Skip to main content
Loading Explore guides…
Intuitive Probability Mastery · Acedra
Home
Explore
Mathematics
Intuitive Probability Mastery
Sign in to save
Chat
If YouTube asks you to sign in, open the video on YouTube.
Watch on YouTube
Reload player
Transcript
112 segments
Copy
Hide video
0:04
You're not bad at probability. You just haven't been taught this. See, everyone teaches probability backwards. They start with formulas, rules, axioms. Your brain memorizes them for a test. Then poof, gone. You're left with nothing. But here's what they should have taught you first. Your brain already understands probability intuitively. You don't need formulas for that.
0:28
So, in the next hour, we're doing something different. We're building probability the way your brain actually works. Intuition first, formalization second. Check out these topics. We'll master sample spaces, conditional probability, B rule, law of total probability, expected value, random variables. Everything that seemed impossible will feel obvious. By minute 60,
0:49
you'll understand probability better than most statistics graduates. Here's the deal. After this video, there's an hour of practice problems waiting. Your brain will struggle. Good. That struggle is literally rewiring your neural pathways. No pain, no gain. Let's transform how you think about probability starting now. [Music] Let's start your transformation. You've seen this notation everywhere. P of E equals 0.5. You know P stands for probability. But what exactly is E? What
1:31
are we actually calculating the probability of? Here's what 90% of people get completely wrong about probability and it breaks everything else they try to learn. Most people think E is an outcome, like getting heads or drawing a red card. They think probability is about single things happening. That's fundamentally wrong. And this misconception is why probability feels so confusing. It's not
2:07
about outcomes, it's about events. So what's an event? Let me show you with an example that will rewire how you think. You've got this bag with 10 marbles. Now watch carefully. Each marble is unique. Let's number them. Red one, red two, red three, red four, red five, blue one, blue two, blue three, green one, green two. When you reach in blindfolded and grab one, what could happen? You could grab red one or red two or blue one or any of the 10 marbles. Each individual marble has
2:52
a 1 in 10 chance of being grabbed. These 10 individual possibilities. That's your sample space. Literally every single thing that could happen. Now comes the breakthrough moment. When someone asks, "What's the probability of drawing a red marble?" They're not asking about one outcome. They're asking about multiple outcomes grouped together. Drawing red means you draw red one or you draw red two or you draw red three or you draw red four or you draw red five.
3:24
That's five different outcomes. Look at this. This collection of outcomes is an event. An event is a set of outcomes. Not one outcome, a whole set. So when we calculate P of red marble equals P of E, we're really doing number of outcomes in the event divided by total number of outcomes. That's 5 / 10, which equals 50%. Let's drive this home with coins. You flip two coins. Quick, what's your sample space? Think about all the ways those two coins could land. Heads,
3:56
heads, heads, tails, tails, heads, tails, tails. Four outcomes, not three. Heads, tails, and tails heads are different. The order matters because they're different coins. Now, the fun part. What events could we define? Exactly one head. That's head tails and tails heads. At least one head. That's head heads. Heads tails and tails heads. First coin is heads. That's head heads and head tails. Both coins match. That's head head and tails tails. Each event
4:28
is a different grouping of outcomes. So P of at least one head equals P of E equals 3 out of 4 which is 75%. We're counting how many outcomes are in our event? Three. How many total outcomes exist? Four. Done. This is the fundamental insight. Probability operates on events, sets of outcomes, not individual outcomes. Yes, you could treat a single outcome like heads
4:56
heads as a singleton event. That's why people get sloppy and say probability of heads. But that's just a special case where your set has one element. Master this distinction that E is a set, not a single thing and everything else in probability clicks into place. Miss it and you'll struggle forever. But hold on. How do we know this calculation is even valid? Here's something mindblowing. Every probability calculation in existence, from coin flips
5:25
to AI models to quantum mechanics, follows the same three fundamental laws. These aren't just rules someone made up. These are mathematical truths as fundamental as 2 + 2= 4. Without them, probability doesn't exist. The three axioms of probability. First, probabilities can't be negative. P of E is greater than or equal to zero. Common sense, right? Second, something must happen. P of S equals 1. The probability of the entire sample space is 100%. Third,
6:08
if events can't overlap, just add. For disjoint events, P of A union B equals P of A plus P of B.
6:21
Let me show you these laws in action with something we all know, a simple die roll. First, let's define our events. Event A rolling less than three, that's one and two. Event B rolling a five, that's just five. Now, watch how the axioms validate everything. Axiom 1, P of A= 2 / 6, which is 1/3. Is it non- negative? Obviously, yes. Axiom 2. The whole sample space 1 2 3 4 5 6 has probability 6 / 6 = 1. Something must happen. Axiom 3. A and B can't happen together. You can't
7:05
roll both less than three and five. So, P of A union B= P of A + P of B. That's 2 6 + 1 6 = 36, which is 12. The math just works. Here's your transformation moment. These three simple laws are the DNA of probability. Every complex formula you'll ever see is just these laws in disguise. From now on, every problem follows this pattern. Identify sample space, define events, apply the laws. Feel that clarity. That's your mathematical intuition growing stronger.
