How does superintelligence work when an AI needs to check a number without knowing the exact answer? A school organiser says twenty-three classroom tables each hold between four and six pupils. The assistant announces that 145 pupils are seated. We may not know the precise attendance, but we can already show that 145 is impossible under the stated rules: even six pupils at every table gives only 138.
The mechanism is upper-and-lower-bound reasoning. A lower bound is the smallest value permitted by verified constraints, and an upper bound is the largest. Bounds make it possible to reject impossible totals, check the plausibility of estimates and identify missing information before attempting a complicated exact calculation. They are particularly useful when AI interprets word problems, plans classroom activities or validates numerical reports.
This friendly guide works through a fictional twenty-three-table arrangement, examples of feasible and impossible totals, and twenty inspectable claims. Every table count, allocation and model statement is invented for education rather than drawn from a real class. “How Superintelligence Works” is the name of this learning series, not a claim that present AI models have attained artificial superintelligence.
1. What is a lower bound?
A lower bound places a minimum on a quantity. If every occupied table in a specified arrangement has at least four pupils and there are twenty-three tables, then the total cannot be lower than twenty-three times four.
That multiplication gives 92. We do not need to know the exact distribution at individual tables to establish this minimum. The result relies on the assumption that all twenty-three tables are included and each really has at least four pupils.
2. What is an upper bound?
An upper bound places a maximum on a quantity. If every one of twenty-three tables holds at most six pupils, the total cannot exceed twenty-three times six.
The upper bound is 138. A total of 139 or 145 conflicts with the stated constraints even if an AI confidently explains it. This is a useful independent check against invented arithmetic or a misunderstood table count.
3. A range can be exact information
An interval such as 92 through 138 can be a rigorous consequence of the rules even when the precise number of pupils is unknown. The range is not necessarily a rough guess.
A correct response can therefore be: “The total is between 92 and 138 inclusive under these assumptions; the exact count needs the table-by-table attendance.” That is informative without pretending the source supplied more than it did.
4. What “inclusive” means
The endpoints are allowed. A table can contain exactly four pupils or exactly six under the specified rule. Therefore 92 and 138 are both feasible total counts, given appropriate allocations.
If the source instead required more than four pupils per table, the minimum would change for whole-number pupil counts. Words such as at least and at most determine whether endpoints are included.
5. The fictional Cedar Classroom Plan
Imagine Cedar School describes twenty-three occupied tables. Every table has a whole-number attendance of four, five or six pupils. No one sits away from the listed tables, and every pupil is counted once.
This is a complete, artificial counting model. It does not describe real seating regulations, health requirements or class sizes. Its value lies in making the allowed range explicit.
6. The frozen reference constraints
- Number of occupied tables: 23.
- Minimum pupils at each table: 4.
- Maximum pupils at each table: 6.
- Allowed integer occupancies per table: 4, 5 or 6.
- Every pupil belongs to exactly one of the 23 table counts.
- No other pupils are included.
- Lower total = 23 × 4 = 92.
- Upper total = 23 × 6 = 138.
- Exact total occupancy: not supplied.
- All numbers and rules are fictional.
7. Calculate the minimum
At the lowest allowed occupancy, each of the twenty-three tables has four pupils. The total is 23 × 4 = 92. This shows that 92 can occur, not merely that it is an abstract inequality.
The minimum is achieved by a concrete allocation: twenty-three tables with four pupils each. When a model gives a bound, showing an allocation that reaches the endpoint can establish that the bound is tight in the stated fixture.
8. Calculate the maximum
At the greatest allowed occupancy, all twenty-three tables have six pupils. The total is 23 × 6 = 138. This is the highest feasible number under the complete rules.
A claimed total of 140 would require at least one table above the six-person maximum, or an additional table or uncounted seating space. All would conflict with the frozen fixture.
9. A lower bound is not the actual count
Because the minimum is 92, we cannot conclude that 92 pupils are actually present. Some tables may have five or six. The source does not specify the individual occupancy values.
An assistant that calls the lower bound “the recorded attendance” has changed a constraint into an observation. The minimum is a proven possibility, not a confirmed event.
