Skip to content

Games Between Programs — The Ruliology of Competition

Source: Stephen Wolfram Writings Date Published: June 4, 2026

The Ruliological Method

Stephen Wolfram applies his signature ruliological approach — the systematic enumeration of all possible strategies rather than cherry-picking human-intuitive ones — to the study of competition between agents. The core question: what happens when you let every possible program compete against every other, without preconceptions about which strategies are "good"?

Key Finding: Computational Irreducibility Dominates

The central result: computational irreducibility is the dominant force in program competition. You generally cannot predict the winner without running the simulation — there is no shortcut, no analytic formula that distills competitive outcomes from strategy descriptions.

Finite-State Machine Strategies

When exploring the space of FSM-based strategies:

  • Larger machines systematically outperform smaller ones by developing customized substrategies for specific opponents
  • The advantage of size is not merely computational power but behavioral specialization

Prisoner's Dilemma Surprises

In the classic Prisoner's Dilemma:

  • Grim Trigger (Machine 30) wins overall — the unforgiving strategy that punishes defection forever
  • Tit-for-Tat — celebrated in Axelrod's tournaments as the canonical cooperative strategy — ranks far lower in this exhaustive search
  • This suggests Axelrod's conclusions may have been artifacts of the limited strategy space he considered, not universal truths

Cellular Automaton Strategies

For cellular automaton-based competition:

  • High payoff strongly correlates with simple behavior — a striking contrast to the FSM results
  • Complex behavior emerges unpredictably in competition between simple rules (e.g., Rule 6 vs Rule 7)
  • The dynamics are often chaotic and non-intuitive even for the simplest rule pairs

Implications

The work challenges foundational assumptions in game theory and suggests that many "well-known" results may be specific to the narrow strategy spaces that human intuition finds natural, rather than reflecting the broader computational landscape of competition.