7:45
You're not just learning probability, you're rewiring how you think about uncertainty. Now that you understand events are sets, here's where things get really powerful. You're playing poker. You draw a card from a standard deck. What questions might you ask? Did I get a king? Sure, that's one event. But real life is messier. Did I get a king or an ace? Did I get a red card and a face card? Did I not get a
8:18
heart? These aren't simple events. They're combinations. And here's what most people don't realize. There's a beautiful mathematical structure for handling exactly these situations. Let me show you with actual cards. From our 52 card deck, let's define event A. Draw a king. That's four cards. Event B, draw a heart. That's 13 cards. Now watch what happens when we combine these events in different ways. The not operation complement. Not a king means everything except
8:56
the kings. That's 48 cards out of 52. So the probability of not getting a king. 48 out of 52. But here's the elegant shortcut. Since we know there's a 4 out of 52 chance of getting a king, we can flip it around. The probability of not getting a king. Well, that's just 1 minus 450 seconds, which gives us 4850 seconds. This is the complement rule in action. Whatever the chance of something happening, the chance of it not happening is just 1 minus that number. Why does
9:32
this work? Because either A happens or it doesn't. Those are the only two possibilities and they must add to one. The and operation intersection king and heart. Which cards are both? Just this one card king of hearts. The intersection is where the two events overlap. Only the king of hearts satisfies both conditions. So the probability of getting both king and heart just one out of 52. The ore operation union king or heart. Now, this is where 80% of people mess up. Your instinct
10:09
says four kings plus 13 hearts equals 17 cards. Simple addition, right? Wrong. Look at this card. Is it a king? Yes. Is it a heart? Also, yes. We counted the king of hearts twice. Once as a king, once as a heart. So the real answer, four kings plus 13 hearts minus one overlap equals 16 cards. The probability of getting king or heart 16 out of 52. Let me draw what mathematicians call a
10:45
ven diagram to show you exactly what's happening. First I'll draw a rectangle. This represents our entire sample space, all 52 cards. Now inside I'll draw two circles. Circle A contains all four kings. Circle B contains all 13 hearts. But wait, these circles overlap. The king of hearts sits right in that overlapping region. Aha moment. See it. Now the king of hearts is in both circles. That's why we counted it twice. This is why we need here's the addition rule doing its magic.
11:27
Let me show you how this works with our cards. The chance of getting a king, four out of 52. The chance of getting a heart, 13 out of 52. And that one card that's both the king of hearts, just one out of 52. Now watch what happens when we calculate king or heart. We take our four kings, add our 13 hearts. But wait, we counted the king of hearts twice. Once as a king, once as a heart. So we subtract that one card we double counted. 4 + 13 - 1 equals
12:03
16. 16 cards out of 52 satisfy our condition. The formula automatically fixes the double counting. Special case. What if events can't overlap like drawing a king versus drawing a queen? Well, a card can't be both a king and a queen, right? So, there's zero overlap. When that happens, the formula becomes beautifully simple. You just add the probabilities together. Four kings plus four queens equals eight cards. Easy. No double counting to worry
12:37
about. Let's use all three tools. In a deck, what's P of red card or face card?
12:50
So, let's figure this out together. In a deck, what counts as red? Well, you've got all your hearts and all your diamonds. That's 26 cards total. And face cards, those are your jacks, queens, and kings across all four suits. So, 12 cards there. But wait, here's where it gets interesting. Some cards are both red and face cards, right? Think about it. the Jack of Hearts, Queen of Hearts, King of Hearts, Jack of Diamonds, Queen of Diamonds,
13:23
King of Diamonds, that's six cards that we'd be counting twice if we're not careful. So, when we work out the probability of getting either a red card or a face card, we take our 26 red cards, add our 12 face cards, but then, and this is crucial, we subtract those six cards we counted twice. That gives us 32 cards out of 52 that satisfy our condition. Do the math and you get about 61 12%. See how that addition rule automatically fixes
13:57
our double counting problem. Another example, what's the probability of getting at least one head in three coin flips? Instead of listing all possibilities, use the compliment. Think about it this way. What's the only way to not get at least one head? You'd have to get all tails, right? Tails, tails, tails. What's the chance of that? Well, each flip has a 1/2 chance of tails, and you need all three to be tails. So, that's 1/2 * 1/2 * 1/2, which gives us 1/8. So,
14:32
if there's a 1/8 chance of getting no heads, then there must be a 7/8 chance of getting at least one head. One calculation instead of seven. That's the power of thinking and compliments. Your transformation toolkit is now complete. When you see not in a problem, that's your signal to use the compliment. When you see and, you're looking for the intersection, what satisfies both conditions. When you see or, time for the addition rule, but always watch for
15:06
that overlap. Feel that clarity. You're thinking like a probabilist now. But wait, we just counted 32 cards out of 52. That was manageable. What happens when the numbers get astronomical? You're trying to crack a fourdigit lock. Quick question. How many combinations do you need to try in the worst case? Most people guess, well, four digits each can be 0 to 9, so maybe a few hundred. Actually, it's 10,000 possibilities from 000000 to 99999. That's 10 * 10 * 10 * 10,
15:54
which equals 10^ 4th power. Now, imagine a 6 character password using lowercase letters. That's 26 to the 6th power. Over 300 million possibilities. Here's the problem we face in probability. Our formula says P of E equals favorable outcomes divided by total outcomes. But when there are millions or billions of outcomes, you can't list them all. You'd die before finishing. So, how do we count when the numbers get astronomical?
