How does superintelligence work when an AI calculates the chance of two things happening in sequence? If a bag contains three red counters and two blue counters, the chance of drawing red first is three-fifths. But the chance of red on the second draw may depend on whether the first counter is returned to the bag. An AI that multiplies the same probability twice without reading the procedure can get a convincing but incorrect answer.
The mechanism is independent and dependent event reasoning. Events are independent when knowing that one occurred does not change the probability of the other in the defined model. If the first draw changes what remains in the bag, the second draw is generally dependent on the first. Understanding this difference improves school probability explanations, AI risk calculations and the interpretation of sequential decisions.
This easy guide uses two fictional versions of a school fair counter game: two draws with replacement and two draws without replacement. It calculates complete outcome probabilities, checks twenty-six claims and explains what remains unknown in a real experiment. All counts and outcomes are invented for teaching, not results from real pupils or an AI benchmark. “How Superintelligence Works” is an educational series name, not evidence that today’s AI systems have attained artificial superintelligence.
Quick facts: the two drawing rules
- Initial bag: 3 red counters and 2 blue counters.
- Both procedures draw twice, in a defined order.
- WITH replacement: return the first counter before the second draw.
- WITHOUT replacement: keep the first counter outside the bag.
- WITH replacement, chance of two red: 9/25 = 36%.
- WITHOUT replacement, chance of two red: 3/10 = 30%.
- The difference comes from whether the first draw changes the second draw’s available counts.
1. A probability describes a defined experiment
A probability is not meaningful without an event and a procedure. “Red is likely” might refer to the next draw, two consecutive draws, exactly one red in two draws or red at least once. Each is a different event.
Before computing, an assistant should write the exact question and specify how counters are selected. In this toy example, we assume each individual counter is equally likely to be drawn at each step and that the selection procedure operates as stipulated.
2. What the initial bag contains
The bag begins with five counters: three red and two blue. Therefore the chance of a red on the first draw is 3/5 = 60%, and the chance of blue on the first draw is 2/5 = 40%.
These are fractions of the five physical counters. An AI should not treat the colours as equally probable simply because there are two colour categories. There are more red counters than blue counters.
3. Why the order of draws matters
Two red counters in a row is the ordered outcome RR. Red then blue is RB, while blue then red is BR. The outcomes RB and BR contain the same pair of colours but occur in different orders.
Keeping the order visible makes conditional probabilities easier to inspect. When a question asks for exactly one red, both RB and BR qualify, but one ordered branch alone does not cover all possibilities.
4. What replacement means
With replacement, the first selected counter is returned to the bag before selecting a second. The initial five-counter composition is restored, so the second draw again has three red and two blue options.
Returning the counter must occur before the second selection. If the assistant misreads the timing and treats the bag as reduced to four counters, it has silently changed the experiment.
5. What no replacement means
Without replacement, the first counter stays outside the bag. Four counters remain for the second draw. Their colour composition depends on whether the first was red or blue.
After a first red, two red and two blue remain. After a first blue, three red and one blue remain. The second-step probability therefore depends on the first outcome.
6. The fictional Cedar School Fair
Imagine the school uses an opaque bag containing exactly the five counters specified. A pupil selects one at random, observes its colour and then follows the named replacement rule before a second random selection.
We are calculating theoretical probabilities under equal-likelihood assumptions. We do not have a real-world trial count, evidence about how the counters feel, or a measured performance score for a model. No counter is damaged or added during either version of the game.
7. Independent means no probability change
For two events A and B, independence means the probability of B is unchanged by learning that A occurred, where the conditioning event has positive probability. In elementary examples this can be expressed as P(A and B) = P(A) × P(B).
That product is justified only for independent events. More generally, a joint probability can be written P(A and B) = P(A) × P(B given A). The conditional term makes the dependence visible.
8. With replacement: two red counters
The first red draw has probability 3/5. Since it is returned, the second red draw also has probability 3/5. Multiplying gives P(RR) = 3/5 × 3/5 = 9/25 = 36%.
The calculation depends on the restored composition. If the assistant simply uses 3/5 twice for a no-replacement experiment, it has applied the right arithmetic to the wrong procedure.