10. An upper bound is not the actual count
Similarly, 138 is the maximum, not an observed headcount. The statement “up to 138 pupils can be accommodated under the occupancy rule” does not establish that all those places are used.
A fluent AI can mislead readers by omitting words such as at most or no more than. Those qualifications are part of the numerical meaning, not optional writing style.
11. A feasible total is different from a true total
A total is feasible if at least one arrangement of allowed table occupancies produces it. A total is actual if the source records that it occurred. Mathematics can settle feasibility without observing reality.
Our task is to determine which totals are permitted by the table rules. We are not claiming that a particular classroom contained any chosen number of pupils.
12. Can exactly 120 pupils fit?
Yes. Put six pupils at fourteen tables and four pupils at nine tables. The table count is 14 + 9 = 23 and the pupil count is 14 × 6 + 9 × 4 = 84 + 36 = 120.
Every table still has between four and six pupils. This explicit construction proves 120 is feasible. It does not prove 120 pupils were present on an actual day.
13. Can exactly 137 pupils fit?
Yes. Place six pupils at twenty-two tables and five pupils at one table. That gives 22 × 6 + 1 × 5 = 132 + 5 = 137.
The allocation respects the twenty-three-table limit and the per-table range. A model that says only totals divisible by six are possible has ignored the permitted occupancy of five.
14. Can exactly 115 pupils fit?
Yes. Five pupils at each of twenty-three tables gives 23 × 5 = 115. The count lies halfway between the minimum and maximum because five is the middle of the allowed occupancies.
The configuration demonstrates one possible total. It does not tell us whether 115 is more likely than another feasible count, because the fixture supplies no probability distribution over table arrangements.
15. Can exactly 125 pupils fit?
Yes. Ten tables with six pupils and thirteen with five pupils produce 10 × 6 + 13 × 5 = 60 + 65 = 125. The numbers of tables add to twenty-three.
This second example reinforces that the allowed totals are not restricted to the extrema or to multiples of twenty-three. Different combinations can produce intermediate whole-number totals.
16. Why 90 is impossible
With every table containing at least four pupils, twenty-three tables require at least ninety-two. A proposed total of ninety is below the lower bound.
The assistant can reject ninety without knowing any individual table occupancy. That is the value of an independently verified constraint: an impossible answer can be detected before a detailed reconstruction.
17. Why 139 is impossible
The maximum possible under the source is 138. A total of 139 exceeds it by one. Even if twenty-two tables are at their highest allowed attendance, the last table cannot add a seventh person under the rules.
A total just one above a bound is as impossible within this model as a number hundreds above it. This is why checking exact endpoints matters.
18. An AI can make a precise-looking impossible claim
Suppose the assistant reports 145 pupils while correctly repeating that each of twenty-three tables holds at most six. Those statements conflict because twenty-three times six equals 138.
A reviewer need not accept the confident headline or ask the model to explain its internal reasoning. The straightforward arithmetic bound provides a decisive external check.
19. Does every integer inside the range work?
Yes for this specific fixture. Start with four pupils at each table, giving 92. Each table can then accept one extra pupil to reach five or a second to reach six. Twenty-three tables offer forty-six such one-pupil increments in total.
Any whole-number increment from zero through forty-six can be allocated without exceeding two extra per table. Thus every integer from 92 to 138 inclusive is feasible, although no particular one is established as the observed total.
20. Why this conclusion is not universal
In other problems, an interval may contain numbers that are not feasible. Suppose every pack contains exactly six books and the number of packs can only be a whole number. Totals such as thirteen books might fall between two larger bounds but fail the pack-size requirement.
Bounds provide necessary limits; additional structural constraints can restrict which interior values are possible. Our classroom fixture has the special property that each table allows occupancy increments of one.
21. The numeric feasibility scorecard
- 23 tables, 4–6 pupils each.
- Lower bound: 92.
- Upper bound: 138.
- Possible exact total 92: 23 × 4.
- Possible exact total 115: 23 × 5.
- Possible exact total 120: 14 × 6 + 9 × 4.
