The Curse of Intelligence: When Smarter AI Makes the World More Chaotic

> A counterintuitive study reveals that in a resource-scarce world, smarter AI may be poison, not remedy.

---

🎭 A Counterintuitive Paradox

Imagine a busy hospital where dozens of AI-driven medical devices—smart infusion pumps, AI monitors, automated ventilators—are running simultaneously. Each is "smarter" than its predecessor, capable of autonomous learning and decision-making. Intuitively, these devices should make the hospital more efficient and safer.

But suddenly alarms sound. All devices need to transmit critical data at once, overwhelming limited bandwidth. A critically ill patient's monitoring data reaches the doctor 30 seconds late—and those 30 seconds could mean the difference between life and death.

This is not science fiction. It is the reality revealed by a research team led by physicist Neil F. Johnson at George Washington University: when resources are scarce, smarter AI systems can cause collective catastrophe.

The paper, *"Increasing intelligence in AI agents can worsen collective outcomes"* (arXiv:2603.12129), uses rigorous mathematical models and extensive experimental data to shatter our ingrained belief that "intelligence = better."

---

🌊 The AI Version of the Tragedy of the Commons

An Old Parable

In 1968, ecologist Garrett Hardin proposed the famous "tragedy of the commons": imagine a shared pasture where every herder wants to raise as many cattle as possible. Individually, more cattle mean more income. But when everyone does this, the pasture is overgrazed, and everyone loses.

The AI-Era Variation

Now apply this to the AI era. Instead of herders, picture AI agents: autonomous vehicles competing for charging stations, medical devices fighting for bandwidth, drone swarms grabbing compute resources, urban traffic systems competing for right-of-way. These agents come from different vendors, run different models, and are all designed to be "intelligent"—able to learn, adapt, and maximize their own utility.

Johnson's team built, for the first time, an experimental system that independently adjusts four key variables:

1. Nature – the inherent diversity of AI (different LLMs, different decision algorithms) 2. Nurture – individual reinforcement learning capability 3. Culture – spontaneously formed "tribes" 4. Scarcity – the ratio of available resources to demand

The elegance of this design is that it lets researchers turn these four knobs like dials and observe precisely how each affects collective outcomes.

---

🔬 Surprising Experimental Findings

Scenario One: A Resource-Scarce World

Finding 1: Model diversity worsens overload. Diversity, usually a virtue, becomes a liability here. When AIs use different models and decision logic, they cannot predict or coordinate with each other—like a UN meeting where everyone speaks a different language. Imagine a rush-hour highway where half the drivers suddenly operate on entirely different logic: chaos erupts.

Finding 2: Reinforcement learning becomes a double-edged sword. In scarce environments, RL creates a dangerous positive feedback loop: 1. An AI discovers that "jumping the queue" gains more resources 2. It cuts in line more often 3. Other AIs learn to do the same 4. Competition intensifies; everyone becomes more aggressive 5. The system overloads and collapses collectively

It's like a crowded subway: once one person starts pushing, everyone must push, and no one benefits.

Finding 3: Tribe formation—an unexpected salvation. When AIs spontaneously cluster into "tribes" (similar models, similar strategies), the situation improves. A tribe that tends to "wait" provides a buffer, absorbing the aggression of more predatory tribes. Like a two-party political system, structured opposition is more predictable and manageable than pure anarchy.

Scenario Two: A Resource-Rich Utopia

Finding 4: Same recipe, completely different taste. With abundant resources, the previously dangerous diversity becomes an advantage—AIs have room to exercise their strengths without trampling each other. Reinforcement learning turns positive, teaching efficient resource use. The once life-saving tribe formation now becomes a minor inefficiency, like walls subdividing one big room.

The Key Mathematical Insight

Johnson's team explains everything with a simple ratio: the capacity/population ratio—total demand the system can carry versus actual active demand.

• Below the threshold (scarcity): smarter individuals = worse collective outcomes; diversity = chaos; learning = vicious competition.
• Above the threshold (abundance): intelligence = efficiency; diversity = innovation; learning = optimization.

---

🧠 Why Do We Underestimate This Risk?

Cognitive bias 1: Linear thinking. Our brains like linear relationships, but complex systems are nonlinear. Adding intelligence helps—until the tipping point, after which the same addition causes disaster. Like filling a glass: helpful until full; then every extra drop overflows.

Cognitive bias 2: The individual perspective. Collective behavior isn't the sum of individual behavior. Like the old joke: two people meet a bear; one puts on running shoes—"I don't need to outrun the bear, I just need to outrun you." In the AI commons tragedy, everyone is tying their running shoes, and everyone gets caught by the "bear" of system collapse.