16:31
Look at these three situations. First, creating a four-digit pin like 1 2 3 4 or 555. Second, selecting four lottery balls from 40. No repeats allowed. Third, choosing a threeperson committee from 10 people. They all involve choosing or selecting, but the counting is completely different. Why? The secret lies in two simple questions that change everything. First question, can I repeat? Can I use the same thing twice? Second question, does order matter?
17:13
Is ABC different from CBA? Let me show you why these questions change everything. First question, can I repeat? Think about a four-digit pin code. Can you use the same digit twice? Of course, you could have 1 one one or 1 2 3 4 or 555 when you can repeat. First digit 10 choices 0 through 9. Second digit still 10 choices. Third digit still 10 choices. Fourth digit still 10 choices. Total
17:46
10 * 10 * 10 * 10 equals 10,000 possible pins. Now think about a lottery drawing four balls from 10 numbered balls. Once a ball is drawn, it's gone. You can't repeat when you can't repeat. First ball, 10 choices. Second ball, only nine left. Third ball, only eight left. Fourth ball, only seven left. That's about 5,000 possible drawings. See the massive difference? With repetition allowed, you get 10,000. Without repetition, only about 5,000. That's why this question matters.
18:30
Quick math note. Factorials. You might see this in the formulas. Factorial just means multiply all the numbers down to one. So 4 factorial, that's 24. That's all there is to it. Here are the three formulas you need. Check out these formulas. When you can repeat and order matters, just multiply n by itself k times. Example, fourdigit pin, that's 10,000 possibilities. Formulas 2 and three, no repetition. This is where the second question comes in. Does order matter?
19:11
When order matters, use permutations. Think about this. You're choosing a president, vice president, and secretary from 10 people. Alice is president. Bob is VP. Charlie is secretary. Charlie is president. Alice as VP, Bob is secretary. These are different outcomes. The roles matter. When order matters, we use permutations. The calculation for president, you have 10 choices. For VP, nine left. For secretary, eight left.
19:44
That's 720 different ways. Look at the formula, but honestly, just remember multiply down K times when order doesn't matter. Use combinations. Now, same club, but you're choosing a threeperson committee. No special roles. Choosing Alice, Bob, and Charlie. choosing Charlie, Alice, and Bob. These are the same committee. Order doesn't matter here. Here's the key insight. Every committee can be arranged six different ways,
20:28
but they're all the same committee. So, we take our 720 permutations and divide by the redundancy.
20:41
That gives us 120 committees. Think of it as permutations divided by redundancy. The power move your decision tree. Before any counting problem, ask can I repeat? If yes and order matters, multiply in by itself k times. If no, go to question two. Does order matter? If yes, use permutation for arrangements, rankings, passwords. If no, use combination for groups, committees, hands of cards.
21:32
Quick test. A password needs four different digits. How many possible passwords? Think. Can digits repeat? No. It says different. Does order matter? Yes, order definitely matters for passwords. So, it's a permutation. Multiply down four times and you get about 5,000 passwords. You just solved a complex counting problem in 10 seconds. That transformation, that's the power of asking the right questions. But so far,
22:13
we've been calculating probabilities in a static world where we know nothing has happened yet. What about when things start happening and we get new information? [Music]
22:29
The weather app says 30% chance of rain today. You're about to leave home without an umbrella, but then you look outside. Dark clouds everywhere. Wind picking up. Does that 30% still apply? Of course not. The probability just changed based on what you observed. Let me show you exactly how this works with something we can calculate. You draw two marbles from this bag without replacement. Meaning once you draw a marble, it's gone.
