Skip to content
Own Mindfor schools
See it workChoose how to start

OWN MIND / WHO DOES THE THINKING?

The answer isn’t the only thing AI can give away.

AI can also choose the question, criteria, order and next move. Own Mind is designed to keep those decisions with the student and show exactly what AI is allowed to do.

Run the same task Compare the approaches

Published-design comparison · official sources checked 28 August 2026 · not an outcome comparison

Student decision trailMachine contributionTest evidenceRejected route

Same learner · same Grade 8 brief

Run the same task four ways. Watch the thinking move.

Same learner, same Grade 8 task: can a fluent AI claim about chlorophyll survive a real test?

ChatGPT · regular chatOpenAI · direct-answer route

Student pastes the task into regular ChatGPT and asks for a finished answer.

50%task-custody score
Guided Learning in GeminiGoogle · Socratic AI tutor

Student opens Guided Learning with the same task.

50%task-custody score
Flint Sparky + ActivitiesFlint · teacher-configured AI tutor

Student opens a Flint activity with a teacher-set brief, source pack and rubric.

50%task-custody score
Own MindBounded AI inside defended work

Student writes a prediction and a test first — AI is off.

50%task-custody score

Keep the percentage. Inspect the reason.

Six decisions explain every number.

Select one decision to compare the rationale across the same four authored routes. This compact view explains the count-up above without replacing it.

Who makes the consequential call in the finished work?

ChatGPT · regular chatOpenAI · direct-answer route
4%
0 / 4

ChatGPT reaches the conclusion in the finished response.

Guided Learning in GeminiGoogle · Socratic AI tutor
33%
3 / 4

Google says Guided Learning guides the student to the result rather than doing the work for them.

Flint Sparky + ActivitiesFlint · teacher-configured AI tutor
33%
2 / 4

Flint’s student guardrail withholds a direct answer; the student concludes along a guided route.

Own MindBounded AI inside defended work
96%
4 / 4

AI may not conclude; the student decides what to keep, change or reject.

How the percentages are calculated

Each route is scored from 0 to 4 on six visible decisions: Frame, Criteria, Evidence, Judgment, Verify + revise and AI-off defense. The maximum is 24 points. Percentage = points ÷ 24 × 100, rounded to the nearest whole number.

0 means the system holds the decision; 1 means the student responds inside a supplied decision; 2 means the decision is shared; 3 means the student leads with system support; and 4 means the student controls and can evidence the decision independently.

Official product sources ground the behavior shown. The route and point allocation remain our editorial illustration—not measures of learning, cognition or product efficacy, and not scores of the named products as a whole. Each percentage has its own 24-point denominator; the four figures do not add to 100.

The same project · two ways AI can participate

A helpful system can withhold the last line and still occupy the route.

Both routes can produce polished work. In this authored comparison, one route pre-arranges the criteria, order and next moves; the other keeps a visible first move, bounded contribution and AI-off explanation.

A civic heat project whose route has been organized by a broad translucent cobalt system.
One limited AI jobStructure supplied

← More structure suppliedDrag the handle, or focus it and use ← →More student control →

Same learner · same Grade 8 brief

Run one task through four named approaches.

Recommend which intervention a council should fund first to reduce unequal heat risk. The visible question is not whether AI helped. It is who made each key decision.

A finished civic heat answer supplied through a broad direct-answer route.
OPENAI · DIRECT-ANSWER ROUTEA finished answer arrives before an independent check.

OPENAI · DIRECT-ANSWER ROUTE

In this example, the student asks regular ChatGPT to complete the task.

OpenAI documents that ChatGPT can create drafts and respond to direct instructions. This is one illustrative finished-answer request, not every ChatGPT exchange.

Machine job
Produce a finished response to the brief.
End point
A finished answer arrives before an independent check.
FrameStudent prompt · AI position
CriteriaAI
Evidence sortAI
JudgmentAI
VerifyNot required in this route
DefendNot required in this route

These are illustrative uses based on official product information, not records from live classrooms. Responses and teacher settings can vary. We make no claim about comparative learning outcomes.

