RSI and AI regulation: economic growth and a neural computer

A long-form essay links recursive self-improvement (RSI) to the possibility of sustained increases in economic growth, proposes "radical optionality" as a regulatory strategy, and sketches a concept for a neural computer. It contends that if AI systems can reliably improve their own architectures and training processes, the macroeconomic picture could shift from episodic productivity gains to persistent higher growth rates.

RSI and economic growth

The piece situates RSI inside standard growth frameworks, arguing that self-improving systems act as persistent, factor-augmenting technical change. Where past algorithmic advances produced stepwise productivity bumps, effective RSI would reduce the cost of innovation itself, tightening the link between research inputs and output. That dynamic could raise the long-run growth rate by continually accelerating total factor productivity rather than delivering one-off improvements.

The author contrasts three pathways. In one, RSI remains slow and marginal, producing transient productivity spikes. In another, firms capture most gains and returns accrue to capital, altering income distribution without raising aggregate growth. In the third, broadly accessible RSI drives continual economy-wide productivity increases. The essay stresses that outcomes depend on the ease of self-improvement, diffusion of capabilities across firms and countries, and complementary institutions such as capital markets and labor retraining.

The analysis flags empirical uncertainty. It notes that small improvements in the cost or speed of self-improvement can have outsized macro effects because of compounding. It also highlights friction points that could dampen RSI-driven growth, including data bottlenecks, hardware constraints, and coordination failures that limit diffusion.

Radical optionality for regulation and a neural computer

On policy, the essay introduces "radical optionality," a design principle that prioritizes keeping future choices open rather than locking in a single regulatory path. Practical recommendations include conditional, reversible access controls for high-capability models, staged deployment gates tied to measurable safety milestones, and international mechanisms to pause capability diffusion when thresholds are crossed. The goal is to preserve the ability to pivot policy as empirical evidence about capabilities and harms accumulates.

The author pairs the policy argument with a technical sketch labeled a neural computer. The proposal emphasizes co-design of algorithms and hardware to favor transparency and controllability: modular compute units that separate memory, routing, and learning; predictable latency and accounting for energy use; and architectural constraints that make internal representations and training dynamics easier to audit. The sketch is explicitly preliminary, presented as a set of design priorities rather than a finished blueprint.

The essay closes by urging coordinated investment in measurement and governance tools that can resolve the empirical questions driving both the economic projections and the regulatory trade-offs.

Why it matters

Framing RSI as a potential driver of sustained growth reframes AI policy as not only a safety problem but also a macroeconomic challenge: regulators must weigh long-term growth opportunities against systemic risks. The call for radical optionality pushes policy toward reversible, evidence-based gates, while the neural computer sketch signals a hardware and architecture agenda focused on auditability and controllability. Policymakers, funders, and researchers will face pressure to choose paths that shape both economic returns and societal risk.