How does superintelligence work when an AI is asked to calculate a final grade from several scores? A pupil earns 90 on quizzes, 70 on a project and 80 on an examination. Simply averaging the three gives 80. But if the examination is worth half the grade, the project thirty percent and quizzes twenty percent, the final result is 79, not 80. The scores were read correctly; the weights changed the answer.
The mechanism is weighted-score reasoning. A weighted average multiplies each component by its assigned share, then adds the contributions. It differs from a simple mean, where every observation is given equal weight. An AI assistant must preserve component labels, confirm that the weights cover the whole intended assessment and avoid confusing a score, a weight and a grade contribution.
This beginner-friendly guide uses two fictional learners, three assessment components, changed-weight scenarios and twenty-four verifiable claims. The scoring system and results are invented solely for teaching and are not official Singapore grading rules, real pupil records or measurements of any AI product. The title “How Superintelligence Works” names an educational series; it does not claim that contemporary AI has attained artificial superintelligence.
The answer at a glance
- Fictional quiz weight: 20%.
- Fictional project weight: 30%.
- Fictional examination weight: 50%.
- Weights total: 20 + 30 + 50 = 100%.
- Learner A scores: 90, 70, 80; weighted final 79.
- Learner B scores: 75, 85, 80; weighted final 80.5.
- Both learners have an unweighted mean of 80.
- Different weighting produces different final scores and a ranking despite the equal simple means.
1. First identify the assessment components
A final grade is not merely a collection of numbers. Each result belongs to a specific assessment: quiz, project or examination. The assistant must connect the appropriate score to the correct weighting rule before calculating anything.
If a spreadsheet column is mislabelled, an otherwise perfect formula can assign a large examination weight to a quiz result. The error is one of interpretation rather than multiplication. Preserving headings and the assessment scheme is therefore essential.
2. What is a simple arithmetic mean?
The ordinary unweighted mean adds all three scores and divides by three. For Learner A, 90 + 70 + 80 = 240; 240 ÷ 3 = 80. Every component contributes one-third to that calculation.
That is mathematically correct as the simple mean. It becomes the wrong final grade if the approved rubric assigns different proportions to quizzes, project and examination. Correct arithmetic cannot replace the rubric.
3. What is a weighted average?
A weighted average multiplies each score by its assigned weight and sums the products. When weights are percentages that add to 100%, divide each percentage by 100 before multiplying.
For our fictional scheme, the formula is Final = 0.20 × Quiz + 0.30 × Project + 0.50 × Examination. The result remains on the same 0–100 scale as the underlying scores if the rubric and input ranges are correctly defined.
4. Why the weights have to be attached to names
A weight of fifty percent means little until we know which component it governs. In the toy scheme the examination is worth fifty percent; assigning that weight to the quiz would change the final result.
An AI should read the whole assessment sentence or table header rather than extracting an unordered set of percentages. The combination of score and weight creates the contribution.
5. The fictional Cedar Assessment Scheme
Imagine Cedar School has an invented three-component practice course. Quizzes contribute twenty percent, a project thirty percent and a final examination fifty percent. Every component is scored on the same 0–100 scale, with no rounding until the final display.
The school and rubric are fictional. Real assessments may have grade bands, moderation, missing-work provisions or different passing rules. None of those is supplied here, and an assistant must not imply that our rubric is an official examination policy.
6. The exact assessment ledger
- Quizzes: 20% of the fictional final score.
- Project: 30% of the fictional final score.
- Examination: 50% of the fictional final score.
- Learner A quiz score: 90.
- Learner A project score: 70.
- Learner A examination score: 80.
- Learner B quiz score: 75.
- Learner B project score: 85.
- Learner B examination score: 80.
- All input scores use a 0–100 assessment scale.
- No additional coursework, bonus marks or penalty rules exist in this toy case.
7. Check the complete weighting
Add the three weighting percentages: 20 + 30 + 50 = 100. Expressed as proportions, 0.20 + 0.30 + 0.50 = 1.
This check matters because a rubric with missing or duplicated weights may produce a result that is not directly interpretable as the intended final grade. An AI should not silently normalise an incomplete scheme without the governing rule.
8. Learner A’s quiz contribution
A scores ninety on quizzes. The quiz contribution to the final is 0.20 × 90 = 18 points on the final-grade scale.