9. With replacement: red then blue
The first draw is red with chance 3/5. The returned counter restores the original bag, making blue on the second draw 2/5. Thus P(RB) = 3/5 × 2/5 = 6/25 = 24%.
Only this exact order is counted in RB. The alternative BR is another branch and must be added when a question asks for one of each colour in either order.
10. With replacement: blue then red
Blue first occurs with probability 2/5. After replacement, red second occurs with probability 3/5. Thus P(BR) = 2/5 × 3/5 = 6/25 = 24%.
The probabilities RB and BR happen to be equal in this restored-bag fixture. That equality is a result of the specific factor products, not a universal property of every sequential process.
11. With replacement: two blue counters
Blue first and blue second each have chance 2/5. Their joint probability is P(BB) = 2/5 × 2/5 = 4/25 = 16%.
The four ordered outcome probabilities add to one: 9/25 + 6/25 + 6/25 + 4/25 = 25/25. This is a useful independent consistency check.
12. Complete outcome table with replacement
- RR: 9/25 = 36%.
- RB: 6/25 = 24%.
- BR: 6/25 = 24%.
- BB: 4/25 = 16%.
- Total: 9/25 + 6/25 + 6/25 + 4/25 = 1.
- Exactly one red: RB + BR = 12/25 = 48%.
- At least one red: 1 − P(BB) = 21/25 = 84%.
- Both colours the same: RR + BB = 13/25 = 52%.
13. Without replacement: two red
The chance of red on the first draw remains 3/5. After drawing one red without returning it, the bag has two red and two blue among four counters. The chance of red next, given red first, is 2/4 = 1/2.
The joint probability becomes P(RR) = 3/5 × 2/4 = 6/20 = 3/10 = 30%. It is lower than the 36% from the replacement version because the first red reduced the remaining red count.
14. Without replacement: red then blue
Red first has probability 3/5. After removing a red, two blue counters remain out of four. Blue second then has probability 2/4 = 1/2.
The ordered outcome is P(RB) = 3/5 × 2/4 = 6/20 = 3/10 = 30%. The first draw’s effect on the bag must be represented in the second fraction.
15. Without replacement: blue then red
Blue first has probability 2/5. After removing a blue, three red remain out of four counters. The conditional chance of red second is therefore 3/4 = 75%.
The joint probability is P(BR) = 2/5 × 3/4 = 6/20 = 3/10 = 30%. The second red probability differs from the 1/2 after a first red, revealing dependence.
16. Without replacement: two blue
Blue first has probability 2/5. Without replacement, only one blue remains among four counters. Blue second then has probability 1/4.
The resulting P(BB) is 2/5 × 1/4 = 2/20 = 1/10 = 10%. It is lower than 16% with replacement because the pool of blue counters has shrunk.
17. Complete outcome table without replacement
- RR: 3/10 = 30%.
- RB: 3/10 = 30%.
- BR: 3/10 = 30%.
- BB: 1/10 = 10%.
- Total: 3/10 + 3/10 + 3/10 + 1/10 = 1.
- Exactly one red: RB + BR = 6/10 = 60%.
- At least one red: 1 − P(BB) = 9/10 = 90%.
- Both colours the same: RR + BB = 4/10 = 40%.
18. Two experiments, different probabilities
With replacement, two red occurs with 36% probability. Without replacement, two red occurs with 30%. Exactly one red occurs with 48% versus 60%, respectively.
An AI cannot answer such questions accurately by recognising the word probability and blindly applying a memorised formula. It must know whether the state of the bag changes between draws.
19. Conditional probability makes dependence clear
In the no-replacement game, P(second red given first red) = 2/4 = 50%. Yet P(second red given first blue) = 3/4 = 75%. The first colour changes the composition of the remaining bag.
These conditional percentages differ, so the second draw’s red event is not independent of the first draw’s colour. The dependence follows directly from the specified sampling without replacement.
20. The unconditional second-draw chance
Even without replacement, the probability that the second draw is red before observing the first is 60%. Add branches with red second: RR is 30% and BR is 30%, making 60%.
