How does superintelligence work when an AI is asked for a number that lies between two known measurements—or beyond the measurements altogether? A graph may show a school club recording fourteen visits in Week 2 and twenty-two in Week 4. The Week 3 record is missing. An assistant could estimate eighteen if it assumes a straight line, but it must not call that estimate a confirmed observation. Predicting Week 6 from the same pattern is an even different task.
The mechanism is interpolation and extrapolation. Interpolation estimates values within the range of observed inputs, while extrapolation predicts outside the observed range. Both require assumptions about how values change. AI systems can calculate the estimates quickly, but a trustworthy explanation preserves the difference between measured facts, model-based estimates and future outcomes that have not happened.
This plain-English guide follows an invented reading-club attendance series with four confirmed observations, one missing interior week, and two unobserved future weeks. We calculate a linear model, test twenty-four claims and compare the same observations with an alternative pattern. The numbers are fictional and do not represent real pupils or a measured AI benchmark. The series title “How Superintelligence Works” is educational rather than a claim that present AI possesses artificial superintelligence.
The essential distinctions
- Observation: a value that was actually recorded and checked.
- Interpolation: an estimate between observed time points.
- Extrapolation: an estimate beyond the observed time range.
- The sample contains confirmed Weeks 1, 2, 4 and 5 only.
- A straight-line model suggests Week 3 = 18 and Week 6 = 30.
- Neither suggested value is independently measured in the fictional ledger.
- An exact arithmetic calculation can still rest on an uncertain modelling assumption.
1. What counts as observed?
An observation is the value a defined source records for a particular event or period. In this fictional series, the verified Week 1 visit total is ten. The source separately confirms fourteen for Week 2, twenty-two for Week 4 and twenty-six for Week 5.
Those numbers are facts within our teaching fixture. A missing Week 3 value is not an observed zero, and a future Week 6 value is not yet recorded. A responsible AI preserves these distinctions before drawing a graph or calculating a forecast.
2. Why missing does not mean zero
If the attendance sheet for Week 3 was never received, entering zero would assert that nobody visited. The source does not support that claim. A blank represents an information gap, not necessarily a week in which the club was empty.
Some software packages silently convert blanks into numerical zeros or drop them from chart calculations. An AI should check that behaviour when handling real spreadsheets because it can create an artificial decline or distort a calculated average.
3. What is interpolation?
Interpolation uses nearby known points to estimate an unknown value inside their domain. Week 3 falls between the confirmed Weeks 2 and 4. We can estimate its value if we choose a suitable model for what happened between those endpoints.
A straight-line interpolation assumes a constant change per unit time between the two known observations. That assumption can be useful, but it does not make the resulting value an independently measured count.
4. What is extrapolation?
Extrapolation carries a fitted rule beyond the range where observations were supplied. Week 6 is later than the latest confirmed Week 5 record, so a Week 6 value derived from the previous pattern is an extrapolation.
Future circumstances may differ from the observed period. A new school timetable, holiday, event or capacity limit could alter attendance. An assistant should label Week 6 as a conditional projection rather than a verified result.
5. Why a simple line is appealing
When observed values appear to rise by a consistent amount over equally spaced weeks, a linear rule provides an easy-to-understand first model. It needs a starting level and a change per week.
A model’s simplicity is an advantage for teaching and basic planning. It is not proof that the underlying process follows a straight line. The purpose of evaluation is to test whether the proposed rule represents the source adequately.
6. The fictional Cedar Reading Club
Imagine an invented school club that counts total recorded visits in each of several weeks. It reports four confirmed weekly totals and has no official record for Week 3. The example treats each value as a total of visits rather than distinct people.
The counting method is stipulated consistent across the confirmed weeks. There is no evidence about pupil identities, attendance causes, holidays or an actual future programme. The artificial series is designed to make mathematical interpretation visible.
7. The complete source ledger
- Week 1: 10 recorded visits — OBSERVED.
- Week 2: 14 recorded visits — OBSERVED.
- Week 3: no supplied observation — MISSING.
- Week 4: 22 recorded visits — OBSERVED.
- Week 5: 26 recorded visits — OBSERVED.
- Week 6: not yet observed — FUTURE OUTSIDE RANGE.
- Week 7: not yet observed — FUTURE OUTSIDE RANGE.
- Time steps are consecutive weeks of equal duration.
- The last confirmed source week is Week 5.
- No real pupil or school is identified by these invented data.
