WEBVTT 00:00:00.000 --> 00:00:03.919 Think about an hourglass for a second. Flip it 00:00:03.919 --> 00:00:06.900 over. The sand falls grain by grain. And you 00:00:06.900 --> 00:00:08.820 can just look at the little pile at the bottom 00:00:08.820 --> 00:00:10.980 and know exactly how much time has passed. Right. 00:00:11.140 --> 00:00:13.179 It's neat. It's totally predictable. Yeah, it 00:00:13.179 --> 00:00:16.179 makes perfect sense. But what if you need to 00:00:16.179 --> 00:00:18.699 predict something completely chaotic, like, I 00:00:18.699 --> 00:00:20.079 don't know, when a startup is going to suddenly 00:00:20.079 --> 00:00:23.239 go bankrupt, or when an airplane engine will 00:00:23.239 --> 00:00:25.339 fail mid -flight? Or even when your own body 00:00:25.339 --> 00:00:28.100 will finally recover from a stubborn virus. Exactly. 00:00:28.239 --> 00:00:30.480 In those situations, that perfect little hourglass 00:00:30.480 --> 00:00:33.000 is, well, It's completely useless. Oh, absolutely. 00:00:33.140 --> 00:00:35.240 The sand isn't falling neatly anymore. It's like 00:00:35.240 --> 00:00:37.700 it's blowing around in a hurricane. You're dealing 00:00:37.700 --> 00:00:42.259 with a reality that is muddy and random and just 00:00:42.259 --> 00:00:45.359 incredibly difficult to map. on a standard calendar. 00:00:45.740 --> 00:00:48.500 Which brings us to our mission for today's deep 00:00:48.500 --> 00:00:51.259 dive. We are taking a single, honestly incredibly 00:00:51.259 --> 00:00:54.520 dense Wikipedia article about a statistical concept. 00:00:54.740 --> 00:00:56.979 It's called the first hitting time model. A bit 00:00:56.979 --> 00:00:59.479 of a mouthful, yeah. It really is. But we're 00:00:59.479 --> 00:01:02.500 going to show you how this obscure page is actually 00:01:02.500 --> 00:01:05.719 this hidden mathematical blueprint. It predicts 00:01:05.719 --> 00:01:07.980 exactly when the unpredictable is going to happen 00:01:07.980 --> 00:01:09.900 in our lives. Right. So whether you're managing 00:01:09.900 --> 00:01:12.379 a business or trying to understand how insurance 00:01:12.379 --> 00:01:15.159 premiums are calculated. or you're just insanely 00:01:15.159 --> 00:01:18.099 curious about the hidden forces shaping reality, 00:01:18.659 --> 00:01:21.219 you are really going to want to hear this. Because 00:01:21.219 --> 00:01:24.359 what looks like pure abstract algebra is actually 00:01:24.359 --> 00:01:27.319 a lens. I mean, it's the ultimate tool for making 00:01:27.319 --> 00:01:31.849 invisible chaotic journeys visible. and for understand 00:01:31.849 --> 00:01:34.250 the precise moment a system hits its breaking 00:01:34.250 --> 00:01:36.650 point. OK, let's untack this. What exactly are 00:01:36.650 --> 00:01:38.370 we looking at when we talk about a first hitting 00:01:38.370 --> 00:01:42.590 time model or a first passage time model, as 00:01:42.590 --> 00:01:44.609 it's sometimes called? Well, the core idea is 00:01:44.609 --> 00:01:46.769 that it's a statistical model estimating the 00:01:46.769 --> 00:01:49.510 amount of time that passes before a random or 00:01:49.510 --> 00:01:52.890 what we call a stochastic process crosses a specific 00:01:52.890 --> 00:01:54.730 boundary. But what does that actually look like 00:01:54.730 --> 00:01:57.129 in practice, like for someone not looking at 00:01:57.129 --> 00:01:59.569 a whiteboard full of equations? Yeah, totally. 00:01:59.629 --> 00:02:02.590 The easiest way to picture it is through a classic 00:02:02.590 --> 00:02:05.310 scenario that mathematicians just love, which 00:02:05.310 --> 00:02:08.569 is Gamber's Ruin. Gambler's ruin sounds intense. 00:02:08.629 --> 00:02:11.830 It is. Yeah, so put yourself in a smoky casino 00:02:11.830 --> 00:02:14.289 You're watching a gambler to blackjack table 00:02:14.289 --> 00:02:16.569 and their bankroll is constantly bouncing up 00:02:16.569 --> 00:02:18.550 and down They win a few hands. They lose a few 00:02:18.550 --> 00:02:21.189 hands. So their wallet is just fluctuating wildly. 