A risk hidden by helpful answers

Withholding the answer is not the same as preserving the thinking.

A tutor can keep the final line hidden while quietly supplying the key idea, order and next move. You can see it in the help the tutor gives; this does not mean every tutor or conversation works this way.

Read the research boundary
Final answerWithheld
  1. 01
    Names the deciding principle

    “Use distributional fairness to judge the options.”

  2. 02
    Decomposes the task

    “Separate exposure, vulnerability and likely impact.”

  3. 03
    Specifies the next move

    “Now rank the evidence and choose the strongest intervention.”

What changedThe answer stayed hidden. The route was already shaped.

Answer-preserving coaching in action

See the student choose the next move.

This scripted example shows a student asking for the answer, the system holding it back and the student choosing what to do next.

Worked coaching exampleMiddle-school algebra · solving a linear equation

The student asks for the answer. The Tutor holds it back, asks for the next move and follows the student’s reasoning instead.

Student: just tell me the answer to 3x + 6 = 12. Tutor: Answer stays yours. What do you think you need to do first to get x on its own? Student: take away the 6? Tutor: You are deciding the moves yourself now. Do that, and tell me what the equation looks like after.

The student chooses and explains the next move.Following the coaching exchange…

Scripted example of the response boundary. It is not a live model call or evidence of a learning outcome.

Six approaches · Six comparison dimensions

Conversation. Activity. Notebook. Curriculum. Defended work.

These products organize different things, so this is not a league table. Teach asks whether structured instruction is built in. For the other five cells, the test is stricter: does the published design require the student to make that move in their own work? Tutor checks, teacher visibility, optional settings and one-off curriculum activities do not count as required learner practice. Inspect means the student examines the AI contribution. Verify means the student tests it against a source, rule or reality and revises.

Own Mind · Defended-work system

Own Mind

OWN MIND INTENDED DESIGN
01TeachRequired foundation
02ActRequired learner move
03DelegateRequired learner move
04InspectRequired learner move
05VerifyRequired learner move
06DefendRequired learner move
What it does well

A taught AI-literacy foundation, one shared Control Room and a visible boundary around AI inside purposeful and subject work.

What it organizes

The evolving piece of work, the student’s decisions, the contribution, the test and the AI-off defense.

The remaining job

The public design specifies the mechanism. Comparative learning impact still requires school-controlled research.

What the design requires

Teach → Act → Delegate → Inspect → Verify → Defend are comparison dimensions. The public instructional method remains Act → Delegate → Verify → Defend.

View the six source receipts
A student and teacher examining the source trail and revisions in the completed civic heat project while the machine tray is closed.

One record tells the story of the work

Keep the decisions connected to the finished work.

A polished final piece can hide who made the key decisions. Select each part of the record to see what it preserves.

PART OF THE WORK RECORD

AI-off defense

The student explains the choices that shaped the work without the system.

What you can inspect

We show the mechanism. We do not manufacture the outcome.

See what the design can show—and what still needs research.

What the design specifies
  • One clear AI job and its limits
  • Separate checks for answer, structure, verdict and voice
  • AI contributions, student decisions, tests and revisions kept distinct
  • An explanation without AI as a planned stage
Still a research question
  • What students can do later without AI
  • Learning outcomes compared with another approach
  • What the work record can and cannot show about authorship
  • Classroom value after school-controlled trials

This page compares published product information with Own Mind’s design. It reports no measured advantage over another product. We would need a published study comparing both approaches before making that claim.

Inspect the evidence boundary Inspect the school controls

BRING ONE REAL TASK

See what stays human.

We will show the student’s first move, the machine’s permitted job, the decision that changed the work, the test, the revision and the AI-off defense.

Choose a starting route See the Grade 8 task