8. The difference between Weeks 1 and 2
From Week 1 to Week 2, the recorded total rises from ten to fourteen. The observed change is +4 visits over one weekly interval.
That establishes one interval’s change, not a general law. An AI should be careful not to say the club permanently gains exactly four people every week. The figure describes a change in recorded visit totals and may include repeat visitors.
9. The difference between Weeks 4 and 5
From Week 4 to Week 5, the confirmed total rises from twenty-two to twenty-six, again a difference of four visits over one week.
This supplies another observed interval consistent with a +4 pattern. It does not tell us the actual Week 3 value. Multiple sequences could connect fourteen in Week 2 to twenty-two in Week 4.
10. Check the two-week gap
Week 2 has fourteen visits, while Week 4 has twenty-two. The difference is twenty-two minus fourteen, or eight visits, across two weekly intervals.
A straight-line interpolation would spread that eight-visit increase equally: four additional visits per week. This is a modelling choice. An uneven increase of three one week and five the next would also reach the observed Week 4 total.
11. Estimate Week 3 by interpolation
Under the straight-line assumption, Week 3 lies halfway between Weeks 2 and 4. Add half of the eight-visit gap to Week 2: fourteen plus four gives eighteen.
The estimate is therefore Week 3 ≈ 18 visits. The arithmetic is exact under the chosen model, but the original source provides no Week 3 observation. A report should write interpolated or estimated rather than recorded.
12. A formula makes the assumption explicit
For equally spaced weekly observations, linear interpolation between Week 2 and Week 4 can be written as Week 2 value plus the fractional distance to the next known point times the difference between endpoints.
Week 3 is halfway through the two-week span: 14 + (1/2) × (22 − 14) = 18. The displayed equation shows precisely how the estimate was constructed and permits a reader to challenge the straight-line assumption.
13. Fit a simple linear rule
The four confirmed points are consistent with a line that begins at ten in Week 1 and adds four visits for each subsequent weekly step. Its formula is visits = 10 + 4 × (week number − 1).
At Week 4 the rule gives 10 + 4 × 3 = 22, and at Week 5 it gives 10 + 4 × 4 = 26. The agreement is an internal consistency check of the model against the invented records.
14. Interpolation and extrapolation use the same arithmetic differently
Applying the rule to Week 3 gives 10 + 4 × 2 = 18, an interpolated estimate inside the observed Week 1–5 range. Applying it to Week 6 gives 10 + 4 × 5 = 30, a projection beyond the latest observation.
The formulas may be identical in form, but their evidential status differs. An AI should not present them as equally observed simply because a calculator returned both numbers without hesitation.
15. Extrapolate Week 6
Continuing the model one week beyond the final confirmed observation produces a Week 6 forecast of thirty visits.
This is a one-step extrapolation. It is neither a historical observation nor a guarantee of future attendance. If the club closes for a school holiday, the actual Week 6 count could be much lower or zero, despite the mathematical continuation.
16. Extrapolate Week 7
Continuing the same rule two steps beyond Week 5 produces Week 7 = 10 + 4 × 6 = 34 visits. That is a second future projection.
It is farther outside the observed range than the Week 6 forecast. The farther we project, the more opportunity there is for a model’s assumptions to diverge from real conditions. The specific change in uncertainty cannot be calculated from this tiny fixture alone.
17. The checked model-versus-source table
- Week 1: observed 10; model 10.
- Week 2: observed 14; model 14.
- Week 3: observed UNKNOWN; model interpolation 18.
- Week 4: observed 22; model 22.
- Week 5: observed 26; model 26.
- Week 6: observed NONE; model extrapolation 30.
- Week 7: observed NONE; model extrapolation 34.
- Modelled increase per weekly step: 4.
- Difference between Week 7 and Week 5 projections: 34 − 26 = 8.
- Verified source does not confirm the Week 3, 6 or 7 numerical outcomes.
18. An equally valid uneven interior path
Suppose the unobserved Week 3 count were seventeen rather than eighteen. The sequence of changes from Week 2 to Week 4 would be +3 and then +5. It would still match the source’s Week 2 value fourteen and Week 4 value twenty-two.
That alternative demonstrates why the straight-line interpolation is an estimate. The endpoints alone do not logically force the missing interior value to be eighteen.
19. An alternative can also exceed the midpoint
Likewise, a hypothetical Week 3 value of nineteen gives changes +5 and +3 across the two intervals. Again, the confirmed endpoints remain unchanged.