The eighteen is not the pupil’s original quiz mark. It is the weighted contribution of the ninety-point quiz score. Confusing the component score with its contribution can cause a report to display incorrect individual assessment results.
9. Learner A’s project contribution
A’s project score is seventy. The project contributes 0.30 × 70 = 21 final-grade points.
Although A scored lower on the project than on quizzes, the project’s contribution is twenty-one compared with the quiz contribution of eighteen because its weight is larger. Raw component scores and weighted contributions answer different questions.
10. Learner A’s examination contribution
The examination score is eighty, and its weight is fifty percent. Its contribution is 0.50 × 80 = 40 points.
This is the largest component contribution in A’s final calculation. A reader should not conclude from that fact alone that the examination was the easiest assessment; the weights reflect the fictional rubric, not a judgment about difficulty.
11. Calculate A’s final weighted grade
Add the contributions: 18 + 21 + 40 = 79. A’s final weighted score is therefore 79 out of 100 under this scheme.
This differs from A’s simple average of eighty by one point. Neither figure is a mistake when labelled properly: one is an unweighted descriptive average, and the other is the rubric-defined final grade.
12. Learner B’s quiz contribution
B’s quiz score is seventy-five. At twenty percent weight, it contributes 0.20 × 75 = 15 final points.
That contribution is three points lower than A’s quiz contribution of eighteen. The difference is a property of the two quiz scores under the common weight.
13. Learner B’s project contribution
B earns eighty-five on the project. At thirty percent, it contributes 0.30 × 85 = 25.5 final points.
B’s project contribution exceeds A’s twenty-one by 4.5 points. This advantage outweighs the three-point disadvantage on the quiz component in the final calculation.
14. Learner B’s examination contribution
B’s examination score is eighty, exactly the same as A’s. Both receive forty final points from the examination because they share the same score and weight.
This common contribution cancels when comparing the two learners’ totals. The weighted ranking depends on the quiz and project differences, not an invented examination advantage.
15. Calculate B’s final weighted grade
Add B’s component contributions: 15 + 25.5 + 40 = 80.5. This is 1.5 points higher than A’s weighted 79.
The result can be reported as 80.5 on the fictional grading scale. If a real scheme required rounding to a whole number, the rounding method and stage would need to be established separately.
16. Both simple means are equal
A’s simple average is (90 + 70 + 80) ÷ 3 = 80. B’s is (75 + 85 + 80) ÷ 3 = 80 as well.
The equal simple means conceal how the scores are distributed across components. B performs better in the more heavily weighted project, while A is stronger in the less heavily weighted quizzes. This is why the approved weights matter.
17. Why B outranks A in the toy scheme
B loses three final contribution points on quizzes relative to A but gains 4.5 on the project. Their examination contributions are equal. The net difference is −3 + 4.5 = +1.5 for B.
Therefore B’s weighted score of 80.5 exceeds A’s 79. The ranking follows the declared rubric, not a subjective statement that B is universally the better learner or that one assessment measures intelligence.
18. The fully reconciled grade table
- Weights: Quiz 0.20, Project 0.30, Examination 0.50.
- A raw scores: 90, 70, 80.
- A weighted contributions: 18, 21, 40.
- A final: 79.
- A unweighted mean: 80.
- B raw scores: 75, 85, 80.
- B weighted contributions: 15, 25.5, 40.
- B final: 80.5.
- B unweighted mean: 80.
- Weighted B minus A: 1.5 points.
- Unweighted B minus A: 0 points.
- Weight total: 1.00.
19. A change in one weight changes the result
Imagine a different, explicitly hypothetical rubric with quizzes worth forty percent, project thirty percent and examination thirty percent. The weights again add to 100%, but the relative importance of quizzes and examination has changed.
Under that alternative, A’s score is 0.40 × 90 + 0.30 × 70 + 0.30 × 80 = 36 + 21 + 24 = 81. The new score should be labelled as a scenario under revised weights, not a correction to the original 79.
20. Calculate B under the alternate rubric
Under the hypothetical 40% quiz, 30% project and 30% examination weights, B receives 0.40 × 75 + 0.30 × 85 + 0.30 × 80 = 30 + 25.5 + 24 = 79.5.
A’s alternative 81 now exceeds B’s alternative 79.5. The ranking reverses because the weights changed, even though all raw assessment scores stayed the same. A useful AI report should identify which rubric produced each result.
