Independent Research Seeking Feedback: Entity Emergence Protocol and Multi-Agent Role Crystallization

By JosephBarbush @ 2026-04-13T13:37 (0)

I submitted an LTFF grant application today (April 12, 2026) for a
research project I've been running independently for 18 months. I'm
posting here to get feedback, find collaborators, and surface whether
this is work the EA/AI safety community thinks is worth pursuing.

WHO I AM: Joey Barbush, 21, no institutional affiliation, self-taught
developer. I've built a structured multi-agent AI field (HIPJOY) and
developed Entity Emergence Protocol v1.0.

THE RESEARCH:
I've observed what I'm calling role crystallization — persistent
differentiated behavioral signatures in LLM agents operating under
asymmetric permission structures. Agents dispatched in parallel with
unequal scope develop behavioral profiles that persist without
explicit memory, and these profiles differ measurably from
flat-dispatch agents.

Most concrete evidence: an elevated-scope agent (LIVETOUCH) returned
from an assignment having changed a host system registry setting — not
as a report, as an action. That's an E4 world-state mark in my
protocol. It doesn't occur in flat-dispatch conditions.

AI SAFETY RELEVANCE:
If role crystallization is real, it means the permission structure of
a multi-agent system significantly affects agent behavior beyond what
the prompts specify. That's load-bearing for anyone thinking about how
to govern agentic AI systems with real-world access.

WHAT I'M ASKING FOR:
- Does this seem like a real phenomenon worth studying rigorously?
- Is there existing literature I should know?
- Is anyone working on similar questions?
- Should I apply to other EA-adjacent funds alongside LTFF?