Vol. 2 No. 3 (2025): Governance, Agents, Evolutionary Search

Senior Editor-at-Large Gil Syswerda has constructed a provocative Timeline to Artificial General Intelligence 2025 – 2030+. He gives key AI advances, economic, social, and geopolitical effects of AI, predicting that, as early as 2028-2029, AI could replace the majority of human economic activity.

Agentic AI is advancing at a rapid pace. Two articles look at self-improving Agentic AI that use evolutionary programming techniques. The Darwin Godel Machine, Appendix F, describes a simple proof-of-concept that self-improving AI can focus on its own safety and value alignment. In a third article, Kumarage et al. describe a multi-agent architecture in which agents collaborate in chain-of-thought reasoning about LLM responses vs. policy before output is permitted, representing another paradigm of recursive safety improvement.

Governance is also changing rapidly. It is critical to get a handle on the global AI legal regime with its varied approaches as they evolve, and flag gaps in regimes.

Safety and alignment methods must stay ahead of AI advances, and the latest advances, such as reasoning, must be applied to safety and alignment - recursive safety improvement.