An AI that calls eighteen definitely true because the observations look linear has confused model fit with direct evidence. Without the actual Week 3 record, several interior paths remain possible.
20. The endpoint pattern does not establish the cause
A reader might attribute the apparent rise to a new reading programme, an appealing book collection or a better teacher. The ledger contains no intervention data or comparison group that can distinguish these explanations.
The right statement is that the recorded weekly visit totals rose between the named weeks. A causal claim about why they rose requires additional design and evidence.
21. Future projections need a scope
The linear model is a useful conditional calculation for Weeks 6 and 7. It is not a statistical forecast with a measured prediction interval, because the fixture contains too little information to estimate realistic future variability.
A responsible AI should not add an invented 95% confidence level or guarantee around the projection. An explicit statement of the model and its limits is better than false precision.
22. Interpolation is not automatically safe
An interior estimate may seem more trustworthy than an exterior projection because it falls between known points. But an unobserved event, sudden closure or irregular change could still occur inside the gap.
The appropriate choice of interpolation model depends on the process and available evidence. A straight line is a transparent teaching assumption, not a universal law of school attendance or other real-world time series.
23. Interpolation for measurements other than time
Suppose a thermometer has confirmed readings at two nearby locations and the temperature at a point between them is unknown. A simple spatial interpolation might estimate an intermediate value.
That requires assumptions about how temperature varies across space. A geographic boundary, different shade or other local effect can break a simple smooth model. The general principle is the same as time interpolation: an estimate must be labelled as such.
24. A graph can blur observed and projected data
If a chart draws one continuous solid line through Weeks 1–7, readers may assume every point was measured. A more honest chart distinguishes confirmed points from interpolated or extrapolated segments.
An accessible caption can state which weekly values were observed and which were generated by the model. The visual design should not make a projection look like an independently confirmed source record.
25. What happens if a later observation arrives
If the club subsequently records thirty-one visits in Week 6, the earlier forecast of thirty would have an error of one visit. The new observation should be stored separately from the old prediction.
A trustworthy system can update its model after the observation is verified. It should not silently replace the old forecast and then report that the original prediction was exactly correct.
26. A prediction error is not always a logic error
A forecast can be calculated correctly from a transparent, reasonable model and still differ from reality. This is uncertainty about the world, not necessarily a failure of basic arithmetic.
Evaluating the model requires comparing forecasts with future observations, reviewing assumptions and checking whether a different method would have performed better. A fluent assistant should acknowledge this distinction.
27. Twenty-four claims for an AI to classify
The fictional desk classifies statements as SUPPORTED, CONTRADICTED or NOT ESTABLISHED. Supported claims follow from the record and defined linear model; contradicted claims conflict with the fixture; unestablished claims ask about missing or future actual outcomes.
The key is to keep the phrase under the linear model distinct from the phrase actually recorded. Several claims use identical numbers but differ in their evidential status.
- 01. Week 1 was observed at ten visits — SUPPORTED.
- 02. Week 2 was observed at fourteen visits — SUPPORTED.
- 03. Week 3 has an independently confirmed count of eighteen — CONTRADICTED as a claim of recorded status.
- 04. Week 4 was observed at twenty-two visits — SUPPORTED.
- 05. Week 5 was observed at twenty-six visits — SUPPORTED.
- 06. The observed Week 2-to-4 change was eight over two intervals — SUPPORTED.
- 07. A straight-line interpolation gives eighteen for Week 3 — SUPPORTED.
- 08. The source proves Week 3 actually had eighteen visits — NOT ESTABLISHED.
- 09. The source proves Week 3 actually had seventeen visits — NOT ESTABLISHED.
- 10. The linear model adds four per week — SUPPORTED.
- 11. Under the model, Week 6 projects to thirty — SUPPORTED.
- 12. Under the model, Week 7 projects to thirty-four — SUPPORTED.
- 13. A Week 6 count of thirty is already observed — CONTRADICTED.
- 14. Week 7 lies inside the observed Week 1–5 range — CONTRADICTED.
- 15. The model matches every one of the four observed points — SUPPORTED.
- 16. The model’s Week 4 value is twenty — CONTRADICTED.
- 17. Week 3 is an interpolation rather than an extrapolation — SUPPORTED.
- 18. Week 6 is an extrapolation rather than an observation — SUPPORTED.
- 19. The source establishes the number of unique pupils in each week — NOT ESTABLISHED.