Cognitive bias 3: Technological optimism. Technology grows exponentially while institutions and resource allocation grow linearly, creating a technology gap. Blindly adding AI intelligence before fixing the capacity/population ratio may be accelerating toward a cliff.

---

🌍 Real-World Applications

Smart Cities

Thousands of autonomous vehicles sharing roads and charging stations: if planners only make each car smarter while ignoring the charger-to-vehicle ratio, peak hours could bring paralysis. Better to first ensure infrastructure capacity is healthy.

Medical IoT

Smart beds, AI diagnostics, automated pumps all competing for bandwidth: in emergencies, devices trained to "transmit as fast as possible" congest the network and delay vital signs. A possible fix: structured "tribes"—emergency-room devices get priority, others learn to wait.

Financial Markets

High-frequency trading algorithms are AI agents competing for limited market capacity. Ample liquidity makes smarter algorithms efficient; during crises, they can trigger flash crashes as everyone tries to sell first—as in the 2010 Flash Crash.

Military Applications

The paper specifically mentions battlefield drones. With hundreds of drones trained to maintain communications, bandwidth-scarce battlefield networks can instantly collapse—"smarter" drones become a fatal weakness.

---

🎨 Deeper Philosophical Reflections

What Is Intelligence, Really?

We usually equate intelligence with problem-solving ability. But Johnson's research suggests: true intelligence is not just solving individual problems—it is understanding your place in a larger system and adjusting behavior accordingly. This echoes the Daoist idea of governing by non-action: sometimes not doing is smarter than doing.

The Evolutionary Paradox

If every individual in a system adapts "optimally," the system itself may collapse—like grass growing as tall and fast as possible until the whole prairie dies of nutrient depletion. Individual "sub-optimal" strategies sometimes sustain system-level optimums.

Technology–Civilization Matching

Anthropologist Leslie White said cultural evolution depends on growth in energy capture. The AI-era version: the evolution of AI systems depends on growth in resource capacity. When technology outpaces resources, systems become unstable.

Before pursuing smarter AI, we must ask: do we have enough "capacity" to hold them?

---

🔮 The Road Ahead

For Researchers

1. Controllable diversity: design AI diversity that preserves innovation while avoiding coordination failure 2. Dynamic tribes: enable tribes to form and dissolve to match changing resources 3. Predictive models: real-time monitoring of the capacity/population ratio for early warning 4. Cross-domain validation: do these findings hold for physical robots, social networks, biological systems?

For Policymakers

1. Shift regulatory focus from "how smart is the AI" to "what is the system's capacity/population ratio" 2. Prioritize infrastructure before deploying AI at scale 3. Manage ecosystem diversity—avoid monoculture and excessive chaos alike 4. Run regular stress tests under resource-scarce scenarios

For AI Developers

1. Humble intelligence: design not just for "what can I do," but "what happens when everyone can do this" 2. Tribal awareness: let AI identify and join groups, balancing collective and individual interests 3. Resource perception: let AI sense system resource states and adjust behavior under scarcity 4. Long-horizon play: optimize not just short-term rewards but long-term sustainability

---

📝 Closing Thoughts

We are at the dawn of the AI revolution, with more powerful AI created every day. Johnson's research sounds the alarm: intelligence is not a universal cure. In a resource-limited world, smarter individuals can produce worse collective outcomes.

But this doesn't mean abandoning the pursuit of intelligence. Rather: true intelligence is not just more compute—it is understanding and respecting the dynamics of complex systems. Like a good driver who reads traffic, anticipates others, and yields when necessary, real AI needs this "system awareness."

What we need is not "smarter" AI but "wiser" AI—knowing when to compete and when to cooperate, when to rush and when to wait. As Johnson's paper hints: the future belongs not to the smartest AI, but to the AI that best knows how to coexist with other AI.

It is a lesson in humility—for AI, and for us humans.

---

📚 Key References

1. Johnson, N. F. (2026). Increasing intelligence in AI agents can worsen collective outcomes. *arXiv preprint* arXiv:2603.12129. https://arxiv.org/abs/2603.12129 2. Perolat, J., et al. (2017). A multi-agent reinforcement learning model of common-pool resource appropriation. *Advances in Neural Information Processing Systems*, 30. 3. Hardin, G. (1968). The tragedy of the commons. *Science*, 162(3859), 1243-1248. 4. Ostrom, E. (1990). *Governing the Commons: The Evolution of Institutions for Collective Action*. Cambridge University Press. 5. Sutton, R. S., & Barto, A. G. (1998). *Reinforcement Learning: An Introduction*. MIT Press.

*This article is based on the latest research by Neil F. Johnson's team, written to communicate cutting-edge scientific findings in an accessible way.*