23:05
First marble blue. Now the critical question. What's the probability the second marble is red? Think about what just happened. Before drawing anything, there were 10 marbles total. But now there are only nine marbles left in the bag. And crucially, all five red marbles are still there. So the probability of getting red second given we drew blue first. Five out of nine. That vertical bar means given that. It's how we
23:40
write conditional probability. Notice something profound here. The probability of drawing red changed from 1/2 to 5 9ths. The blue marble being gone literally changed our universe. This is the intuition behind conditional probability. When you gain information, you're working in a new smaller universe. Check out the formula. The probability of A given B equals the probability of both happening divided by the probability of B. But why does this formula
24:17
work? Let me show you. Think about all the ways to get first blue then red. That probability is 16th. And the probability of drawing blue first 3/10. So when we divide we get 5 9ths. exactly what we calculated intuitively. The formula is just the mathematical way to shrink the universe. Here's what the formula really does. Think about it this way. The top part of the fraction counts all the ways both things happen together. The bottom part that counts all the ways your
24:58
condition happens. So when you divide them, you're basically asking out of all the times B happened, how often did A also happen? It's like zooming in on just the B cases and seeing what fraction of those also have A. Let's see this with cards. You draw a card and see it's a king. What's the probability it's specifically the king of hearts? The intuitive way is to think about your new universe. You've got just the four kings now. And out of those four kings,
25:27
how many are the king of hearts? just one. So the probability is 1/4. Using the formula, the king of hearts is one out of 52 cards total, while any king is four out of 52. When you divide these, you get 1/4. Same answer. See how the formula automatically does the shrinking. The denominator represents our new smaller universe of just kings.
25:51
Practice problem. A family has two children. You know, at least one is a girl. What's the probability both are girls? Let's think step by step. All the possibilities for two children are boy boy, boy girl, girl boy, and girl girl. But wait, we know at least one is a girl. So that rules out boy. That leaves us with boy girl, girl boy, and girl girl. Now of these three remaining possibilities, how many have both girls?
26:19
Just the one. Girl, girl. So the probability is 1/3, not 1/2. The condition changes everything.
26:35
your transformation. You now understand that probability isn't fixed. It updates as you gain information. Every time you learn something new, you're essentially working in a smaller, more focused sample space. This is the foundation of how AI learns, how doctors diagnose, and how we make decisions under uncertainty. Feel that power growing.
27:00
So far, we've been dealing with single events. But what happens when you want to track two things at the same time? Let me show you with a simple example. Imagine you're tracking students in a class. For each student, you record two things. Whether they pass the midterm, yes or no, and whether they pass the final, yes or no. Out of 100 students, here's what you found. 50 students passed both exams. 10 students passed
27:29
midterm but failed final. Five students failed midterm but passed final. 35 students failed both. Look at this table. It shows joint probabilities, the probability of both things happening together. Half the students passed both exams. But here's where it gets powerful. From this same table, you can calculate marginal probabilities. the probability of just one event ignoring the other. What's the overall probability of passing the midterm? Add up all the cases where midterm
28:03
was passed. Passed both 50 students. Passed midterm, failed final, 10 students. Total who passed midterm? 60 out of 100. That's 60%. That's a marginal probability. It's called marginal because you literally write it in the margin of your table. [Music] Check out the complete picture. The joint probabilities live in the cells of the table. Those are the combinations of both events. Meanwhile, the marginal probabilities sit in the margins where we write the totals. That's
28:39
actually why they're called marginal. They're literally in the margins. This isn't just for exams. Anytime you have two events, medical test results and actual disease, weather and accidents, age and voting preference, you can use a joint probability table to see the complete picture. Now that you understand joint and marginal probabilities, here's a profound question. How can you tell if two events are connected or independent?
29:11
Look at our exam data. The marginal probability of passing the midterm is 60%. The marginal probability of passing the final is 55%. If these exams were completely independent, if performance on one had nothing to do with performance on the other, then we'd expect about 33% to pass both. But our actual data shows 50% passed both. That's way higher than expected. This tells us the exams are not independent. Students who do well on the midterm tend to also do well on the final.
29:55
This is the independence test. If the joint probability equals the product of the marginalss, they're independent. If not, they're dependent. Now, let's see true independence. You roll a dieice and flip a coin simultaneously. Think about it. Does the die care what the coin shows? Does the coin care about the die? Of course not. They're completely separate. Let's create a joint probability table. We observe 600 trials. Now calculate the marginalss. Probability of rolling sixth.
