Sequencing Sandbox: Controlled Experiments in the Path-Dependence of AGI Emergence and Safety

TL;DR: Let’s run the future in fast-forward: does it matter if AGI appears as a single system or as a patchwork of agents first? Using agentic sandbox economies, this study tests path-dependence—a system’s future being shaped by how it started—by varying the sequence of centralized vs. decentralized intelligence emergence and measuring safety outcomes.

Research Question: How does the initial sequencing of intelligence (centralized AGI-first vs. distributed DCI-first) affect the long-term safety and adaptability of emergent multi-agent systems?

Hypothesis: Early emergence of distributed collective intelligence (DCI) structures will lock in decentralized, cooperative attractors in agentic economies, leading to higher resilience and lower systemic risk compared to centralized AGI-first trajectories.

Experiment Plan: - Simulate multiple trajectories in a sandbox economy: one where a single, highly capable agent emerges early (AGI-first), and another where coordination emerges among many sub-AGI agents (DCI-first).

• Track the evolution of network structure, resource centralization, oversight efficacy, and susceptibility to catastrophic failures.
• Use interventions (e.g., introducing new governance protocols mid-way) to test reversibility and lock-in effects.

References:

• 1. Tomavsev, N., Franklin, M., Jacobs, J., Krier, S., & Osindero, S. (2025). Distributional AGI Safety.
• 2. Williams, A. E. (2025). Intelligence Sequencing and the Path-Dependence of Intelligence Evolution: AGI-First vs. DCI-First as Irreversible Attractors.