Learning Question
What final mental model should guide the user’s use of AI as a thinking partner?
The Final Model
AI should be understood as a mirror, critic, scaffold, and tool.
As a mirror, AI reflects the user’s current thinking back in a more visible form.
As a critic, AI tests assumptions, boundaries, claims, plans, and decisions.
As a scaffold, AI helps the user move through a difficult learning or reasoning process without replacing the process.
As a tool, AI performs useful operations such as rewriting, comparing, summarizing, searching, editing, testing, or acting in a workspace when given the right context and permissions.
None of these roles makes AI the owner of judgment. The user remains responsible for goals, standards, verification, and final decisions.
The Complete Thinking Loop
The durable loop is:
current model
-> correction
-> layer separation
-> boundary testing
-> opposition
-> judgment criteria
-> practice or verification
-> durable knowledge captureThe loop does not have to be used in full every time. It is a mental model for choosing the next useful move.
If the user has no model, build a first model.
If the model is vague, ask for correction.
If the model collapses levels, separate layers.
If the model feels too smooth, test boundaries.
If the user is about to choose, use opposition.
If the explanation is valuable, extract judgment criteria.
If the topic matters, verify and practice.
If the insight is durable, preserve it.
Role 1: AI as Mirror
A mirror makes the user’s thinking visible.
This role is useful when the user has a rough idea but cannot see its structure.
Good mirror prompts include:
Here is my current understanding.
Reflect it back as a structured model.
Identify the main claim, assumptions, vague terms, and missing boundaries.The mirror role should not flatter the user. It should make the model inspectable.
Role 2: AI as Critic
A critic tests the model.
This role is useful when the user needs confidence that a claim, plan, design, or decision can survive scrutiny.
Good critic prompts include:
Challenge this from the strongest reasonable opposing view.
Rank objections by practical importance.
Separate blockers from manageable risks.The critic role should not become performative negativity. It should improve the decision or clarify why the decision is weak.
Role 3: AI as Scaffold
A scaffold supports work that the user still performs.
This role is useful during learning, writing, and problem solving. AI can break a task into steps, provide hints, ask questions, and explain missing prerequisites.
The scaffold should be removed gradually. If AI always completes the hard part, the user may not build the skill.
Good scaffold prompts include:
Do not give the full answer yet.
Give the next hint and ask me to continue.
Then correct my attempt.Role 4: AI as Tool
A tool performs a specific operation.
This role is useful when the task is concrete: summarize this source, compare these options, rewrite this paragraph, inspect this file, run this test, or update this document.
The tool role needs clear inputs, constraints, and verification.
Good tool prompts include:
Make this change.
Keep the scope limited to these files.
Verify these conditions afterward.
Report what changed and what was not verified.The more the AI can affect real state, the more explicit the user should be about permissions and success criteria.
The Division of Responsibility
The user owns:
- the goal
- the context that matters
- the standard of correctness
- the willingness to verify
- the final judgment
- the decision to preserve or discard output
AI helps with:
- reflection
- correction
- generation
- comparison
- opposition
- structure
- transformation
- tool-mediated work
Confusion appears when these responsibilities are swapped.
The Practical Rule
Use AI differently depending on the desired outcome.
If the desired outcome is information, ask for an answer and verify as needed.
If the desired outcome is understanding, expose the current model and ask for correction.
If the desired outcome is judgment, define criteria and ask for opposition.
If the desired outcome is learning, combine sources, practice, feedback, and verification.
If the desired outcome is durable knowledge, extract the durable insight rather than storing the conversation.
If the desired outcome is action, provide scope, constraints, permissions, and verification steps.
What This Collection Does Not Resolve
This collection does not make AI reliable by itself.
It does not remove the need for expertise, sources, tests, reviewers, or practice.
It does not promise that every AI interaction should be deep.
It provides a way to decide how AI should participate in the user’s thinking.
Core Mental Model
AI is not a replacement mind. It is an external thinking instrument.
Used poorly, it produces fluent shortcuts that can weaken judgment. Used well, it makes the user’s current model visible, corrects it, tests it, supports practice, and helps preserve durable insight.
Final Summary
The strongest use of AI is not getting to an answer faster. It is becoming harder to confuse later.