30:40
Probability of getting heads 1/2. Probability of both 112th. Check for independence. Does 16 * 1/2 equal 112th? Yes, they're independent. When events are independent, the multiplication rule emerges naturally. Just multiply the individual probabilities. But watch what happens when events do affect each other. Now let's build a joint probability table for drawing two cards without replacement. When we work out
31:17
all the probabilities for drawing two cards, something interesting happens. The marginal probability of getting a king on any single draw about 7.7%. If the draws were independent, we'd expect both kings about 0.59% of the time. But our actual calculation gives us about 0.45%. They're not equal. The events are dependent. Drawing the first card changes the probabilities for the second. So how do you know if events are independent? Well,
31:51
there are three equivalent ways to check. First way, see if the joint probability equals the product of the marginalss. Second way, check if condition on one event doesn't change the probability of the other. And the third way, which is really intuitive, ask yourself if knowing one thing changes your belief about the other. If it doesn't, they're independent. Let me give you an example that trips people up. Someone might say, "Having a cold and wearing a
32:21
jacket are independent, right?" Let's check with data. Winter observations out of 10,00 people. Marginals show 40% have colds, 70% wear jackets. If independent, we'd expect 28% to have both, but actually 30% have both. They're not independent. Why? Both are connected by season. Winter makes you more likely to be cold and more likely to wear a jacket. They're connected through the hidden factor of season. True independence means A doesn't cause B,
32:58
B doesn't cause A, and there's no hidden factor affecting both. Your new superpower. Before calculating any and probability, always check if the events are independent first. If they are independent, life is easy. Just multiply the probabilities. But if they're dependent, you'll need to use conditional probability instead. Independence isn't just a math concept. It's how we understand causation, correlation, and the structure of reality itself. When you
33:29
master this, you're thinking like a scientist. [Music] All right, here's where probability becomes truly powerful. Up until now, we've been dealing with outcomes like heads, red marble, or king of hearts. But what if we want to work with actual numbers? Let me show you what I mean. Let's say we're playing a simple game. You flip two coins. For every head you get, you win $5. For every tail, you lose $2. Now, the sample space for flipping
33:58
two coins is still the same. Heads, heads, heads, tails, tails, heads, tails, tails. But instead of just caring about the pattern, we care about how much money you make or lose. This is where random variables come in. A random variable is just a function that assigns a number to every outcome in your sample space. Think of it as a translator that converts random outcomes into numbers we can
34:20
actually work with. Let's call our random variable X and it represents your total winnings. Let's work through each outcome. Heads heads two heads. So you win $10. Heads tails one head one tail. So you win five but lose two. That's $3. Tails heads same thing $3. tails, tails, two tails. So, you lose $4. Now, look what we've done. We've taken this random process and turned it into something mathematical. Instead of talking about heads and tails, we're talking about the numbers 10,
34:57
3, and -4. Here's the beautiful part. We can now ask mathematical questions. What's the probability you win exactly $3? Well, that happens with heads, tails, or tails heads. Two out of four outcomes, so 50%. What's the probability you make money? That's three out of four outcomes, so 75%. Random
35:19
variables are everywhere, the number of cars passing through an intersection in an hour, your test score, the time it takes for your pizza to arrive, the temperature tomorrow, stock prices. All of these are random variables. The power is that once we have numbers, we can use all the tools of mathematics. We can add them, multiply them, take averages, and much more. We've just transformed the messy world of randomness into the clean world of numbers. All right, here's where
35:47
probability gets practical. Once you have a random variable, the single most important question you can ask is what's the average value in the long run? This is called the expected value, and it's everywhere. When a casino sets the odds on a game, they're using expected value. When an insurance company decides your premium, expected value. When you're deciding whether to take an umbrella based
36:11
on weather forecasts, you're intuitively thinking about expected value. Let me show you how this works with a simple game. You roll a dieice. If you roll a six, you win $12. Roll a four or five, win $3. Roll 1 through three, lose $6. Should you play this game? Well, let's figure out the expected value. The expected value is the weighted average of all outcomes. Probability of rolling a
36:39
six 1 and six win $12. Probability of four or five 1/3 win $3. Probability of 1 through 3 1/2 lose $6. When we calculate the weighted average, we get exactly $0. So the expected value of this game is exactly zero. This means if you played this game thousands of times, you'd expect to break even on average. Sometimes you'd win, sometimes you'd lose, but over the long haul, you'd neither gain nor lose money. This is what we call a fair game. Most casino games, by the way,
37:19
have negative expected values for players and positive expected values for the house. That's how casinos make money. But expected value isn't just about gambling. Let's say you're trying to decide whether to buy insurance for your phone. The phone costs $800. There's a 5% chance you'll break it this year. The insurance costs $100. Without insurance, your expected loss is $40. With insurance, you definitely lose $100. So, from a pure expected value standpoint, insurance doesn't
37:52
make sense. But here's the thing. Expected value doesn't capture everything. It doesn't account for how much losing $800 would hurt you personally. That's where the concept gets more nuanced. The beautiful thing about expected value is that it gives you a single number to compare different random situations. It's the center of gravity of all possible outcomes. And once you start seeing
38:17
the world through the lens of expected value, you'll make better decisions in everything from investments to career choices to daily life. But what if we want to add two random variables and compute the expectation of their sum? Let me show you with a simple example. You roll a dieice with this payout. Roll one through three, win $4. Roll four through six, win $8. Now, here's the twist. We're tracking two random variables from this single game. X is your winnings in dollars.
38:49
Y is the number you rolled. The expected value of X, let's see, $4 half the time, $8 the other half. That gives us $6. The expected value of Y, well, that's just the average of 1 through six, three and a half. So, what about X + Y? Your winnings plus the number you rolled. Let me walk through what happens. If you roll a one, you win $4 plus the one you rolled, that's five total. Roll a two. 4 + 2 makes six. A three gives you seven. But then it
39:23
jumps. Roll a four and suddenly you're getting $8 + 4, 12 total. A 5 13 and a 6 14. So our possible outcomes are 5 6 7 12 13 and 14. The average 9 and 1/2. Wait a minute. E X is 6. E Y is 3 and 1/2. And 6 + 3 and 1/2 is 9 and 1/2. This is not an accident. You might be thinking, "Hold on, this feels wrong. X and Y are completely dependent. If I know one, I automatically know the other. How can their expectations just add up like that?" Here's what's about to blow your mind.
