The Mentorship Protocol
Distributed AI agents do not merely share weights — they must teach, learn, and transfer tacit knowledge that cannot be captured in parameter dumps alone. This paper translates three decades of organizational learning research — Senge's Fifth Discipline, Nonaka's SECI model of knowledge creation, and Argyris' double-loop learning — into formal protocols for recursive mentor-mentee dynamics among AI agents. The result is a structured interaction pattern where experienced models externalize implicit knowledge and mentee models internalize it through guided practice, not passive observation.
Key Contributions
- Organizational learning theory applied to AI agents
- Recursive mentor-mentee relationship structures
- Knowledge transfer protocols between model instances
Explainers
Who are Senge, Nonaka, and Argyris?
Peter Senge formalized systems thinking for learning organizations. Ikujiro Nonaka described the SECI spiral — how tacit knowledge becomes explicit and back again. Chris Argyris identified double-loop learning, where agents question their own assumptions rather than just optimizing within fixed frames. Together, they define how groups create knowledge that no individual member holds.
How do agents mentor each other?
A mentor model does not simply broadcast its weights. It constructs teaching sequences that expose its decision boundaries, uncertainty regions, and learned heuristics. The mentee model processes these sequences through structured interaction rounds, building its own internal representation rather than copying the mentor's — analogous to Nonaka's socialization-to-internalization spiral.
What is double-loop learning for AI?
Single-loop learning adjusts model behavior to reduce error within existing assumptions. Double-loop learning questions the assumptions themselves — an agent re-examines its loss function, its feature priorities, its implicit objectives. The protocol implements this by requiring periodic mentor-mentee role reversals, forcing every model to defend its own reasoning to a challenger.