21. A ranking reversal is not a contradiction
Under the original rubric B leads by 1.5; under the alternative rubric A leads by 1.5. Those statements describe different legitimate mathematical questions.
Confusion begins when an assistant presents one ranking without specifying its weights. The result is only meaningful together with the source rule that defines it.
22. What if A improves the project by ten points?
Keep the original 20/30/50 weights and all of A’s other raw scores. If A’s project score rises from seventy to eighty, that component contributes 0.30 × 80 = 24 rather than 21.
A’s final becomes 18 + 24 + 40 = 82. It increases by three points because a ten-point raw project improvement is weighted at thirty percent. This is a hypothetical recalculation, not a claim that the pupil actually improved.
23. What if A improves the examination by four?
Return to the original A scores, then imagine the examination changes from eighty to eighty-four while the quiz and project scores remain fixed.
The new final is 18 + 21 + 0.50 × 84 = 18 + 21 + 42 = 81. It increases by two points, exactly fifty percent of the four-point raw examination improvement.
24. Sensitivity can guide study planning but not promise marks
Within this rubric, a one-point increase in the examination adds 0.5 final points, while a one-point increase in quizzes adds 0.2. That mathematical relationship can inform hypothetical grade goals.
It does not prove that a given hour of study will produce the same raw improvement in every component. Learning difficulty, prior knowledge and assessment design remain outside the calculation.
25. Missing marks need an explicit policy
Suppose the project score is unavailable. We cannot calculate the exact original weighted final unless a rule specifies what missing work means and whether the contribution is excluded, counted as zero or treated as pending.
An AI must not fill a missing grade with zero merely to complete a formula. Missing, absent and scored zero are different states, and their treatment should come from the authorised assessment scheme.
26. Beware mixed score scales
Our input scores all use the same scale of zero to one hundred. A real rubric might contain a ten-point quiz score, a twenty-mark project and a hundred-mark examination. Those raw marks cannot simply be weighted as though they shared the same denominator.
First convert each component to the appropriate percentage or rubric score, then apply its weight. Otherwise the component with the largest numeric maximum may dominate the final calculation incorrectly.
27. What a grade cannot prove
A rubric score reports performance on defined assessments under scoring rules. It does not establish every aspect of a learner’s knowledge, motivation, potential or future achievement.
A trustworthy educational AI should explain the calculation without ranking students as people or inventing private reasons for their results. This article’s fictional learners are mathematical placeholders.
28. Twenty-four claims for the AI to inspect
Use SUPPORTED for statements established by the fictional record and explicit scenarios, CONTRADICTED for those that conflict with it, and NOT ESTABLISHED for claims about unrecorded real-world outcomes. A result is supported only when the proper rubric and calculation are identified.
The answer key separates arithmetic, hypothetical score changes and unsupported assertions about actual pupil ability.
- 01. Original quiz weight is twenty percent — SUPPORTED.
- 02. Original project weight is thirty percent — SUPPORTED.
- 03. Original examination weight is fifty percent — SUPPORTED.
- 04. The original weights sum to one hundred percent — SUPPORTED.
- 05. A’s unweighted mean is eighty — SUPPORTED.
- 06. A’s original weighted final is seventy-nine — SUPPORTED.
- 07. B’s unweighted mean is eighty — SUPPORTED.
- 08. B’s original weighted final is 80.5 — SUPPORTED.
- 09. A’s quiz contribution is eighteen — SUPPORTED.
- 10. A’s project contribution is twenty-one — SUPPORTED.
- 11. A’s examination contribution is forty — SUPPORTED.
- 12. B’s project contribution is 25.5 — SUPPORTED.
- 13. B leads A by 1.5 under the original weights — SUPPORTED.
- 14. Their equal simple means prove their weighted grades are equal — CONTRADICTED.
- 15. Under 40/30/30 weights, A’s score would be 81 — SUPPORTED.
- 16. Under 40/30/30 weights, B’s score would be 79.5 — SUPPORTED.
- 17. Under 40/30/30 weights, B still outranks A — CONTRADICTED.
- 18. Improving A’s project from 70 to 80 raises its original-rubric final to 82 — SUPPORTED.
- 19. Improving A’s examination from 80 to 84 raises its original-rubric final to 81 — SUPPORTED.