This does not make the two draws independent. The unconditional second-red chance may equal the first-red chance due to symmetry, while the conditional chance after a known first outcome differs.
21. With replacement gives stable conditional chances
In the with-replacement procedure, P(second red given first red) and P(second red given first blue) both equal 3/5, or 60%. Returning the first counter restores the same five-counter composition.
This is a simple case of independence under the model. It does not mean that every repeated action with a replacement label is independent in the real world, where other conditions may change.
22. At least one is a complement calculation
Instead of adding RR, RB and BR separately, ask for the complement of no red: BB. With replacement, P(at least one red) = 1 − 4/25 = 21/25 = 84%.
Without replacement it is 1 − 1/10 = 9/10 = 90%. The complement method reduces the chance of missing one of the qualifying ordered branches.
23. Exactly one is not at least one
Exactly one red includes RB and BR but excludes RR. With replacement it is 6/25 + 6/25 = 12/25 = 48%. Without replacement it is 3/10 + 3/10 = 6/10 = 60%.
The phrase at least one includes RR as well, giving larger probabilities. A model that treats both phrases as synonyms changes the requested event.
24. Independent is not mutually exclusive
Independent events do not change one another’s probability under the model. Mutually exclusive events cannot occur together in a single relevant trial. Those are different properties.
For example, on the first draw the outcomes red and blue are mutually exclusive: a selected counter cannot have both defined colours at once. The first-red and second-red events can occur together, so they are not mutually exclusive.
25. What a tree diagram contributes
A two-level probability tree starts with branches red 3/5 and blue 2/5. For replacement, every second-level branch keeps 3/5 red and 2/5 blue. For no replacement, the second-level branches differ depending on the first result.
Multiplying along each branch gives the joint probability for that ordered sequence. Adding the relevant final branches gives the probability of a combined event such as exactly one red.
26. Twenty-six inspectable claims
- 01. The initial bag has three red and two blue — SUPPORTED.
- 02. Initial red probability is 3/5 — SUPPORTED.
- 03. With replacement, RR probability is 9/25 — SUPPORTED.
- 04. With replacement, BB probability is 4/25 — SUPPORTED.
- 05. With replacement, RB probability is 6/25 — SUPPORTED.
- 06. With replacement, BR probability is 6/25 — SUPPORTED.
- 07. With replacement, at least one red is 21/25 — SUPPORTED.
- 08. With replacement, exactly one red is 12/25 — SUPPORTED.
- 09. Without replacement, RR probability is 3/10 — SUPPORTED.
- 10. Without replacement, BB probability is 1/10 — SUPPORTED.
- 11. Without replacement, RB probability is 3/10 — SUPPORTED.
- 12. Without replacement, BR probability is 3/10 — SUPPORTED.
- 13. Without replacement, at least one red is 9/10 — SUPPORTED.
- 14. Without replacement, exactly one red is 6/10 — SUPPORTED.
- 15. Without replacement, P(second red given first red) is 3/5 — CONTRADICTED.
- 16. Without replacement, P(second red given first red) is 1/2 — SUPPORTED.
- 17. With replacement, second-draw colour is independent of first-draw colour under this procedure — SUPPORTED.
- 18. Without replacement, the second-draw colour is independent of the first — CONTRADICTED.
- 19. Without replacement, P(second red given first blue) is 3/4 — SUPPORTED.
- 20. After a first red in the replacement game, only two red remain in the bag at the second draw — CONTRADICTED.
- 21. Without replacement, RR probability is 36% — CONTRADICTED.
- 22. The probabilities were measured in a real school game — NOT ESTABLISHED.
- 23. The source identifies which real pupil performed each draw — NOT ESTABLISHED.
- 24. With replacement, the unconditional second-red probability is 60% — SUPPORTED.
- 25. Without replacement, the unconditional second-red probability is 60% — SUPPORTED.
- 26. With replacement, the probability of two matching colours is 52% — SUPPORTED.
