1 00:00:05,200 --> 00:00:08,840 Speaker 1: Why do we find movies so compelling given that they're 2 00:00:08,840 --> 00:00:13,319 Speaker 1: just a series of photographs flash rapidly. What does this 3 00:00:13,480 --> 00:00:18,159 Speaker 1: tell us about the slow speed of human brains? What 4 00:00:18,400 --> 00:00:22,120 Speaker 1: percentage of stock market trades are taken care of by 5 00:00:22,239 --> 00:00:25,919 Speaker 1: algorithms nowadays at timescales. 6 00:00:25,480 --> 00:00:28,000 Speaker 2: That humans could not even conceive of. 7 00:00:28,600 --> 00:00:31,440 Speaker 1: And what is it like to have the speed and 8 00:00:31,640 --> 00:00:37,080 Speaker 1: power of a computer and be dealing with slow humans. 9 00:00:39,880 --> 00:00:43,280 Speaker 1: Welcome to Inner Cosmos with me, David Eagleman. I'm a 10 00:00:43,360 --> 00:00:47,400 Speaker 1: neuroscientist and in these episodes we sail deeply into our 11 00:00:47,520 --> 00:00:51,920 Speaker 1: three pound universe to understand some of the most surprising 12 00:00:52,080 --> 00:01:06,600 Speaker 1: aspects of our lives. Today's episode is about speed, the 13 00:01:06,640 --> 00:01:10,800 Speaker 1: speed at which we humans operate and the speed at 14 00:01:10,800 --> 00:01:14,880 Speaker 1: which our machines operate, and the future of this as 15 00:01:14,920 --> 00:01:17,360 Speaker 1: the divergence grows even larger. 16 00:01:17,680 --> 00:01:18,720 Speaker 2: So let's start with this. 17 00:01:19,000 --> 00:01:24,039 Speaker 1: Imagine an extraterrestrial planet where they have an active society. 18 00:01:24,440 --> 00:01:28,520 Speaker 1: These aliens have politics and division of labor, and philosophers 19 00:01:28,520 --> 00:01:31,360 Speaker 1: and actors and athletes and artists, everything that makes up 20 00:01:31,680 --> 00:01:36,280 Speaker 1: an active society. Okay, So, now imagine a second planet, 21 00:01:36,360 --> 00:01:40,000 Speaker 1: Planet B. We land on this second planet in the 22 00:01:40,040 --> 00:01:44,600 Speaker 1: near future, and we discover nothing like a society there. Instead, 23 00:01:44,640 --> 00:01:47,520 Speaker 1: there are just things like trees, like these very old 24 00:01:47,600 --> 00:01:50,280 Speaker 1: growths that don't really do much of anything. And we 25 00:01:50,480 --> 00:01:52,720 Speaker 1: the astronauts, we stay there for a month, we take 26 00:01:52,720 --> 00:01:55,240 Speaker 1: a lot of measurements, and finally we blast off when 27 00:01:55,280 --> 00:01:58,160 Speaker 1: we go home. So two planets, one with an active 28 00:01:58,240 --> 00:02:02,600 Speaker 1: society and the other with nothing but tree like things. 29 00:02:03,120 --> 00:02:08,040 Speaker 1: Now imagine that these two descriptions are of the same planet. 30 00:02:08,120 --> 00:02:11,760 Speaker 1: What we have stumbled on is a society of creatures 31 00:02:12,320 --> 00:02:17,040 Speaker 1: that operate at a very different time scale from us. 32 00:02:17,360 --> 00:02:20,960 Speaker 1: They move so slowly that we just can't see it. 33 00:02:21,360 --> 00:02:25,320 Speaker 1: They have an active society, it just operates a trillion 34 00:02:25,480 --> 00:02:29,720 Speaker 1: times slower than ours. And the question is, would we 35 00:02:29,840 --> 00:02:34,720 Speaker 1: ever even notice that these tree people have a society 36 00:02:34,760 --> 00:02:38,400 Speaker 1: happening in what we would think of as super slow motion. 