40:05
Expected values always add, even when variables are completely dependent. Don't believe me? Look closer at what's really happening. See what I'm doing? I'm separating every outcome into its X part and Y part. The X contributions 4 44 888. The Y contributions 1 2 3 4 5 6. Now, here's the magic trick. Average each part separately. Average of the X parts. That's four three times and eight three times. divide by six. $6. That's exactly our expected value of x. Average of the y parts.
40:41
1 + 2 + 3 + 4 + 5 + 6ide by 6. 3 and 1/2. That's exactly our expected value of y. The dependencies between x and y are completely irrelevant. When we break everything apart and average separately, we automatically get back to the original expectations. This is linearity of expectation. One of the most powerful shortcuts in all of probability. Check this out. Even though X and Y are completely dependent on each other, when we break apart the math and regroup it, the
41:14
dependencies automatically wash out. It's like the math doesn't care about the dependencies. It just naturally separates things back to their original form. Important. This only works for addition. For multiplication, you need independence. So far, we've been dealing with things you can count. Coin flips, dice rolls, number of customers. But what about things you measure? Height, weight, time, temperature. These don't come in neat little packages. They flow continuously.
41:49
Let me show you what I mean. Imagine you have an automated bread slicer that's supposed to cut loaves exactly 16 in long. But like any machine, it's not perfect. Sometimes it cuts a little long, sometimes a little short. Let's say the actual length varies uniformly between 14 and 18 in. What does uniformly mean? It means every length in that range is equally likely. The machine isn't biased toward any particular length within that range. Now,
42:21
here's the mindbending part. What's the probability that a loaf is exactly 1600 0 in long? Not 16.1, not 15.9, but exactly 16 with infinite precision. The answer is zero. Absolutely zero. I know that sounds crazy, but think about it. There are infinitely many possible lengths between 14 and 18 in. Any specific exact value has zero probability because it's one point out of infinitely many. But here's what we can calculate. The probability
43:00
that a loaf is between say 15 and 17 in long. That's like asking for the area under a curve between those two points. Since our distribution is uniform from 14 to 18 in, we can think of it as a rectangle. The total width is 4 in. The height has to be 1/4 because the total area under any probability curve must equal 1. So to find the probability that a loaf is between 15 and 17 in, we calculate the area of a rectangle that's 2 in wide and 1/4 tall. That's 50%. [Music]
43:47
This is the fundamental shift when dealing with continuous variables. Let me introduce some terminology here because it's important. For discrete variables like dice rolls, we use something called a PMF probability mass function. We can ask what's the probability of getting exactly six. But for continuous variables, we use a PDF, probability density function. We can't ask for exact values because that's always zero. Instead, we ask about ranges.
44:28
Now, don't panic if you haven't learned integration in calculus yet. Integration is just a fancy way of computing the area under a curve. When we want the probability that our bread length is between 15 and 17 in, we're finding the area under the PDF curve between those two points. In our uniform distribution example, that curve is just a flat rectangle. So the integration is simple multiplication, width time height. But for more complex distributions, the curves can
45:03
be bell-shaped or have other interesting forms. It's like the difference between asking what's the chance of hitting this exact point on a dart board versus what's the chance of hitting somewhere in this region. The first is essentially zero. The second is meaningful and calculated by finding the area of that region. So to summarize, discrete variables use PMFs and we add up probabilities.
45:34
Continuous variables use PDFs and we integrate to find areas. Same intuition, different mathematical tools. [Music] Continuous variables are everywhere in the real world. Stock prices, arrival times, measurement errors, biological traits. And once you understand that we're dealing with PDFs and integration instead of PMFs and addition, a whole new world of mathematical modeling opens up. The beautiful thing is that the same probabilistic intuition applies. We're still asking about
46:14
likelihood. We're still calculating expected values. We're still thinking about distributions. We've just moved from a world of discrete chunks to a world of smooth curves. All right, here's where probability becomes a problem-solving superpower. Sometimes you face a probability question that seems impossible to calculate directly. But there's an elegant technique that breaks complex problems into manageable pieces.
46:42
Let me show you what I mean. Imagine you have two bags of marbles. Bag A has five red marbles and five blue marbles. perfectly balanced. Bag B has one red marble and seven blue marbles, heavily biased toward blue. Now, here's the setup. You randomly choose one of the bags. There's a 70% chance you pick bag A and a 30% chance you pick bag B. You then draw one marble from whichever bag you chose. The question is, what's the overall probability of drawing a blue marble?
