From Detecting Agency to Doing Work: Self-Caused Credit Builds a Durable Behavioral Self in a Minimal Spiking Agent
arXiv:2606.30191v1 Announce Type: cross Abstract: How does an agent that can tell self from world come to be durably shaped by that distinction? Recent work shows that a predictive system can detect its own agency (Ye, 2026), but detecting agency does not explain durable, self-shaped behavior. We show that agency-gated slow credit -- a conjunctive term Own*Agency*Salience driving a slow parameter update -- produces post-unload behavioral residue: on a spiking substrate (Nengo LIF/PES), a learned self-preserving choice survives episodic buffer removal (retained fraction 0.96, N=50) and collapse
This research builds on recent work demonstrating agency detection in predictive systems, extending a foundational understanding of how AI agents can develop a 'self' and durable behavioral patterns.
A strategic reader should care because this research explores the fundamental mechanisms for agent self-preservation and sustained autonomous behavior, which is critical for future sophisticated AI applications and interaction.
This research suggests a pathway for AI agents to develop persistent, self-preserving behaviors even after initial learning episodes, moving beyond mere agency detection to durably shaped action.
- · AI researchers
- · AI ethics and safety organizations
- · Developers of autonomous systems
This research contributes to the theoretical underpinnings for more robust and long-term autonomous AI agents.
Improved understanding of 'self' in AI could lead to agents capable of more complex decision-making and self-maintenance, potentially accelerating the development of general AI.
The emergence of AI agents with durable 'self-preservation' instincts could necessitate new regulatory frameworks and ethical considerations for their deployment and interaction with humans.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG