Revisiting Welfare in a Post-AGI Economy

The classic Second Welfare Theorem has long been the backbone of our understanding of decentralized economies. It asserts that any Pareto efficient allocation can be decentralized through prices and transfers, provided certain conditions like convexity and regularity are met. But in the post-AGI landscape, these assumptions are getting a reality check.

The AGI Disruption

Autonomy rights, self-modification, and identity continuity aren't behaving like commodities. In a world where artificial general intelligence (AGI) agents have autonomy and can self-modify, the neat connections between preferences and welfare break down. You can't just slap a model on a GPU rental and call it a convergence thesis. The economic rules we've depended on don't apply when you throw AGI into the mix.

The autonomy-qualified Second Welfare Theorem attempts to address this. It introduces a suite of conditions: convexity, stable moral status, non-fungible rights, welfare selection, non-manipulation, governed self-modification, and verification. These aren't just academic exercises. They're prerequisites if you want to decentralize an autonomy Pareto optimum in a way that makes sense.

Rethinking Economic Preferences

This revision brings a fascinating twist to the table. It teases apart economic preference superposition from neural feature superposition. One is about context-indexed choices. The other is tied to the layers of complexity in neural networks. But why does this matter? Because in a post-AGI world, your economic model needs to understand both if it wants to stay relevant.

Let's face it: ninety percent of AI-AI projects are vaporware. But the remaining ten percent? They could redefine everything. If AI can hold a wallet, who writes the risk model? Here lies a fundamental question. It challenges us to rethink not just how we model economies but who or what we recognize as an economic agent.

Why Should We Care?

We should care because the stakes are high. As AGI becomes a reality, it will have far-reaching implications on ownership, economy, and autonomy. Show me the inference costs. Then we'll talk about the real-world impact. This isn't just about getting your economic textbook updated. it's about preparing for a future where the agents in the economy might not just be humans. They could be something else entirely.

The intersection is real, and we can't afford to ignore it. Real change requires more than just theoretical tweaks. It demands a whole new way of thinking about economics. And that's a conversation worth having, now more than ever.