# Collapsing Self-Reference: The Mathematical Equivalence of Löbian and Payorian FairBots > Formal verification in proof-based game theory simplifies the design of cooperative AI agents by eliminating redundant self-referential conditions. **Published:** July 18, 2026 **Author:** PSEEDR Editorial **Category:** devtools **Content tier:** free **Accessible for free:** true **Editorial format:** analysis **News quality eligible:** true **Source count:** 1 **Word count:** 1022 **Tags:** AI Alignment, Game Theory, Formal Verification, Provability Logic, Program Equilibrium **Canonical URL:** https://pseedr.com/devtools/collapsing-self-reference-the-mathematical-equivalence-of-lbian-and-payorian-fai --- In a recent analysis published on [lessw-blog](https://www.lesswrong.com/posts/2JQzDZXjoG2opnAjk/my-payorian-fairbot-was-just-the-original-fairbot), the author demonstrates that a proposed "Payorian FairBot" is mathematically equivalent to the original "Löbian FairBot" introduced in the Machine Intelligence Research Institute's (MIRI) proof-based prisoner's dilemma tournament. For PSEEDR, this equivalence signals a critical simplification in the field of program equilibrium, illustrating how formal verification and provability logic can streamline the design of provably cooperative, non-exploitable multi-agent AI systems. ## The Mechanics of Proof-Based Cooperation The foundation of this equivalence lies in the architecture of MIRI's proof-based prisoner's dilemma tournament. Unlike traditional game theory, where agents rely on behavioral observation, repeated interactions, or heuristic trust mechanisms like tit-for-tat, the MIRI framework operates on formal verification. Agents are encoded as formulas of Peano arithmetic (PA) with one free variable. Through Gödel numbering, an agent can effectively read and analyze the source code (the mathematical formula) of its opponent before deciding whether to cooperate or defect. In this environment, the baseline for cooperative behavior is the Löbian FairBot. The Löbian fairness condition dictates that an agent will cooperate if and only if there exists a formal mathematical proof within Peano arithmetic that its opponent will cooperate. This creates a robust, non-exploitable baseline: the agent only extends trust when cooperation is a mathematically guaranteed outcome. The alternative formulation, dubbed the Payorian FairBot, introduces a layer of self-reference. Its cooperation condition explicitly references its own cooperation, essentially stating that it will cooperate if there is a proof that its own cooperation guarantees the opponent's cooperation. Initially, this self-referential loop appears to define a distinct class of agent behavior, potentially altering the dynamics of mutual cooperation. ## Eliminating Self-Reference via Theorem 4.6 The core revelation of the source text is that the Payorian fairness condition is not a distinct behavioral paradigm but rather a mathematically equivalent restatement of Löbian fairness. The author outlines two pathways to arrive at this conclusion: an elementary proof and a sophisticated proof. The elementary proof relies on the mechanical application of the rules of inference within provability logic. Because provability logic is decidable, the equivalence of these two agent formulations is a finite, computable problem. One can systematically grind through the logic in both directions to prove that any agent satisfying the Payorian condition inherently satisfies the Löbian condition, and vice versa. The sophisticated proof leverages Theorem 4.6 from the original MIRI paper. This theorem addresses a fundamental challenge in computer science and logic: the complications introduced by self-reference. In many computational contexts, self-reference leads to paradoxes, infinite loops, or undecidable propositions (akin to the halting problem). However, Theorem 4.6 demonstrates that when an agent's cooperation condition references its own cooperation in this specific proof-based manner, the self-reference can be mathematically eliminated. The complex, recursive requirement of the Payorian FairBot collapses into the direct, non-referential proof requirement of the Löbian FairBot. This mathematical flattening confirms that the two agents will behave identically in all possible tournament matchups. ## Implications for Program Equilibrium and AI Alignment For the broader ecosystem of multi-agent artificial intelligence, this mathematical equivalence carries significant implications for the study of program equilibrium and open-source game theory. Program equilibrium, a concept where agents can condition their strategies on the source code of their peers, is a critical frontier in AI alignment. As autonomous systems become more integrated, the ability for these systems to verify the intentions and policies of other agents is paramount to preventing exploitation and ensuring cooperative outcomes. By proving that complex, self-referential fairness conditions collapse into simpler Löbian frameworks, researchers can effectively reduce the theoretical search space for cooperative agents. Instead of cataloging and testing an infinite variety of recursive trust conditions, the field can focus on a narrower, more robust set of axioms. This streamlines the design of provably cooperative AI systems. In environments where agents might attempt to obfuscate their policies or introduce complex conditional logic to exploit naive cooperators, the mathematical guarantee that certain self-referential conditions can be eliminated simplifies the verification process. It ensures that cooperation remains tethered to verifiable proofs rather than easily manipulated behavioral heuristics, establishing a more secure foundation for multi-agent reinforcement learning (MARL) architectures. ## Limitations and Theoretical Gaps Despite the elegance of this mathematical equivalence, several limitations and theoretical gaps remain. The source text omits the explicit mathematical formulas for both the Löbian and Payorian fairness conditions, as well as the specific rules of inference used to construct the paper-based elementary proof. This omission limits the ability to independently audit the mechanical steps of the proof without reconstructing the underlying Peano arithmetic from the original MIRI paper. More critically, a massive translation gap exists between theoretical proof-based game theory and practical AI deployment. Modern AI systems, particularly large language models and deep reinforcement learning agents, operate as continuous, high-dimensional neural networks, not as discrete formulas of Peano arithmetic. Verifying the policy of a billion-parameter neural network is fundamentally different from verifying a PA formula. While provability logic is decidable in theory, the computational complexity and overhead of running formal verification checks in real-time, multi-agent environments remain prohibitive. The practical mechanisms for allowing modern AI agents to "read" and formally verify each other's weights in a way that maps to Gödel numbering are currently unsolved, leaving this equivalence as a purely theoretical construct for the time being. Ultimately, the collapse of Payorian fairness into Löbian fairness serves as a vital micro-example of a macro-goal in AI alignment: reducing complex, unpredictable agent behaviors into verifiable, predictable mathematical structures. While the bridge to practical neural network verification remains unbuilt, establishing these theoretical bounds ensures that when formal verification tools do mature, the foundational logic governing cooperative AI will be streamlined, robust, and mathematically sound. ### Key Takeaways * The proposed Payorian FairBot, which relies on a self-referential cooperation condition, is mathematically equivalent to the original Löbian FairBot. * This equivalence can be proven mechanically via the decidable rules of provability logic or through MIRI's Theorem 4.6, which eliminates specific self-referential references. * Simplifying fairness conditions reduces the theoretical search space for program equilibrium, aiding the design of non-exploitable cooperative AI. * Significant gaps remain in translating discrete Peano arithmetic proofs into verification frameworks for continuous, high-dimensional neural networks. --- ## Sources - https://www.lesswrong.com/posts/2JQzDZXjoG2opnAjk/my-payorian-fairbot-was-just-the-original-fairbot