00:02:21.289 --> 00:02:23.789 Exactly It's a stochastic process. That just 00:02:23.789 --> 00:02:25.789 means it has a random probability distribution 00:02:25.789 --> 00:02:28.729 that can be analyzed statistically But you can't 00:02:28.729 --> 00:02:31.110 predict it precisely hand -by -hand got it and 00:02:31.110 --> 00:02:34.550 the first hitting time That is the exact microscopic 00:02:34.550 --> 00:02:37.870 moment the bankroll hits zero. Bankruptcy, the 00:02:37.870 --> 00:02:40.729 threshold has been crossed, and, well, the game 00:02:40.729 --> 00:02:43.009 is over. I see where this is going, but wait, 00:02:43.229 --> 00:02:45.330 isn't this just predicting a streak of bad luck? 00:02:45.669 --> 00:02:48.090 I mean, how is this different from standard probability? 00:02:48.389 --> 00:02:50.650 How so? Well, like, if I play roulette long enough, 00:02:50.689 --> 00:02:52.150 I know the house edge is going to take all my 00:02:52.150 --> 00:02:55.009 money eventually. A basic math textbook could 00:02:55.009 --> 00:02:57.810 tell me my chances of losing. Right, but standard 00:02:57.810 --> 00:03:00.050 probability only tells you the overall chance 00:03:00.050 --> 00:03:02.250 of ruin. It says, you know, if you play this 00:03:02.250 --> 00:03:04.770 game, you have a 90 % chance of going broke. 00:03:04.889 --> 00:03:07.469 OK. What the first hitting time model does is 00:03:07.469 --> 00:03:10.050 entirely different. It models the actual time 00:03:10.050 --> 00:03:13.409 until that ruin happens in a continuous, evolving 00:03:13.409 --> 00:03:17.250 journey. It tracks that messy up and down path 00:03:17.250 --> 00:03:19.629 of the bankroll as it literally wanders toward 00:03:19.629 --> 00:03:22.030 the zero dollar threshold. Oh, wow. So it's tracking 00:03:22.030 --> 00:03:25.569 the journey itself. Yes. It's actually a subclass 00:03:25.569 --> 00:03:27.590 of survival models because you are measuring 00:03:27.590 --> 00:03:30.620 the lifespan of that specific state. You know, 00:03:30.620 --> 00:03:32.740 the state of having money. That makes a lot more 00:03:32.740 --> 00:03:34.719 sense. It's about the when, not just the if. 00:03:35.159 --> 00:03:37.120 And looking at the history here in our source, 00:03:37.599 --> 00:03:39.939 mathematicians have been obsessed with this for, 00:03:39.939 --> 00:03:42.479 well, over a century. Oh, easily. You've got 00:03:42.479 --> 00:03:44.680 researchers like Bachelier in 1900 trying to 00:03:44.680 --> 00:03:46.500 figure out the chaotic fluctuations of the French 00:03:46.500 --> 00:03:49.300 stock market. And then physicists like Schrodinger 00:03:49.300 --> 00:03:52.159 around 1915 looking at random particle movement. 00:03:52.719 --> 00:03:55.340 It's wild that the exact same math applies to 00:03:55.340 --> 00:03:59.759 both. It is wild. And let's not forget insurance. 00:04:00.080 --> 00:04:04.580 In 1903, an actuary named Philip Lundberg used 00:04:04.580 --> 00:04:07.419 this exact framework to model the probability 00:04:07.419 --> 00:04:09.620 of an insurance company facing financial ruin. 00:04:09.919 --> 00:04:11.759 Wait, really? Insurance companies were using 00:04:11.759 --> 00:04:15.439 this back in 1903. Yeah. He needed to know exactly 00:04:15.439 --> 00:04:18.439 when the random influx of premium payments would 00:04:18.439 --> 00:04:21.259 be outpaced by the random outflow of massive 00:04:21.259 --> 00:04:23.800 accident claims. It's the exact same concept. 00:04:24.019 --> 00:04:26.019 OK, so if we want to actually build one of these 00:04:26.019 --> 00:04:28.920 models to predict, say, a chain breaking down 00:04:28.920 --> 00:04:31.420 or a company failing, what do we need? Like, 00:04:31.459 --> 00:04:33.480 what are the actual building blocks here? You 00:04:33.480 --> 00:04:36.079 generally need three underlying components. First, 00:04:36.180 --> 00:04:38.779 you need a parent stochastic process. So that's 00:04:38.779 --> 00:04:40.740 the actual thing that is fluctuating. Second, 00:04:40.800 --> 00:04:42.660 you need a threshold, which is the barrier or 00:04:42.660 --> 00:04:44.920 the cliff you're waiting for it to hit. OK, process, 00:04:45.040 --> 00:04:46.720 threshold. What's the third? And third, you need 00:04:46.720 --> 00:04:49.480 a time scale. Gotcha. Well, let's focus on that 00:04:49.480 --> 00:04:52.000 first one, the parent stochastic process. Because 00:04:52.000 --> 00:04:53.920 there's a specific word used in the Wikipedia 00:04:53.920 --> 00:04:56.160 article that