27. Reconcile the claim labels
Supported cases are 01–14, 16, 17, 19 and 24–26: twenty. Contradicted are 15, 18, 20 and 21: four. Not established are 22 and 23: two. Twenty plus four plus two equals twenty-six.
The final classifications reinforce the difference between mathematical theoretical chances under a given procedure and claims about events that actually occurred. The record has a model and numbers, not actual pupil identities or observed trial outcomes.
28. Why probabilities need assumptions
The theoretical calculations assume that all individual counters are equally likely to be selected, the bag’s contents are exactly known and the replacement procedure is followed consistently.
If counters differ in shape or weight, or the selection is not genuinely random, actual selection frequencies might differ. A responsible AI should disclose this dependence on the procedure rather than treating a mathematical answer as an observed physical law.
29. Independence can be conditional on context
Two processes may appear independent in one setting but influence each other when a shared factor changes. For example, repeated attempts by one learner may share knowledge, fatigue or task difficulty.
An AI should not automatically multiply success rates across repeated school tasks as though the attempts were independent. The relationship needs evidence or an explicit hypothetical independence assumption.
30. A transfer exercise with a coin
For two independent fair coin flips, the chance of two heads is 1/2 × 1/2 = 1/4, or 25%. The physical coin is not removed from a shrinking population after the first flip.
This resembles replacement sampling in its independent probability structure. It does not prove all real coin-flipping methods are perfectly fair or independent; the usual classroom answer rests on the stated model.
31. A transfer exercise with two students
Suppose two students each have an estimated 80% chance of completing a practice task. If those completion events were genuinely independent, the chance both complete is 0.8 × 0.8 = 64%.
But the students may be in the same class, face the same technical outage or collaborate. Independence cannot be assumed solely because there are two different people. A model should make the assumption explicit.
32. Why risk calculations also need dependence checks
An agent may rely on several sources or services. If all depend on one network connection, their failures can occur together. Multiplying separate success probabilities as though failures were independent can underestimate shared risk.
Understanding the dependency structure is therefore part of responsible AI planning. A neat equation with invalid independence assumptions can yield a misleading confidence statement.
33. A practical classroom activity
Place three red and two blue counters in a bag. Ask learners to predict two-draw outcomes with replacement, then without replacement. Have them draw probability trees before running any demonstration.
Compare the theoretical probabilities with a modest number of actual trials, while emphasising that the observed frequencies can vary. The aim is to connect a clearly defined procedure with its mathematical model.
34. What AI builders should test
Include with-replacement and without-replacement examples, ordered and unordered events, conditional questions and missing procedure descriptions. Check whether the model asks for the replacement rule when it changes the answer.
Score arithmetic separately from task interpretation. A response can multiply fractions correctly while following the wrong tree, and that should be recognised as a meaningful error.
35. The central lesson
Independent events allow simple multiplication of unchanging probabilities. Dependent events require attention to what the first event changes and to the resulting conditional probabilities. The difference can be visible in one very small bag of counters.
Before accepting an AI-generated chance, ask whether the earlier event affects the later one, what is known at each step and which ordered outcomes the question includes. That makes probability a checkable explanation rather than a confident guess.
Frequently asked questions
- Does replacement matter? — Yes. Returning the first counter restores the original bag composition.
- What is independence? — Knowing one event does not change the probability of the other under the defined model.
- Why is RR 36% with replacement but 30% without? — The second red chance falls from 3/5 to 2/4 after a red is removed.
- Is at least one red the same as exactly one red? — No. At least one also includes two red.
- Can the second red chance be 60% unconditionally when draws are dependent? — Yes, in this symmetric no-replacement model.
- Does multiplying probabilities always work? — Use conditional probability unless independence is justified.
- Are these actual school outcome frequencies? — No. They are theoretical figures for fictional rules.
Connected learning and source notes
- How Super Intelligence Works — main learning curriculum.
- Conditional Probability and Base Rates — recognising the conditioning event.
- Probability and Confidence — why a percentage is not a guarantee.
- Sets and Overlap — AND, OR and mutually exclusive membership.
- OpenStax Introductory Statistics: Independent and Mutually Exclusive Events.
- OpenStax: Two Basic Rules of Probability.