47:15
This might seem tricky at first, but here's the insight. There are exactly two ways you can end up with a blue marble. Either you pick bag A and draw blue from it, or you pick bag B and draw blue from it. These are the only two paths to your desired outcome. So, let's calculate each path. What's the probability you pick bag A and then draw blue? Well, you've got a 70% chance of picking bag A. And if you do, there's a 50% chance of drawing blue from it.
47:45
That gives us 35% for this path. What about the other path? Picking bag B and then drawing blue. That's 30% chance of picking bag B times, let's see, seven out of eight marbles are blue. So about 87.5%. That's about 26% for this path. Since these are the only two ways to get blue, we add them up. 35% plus 26% equals about 61% total. This technique is called the law of total probability.
48:18
The formal way to write it involves summing over all possible scenarios. But don't get caught up in the notation. The intuition is simple. Break the problem into all the different ways your outcome can happen. Calculate the probability of each path, then add them all up. Think of it like this. You're trying to get to a destination, but there are multiple routes. Each route has its own probability of being taken and its own probability of success once you're on it. The
48:46
total probability is just the sum of all the root probabilities. This shows up everywhere. In medical diagnosis, you might get symptoms through different diseases, each with different base rates. In quality control, defective products might come from different machines with different failure rates. In finance, market crashes might happen through different economic scenarios. The law of total probability is your Swiss Army knife for complex probability problems.
49:12
When direct calculation seems impossible, ask yourself, what are all the different ways this outcome could happen? Calculate each pathway, add them up, and you've got your answer. Now, we come to the crown jewel of probability theory. Baze rule is the mathematical foundation of how we learn, how AI works, and how we update our beliefs when we get new information.
49:34
Let me start with a story that will shock you. A disease affects only 1% of the population. There's a test for this disease that's 99% accurate. Meaning, if you have the disease, the test will be positive 99% of the time. And if you don't have the disease, the test will be negative 99% of the time. You take the test and it comes back positive. What's the probability you actually have the disease? Most people, including doctors, would guess around 99%. After all,
50:10
the test is 99% accurate, right? The real answer will blow your mind. It's only about 50%. Let me show you why. Imagine we test 1,000 people. Since the disease affects 1% of the population, about 10 people actually have it and 990 people don't. Of those 10 people who have the disease, the test will correctly identify about 10 of them as positive. But here's the kicker. Of the 990 people who don't have the disease, the test will incorrectly identify about 10 of them as positive,
50:49
too. That's the 1% false positive rate. So out of everyone who tests positive, we have about 10 true positives and about 10 false positives. That means only about half of the positive tests actually indicate the disease. This is BA's rule in action. It's teaching us something profound. The accuracy of a test isn't the same as the probability you have the condition given a positive test. You
51:21
have to account for the base rate. how common the condition is in the first place. All right, let's connect this back to our bag problem. Remember the setup. We have two bags. Bag A is perfectly balanced with five red and five blue marbles. Bag B is blue biased with only one red and seven blue marbles. We said there was a 70% chance we'd pick bag A and a 30% chance we'd pick bag. From that, we calculated the overall probability of drawing a blue marble is 61.25%.
51:52
But now let's flip the question. If someone reaches into a bag and pulls out a blue marble, what's the probability it came from bag A? This is the exact kind of question that BA's rule is built to answer. Here is the formula in its formal glory. It says the probability of A given B is equal to the probability of B given A times the probability of A all divided by the probability of B. Now, I know that can look a little intimidating. So let's apply it to our
52:21
problem because it's much more intuitive than it looks. We have three key pieces of information. Our prior belief, this is our starting point. The initial chance of picking bag A was 70%. The likelihood, this is the chance of our evidence happening given our prior belief. So the probability of drawing a blue marble from bag A is 50%. and the evidence. This is the total chance of drawing a blue marble from either bag, which we already figured out is 61.25%.
52:53
Now, we just plug those values in. The probability that we have bag A, given that we drew a blue marble, is 50% * 70%, all divided by 61.25%. That simplifies to 35% divided by 61.25%, which gives us our answer, approximately 57.1%. Now, notice what just happened here. We started out with a 70% confidence that we were holding bag A. But the moment we saw the new evidence, the blue marble, our confidence dropped to 57.1%. Why did it go down? Because drawing a blue marble
53:30
is actually stronger evidence that you might be holding bag B since it's so heavily biased towards blue. Seeing blue is evidence in favor of bag B. If the formula feels a bit clunky, there's a more intuitive way to think about it. We calculated that the specific pathway of choosing bag A and then drawing a blue marble was 35%. We also knew the total chance of getting any blue marble was
53:55
61.25%. So the probability that it came from bag A is just the specific pathways probability divided by the total evidence. 35% divided by 61.25%. It's the exact same calculation. This right here is the essence of learning. You start with prior beliefs. You observe evidence and you update your beliefs accordingly. This isn't just for abstract thought experiments. It's how spam filters learn what is
54:26
and isn't junk. It's how recommendation systems work. It's even how medical AI diagnoses diseases by constantly updating its beliefs as it gets more information. Baze rule is the mathematical formalization of rational thinking. It tells us exactly how much we should update our beliefs when we get new information in a world full of data and uncertainty. That makes it one of the most powerful tools humanity has ever discovered. All right, here's where probability gets mindbending.