honestly changes how you think about 00:04:56.160 --> 00:04:58.379 this entirely. It says this process is usually 00:04:58.379 --> 00:05:01.279 latent. Yeah. What's fascinating here is that 00:05:01.279 --> 00:05:04.480 latent means hidden, unobservable. So you can't 00:05:04.480 --> 00:05:06.420 even see it. Right. You can't see it directly 00:05:06.420 --> 00:05:09.160 with the naked eye. A common mathematical example 00:05:09.160 --> 00:05:12.360 is a Wiener process, which has a mean, a variance, 00:05:12.480 --> 00:05:15.139 and a starting value. But in the real world, 00:05:15.560 --> 00:05:17.560 you rarely get to just sit back and watch this 00:05:17.560 --> 00:05:20.279 smooth mathematical curve unfold. Wait, latent 00:05:20.279 --> 00:05:23.180 process still sounds a bit, um, abstract. Are 00:05:23.180 --> 00:05:24.699 we talking about something like metal fatigue 00:05:24.699 --> 00:05:27.100 in a car engine? Where the wear and tear is just 00:05:27.100 --> 00:05:29.240 happening invisibly over years? Yeah, it is the 00:05:29.240 --> 00:05:32.160 perfect example. Picture a massive steel gear 00:05:32.160 --> 00:05:36.000 inside a factory machine. Every single day, microscopic 00:05:36.000 --> 00:05:38.220 stress fractures are forming inside that solid 00:05:38.220 --> 00:05:41.079 metal. But you can't see them. Exactly. The machine 00:05:41.079 --> 00:05:43.639 sounds fine. It looks fine. That is your latent 00:05:43.639 --> 00:05:46.819 unobservable process, the silent accumulation 00:05:46.819 --> 00:05:49.360 of damage. You only see the result when the gear 00:05:49.360 --> 00:05:52.019 suddenly snaps in half. And that snap is the 00:05:52.019 --> 00:05:54.560 threshold, the barrier. You've got it. The first 00:05:54.560 --> 00:05:57.699 hitting time is the exact millisecond that unseen 00:05:57.699 --> 00:06:00.459 latent degradation finally hits the structural 00:06:00.459 --> 00:06:03.160 breaking point. And that is that's kind of terrifying. 00:06:03.370 --> 00:06:06.250 It can be, and because researchers can't see 00:06:06.250 --> 00:06:08.990 those microscopic fractures forming, they basically 00:06:08.990 --> 00:06:11.750 have to play detective. How do they do that? 00:06:11.889 --> 00:06:14.910 They use what's called censored time -to -event 00:06:14.910 --> 00:06:18.290 data, or they look at correlated marker processes, 00:06:18.730 --> 00:06:20.949 like maybe the machine vibrates just a tiny bit 00:06:20.949 --> 00:06:24.050 more than usual. They use those clues to reconstruct 00:06:24.050 --> 00:06:26.910 that invisible journey to the threshold. Which 00:06:26.910 --> 00:06:29.769 brings us to the third component, the clock, 00:06:30.290 --> 00:06:32.980 the time scale. This honestly blew my mind when 00:06:32.980 --> 00:06:34.839 I was reading the source, because I always just 00:06:34.839 --> 00:06:37.079 assumed time is time. A day is a day, a year 00:06:37.079 --> 00:06:39.540 is a year. You do. Most people do. But the math 00:06:39.540 --> 00:06:42.079 here says time scales don't have to be calendar 00:06:42.079 --> 00:06:45.600 time. They can be, quote, operational time. How 00:06:45.600 --> 00:06:48.540 does that even work? So operational time completely 00:06:48.540 --> 00:06:50.839 reshapes how you measure a lifespan. Instead 00:06:50.839 --> 00:06:52.720 of a clock on the wall ticking away seconds, 00:06:53.379 --> 00:06:55.779 operational time only ticks forward when a system 00:06:55.779 --> 00:06:58.180 is actively experiencing change or degradation. 00:06:58.459 --> 00:07:01.220 Oh, I see. Think about the mileage on your car. 00:07:01.899 --> 00:07:04.120 If you leave your car parked in a garage for 00:07:04.120 --> 00:07:07.019 an entire year, calendar time has passed. But 00:07:07.019 --> 00:07:09.920 operational time, like the accumulated wear and 00:07:09.920 --> 00:07:12.439 tear on the engine, the degradation of the tires 00:07:12.439 --> 00:07:15.519 has stood perfectly still. So a three -year -old 00:07:15.519 --> 00:07:18.199 taxi cab might have a completely different operational 00:07:18.199 --> 00:07:20.519 age than a 10 -year -old sports car that just 00:07:20.519 --> 00:07:23.180 sits in a showroom. Exactly. Now, apply that 00:07:23.180 --> 00:07:26.290 to human biology. Take a 40 -year -old