The AI race is a N-player prisoner’s dilemma

The Suicide Pact: Why the AI Race is Unstoppable

If you listen to the media, the AI race is a story about ego, hype, and bubbles. They paint CEOs like Larry Page and Sam Altman as reckless gamblers chasing a shiny new toy.

But this view is dangerously naive. When Google co-founder Larry Page allegedly told confidants he would “rather go bankrupt than lose this race,” he wasn't being melodramatic. He was being mathematically rational.

To understand why, you have to strip away the technology and look at the game theory. We are trapped in a three-layer logical prison that ensures acceleration is the only valid move.

It is a Fractal Multiplayer Prisoner’s Dilemma, played with Dollar Auction payoffs, governed by the Sorites Paradox.

Here is why we can’t stop.

1. The Structure: A Fractal Prisoner’s Dilemma

Most people understand the standard Prisoner’s Dilemma: two people act in self-interest and end up with a worse outcome than if they cooperated.

But the AI race is a Fractal version of this. It isn't just happening between Microsoft and Google. It is happening at every scale of the economy simultaneously, creating a trap that no individual or group can escape.

• The Employee (Micro Level): You are a copywriter or coder. You know AI might eventually make your profession obsolete. But today, right now, using AI makes you 10x faster than the colleague sitting next to you.
  • The Trap: If you don't use it, you are fired for low productivity. If you do use it, you train the system that will eventually replace you.
  • The Move: You defect (use AI) to survive the quarter, accelerating your own long-term obsolescence.
• The Corporation (Mesa Level): This is where the CEOs sit. If Google pauses AI development to ensure safety, they don't get a “participation trophy” for ethics. OpenAI, Anthropic, or Meta simply capture 100% of the future market. In a winner-take-all cognitive economy, the second-best AI isn't a competitor; it is zero.
  • The Move: Total capital commitment. You must build the model, even if it breaks the bank.
• The Nation State (Macro Level): The US government cannot regulate AI into safety because if they do, China achieves “Unit Cost Dominance” in intelligence. The economic and military advantage of automated cognition is so high that any nation that pauses development instantly becomes a vassal state to the one that didn't.

At every level, “Defection” (acceleration) is the dominant strategy. To stop is to die.

2. The Stakes: Dollar Auction Payoffs

Why don't the players just fold? In a normal game, if the stakes get too high, you walk away.

But this isn't a normal game. It is a Dollar Auction.

In a standard auction, if you bid $10 and lose, you keep your $10. In a Dollar Auction, the second-highest bidder must pay their full bid but gets nothing.

Google, Microsoft, and Meta have sunk tens of billions into GPU clusters and data centers. These are “sunk costs” in the most brutal sense.

• If they win: They own the infrastructure of the future economy.
• If they stop: Their billions are incinerated, their stock goes to zero, and they are crushed by the winner.

This explains Larry Page’s calculus. To a logician, “Going Bankrupt” and “Losing the Race” are the exact same outcome: Zero.

• Fold now: Guaranteed Zero.
• Keep bidding (even risking bankruptcy): Non-zero chance of winning everything.

The rational move in a Dollar Auction is to bid every single cent you have until you either win or die. There is no “cutting your losses.”

3. The Lock: The Sorites Paradox

This is the piece most analysts miss. Even if all the CEOs met in a room and agreed to a truce—”We will stop before AI becomes dangerous”—they couldn't do it.

This is the Sorites Paradox, also known as the Paradox of the Heap.

• If you have a pile of sand and remove one grain, is it still a heap? Yes.
• If you remove another, is it still a heap? Yes.
• At what specific grain does it stop being a heap? There is no line.

AI development works the same way. There is no clear line where “Helpful Autocomplete” becomes “Dangerous Agent.”

• Is a spell-checker dangerous? No.
• Is a paragraph drafter dangerous? No.
• Is an email auto-responder dangerous? No.

Because integration is continuous, you cannot write a treaty to ban “Dangerous AI” without banning the calculator. Every competitive step is just “one more grain of sand.” Companies will push right up to the invisible edge of safety, and because the edge is undefined, they will inevitably cross it.

Why The Other Models Are Wrong

You will hear pundits and optimists try to frame this race using other game theory models. They are wrong because they are bargaining with the math.

The “Stag Hunt” Fallacy

• The Theory: “If we all coordinate, we catch the Stag (a shared big win). If we defect, we only get a Hare (small win). So we should trust each other!”
• The Reality: Stag Hunt assumes there is a winning state for cooperation. In the AI Dollar Auction, there is no Stag. If you coordinate to pause, you don't win; you get eaten by a third party (open source or a rogue state). Furthermore, the payoffs are binary (Survive/Die), not incremental. There is no shared win.

The “Cold War” (MAD) Fallacy

• The Theory: “It's like nukes. We will build them until we reach Mutually Assured Destruction, then we will freeze in a stable stalemate.”
• The Reality: Nuclear weapons are static; they sit in silos. AI is fluid; it integrates into the economy to produce GDP. You cannot “freeze” an arms race when the weapon is also your engine of economic growth. If you freeze, you go broke. There is no stable stalemate in a Dollar Auction.

The “Demon” Fallacy (Roko's Basilisk)

• The Theory: “We must build AI because a future super-intelligence might torture us if we don't.”
• The Reality: This is sci-fi theology, not economics. The threat isn't a future digital demon; the threat is Unit Cost Dominance. The “monster” chasing Larry Page isn't a Basilisk; it's a quarterly earnings report showing his competitors can produce intelligence for $0.00 while he pays $30.00.

The “Moloch” Half-Truth

• The Theory: “We are sacrificing our values to the god of competition (Moloch).”
• The Reality: This is the closest model, as it correctly identifies the trap. However, without the Dollar Auction nuance, it fails to explain the speed. Moloch implies a drift into bad outcomes; the Dollar Auction explains the panicked escalation—the reason why burning cash at the risk of bankruptcy is the only rational move.

Conclusion

The AI race isn't a bubble; it's a suicide pact.

• The Prisoner's Dilemma forces us to run.
• The Dollar Auction forces us to spend everything we have.
• The Sorites Paradox ensures we cannot define a finish line.

Larry Page is right. In this game, bankruptcy isn't a risk to be managed. It is the cost of entry. And the only thing worse than burning all your cash is letting someone else light the match.

Ben

Ben Luong is a technical marketing consultant who operates where AI falls short. In a world flooded with cheap, mediocre code and automated strategies, he provides the expert integration, verification, and strategic accountability required to make modern marketing stacks profitable. He specialises in architecting Google Ads, SEO, and GA4 into a single, high-performance system that is accountable to the bottom line.