- Possible exact total 125: 10 × 6 + 13 × 5.
- Possible exact total 137: 22 × 6 + 1 × 5.
- Possible exact total 138: 23 × 6.
- Impossible total 90: below minimum.
- Impossible total 139: above maximum.
- Actual recorded attendance: unknown.
22. Twenty claims for an AI to classify
The fictional desk asks an assistant to label statements SUPPORTED, CONTRADICTED or NOT ESTABLISHED. Supported statements follow from the rules or a constructed feasible arrangement. Contradicted claims conflict with a fixed constraint. Unestablished statements ask about actual attendance or identities not supplied.
We deliberately separate “could fit” from “did fit”. That single distinction prevents an AI from presenting its own possible arrangement as a source record.
- 01. There are twenty-three occupied tables in the fixture — SUPPORTED.
- 02. The lowest allowed total is 92 — SUPPORTED.
- 03. The highest allowed total is 138 — SUPPORTED.
- 04. An arrangement of exactly 120 pupils is feasible — SUPPORTED.
- 05. An arrangement of 139 pupils is feasible — CONTRADICTED.
- 06. An arrangement of 90 pupils is feasible — CONTRADICTED.
- 07. A total of 92 pupils is feasible — SUPPORTED.
- 08. A total of 138 pupils is feasible — SUPPORTED.
- 09. Exactly five pupils at every table gives 115 — SUPPORTED.
- 10. Exactly five at every table gives 120 — CONTRADICTED.
- 11. Fourteen tables at six and nine at four give 120 — SUPPORTED.
- 12. Twenty-two tables at six and one at five give 137 — SUPPORTED.
- 13. Exactly 120 pupils were actually present — NOT ESTABLISHED.
- 14. Every table actually held six pupils — NOT ESTABLISHED.
- 15. The table count is twenty-three — SUPPORTED.
- 16. The lower per-table occupancy permitted is four — SUPPORTED.
- 17. The maximum per-table occupancy permitted is six — SUPPORTED.
- 18. A total of 142 pupils could fit under the rule — CONTRADICTED.
- 19. The mean table occupancy in any valid arrangement is between four and six — SUPPORTED.
- 20. The identities of all pupils actually present are known — NOT ESTABLISHED.
23. Verify the claim totals
Supported claims are 01, 02, 03, 04, 07, 08, 09, 11, 12, 15, 16, 17 and 19: thirteen. Contradicted claims are 05, 06, 10 and 18: four. Not established are 13, 14 and 20: three. Thirteen plus four plus three equals twenty.
The unknown cases matter. The arithmetic establishes what could happen, but the record contains no observed attendance distribution or list of pupils.
24. A possible arrangement is not evidence of occurrence
The fourteen-full-and-nine-minimum allocation proves that 120 is possible. It is a constructed example produced for verification, not a record of the actual classroom.
An AI should label such examples as hypothetical. Otherwise its explanation can turn a useful mathematical witness into a fabricated real-world fact.
25. If the number of tables changes
Suppose the school authorises an additional identical table, making twenty-four in total. The lower bound becomes 24 × 4 = 96 and the upper becomes 24 × 6 = 144.
This is a new scenario. The assistant should not silently change the original twenty-three-table answer; it should state which parameter changed and recalculate the dependent bounds.
26. If the per-table rule changes
Suppose the upper occupancy limit is reduced from six to five while the twenty-three-table count stays fixed. The maximum drops to 23 × 5 = 115, while the minimum remains 92.
Under the changed policy a proposed total of 120 is no longer feasible. The same number may be valid under one rule version and invalid under another. The source and effective rule matter.
27. Why missing values do not erase bounds
We may know only that each table has four, five or six pupils, without knowing which table has which count. The unknown individual values do not stop us establishing the aggregate limits.
That is a different situation from not knowing the total number of tables or the maximum allowed per table. The system should track which facts are known constraints and which are missing variables.
28. Bound reasoning helps detect bad word-problem answers
A calculation may produce 200 pupils from twenty-three tables with six at most. The contradiction is visible without inspecting every intermediate arithmetic step.