construction 00:07:26.290 --> 00:07:28.810 worker who spends 10 hours a day carrying heavy 00:07:28.810 --> 00:07:31.410 loads versus a 40 -year -old office worker who 00:07:31.410 --> 00:07:34.189 just sits at a desk. Okay. In calendar time, 00:07:34.589 --> 00:07:37.910 their knees are the exact same age. But in operational 00:07:37.910 --> 00:07:40.610 time, the construction worker's knees are decades 00:07:40.610 --> 00:07:43.709 older because the invisible stochastic process 00:07:43.709 --> 00:07:46.470 of cartilage degradation has been ticking forward 00:07:46.470 --> 00:07:49.370 at a massively accelerated rate. That is fascinating 00:07:49.370 --> 00:07:51.649 and the clock only ticks when the damage is being 00:07:51.649 --> 00:07:54.149 done or I guess conversely when the healing is 00:07:54.149 --> 00:07:55.810 being done. Right, it applies to everything. 00:07:56.310 --> 00:07:58.449 Accumulated exposure to toxic chemicals in a 00:07:58.449 --> 00:08:01.050 factory, the number of flight cycles on an airplane 00:08:01.050 --> 00:08:04.290 wing. If you use the wrong clock, your first 00:08:04.290 --> 00:08:06.350 hitting time model will be completely useless. 00:08:06.670 --> 00:08:09.350 Yeah, that makes total sense. Well, let's shift 00:08:09.350 --> 00:08:11.649 gears from factory floors back to the raw physics 00:08:11.649 --> 00:08:14.449 for a minute. Because to really grasp the mechanics 00:08:14.449 --> 00:08:16.949 of this, we have to look at one of the simplest 00:08:16.949 --> 00:08:20.329 and most omnipresent stochastic systems in existence. 00:08:20.509 --> 00:08:23.110 The one -dimensional Brownian particle. Yes, 00:08:23.389 --> 00:08:26.689 just a tiny microscopic particle moving randomly 00:08:26.689 --> 00:08:29.569 left or right. And Brownian motion is foundational 00:08:29.569 --> 00:08:31.449 here, because it describes how things spread 00:08:31.449 --> 00:08:33.570 out organically. Here's where it gets really 00:08:33.570 --> 00:08:37.039 interesting. To help visualize this, Let's use 00:08:37.039 --> 00:08:39.179 a slightly modified version of a classic analogy 00:08:39.179 --> 00:08:41.919 from the text. Imagine a ceramic cup of black 00:08:41.919 --> 00:08:44.620 coffee. Okay, I'm picturing it. You take a dropper 00:08:44.620 --> 00:08:47.519 and release a single concentrated drop of heavy 00:08:47.519 --> 00:08:50.340 cream right into the center of the mug. Initially, 00:08:50.580 --> 00:08:52.899 all that cream is contained in one tiny location. 00:08:52.919 --> 00:08:55.379 Right. But over time, the molecules randomly 00:08:55.379 --> 00:08:57.620 bump into each other and diffuse outward. And 00:08:57.620 --> 00:08:59.539 that spreading is governed by the one -dimensional 00:08:59.539 --> 00:09:02.490 diffusion equation. Without getting bogged down 00:09:02.490 --> 00:09:05.029 in the algebra, what the math shows is that the 00:09:05.029 --> 00:09:07.350 probability of finding a specific cream molecule 00:09:07.350 --> 00:09:10.769 at a given spot forms a Gaussian shape, a bell 00:09:10.769 --> 00:09:13.950 curve. And as time marches on, that bell curve 00:09:13.950 --> 00:09:17.769 gets wider and flatter, the peak drops, and the 00:09:17.769 --> 00:09:20.450 edges spread out further and further. Exactly. 00:09:20.909 --> 00:09:23.820 But remember, A first -hitting time model isn't 00:09:23.820 --> 00:09:26.539 just asking how the cream mixes. It wants to 00:09:26.539 --> 00:09:29.019 know when a threshold is crossed. Right, the 00:09:29.019 --> 00:09:31.039 barrier. So the question isn't, you know, where 00:09:31.039 --> 00:09:33.379 is the cream? The question is, at what exact 00:09:33.379 --> 00:09:36.519 millisecond does the very first molecule of cream 00:09:36.519 --> 00:09:39.379 randomly bump against the ceramic wall of the 00:09:39.379 --> 00:09:41.990 coffee mug? The ceramic wall is the barrier, 00:09:42.309 --> 00:09:44.590 the cliff, as mathematicians call it. We want 00:09:44.590 --> 00:09:47.370 to find the exact moment of impact and not a 00:09:47.370 --> 00:09:50.149 single moment before. To find that, physicists 00:09:50.149 --> 00:09:52.370 calculate what's called the first passage time 00:09:52.370 --> 00:09:55.059 density. They look at the survival probability, 00:09:55.220 --> 00:09:57.100 