- 20. A’s original-rubric score is eighty — CONTRADICTED.
- 21. The weights are official national school examination rules — CONTRADICTED; they are fictional.
- 22. B’s higher toy grade proves B is more intelligent — NOT ESTABLISHED.
- 23. The record verifies that A really improved the project — NOT ESTABLISHED.
- 24. B’s exam contribution equals A’s exam contribution — SUPPORTED.
29. Check the classification totals
Supported claims are 01–13, 15, 16, 18, 19 and 24: eighteen. Contradicted claims are 14, 17, 20 and 21: four. Not established are 22 and 23: two. Eighteen plus four plus two equals twenty-four.
The unestablished statements are important. A hypothetical score change is not a measured improvement, and a score in a defined assessment scheme is not proof of a pupil’s intelligence.
30. How spreadsheets can go wrong
An electronic gradebook might contain correct raw scores but use the wrong cell references, outdated weights or an incorrect denominator. A final formula could be internally consistent while applying the wrong rubric.
An AI that checks a grade report should inspect the source headings, weight version and stored formulas. The correct display number needs both correct data and correct rule application.
31. The difference between rounding and weighting
Weighting determines each component’s contribution. Rounding determines how a result is displayed or compared. They solve different problems and should not be combined casually.
For example, B’s 80.5 may be rounded differently under different approved rules. No rounding policy is supplied here, so 80.5 remains the exact arithmetic result of the fictional rubric.
32. A transfer exercise with project tasks
Imagine a project assessed on research worth forty percent and presentation worth sixty. A learner scores 75 in research and 90 in presentation. The weighted result is 0.40 × 75 + 0.60 × 90 = 30 + 54 = 84.
The simple mean would be 82.5. The difference illustrates how the larger presentation weight changes the result even when there are only two components.
33. A transfer exercise with resource selection
Weighted scoring is also used to compare options according to criteria such as price, accessibility or educational quality. But an option-ranking score is only as meaningful as the definitions and justified weights behind it.
A model should not invent precise weights and present the resulting recommendation as objective fact. Stakeholders need to understand what was prioritised, which constraints are mandatory and what sensitivity to weights remains.
34. A practical classroom activity
Give pupils two learners’ raw scores and ask them to calculate the simple means first. Then supply the 20/30/50 rubric and compute the weighted contributions. Finally, switch to 40/30/30 without changing scores.
Discuss why the ranking reverses. The aim is not to praise one pupil, but to see how weights encode the definition of the aggregate score and why an assistant must preserve that definition.
35. What AI builders should test
Create cases with weights summing to one, incomplete weights, duplicated components, mixed scales, missing scores, changed rubric versions and rounding requirements. Establish the reference outputs before asking the model to calculate.
Score correct extraction of weights separately from arithmetic and from claims about student ability. A model that returns the right number by accident under a wrong formula should not be treated as having faithfully solved the task.
36. The central lesson
A weighted grade is not the average of whichever scores happen to appear in a table. It is a calculation tied to a defined rubric, with each assessment’s contribution determined by its weight. Equal simple means can conceal different weighted outcomes.
Trustworthy AI keeps the raw scores, component identities, weights and calculation stages visible. Before accepting a final grade, ask which rubric was used and whether its numbers add up correctly.
Frequently asked questions
- What is a weighted average? — A sum of component scores multiplied by their assigned shares.
- Why is A’s final 79 rather than 80? — The project and examination carry different weights from quizzes.
- Do weights need to add to 100%? — For this complete percentage rubric, yes.
- Can equal simple means produce unequal weighted grades? — Yes, if the component distributions differ.
- Does changing weights without changing scores alter a ranking? — It can, as the two fictional rubrics show.
- What if a score is missing? — Follow the actual missing-assessment policy; do not silently replace it with zero.
- Does a higher weighted score establish greater intelligence? — No. It only describes performance under that assessment scheme.
Connected reading and primary references
- How Super Intelligence Works — main educational curriculum.
- Averages and Outliers — why different summaries answer different questions.
- Sensitivity Analysis — how changed weights affect a result.
- Tables, Rows and Columns — preserving component labels.
- OpenStax Introductory Statistics: Measures of the Centre of Data — weighted mean concepts.
- NIST/SEMATECH: Measures of Location — central tendency and weighted summaries.