Divide-and-Conquer Dynamics in AI-Driven Disempowerment

Peter S. Park, Max TegmarkarXiv 2023 · DOI · arXiv

Abstract

This paper by Peter S. Park and Max Tegmark (MIT) applies game theory and dynamical systems to explain why collective resistance to AI-driven job displacement is failing. Using a "generalized gambler’s ruin" model, they demonstrate that AI companies benefit from "divide-and-conquer" dynamics, and that complete unity is the only mathematically optimal strategy for humans to prevail against a sufficiently powerful AI threat.

Introduction: The Coordination Failure

As artificial intelligence begins to automate the livelihoods of creative professionals, the "pro-human movement" faces a critical coordination failure. While individual groups advocate for specific protections, such as copyright or labor rights, the broader movement remains fragmented. Park and Tegmark address this fundamental issue: why don't "future victims" of AI support "current victims" in a unified front? They argue that AI companies (the "AI replacement movement") utilize influence campaigns to downplay future risks, effectively isolating current victims and ensuring that the opposition is "defeated in detail"—a military term for losing a conflict piece by piece because forces are divided.

Theoretical Framework

The researchers construct a two-part mathematical framework to analyze the conflict between the pro-human movement and the AI replacement movement.

The Battle Model (Dynamical Systems)

The conflict is represented as a generalized gambler’s ruin model on a Markov chain. The state space represents "battlefields" (e.g., legal, political, or social arenas). Movement 1 (Pro-human) and Movement 2 (AI replacement) win or lose ground based on their relative "force strength" ($s$) and their degree of disunity ($m$ and $n$ parts). Proposition 1 proves that when the AI threat ($s$) is large, the probability of human victory is maximized only when the movement is completely united ($m=1$). As disunity increases, the win probability drops toward zero.

The Cooperation Model (Game Theory)

The authors analyze the conditions for a Nash equilibrium where members of the pro-human movement choose whether to pay a personal cost ($c$) to unite or "wait-and-see." The model incorporates psychological and strategic variables:

  • $r$ (Myopia): How much a member discounts future well-being.
  • $b$ (Benefit): The perceived value of winning the conflict.
  • $\hat{s}$ (Perceived Strength): How strong the member thinks the AI threat is compared to its actual strength.

Strategic Vulnerabilities: The Five Psychological Weaknesses

The paper identifies five traits that AI companies can exploit to prevent human unity:

  • Myopia: Valuing present comfort over future survival.
  • Naivety: Underestimating the severity of the AI threat.
  • Collaborationism: Believing one will personally benefit from the AI takeover.
  • Defeatism: Believing AI dominance is inevitable, so resistance is a waste of resources.
  • Complacency: Believing the threat is too far off to worry about now.

Industry leaders often frame the debate as "current harms" (bias, copyright) vs. "future harms" (existential risk). The model suggests this is a strategic divide-and-conquer tactic to prevent a unified front.

Results and Historical Corroboration

The "Wait-and-See" Threshold

The paper derives a key inequality (Equation 19) that determines when a rational actor will defect from a coalition. If the cost of joining the movement is perceived as higher than the discounted future benefit of winning, the movement collapses into disunity. The authors highlight the rapid pace of AI progress—citing the leap in Midjourney’s capabilities between 2022 and 2023—as a warning that "waiting" for even a year can result in a total loss of bargaining power.

Historical Parallels

The model draws on historical-empirical records to corroborate its predictions. Just as Napoleon defeated larger armies by splitting them in his Italian campaign, AI companies can prevail over a much larger human population by ensuring that different sectors (artists vs. lawyers vs. coders) do not realize they are "in the same boat." The authors also cite corporate PR tactics, such as the "Keep America Beautiful" campaign, which shifted blame for pollution from industry to individuals, as a precedent for modern AI influence campaigns.

Conclusion and Future Directions

While the model assumes simplified rationality and binary movements, it provides a stark warning about the necessity of unity. The authors suggest applying this model to specific "battlefields" like the WGA/SAG-AFTRA strikes and political advocacy for AI regulation. They emphasize that the "pro-human movement" must create a sense of urgency to counter the "wait-and-see" tendency before the window for effective resistance closes.

Summary updated

BibTeX
@misc{park2023divide,
  author = {Peter S. Park and Max Tegmark},
  title = {{Divide-and-Conquer Dynamics in AI-Driven Disempowerment}},
  year = {2023},
  eprint = {2310.06009},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2310.06009},
}
Figure