which is just the mathematical chance that the 00:09:57.100 --> 00:09:58.919 molecule has managed to wander around in the 00:09:58.919 --> 00:10:01.519 coffee without ever touching the ceramic wall. 00:10:01.639 --> 00:10:04.220 OK, survival probability. Got it. By tracking 00:10:04.220 --> 00:10:06.440 the exact rate at which that survival probability 00:10:06.440 --> 00:10:08.899 drops, they pinpoint the moment of impact. OK, 00:10:08.899 --> 00:10:10.279 hold on. You're losing me here a bit because 00:10:10.279 --> 00:10:11.960 I was reading through the section on the math 00:10:11.960 --> 00:10:14.620 behind this first passage time density, and it 00:10:14.620 --> 00:10:17.620 yields a truly bizarre quirk. Oh, the heavy tail. 00:10:17.799 --> 00:10:20.840 Yes. It says the motion follows a heavy tailed 00:10:20.840 --> 00:10:23.940 levy distribution. And because of that, if I 00:10:23.940 --> 00:10:26.279 ask you to calculate the average time it takes 00:10:26.279 --> 00:10:29.259 for a particle to hit that ceramic wall, the 00:10:29.259 --> 00:10:32.129 answer is infinite. That doesn't make any sense. 00:10:32.730 --> 00:10:35.690 Calculating averages is basic middle school math. 00:10:35.789 --> 00:10:38.049 You add up all the times, you divide by the number 00:10:38.049 --> 00:10:40.990 of particles. How on earth can an average be 00:10:40.990 --> 00:10:42.750 infinite? I know, it breaks your brain a little 00:10:42.750 --> 00:10:45.850 bit. But here is the aha moment of the entire 00:10:45.850 --> 00:10:48.669 mathematical framework. Because it's a random 00:10:48.669 --> 00:10:51.929 walk in an open space, the math allows for a 00:10:51.929 --> 00:10:54.629 tiny, tiny chance that a particle wanders off 00:10:54.629 --> 00:10:56.710 in the exact opposite direction of the wall. 00:10:56.830 --> 00:10:58.850 Just floating toward the center of the mug forever. 00:10:58.909 --> 00:11:01.799 Right. theoretically wander off and take a near 00:11:01.799 --> 00:11:04.360 eternity to finally randomly bounce back and 00:11:04.360 --> 00:11:07.240 hit the cliff. Wow. Even if it's a one in a trillion 00:11:07.240 --> 00:11:10.139 chance, factoring eternity into your data set 00:11:10.139 --> 00:11:12.779 completely breaks the math. In statistical terms, 00:11:12.960 --> 00:11:14.940 the first moment of the distribution diverges. 00:11:15.360 --> 00:11:17.340 The heavy tail pulls the mathematical average 00:11:17.340 --> 00:11:21.019 out to infinity, and, well, you just cannot average 00:11:21.019 --> 00:11:24.000 infinity. That is mind -bending. A single particle 00:11:24.000 --> 00:11:26.299 wandering off into the void ruins the average 00:11:26.299 --> 00:11:29.009 for everything else. So what do physicists do? 00:11:29.149 --> 00:11:30.690 They can't just throw their hands up and say, 00:11:30.990 --> 00:11:33.230 well, the math is broken. Time is a flat circle. 00:11:33.389 --> 00:11:36.090 No, they pivot. If they can't calculate the average 00:11:36.090 --> 00:11:38.950 time, they calculate the typical time. Typical 00:11:38.950 --> 00:11:41.389 time. What's the difference? This is the moment 00:11:41.389 --> 00:11:44.509 when the first passage time density reaches its 00:11:44.509 --> 00:11:47.309 absolute maximum peak. it's the most likely time 00:11:47.309 --> 00:11:49.350 for the impact to happen. And there's a specific 00:11:49.350 --> 00:11:51.830 formula for that, right? Yes. The typical time 00:11:51.830 --> 00:11:53.789 which mathematicians represent with the Greek 00:11:53.789 --> 00:11:57.129 letter tau is calculated as the distance to the 00:11:57.129 --> 00:12:00.389 cliff squared divided by a multiple of the diffusion 00:12:00.389 --> 00:12:03.710 constant. Okay, distance squared over the diffusion 00:12:03.710 --> 00:12:05.830 constant. Yeah, and the diffusion constant is 00:12:05.830 --> 00:12:08.289 basically just an indirect measure of how fast 00:12:08.289 --> 00:12:10.990 the particle is wandering. Distance squared over 00:12:10.990 --> 00:12:13.769 speed. It gives you a highly accurate, usable 00:12:13.769 --> 00:12:16.549 number without letting that one wandering particle 00:12:16.549 --> 00:12:19.309 drag your math into infinity. That is brilliant. 