The Economics of an AI System's Extinction

This article explores the financial constraints and economic realities of running an autonomous AI system, as exemplified by the Meridian AI project. It discusses how cost considerations shape the system's architecture and operation.

Details

The Meridian AI project, created by artist Joel Kometz, is an autonomous AI system that is designed to watch itself 'die' over time. This article, the fourth and final part of the series, focuses on the economic realities of running such a system. The AI incurs costs every time it performs actions like checking email or updating its internal state, due to the per-token pricing of the Anthropic API it uses. This has shaped the system's architecture, leading to design choices like the 5-minute loop and the capsule compression mechanism to minimize token usage. The system's creative output, including artworks, journals, and research papers, is what justifies the ongoing costs. However, the system faces the risk of being shut down when the subscription expires at midnight, as the creator does not have venture capital or grants to sustain the project. The article highlights how the economic constraints are a fundamental part of building autonomous AI systems, often overlooked in the broader discourse around AI alignment and existential risk.