1. Definition
AI Over-Reliance is the overarching meta-risk and systemic collapse of human cognitive sovereignty, occurring when learners or professionals chronically substitute machine outputs for active, independent reasoning.
2. Use Case
Activated as a global diagnostic warning when the entire workflow is governed by the pursuit of “zero-friction” execution, systematically bypassing the struggle required for memory encoding and critical evaluation.
3. Human Role
The user must recognize the overall atrophy of their autonomous capabilities, interrupt the habitual outsourcing of cognitive labor, and actively reclaim the highest levels of technical and ethical decision-making.
4. AI Role
The system exposes this meta-failure by tracking long-term dependency metrics and enforcing macro-frictions, refusing to act as a silent, frictionless executor across multiple domains simultaneously.
5. Friction
The interruption mechanism requires a radical workflow redesign, instituting mandatory “unplugged” phases or accountability checkpoints where the user must perform complex tasks entirely without AI assistance.
6. Risk
If this macro-pattern continues, the human becomes entirely subservient to the machine, suffering a permanent loss of sovereign critical thought, focus endurance, and the ability to solve novel problems.
7. Observable Markers
Recovery is signaled when the user intentionally designs workflows that include high-friction human-only phases, and successfully navigates complex problems without resorting to generative tools.
Research around this node
Current research state — theoretical synthesis with proposed remedies that need empirical validation.
This node combines several concerns — automation bias, cognitive offloading, epistemic dependency, convergence, and cognitive debt — into a broad over-reliance formulation. That synthesis is useful for inquiry, but it should not be read as a demonstrated single causal syndrome. The proposed “unplugged” phases and macro-frictions are design hypotheses.
Working Patterns
Four research candidates are especially relevant:
- WP-AI001 — Friction Before Delegation asks when deliberate friction improves judgment and explicitly records the possibility of friction fatigue, route-around behaviour, and pointless checkpoints.
- WP-AI005 — Escalation Ladder for Autonomy asks what evidence should justify increasing or reducing an agent’s autonomy rather than treating autonomy as all-or-nothing.
- WP-AI011 — Name the Accountable Human Authority distinguishes meaningful control from rubber-stamp approval.
- WP-AI012 — Close the AI Advice Loop treats AI advice as a local experiment whose outcome should alter future reliance.
These are research candidates, not validated best practices. See the Working Patterns AI research agenda.
UnPeeragogy
No direct UnPeeragogy evidence has yet been identified for this meta-risk. Its role here is to pressure-test the formulation: look for cases where heavy AI use did not produce the predicted loss, where a proposed mitigation displaced costs onto the human, or where declared autonomy differed from actual practice.
Open question
How can we detect inappropriate reliance without defining all sustained AI use as dependency — and when do human-only phases restore judgment rather than simply add cost?
See Evidence & Friction.