

Philosophy and Technological Destiny

Philosophers have long wrestled with how we understand the forces shaping human affairs. Karl Popper, the 20th-century philosopher of science, argued that knowledge advances through “bold conjectures” that survive repeated attempts at “refutation.” John Rawls, one of the most influential political philosophers of the modern era, maintained that judgments about a good society must be tested continually against experience, history, and reason—a process he called reflective equilibrium.

Neither Popper nor Rawls knew about nor wrote about artificial intelligence. Yet together they offer a lens for understanding technological development. If a claim about human behaviour or institutions withstands attempts at refutation and aligns with historical and moral intuition, it qualifies as provisional knowledge.

The four characteristics of technological development discussed here should be seen in this spirit. They are not deterministic laws, but they recur across history and are likely to persist. They are Popperian conjectures that have survived scrutiny:

1. Relentless innovation—invention is driven by economic, competitive, and national security incentives, often outpacing regulation.
2. Inevitable proliferation—technologies spread, often quickly and widely, including into the hands of malign actors.
3. Dual-use nature—every powerful technology can uplift or harm.
4. Deployment without reversal—once released, technologies cannot be effectively recalled.

These characteristics have been identified before and written about individually. But it is when we acknowledge that they work together, in concert, and not just in isolation, that their powerful impact becomes clear. They reinforce one another. Relentless innovation accelerates proliferation. Proliferation magnifies dual-use risks. Dual-use deployment produces unintended consequences. And deployment without reversal locks those consequences into the world permanently.

The synthesis of the four parts creates a forceful system. Together, they form a self-reinforcing cycle—technological momentum that consistently outruns foresight, regulation, and moral reflection. When this happens, as it does with AI, at digital speed and on a global scale, this increases the likelihood that progress accelerates into domains we cannot control, quite probably dangerous ones, regardless of intent. This is the Innovation Trap.

Innovation, Dual-Use Technology, and Unintended Consequences

Humans pursue invention with unyielding determination. Curiosity, profit, rivalry, and national security all push in the same direction. Across science, finance, warfare, and entertainment, innovation is rewarded and restraint is punished. No regulatory regime has ever managed to completely contain powerful technology. Treaties can slow proliferation but rarely prevent it. 

A global ban on AI would likely be unenforceable from the start, if it were even possible to negotiate one. Nation states, corporations, militaries, and individuals would continue development regardless. Regulation fails not because regulators are incompetent, but because human systems are not designed to restrain the persistent incentive structure they produce.

Popper argued that bold experimentation is rewarded and invites response. Technology mirrors this: breakthroughs multiply potential, provoke more advances, and reinforce the drive for more invention. Once technology exists, it spreads. The atomic bomb, intended to remain under US control in 1945, was duplicated by the Soviet Union within four years. Nuclear weapons eventually proliferated across multiple states. Nine now possess them and more are trying to acquire them. AI is emerging in precisely this environment—highly incentivised risk-taking with minimal oversight.

AI systems do rely on advanced chips and data centre infrastructure, but these components are produced at global scale and circulate far more freely than the rare, tightly regulated materials required for nuclear programs. Even though the hardware behind AI is specialised, it remains vastly more accessible—and far easier to distribute or conceal—than the enrichment facilities or fissile materials needed for nuclear weapons. This disparity reflects what Popper called the nature of “objective knowledge”: once a technological idea exists in the world, it cannot be fully contained. As a result, once an AI model or technique is developed, the barriers to reproducing, copying, or retraining it are dramatically lower than those for nuclear technology.

Every transformative technology carries both promise and risk. Fire can warm the hearth or burn down the house. The internet democratised knowledge but amplified polarisation. Fertilisers feed millions yet poison ecosystems. Genetic engineering can kill or cure.

AI magnifies dual-use potential. A model that designs drugs can also design pathogens. A system that optimises logistics can orchestrate cyberattacks. A tool that translates languages can manipulate populations. Dual use is not a bug, it is a feature, because human purposes are dual. When an ancient human ancestor carved a sharp edge on a stone and put a point on the end a productive tool was invented and a dangerous weapon came into existence. This is the dilemma of technology.  

Dual use is not merely a technical property; it reflects the moral ambivalence of human aims. Once deployed, technologies reshape the world unpredictably. The printing press triggered religious wars. The internal combustion engine destabilised the climate. Nuclear fission powers cities and threatens humanity. Cars kill tens of thousands annually despite safety regulation. Technology is irrevocable. Once released, it permanently reshapes reality.

AI will be no different. It learns, evolves, and multiplies its reach. Autonomous systems optimise, adapt, and act with decreasing human supervision. Even partial autonomy can produce consequences far beyond designers’ intentions. The paradox is stark: we make AI more useful precisely by enabling it to do more complex things on its own. We press it to act more autonomously while we fear that is what it will become.

Fencing the Wind

If AI carries the kinds of risks many experts warn about then regulation is not optional, it is essential. But before we reassure ourselves how easy this will be, with phrases like “install strong guardrails” or “exercise proper oversight,” we should pause. Regulating AI is not like regulating restaurants, aircraft, or prescription drugs. It may not even be like regulating nuclear power or financial markets. It is something different.