Data makes societies less intelligent

Data makes societies less intelligent

This is the third in a series of five articles highlighting resilience in the era of artificial intelligence — ED.

In the previous column, I argued that the artificial intelligence (AI) era will be defined less by access to technology than by what I called collective learning capacity. This is the continuous cycle through which organizations and societies detect signals, interpret them, decide on a course of action and learn from the results. I closed by suggesting that the first place this capacity is breaking down is also the most fundamental: the basic ability to tell signal from noise.

In this column, I explore why, paradoxically, more information is currently making modern societies less, not more, intelligent.

For most of human history, information was overwhelmingly scarce. The civilizations and institutions that gathered, preserved and transmitted data most effectively consistently outperformed those that did not. Our entire architecture of progress — from universities to corporate hierarchies — was built on the assumption that more information would naturally and inevitably produce better outcomes. The goal was always to acquire more.

That assumption is now breaking.

We generate more information in a single day than was produced over centuries of earlier history. We possess unprecedented analytical capacity, real-time data streams and global connectivity. And yet, institutional trust, public cohesion and shared understanding of basic facts appear strained rather than strengthened, despite this abundance.

Something has fundamentally inverted.

The first reason for this inversion is structural. Above a certain volume, additional information does not improve a system's grasp of its environment; it actively degrades it. A signal is information that carries actionable meaning relative to a decision, but noise is information that does not. When noise grows faster than signal, a system's effective intelligence falls even as its total information base rises. This is not a metaphor, but a measurable property of any information-processing system.

The math is straightforward. The volume of available data has expanded by orders of magnitude over recent decades. The number of real-world decisions and underlying events that generate meaningful signal has not. The share of any information environment that carries decision-relevant meaning must therefore be falling, even if we cannot measure it directly.

Decision science has documented this pattern for decades. Studies of expert decision-making across clinical diagnosis and financial trading consistently show that beyond a specific threshold, additional information induces cognitive overload, generates false confidence and makes it impossible to weigh contradictory inputs correctly. What AI changes is not the pattern itself, but the inevitable scale at which all of us are now exposed to.

Until recently, this signal degradation was driven mostly by human behaviors and early algorithms — by platforms optimizing for engagement, by content saturation and the slow erosion of editorial gatekeeping.

AI is now accelerating this degradation exponentially.

Generative systems produce text, images, video and synthetic data at a scale no prior technology has ever approached. A rapidly growing share of our public and private information is machine-generated: automated summaries, synthetic commentary and AI-laundered citations. Because these models mimic human fluency, each item looks credible. But a vast number are confidently wrong, devoid of context or fabricated entirely. Distinguishing among them at speed is now beyond what most human readers and institutional filters can reliably do.

The result is that signal is becoming harder to isolate, not easier, even as analytical capacity grows.

The harder problem comes next. Even if a signal could be perfectly isolated, modern societies increasingly struggle to interpret it together.

Interpretation depends entirely on shared frames: common definitions, trusted experts and agreed-upon methods. However, our hyper-abundant information environments have steadily fragmented those frames. Each camp now has its own curated data, its own preferred analysts and its own custom AI tailored to confirm its preexisting reading of events. Two communities can examine the exact same dataset and arrive at totally incompatible conclusions, with each side feeling internally and mathematically justified. The result is a strained society in which most factions can find enough data to defend their preferred positions. Interpretation no longer reliably converges; it tends to diverge.

The final failure point in this cycle is the most subtle because it is the least visible. When an information environment is profoundly polluted, the feedback loop itself becomes unreliable.

Feedback is the mechanism through which systems learn whether their decisions worked. It depends on observing outcomes accurately and tracing them back to root causes. In a degraded environment, both are compromised. We increasingly rely on AI systems to produce confident summaries of results they cannot verify in the physical world. Metrics proliferate, but their relationship to underlying reality steadily weakens. Without honest feedback, the learning loop stops functioning. Decisions are made, but their true consequences are never properly absorbed. The cycle keeps turning, but the system learns nothing.

Korea is by no means exempt from this dynamic. A society characterized by extraordinary digital throughput is also uniquely exposed to information saturation, fragmented interpretation and feedback distortion. The country's immense strengths in producing and transmitting information that may, in time, accelerate the very breakdown this piece describes. I will return to this specific risk in the final installment.

Ultimately, what I have described here is learning inequality at its very first failure point. The organizations and societies that will pull ahead are those that figure out how to preserve signal, share interpretation and maintain honest feedback under conditions of AI-driven noise.

Because as we move deeper into this era, the ultimate strategic truth becomes clear: What matters is not intelligence, but the capacity to learn collectively.

We must discard the illusion that accumulating more information is the same thing as generating intelligence.

Additionally, signal detection and interpretation are only the first half of the cycle. Even if a system understands its environment perfectly, it still must act on that understanding. In most modern organizations, this is where the harder, more entrenched bottleneck sits.

Analysis has become cheap. Action has not.

That is where this series turns next.

Charles Chang is a PhD. candidate in AI Convergence and a security resilience consultant based in Seoul, with extensive experience spanning government and corporate leadership.