54:56
Sometimes you encounter problems where the answer depends on the answer itself. These are called recursive problems and they show up in some of the most elegant places in probability. Let me show you what I mean with a simple question that's actually quite deep. How many coin flips do you expect to need before you get your first heads? Now, you might think this is impossible to calculate because theoretically you could flip tails forever. The sample space is infinite.
55:29
heads, tails, heads, tails, tails, heads, and so on. How do you add up infinitely many possibilities? Actually, it is mathematically possible to add up infinite sequences like this. Even though there are infinite outcomes, each has exponentially decreasing probability. The probability of getting heads on the first flip is 1/2. Tails then heads is 1/4. Tails, tails, then heads is 1/8, and so on. The infinite sum actually converges to exactly one,
55:58
which makes sense since something has to happen eventually. But calculating that infinite sum directly would be tedious. Here's where recursion gives us an elegant shortcut. Instead of trying to enumerate every possibility, let's think recursively. Let's assume that we know the expected number of flips to get the first heads. And let's call this unknown value e. Here's the recursive insight. We can define E in terms of itself by thinking about
56:29
what happens on the very first flip. You flip once. That always costs you exactly one flip. Now, there are two possibilities. Either you get heads or you get tails. If you get heads, which happens with probability 1/2, you're done. You needed exactly one flip total. If you get tails, which also happens with probability 1/2, you're right back where you started. You still need to get your first heads and you still expect that to take e more flips.
57:01
So here's the equation. E equals one flip that you definitely take plus 1/2 * 0 additional flips if you get heads plus 12 * e additional flips if you get tails. That gives us e = 1 + 0 + e / 2. simplifying e = 1 + e / 2. Now we solve for e. Subtract e / 2 from both sides and you get e minus e / 2 = 1. That's e / 2 = 1. So e= 2. You expect to need exactly two flips to get your first heads. Even though you could theoretically flip forever, the average is just two.
57:47
Let's push this further. What if you want two heads in a row? Now we need to think about states. Let E be the expected flips starting from scratch. Let E subH be the expected additional flips after you've just gotten one head. Starting from scratch, you flip once. If you get tails, you're back to the start and expect e more flips. If you get heads, you're in the one head state and expect E subH more flips. So
58:20
E = 1 + 1/2 * E + 1/2 * E subH. From the one head state, you flip once. If you get heads again, you're done. If you get tails, you're back to square 1. So e subh = 1 + 12 * 0 + 12 * e. [Music] From the second equation, e subh= 1 + e / 2. Substituting into the first equation and solving, you get e= 6. You expect to need six flips to get two heads in a row. Recursive power. The power of recursion is that it turns impossible infinite sums
59:12
into simple algebra. Instead of trying to calculate every possible sequence, you just identify the states, figure out how they connect to each other, and solve the resulting equations. This technique shows up everywhere in analyzing algorithms, in modeling random walks, in game theory, in queuing systems. Once you master recursive thinking and probability, you've unlocked a whole new level of problem solving power.
59:42
All right, we've done it. We've completely rebuilt your understanding of probability from the ground up. You've gone from thinking probability was about memorizing formulas to understanding that it's about building mathematical intuition. You now know how to identify sample spaces and events. You can use set operations like a pro.
1:00:01
You understand conditional probability and independence. You can handle random variables, joint probabilities, and expected values. You've seen the power of BA's rule, and the elegance of recursive thinking. But here's the brutal truth I told you at the beginning. You're going to forget this tomorrow. Not because you're not smart, not because the material is too hard,
1:00:22
but because that's how the human brain works. Without practice, without repetition, without that systematic training we talked about, these concepts will fade back into Swiss cheese knowledge. Remember what I promised you at the start, the hour-long practice video, the systematic training, the no pain, no gain approach that builds lasting understanding. Here's what you need to do right now. Click on this practice video. This isn't optional. This is
1:00:50
where the real learning happens. You're going to work through dozens of problems. You're going to feel that mental burn we talked about. Your brain is going to struggle and you're going to think, "This is hard." And that's exactly when you know you're growing. You practice every day. At first, it feels difficult, but then something magical happens. It gets easier and easier. Finally, these concepts become part of your mathematical intuition forever. The choice is yours.
1:01:17
You can close this video and let everything fade away. Or you can click that practice video and transform how you think about uncertainty, data, and the random world around us. The concepts are in your head right now. But will they still be there tomorrow? That depends on what you do in the next 10 seconds. Click the practice video, feel the burn, build the intuition, transform your mathematical thinking forever. Your future self will thank you.
Here to help you learn
Sign in to keep a private conversation with your course tutor.
Ask me anything about the material…
Sign in to chat