HOW INTELLIGENCE WORKS · COMMITMENT REVISION · eduKateSG
How Intelligence Knows When to Abandon a Chosen Course Without Letting Sunk Effort Hold It Hostage
Commitment revision is the intelligence process that reopens an already chosen action when new evidence changes its expected value, separating the future merits of continuing from the past costs that cannot be recovered.
Committed course → world return → deviation → remaining value → switching cost → sunk cost removed from comparison → continue, modify, pause or exit.
This article belongs to the How Intelligence Works series. Belief Revision changes what the mind thinks is true. Cognitive Flexibility switches frames or routes. Commitment revision owns the action question: after we have already chosen and invested, when should we stop treating persistence as a virtue?
Past Effort Can Distort Future Choice
Once time, money, status or identity has been invested, abandoning the route can feel like admitting that the investment was wasted. That feeling can keep a failing plan alive even when continuing creates larger future losses.
Commitment revision separates history from the decision that remains.
The rational question is not “How much have we already spent?” but “From this point forward, which available course creates the best expected return?”
1. Separate Sunk Cost From Remaining Cost
Money, time and effort already spent cannot be recovered by continuing. Only future costs and future benefits belong in the next decision.
The past matters for learning and accountability, but not as a reason to add more loss merely to justify what came before.
2. Revision Needs a Trigger
| Trigger | What it means |
|---|---|
| Forecast failure | Expected milestones are repeatedly missed |
| Changed environment | The original assumptions no longer hold |
| New alternative | A better route becomes available |
| Escalating cost | Remaining investment rises beyond expected benefit |
| Boundary breach | Safety, ethics or authority limits are crossed |
| World return | Observed consequences contradict the original plan |
3. Commitment Revision Is Not Indecision
A system that constantly reopens every choice cannot execute. Strong commitment revision therefore preserves stable action until meaningful evidence reaches a predefined review threshold.
The goal is commitment with a correction channel, not permanent hesitation.
4. Precommitment Makes Later Revision Easier
Before investing deeply, specify what evidence would cause pause, review or exit. This reduces the temptation to move the goalposts after identity and resources become attached.
The best time to define an exit condition is before you need it.
5. Commitment Revision in Mathematics
A solver may spend several minutes on one algebraic route and discover that the representation is becoming more complicated rather than simpler. Prior effort does not make the route better.
Strong problem solving can abandon the branch, preserve what was learned and restart from a cleaner representation.
6. Commitment Revision in Learning
A student may continue a study method because many hours have already been spent on it. If the method produces familiarity without retrieval, more of the same may only deepen the illusion.
Revision means using evidence of learning—not invested hours—to decide whether the method should continue.
The failed method still has value if it teaches what the learner now needs instead.
7. Commitment Revision in Projects and Operations
Projects accumulate budgets, reputations and teams. That makes stopping socially harder as evidence deteriorates.
Independent review, staged funding and milestone gates create formal opportunities to compare the remaining value of continuing against alternatives.
8. Revision Can Mean Modify, Not Only Quit
The decision is not binary. A course can continue unchanged, be narrowed, staged, paused, redesigned or terminated.
Commitment revision therefore searches for the best remaining route rather than equating correction with abandonment.
9. Commitment-Revision Failure Atlas
| Failure | What happens | Repair |
|---|---|---|
| Sunk-cost capture | Past investment is used to justify future loss | Compare only remaining futures |
| Identity lock | Changing course feels like personal failure | Separate person from plan |
| Moving goalposts | Review criteria change to protect commitment | Predefine triggers |
| Premature abandonment | Normal difficulty is mistaken for plan failure | Use milestone evidence |
| Binary revision | Only persist or quit are considered | Generate modify and stage options |
| Escalation inertia | More resources are added because earlier resources were added | Recalculate marginal return |
| Learning loss | Abandoned route is erased rather than analysed | Preserve failure provenance |
10. Commitment Revision and Belief Revision Are Different
Belief revision updates the model of what is true. Commitment revision updates what the system will continue doing.
A person can accept that the world changed yet still fail to change action because investment, identity or coordination costs hold the old course in place.
11. Teams Need Legitimate Exit Routes
Teams persist too long when stopping threatens status, budget, ownership or face. A legitimate review process allows evidence to trigger change without requiring personal defeat.
A system that cannot exit a bad commitment is not stable. It is trapped.
12. Institutions Need Stage Gates and Sunset Conditions
Stage gates, review dates, sunset clauses and independent audits create scheduled moments where continuation must be justified by current evidence rather than historical investment.
These mechanisms turn revision from an act of rebellion into part of normal governance.
13. Artificial Intelligence and Commitment Revision
AI agents can become path-dependent after selecting a plan, tool sequence or hypothesis. Later evidence may show that the route is failing, yet the agent may continue because earlier steps created local momentum.
Reliable agents should periodically recompute the expected value of continuing from the current state, ignore irrecoverable compute or tool cost except as learning data, and preserve fallback or rollback options.
Persistence is useful only while the route remains justified.
14. The Commitment Revision Audit
- Commitment: What course is currently active?
- Original thesis: Why was it chosen?
- World return: What has happened since?
- Trigger: Has a review condition been met?
- Sunk cost: Which past costs cannot be recovered?
- Remaining benefit: What value can continuation still create?
- Remaining cost: What must still be spent?
- Alternatives: What other routes are now available?
- Switching cost: What does changing course require?
- Decision: Continue, modify, pause, stage or exit?
15. CivDJ Reading: Do Not Keep a Bad Mix Because It Took Hours to Build
In the CivDJ frame, a working mix may have required significant tuning, but the Receiver and World Return remain the final test.
If the room changes or the mix consistently fails the Receiver, the hours already spent are not a reason to preserve the wrong state.
Keep the learning from the old mix. Do not keep the failure merely to honour the effort.
16. Return to the Decision That Still Remains
Commitment revision protects intelligence from becoming hostage to its own history.
Past effort can be honoured through learning, documentation and responsibility. It does not need to be honoured through additional loss.
The mature system can commit strongly, listen to the return and change course when the future no longer justifies the path.
Related canonical owner: How The World Works | Sunk Costs.