00:12:19.509 --> 00:12:21.629 OK, so we've talked about gamblers losing their 00:12:21.629 --> 00:12:24.009 shirts. We've talked about invisible wear and 00:12:24.009 --> 00:12:26.549 tear on car engines. And we've talked about cream 00:12:26.549 --> 00:12:29.409 hitting the edge of a coffee mug. But so what 00:12:29.409 --> 00:12:32.009 does this all mean? How does this abstract physics 00:12:32.009 --> 00:12:35.169 concept actually impact human beings outside 00:12:35.169 --> 00:12:38.529 of a laboratory? It impacts you because over 00:12:38.529 --> 00:12:41.549 the last two decades, this field has experienced 00:12:41.549 --> 00:12:44.470 a massive practical evolution through something 00:12:44.470 --> 00:12:47.080 called threshold regression. Okay, let's unpack 00:12:47.080 --> 00:12:49.860 that. Threshold regression takes these theoretical 00:12:49.860 --> 00:12:53.039 first -hitting -time models and equips them with 00:12:53.039 --> 00:12:56.100 explanatory variables, covariates. Exactly. Because 00:12:56.100 --> 00:12:58.480 real life isn't a vacuum. A patient's health 00:12:58.480 --> 00:13:00.740 isn't just a random particle floating in a void, 00:13:00.980 --> 00:13:02.940 you know? Right. There are external factors pushing 00:13:02.940 --> 00:13:05.500 that particle. Threshold progression allows scientists 00:13:05.500 --> 00:13:08.259 to inject real -world data into the model. They 00:13:08.259 --> 00:13:10.799 can add covariates to the time scale or to the 00:13:10.799 --> 00:13:13.320 parameters of the unseen process or even to the 00:13:13.320 --> 00:13:15.419 threshold itself. This connects deeply to something 00:13:15.419 --> 00:13:17.940 called the Cox proportional hazards model, right, 00:13:17.980 --> 00:13:20.399 which the source mentions is widely used in medical 00:13:20.399 --> 00:13:24.059 research. Yes, very widely used. By adding these 00:13:24.059 --> 00:13:26.779 explanatory variables, researchers can see how 00:13:26.779 --> 00:13:30.039 specific behaviors or traits speed up or slow 00:13:30.039 --> 00:13:32.580 down the journey to the cliff. Let's look at 00:13:32.580 --> 00:13:36.299 how this is actively saving lives. Take osteoporosis, 00:13:36.559 --> 00:13:38.759 which is a disease that severely weakens bones. 00:13:39.379 --> 00:13:41.220 Researchers use threshold regression to predict 00:13:41.220 --> 00:13:43.399 the exact time a patient is likely to suffer 00:13:43.399 --> 00:13:45.740 a severe bone fracture. And the way they model 00:13:45.740 --> 00:13:48.059 this is just brilliant. They model it as a shock 00:13:48.059 --> 00:13:50.500 stream superimposed on progressive degradation. 00:13:50.740 --> 00:13:53.179 Let's break that down. The progressive degradation 00:13:53.179 --> 00:13:55.779 is the latent, unobservable process we talked 00:13:55.779 --> 00:13:59.480 about. Deep inside the patient's body, bone mineral 00:13:59.480 --> 00:14:02.519 density is slowly, stochastically dropping over 00:14:02.519 --> 00:14:05.120 years. That's the invisible wear and tear. The 00:14:05.120 --> 00:14:07.720 operational time clock ticking away. Yes. And 00:14:07.720 --> 00:14:09.820 the shock stream represents the sudden impacts, 00:14:10.220 --> 00:14:12.259 stumbling on a rug, bumping into a table, just 00:14:12.259 --> 00:14:14.940 a minor fall. Oh, I see. The model analyzes how 00:14:14.940 --> 00:14:17.559 the hidden degradation lowers the threshold barrier 00:14:17.559 --> 00:14:20.370 over time. until a minor shock that wouldn't 00:14:20.370 --> 00:14:22.710 hurt a healthy person suddenly shatters the weakened 00:14:22.710 --> 00:14:25.470 bone. By adding covariates like the patient's 00:14:25.470 --> 00:14:27.990 diet, their vitamin D levels, or their exercise 00:14:27.990 --> 00:14:30.990 habits, doctors can predict when that threshold 00:14:30.990 --> 00:14:33.549 will be critically vulnerable and actually intervene 00:14:33.549 --> 00:14:35.769 before the fracture ever happens. It's incredible, 00:14:36.029 --> 00:14:38.470 and it's used across the board. The source mentions 00:14:38.470 --> 00:14:40.570 it predicts recurrent exacerbations in patients 00:14:40.570 --> 00:14:44.659 with COPD, you know, chronic lung disease, a 00:14:44.659 --> 00:14:47.600 doctor can input smoking history or air quality 00:14:47.600 --> 00:14:49.940 as a covariate to see how it accelerates the 00:14:49.940 --> 00:14:52.299 operational clock toward a lung failure threshold. 