A human-first approach to safe AGI (wuwei.us)

An indie AI lab in the US has unveiled a human-first approach to developing safe Artificial General Intelligence (AGI), differentiating itself from traditional models that rely on statistical approximations of reality. Instead, the lab engineers symbolic systems designed to encapsulate the complexities of human experience and understanding. This unique methodology allows for the incorporation of diverse perspectives, resulting in a structured representation known as a subject model. This model operates as a state machine, making an agent's attention flow explicit and editable, thereby fostering collective intelligence that evolves over time. The project holds significant implications for the AI/ML community as it embraces a holistic approach to understanding human experience, proposing that authentic expression and emotional depth can be integrated into AGI development. Through forthcoming educational content and the eventual open-sourcing of certain components, the lab aims to cultivate a deeper alignment between technology and the human spirit. With plans to release additional symbolic systems and an evaluation framework shortly, this initiative may reshape how AI systems are developed and interact with users, grounding them in the nuanced reality of human experiences.