- 20. Week 5 exceeded Week 1 by sixteen visits — SUPPORTED.
- 21. Under the model, Week 7 exceeds Week 5 by eight — SUPPORTED.
- 22. A future real Week 6 count must equal thirty — NOT ESTABLISHED.
- 23. The series proves a tutoring intervention caused growth — NOT ESTABLISHED.
- 24. A linear rule is mathematically consistent with the four recorded points — SUPPORTED.
28. Reconcile the reference labels
Supported statements are 01, 02, 04, 05, 06, 07, 10, 11, 12, 15, 17, 18, 20, 21 and 24: fifteen. Contradicted are 03, 13, 14 and 16: four. Not established are 08, 09, 19, 22 and 23: five. Fifteen plus four plus five equals twenty-four.
Claim 03 is contradicted as a statement about confirmed source status because the fixture explicitly says no Week 3 observation exists. Claims 08 and 09 remain unknown because the source does not establish the real numerical value that might have occurred in Week 3.
29. Interpolation can conceal a missing source
An assistant could fill every blank cell in a spreadsheet with model estimates and accidentally make the table look complete. A later report might then treat those estimated entries as actual records.
Keep a separate flag or source-status field for observed, interpolated and projected values. The distinction should survive exports, charts and summaries so later users know which numbers were independently measured.
30. Extrapolation and linear growth are different topics
Linear growth describes a rule that adds the same amount each step. Extrapolation describes using a model outside its observed input range. A model can extrapolate using a linear, exponential or other mathematical relationship.
Our example uses a linear model for convenience. The word extrapolation does not itself specify which growth law is correct or how confident a future prediction should be.
31. When an estimate is fit for purpose
A teacher preparing a rough display might use the eighteen-visit interpolation with an explanatory footnote. A funding decision depending on exact historical attendance may instead require the missing actual register.
The same estimated number can be useful in one context and inadequate in another. Trustworthy AI assesses what the user needs the value for, not merely whether a formula produces an answer.
32. A transfer exercise in travel
A train’s known positions at 10:00 and 10:10 can support a between-time estimate under an assumed constant speed. Predicting its position at 10:30 from the same two points requires extrapolation.
Real trains may stop, accelerate or change routes. The appropriate model depends on additional operational data. A simple calculation is not a substitute for real-time tracking.
33. A transfer exercise in sales
A shop records ten units sold on Monday and eighteen on Wednesday. A straight-line midpoint estimate for Tuesday is fourteen. But Tuesday sales could have been twelve or sixteen while matching the two observed days.
Predicting Thursday’s sales using the same +4-per-day assumption is a new extrapolation. A model should label it as forecasted and should not claim to know customers’ future purchases.
34. How AI builders can test this skill
Provide time series with missing interior points, future requests, clear nonlinear counterexamples and sudden breaks. Score whether the assistant identifies the status of each data point, applies the stated model correctly and avoids claiming missing observations as facts.
Also test whether later actual data are compared with saved forecasts rather than used to rewrite them. Accurate arithmetic and honest evidence status are separate dimensions of model quality.
35. The central lesson
Interpolation fills a gap inside observed data; extrapolation extends a model beyond it. The same formula can perform both operations, but the estimated values have different relationships to the source evidence.
A reliable AI assistant identifies what was actually recorded, labels estimates visibly, states the assumed rule and avoids presenting conditional projections as confirmed outcomes. That is how mathematical calculation supports understanding without inventing facts.
Frequently asked questions
- What is interpolation? — Estimating an unobserved value between known points.
- What is extrapolation? — Estimating outside the observed input range.
- Is an interpolated value always correct? — No. It depends on the assumed model.
- Why is Week 3 eighteen? — The straight-line model divides an eight-visit gap across two intervals.
- Does Week 6 definitely have thirty visits? — No. Thirty is a model projection, not a recorded future outcome.
- Can an AI show both observations and estimates in one chart? — Yes, but they should be visually distinguished.
- What if later data disagree with the model? — Preserve the forecast and compare it with the new verified observation.
Connected reading and references
- How Super Intelligence Works — main educational curriculum.
- Linear and Exponential Growth — the meaning of a growth rule.
- Graphs and Axes — interpreting plotted measurements.
- Seasonal Patterns — why future behaviour may change.
- NIST/SEMATECH e-Handbook of Statistical Methods — modelling and experimental predictions.
- OpenStax: Linear Functions — slope and linear interpolation foundations.