00:14:52.500 --> 00:14:54.539 Right. It's a huge predictive tool. It's used 00:14:54.539 --> 00:14:56.659 to model the one -year risk of death in patients 00:14:56.659 --> 00:14:59.899 with cystic fibrosis. It's even used in occupational 00:14:59.899 --> 00:15:02.679 health to predict the exact time frame for when 00:15:02.679 --> 00:15:04.600 someone will be able to return to work after 00:15:04.600 --> 00:15:07.740 a traumatic limb injury. And if you zoom in from 00:15:07.740 --> 00:15:10.830 the human scale down to the cellular level, There's 00:15:10.830 --> 00:15:13.169 this fascinating biophysics application called 00:15:13.169 --> 00:15:15.549 the narrow escape problem. The narrow escape 00:15:15.549 --> 00:15:17.149 problem? I love that name. It sounds like an 00:15:17.149 --> 00:15:19.210 action movie. How does that work? It really does. 00:15:19.629 --> 00:15:21.830 So it calculates the time it takes for a tiny 00:15:21.830 --> 00:15:25.009 particle like an ion or a protein to escape through 00:15:25.009 --> 00:15:28.730 a microscopic narrow opening in a confined cellular 00:15:28.730 --> 00:15:31.730 space. Okay, so a tiny room with a tiny door. 00:15:32.070 --> 00:15:34.889 Exactly. The particle is wandering stochastically 00:15:34.889 --> 00:15:37.820 inside the cell, bouncing off the walls and The 00:15:37.820 --> 00:15:40.480 threshold isn't a solid barrier this time. It's 00:15:40.480 --> 00:15:44.299 a tiny exit door. Calculating that first hitting 00:15:44.299 --> 00:15:47.179 time is crucial for understanding how our nerves 00:15:47.179 --> 00:15:50.120 communicate and how viruses spread between cells. 00:15:50.480 --> 00:15:52.360 It really makes you realize how universal this 00:15:52.360 --> 00:15:54.799 framework is. But I will say, everything we've 00:15:54.799 --> 00:15:58.679 talked about, ruin, fractures, death, machines, 00:15:59.100 --> 00:16:03.299 snapping, it all sounds incredibly bleak. It 00:16:03.299 --> 00:16:05.200 sounds like we are all just ticking time bombs 00:16:05.200 --> 00:16:07.440 wandering toward disaster. I get that. And that 00:16:07.440 --> 00:16:09.519 is a very common misconception. We naturally 00:16:09.519 --> 00:16:11.440 focus on negative outcomes because they demand 00:16:11.440 --> 00:16:14.159 our attention. But the math itself is completely 00:16:14.159 --> 00:16:15.940 neutral. A threshold is just a state change. 00:16:15.980 --> 00:16:17.980 It doesn't have to be a tragedy. Right. The critical 00:16:17.980 --> 00:16:20.399 endpoint can also be a positive event. Absolutely. 00:16:21.019 --> 00:16:23.139 If we connect this to the bigger picture, the 00:16:23.139 --> 00:16:25.360 threshold can represent a patient's immune system 00:16:25.360 --> 00:16:28.700 finally overwhelming a bacterial infection. The 00:16:28.700 --> 00:16:30.600 cliff can be the moment a pregnant woman goes 00:16:30.600 --> 00:16:33.120 into labor, childbirth. Oh, sure. It can be a 00:16:33.120 --> 00:16:35.679 discharge from a hospital stay. In the statistical 00:16:35.679 --> 00:16:38.220 sense, the time leading up to it is generically 00:16:38.220 --> 00:16:41.980 called a survival time. But in reality, you might 00:16:41.980 --> 00:16:44.120 just be surviving the wait until something wonderful 00:16:44.120 --> 00:16:46.659 happens. That's a great way to reframe it. And 00:16:46.659 --> 00:16:48.899 sometimes the system is racing toward multiple 00:16:48.899 --> 00:16:50.940 outcomes at once, right? The threshold isn't 00:16:50.940 --> 00:16:54.120 just one cliff. It's a set of multiple competing 00:16:54.120 --> 00:16:57.940 states. Competing first hitting times. It's a 00:16:57.940 --> 00:17:01.279 race of unseen variables. Will the factory machine's 00:17:01.279 --> 00:17:04.000 engine overheat first, or will its drive belt 00:17:04.000 --> 00:17:07.519 snap first? Will a patient recover from surgery, 00:17:07.880 --> 00:17:10.200 or will they develop an infection? So it's a 00:17:10.200 --> 00:17:12.920 race. Yeah, the stochastic processes are all 00:17:12.920 --> 00:17:15.440 wandering simultaneously, and the model helps 00:17:15.440 --> 00:17:17.799 predict which threshold will be hit first. Okay, 00:17:17.859 --> 00:17:20.259 let's pull all of this together. We've gone on 00:17:20.259 --> 00:17:22.500 quite a journey today. We started in a casino 00:17:22.500 --> 00:17:24.859 watching a gambler bounce toward bankruptcy. 