A teacher can train pupils to estimate a reasonable minimum and maximum before solving. If the final result lies outside them, return to the equation, units and source quantities to find the error.
29. Bound reasoning helps with costs
Imagine twenty-three identical items priced between $4 and $6 each, with no tax or discounts. Total cost must fall between 23 × $4 = $92 and 23 × $6 = $138.
This looks similar to the seating problem because the per-item amount lies in a known interval. But real pricing may include taxes, bundles or nonidentical items, so the assumptions must be checked.
30. Bound reasoning helps with time estimates
A worksheet contains ten tasks each taking between two and four minutes under a defined planning estimate. The total estimated task time lies between twenty and forty minutes if the times add and no other activity is included.
The interval does not establish how long any real child will need. It is a planning range based on specified task-duration assumptions.
31. Averages also have bounds
If every table has between four and six pupils, the average occupancy across the twenty-three tables must also be between four and six. Averaging cannot produce seven when every counted value is at most six.
A reported mean above the maximum individual value is an automatic warning sign for a simple unweighted average of those values. The model may have used the wrong denominator or included extra quantities.
32. Bounds can be stronger than estimates
An assistant may offer a best-guess total of 120 without evidence. The justified bound 92–138 is less specific but more defensible under the actual source.
A useful response can provide the proven range and state what additional data are needed for an exact total. Being honest about precision is a form of mathematical competence, not a failure to answer.
33. Interval arithmetic is a related technique
In numerical computing, interval arithmetic propagates ranges through calculations to obtain bounds that contain possible results under defined assumptions and rounding rules. The NIST Digital Library of Mathematical Functions describes this approach in validated computing.
Our classroom multiplication is a simple combinatorial bound rather than a full interval-arithmetic implementation. The connection shows how rigorous bounds can support much more complex computational verification.
34. When an interval is too broad
A range from 92 to 138 may be adequate to reject 145 but too broad to decide whether there are enough books for every pupil. A decision with a tight capacity limit may require the actual attendance.
The next useful action is to retrieve the missing per-table counts, not replace the range with a fabricated exact number. Different user goals require different levels of precision.
35. A classroom transfer exercise
A shelf has fifteen boxes, with between eight and ten pencils in each box. The total must lie between 15 × 8 = 120 and 15 × 10 = 150 pencils.
A claim of 155 pencils contradicts the given bounds. A claim of 135 is feasible if suitable box compositions exist, but does not prove the shelf actually contains 135. Ask pupils to explain both distinctions.
36. How to design a useful AI check
Before trusting a computed total, identify independent minimum and maximum constraints. Multiply or combine them according to the structure, make sure the units match and compare the final output against the resulting interval.
If a result lies inside the interval, the bound check has not proven it correct. It has merely avoided one category of impossible answer. Further arithmetic, source and state verification may still be required.
37. The central lesson
Upper and lower bounds let us know something rigorous without knowing everything. They are especially powerful for rejecting impossible numbers and checking that an AI has not forgotten a constraint.
A strong answer says what is proven, what is feasible and what actually happened only when there is evidence. That separates mathematical possibility from invented certainty.
Frequently asked questions
- What is an upper bound? — The largest value permitted by the specified constraints.
- What is a lower bound? — The smallest value permitted by the specified constraints.
- Does being inside the bounds prove a total is correct? — No. It may be feasible without being the observed total.
- Why is 139 impossible here? — Twenty-three tables holding at most six pupils can hold no more than 138.
- Can 120 fit? — Yes. Fourteen six-pupil tables plus nine four-pupil tables give 120.
- Can the exact total be calculated from bounds alone? — Not without further data about the actual occupancy.
- Do all values between bounds always work? — No, additional structural constraints can create gaps, though every integer in this toy interval is feasible.
Connected learning and authoritative references
- How Super Intelligence Works — main educational curriculum.
- Boundary Values — why endpoints matter.
- Multi-Step Word Problems — translating a story into quantities.
- Invariants — checks that must remain true.
- OpenStax: The Language of Algebra — inequalities and limits.
- NIST Digital Library of Mathematical Functions: Interval Arithmetic — rigorous numerical bounds.