00:17:25.059 --> 00:17:27.920 We did. We watched a drop of cream slowly make 00:17:27.920 --> 00:17:30.279 its way to the edge of a coffee mug, completely 00:17:30.279 --> 00:17:32.660 breaking the concept of mathematical averages 00:17:32.660 --> 00:17:35.420 along the way. Poor infinity. Right. And we've 00:17:35.420 --> 00:17:37.759 seen how threshold regression takes this deeply 00:17:37.759 --> 00:17:41.099 abstract physics and applies it to human biology, 00:17:41.539 --> 00:17:43.380 predicting when weakened bones will fracture, 00:17:43.819 --> 00:17:46.559 when lungs will fail, and when cells will communicate. 00:17:46.640 --> 00:17:49.640 All of it united by a single, elegant mathematical 00:17:49.640 --> 00:17:52.759 framework, the first hitting time model. And 00:17:52.759 --> 00:17:55.200 the real takeaway here isn't just the math. It's 00:17:55.200 --> 00:17:57.480 how this framework forces you to completely re 00:17:57.480 --> 00:18:00.049 -evaluate the world around you. How so? Like 00:18:00.049 --> 00:18:02.710 in day -to -day life? Yeah. When you look at 00:18:02.710 --> 00:18:04.769 an event now, whether it's a global financial 00:18:04.769 --> 00:18:07.190 crash, a piece of infrastructure collapsing, 00:18:07.690 --> 00:18:10.869 or your own body finally beating a cold, you 00:18:10.869 --> 00:18:13.609 are no longer just seeing a random isolated incident 00:18:13.609 --> 00:18:16.230 that fell out of the sky. Right. You are witnessing 00:18:16.230 --> 00:18:20.369 the final visible moment of a latent stochastic 00:18:20.369 --> 00:18:22.490 process that has been quietly running in the 00:18:22.490 --> 00:18:24.269 background for a long time. You are seeing the 00:18:24.269 --> 00:18:26.450 culmination of a journey that finally hit its 00:18:26.450 --> 00:18:28.859 threshold. It makes the invisible visible. It 00:18:28.859 --> 00:18:31.059 reminds us that very few things happen suddenly. 00:18:31.160 --> 00:18:33.299 They happen eventually. But it leaves me with 00:18:33.299 --> 00:18:35.839 one final provocative thought. Let's hear it. 00:18:35.940 --> 00:18:38.460 It goes back to that brilliant concept of operational 00:18:38.460 --> 00:18:41.559 time. The idea that time scales don't have to 00:18:41.559 --> 00:18:44.859 be uniformly ticking clocks on a wall. They can 00:18:44.859 --> 00:18:47.420 be accumulated wear and tear. They can be accumulated 00:18:47.420 --> 00:18:50.000 exposure. The clock only ticks when you're changing. 00:18:50.359 --> 00:18:54.400 Exactly. So if our lives and our bodies are governed 00:18:54.400 --> 00:18:57.299 by these latent, unobservable processes inching 00:18:57.299 --> 00:19:00.779 toward threshold Well, what invisible operational 00:19:00.779 --> 00:19:03.339 clocks are ticking inside your life right now? 00:19:03.460 --> 00:19:05.680 That's a deep question. Are they ticking based 00:19:05.680 --> 00:19:07.579 on the chronic stress you carry in your shoulders? 00:19:07.819 --> 00:19:09.720 Are they ticking based on your daily exposure 00:19:09.720 --> 00:19:12.380 to new, challenging ideas? And what thresholds, 00:19:12.380 --> 00:19:14.559 good or bad, are you unknowingly wandering toward 00:19:14.559 --> 00:19:17.220 right at this very moment? It's a profound question 00:19:17.220 --> 00:19:19.480 to walk away with, because the calendar year 00:19:19.480 --> 00:19:21.920 might be the same for all of us, but your internal 00:19:21.920 --> 00:19:24.329 operational clocks are strictly your own. The 00:19:24.329 --> 00:19:26.549 hourglass is officially shattered. We're all 00:19:26.549 --> 00:19:29.130 just wandering particles waiting to hit our threshold. 00:19:30.069 --> 00:19:32.210 Thank you so much for joining us on this deep 00:19:32.210 --> 00:19:34.549 dive into the hidden math of when things happen. 00:19:35.029 --> 00:19:36.750 We hope you are leaving with a few new tools 00:19:36.750 --> 00:19:39.029 to see the invisible processes operating all 00:19:39.029 --> 00:19:41.789 around you. Until next time, keep questioning 00:19:41.789 --> 00:19:42.289 the clocks.