37 00:02:38,800 --> 00:02:42,360 Speaker 1: A conversation over a cup of coffee for them takes 38 00:02:42,600 --> 00:02:47,119 Speaker 1: eons for us. So would we ever decode their language? 39 00:02:47,160 --> 00:02:50,480 Speaker 1: Would we ever even have an opportunity to discover that 40 00:02:50,520 --> 00:02:53,560 Speaker 1: there's a whole world going on there, but at a 41 00:02:53,720 --> 00:02:58,400 Speaker 1: different temporal scale. So that's what today's episode is about. 42 00:02:58,800 --> 00:03:03,799 Speaker 1: Hidden landscapes, parts of the time domain that are totally 43 00:03:03,840 --> 00:03:06,799 Speaker 1: invisible to us. Now you've heard me talk in other 44 00:03:06,880 --> 00:03:10,400 Speaker 1: episodes about all the information in the world that we 45 00:03:10,520 --> 00:03:14,080 Speaker 1: just don't see. So is one example. We only see 46 00:03:14,520 --> 00:03:18,080 Speaker 1: a small fraction of the light that is out there, 47 00:03:18,320 --> 00:03:21,960 Speaker 1: and that little slice we call visible light. That's what 48 00:03:22,000 --> 00:03:25,799 Speaker 1: we think of as red, orange, yellow, green, blue, indigo, violet. 49 00:03:26,080 --> 00:03:28,799 Speaker 1: And it seems like that's all the light that's out there. 50 00:03:28,840 --> 00:03:31,840 Speaker 1: You open your eyes, you see reds and greens and 51 00:03:31,880 --> 00:03:34,640 Speaker 1: blues and golds, and you think, okay, that's everything. And 52 00:03:34,680 --> 00:03:39,240 Speaker 1: so it came as a surprise when scientists eventually realized 53 00:03:39,680 --> 00:03:43,200 Speaker 1: that all these other signals that had been detected, things 54 00:03:43,240 --> 00:03:46,560 Speaker 1: that are totally invisible to us, like infrared light or 55 00:03:46,680 --> 00:03:50,360 Speaker 1: radio waves or X rays or ultraviolet or microwaves, all 56 00:03:50,400 --> 00:03:54,040 Speaker 1: of these are light. They're exactly the same thing, is 57 00:03:54,040 --> 00:03:56,720 Speaker 1: what we call visible light. It's just that we don't 58 00:03:56,720 --> 00:03:59,840 Speaker 1: have the receptors in our eyes to pick up on them, 59 00:04:00,400 --> 00:04:03,360 Speaker 1: and so they're totally invisible to us. And it turns 60 00:04:03,400 --> 00:04:07,480 Speaker 1: out that what we call visible light is a very 61 00:04:07,680 --> 00:04:10,840 Speaker 1: tiny part of the light spectrum. It's less than a 62 00:04:11,080 --> 00:04:14,840 Speaker 1: ten trillionth of it. And once we understood this, we 63 00:04:14,920 --> 00:04:19,599 Speaker 1: started to build technologies to burrow into those other parts 64 00:04:19,640 --> 00:04:24,000 Speaker 1: of the spectrum. For example, we make little machines that 65 00:04:24,120 --> 00:04:27,920 Speaker 1: sit in the dashboards of our cars, and these see 66 00:04:28,080 --> 00:04:31,680 Speaker 1: in exactly the frequency that we call radio waves, and 67 00:04:31,720 --> 00:04:35,680 Speaker 1: so we can transmit information in that frequency. Or we 68 00:04:35,720 --> 00:04:40,520 Speaker 1: build night vision security cameras which see in the infrared range, 69 00:04:40,520 --> 00:04:43,560 Speaker 1: so that we can detect in that frequency outside of 70 00:04:43,560 --> 00:04:46,000 Speaker 1: what we can naturally see. And we build X ray 71 00:04:46,080 --> 00:04:49,839 Speaker 1: machines and telescope dishes, and these all pick up on 72 00:04:50,080 --> 00:04:54,120 Speaker 1: other frequencies that are invisible to us. We build all 73 00:04:54,240 --> 00:04:58,159 Speaker 1: kinds of devices to see what we can't naturally see, 74 00:04:58,200 --> 00:05:01,840 Speaker 1: and this has been the gen of scientific development over 75 00:05:01,880 --> 00:05:05,720 Speaker 1: the last few hundred years. It's the understanding that the 76 00:05:05,760 --> 00:05:10,640 Speaker 1: reality we inhabit is actually just a small fraction of 77 00:05:10,720 --> 00:05:13,960 Speaker 1: what is out there, and there are many other things 78 00:05:14,160 --> 00:05:18,159 Speaker 1: that can be taken advantage of. So we're very used 79 00:05:18,200 --> 00:05:23,320 Speaker 1: to communicating in landscapes that are outside our biology. Our 80 00:05:23,320 --> 00:05:27,680 Speaker 1: biology allows us to detect something like visible light, and 81 00:05:27,720 --> 00:05:29,720 Speaker 1: then we figure out from there that there are other 82 00:05:30,240 --> 00:05:33,680 Speaker 1: flavors of light that are beyond our biology, and we. 83 00:05:33,600 --> 00:05:34,880 Speaker 2: Take advantage of those. 84 00:05:35,560 --> 00:05:39,240 Speaker 1: So I want to use this as an analogy and 85 00:05:39,279 --> 00:05:42,680 Speaker 1: an inroad for us thinking about somewhere else that we 86 00:05:42,760 --> 00:05:46,080 Speaker 1: do this, and that is in the domain of time. 87 00:05:47,160 --> 00:05:50,280 Speaker 1: The question is can we use different ranges of time 88 00:05:50,400 --> 00:05:55,359 Speaker 1: to build technologies, perhaps a range of time that is 89 00:05:55,480 --> 00:05:59,920 Speaker 1: essentially invisible in our normal perception. When I was a tea, 90 00:06:00,760 --> 00:06:04,360 Speaker 1: I visited MIT and I met with Harold Edgerton who 91 00:06:04,360 --> 00:06:07,400 Speaker 1: was known as Doc Edgerton. And he was a pioneer 92 00:06:07,600 --> 00:06:11,680 Speaker 1: in high speed photography. And what he would do is 93 00:06:11,880 --> 00:06:16,480 Speaker 1: capture moments that were otherwise totally invisible to us. So 94 00:06:16,680 --> 00:06:21,400 Speaker 1: he froze ultrafast events using a quick strobe light and 95 00:06:21,640 --> 00:06:25,560 Speaker 1: ultra high speed photography. One of his most famous photographs 96 00:06:25,720 --> 00:06:27,680 Speaker 1: was taken in the nineteen sixties and you may have 97 00:06:27,720 --> 00:06:32,440 Speaker 1: seen this. It shows a bullet piercing an apple. The 98 00:06:32,480 --> 00:06:36,159 Speaker 1: image captures the exact moment of impact with the apple 99 00:06:36,240 --> 00:06:40,000 Speaker 1: exploding into fragments, and the exposure time for images like 100 00:06:40,040 --> 00:06:43,680 Speaker 1: this it was less than a microsecond. Another one of 101 00:06:43,680 --> 00:06:48,640 Speaker 1: his iconic images is the milk drop coronet that shows 102 00:06:48,680 --> 00:06:51,960 Speaker 1: a drop of milk hitting a surface and forming this 103 00:06:52,480 --> 00:06:57,360 Speaker 1: crown like splash, and the detail in this splash, frozen 104 00:06:57,440 --> 00:07:00,159 Speaker 1: in time, gave all kinds of insights into flu with 105 00:07:00,279 --> 00:07:04,880 Speaker 1: dynamics and surface tension. So the temporal resolution of these 106 00:07:04,920 --> 00:07:09,400 Speaker 1: photos allowed everyone to study and to appreciate the beauty 107 00:07:09,960 --> 00:07:14,880 Speaker 1: and the complexity of fast phenomenon. So he could freeze 108 00:07:15,000 --> 00:07:18,120 Speaker 1: the motion of bullets in flight or splashing milk drops, 109 00:07:18,360 --> 00:07:22,760 Speaker 1: and these images revealed details that were previously hidden to 110 00:07:23,000 --> 00:07:26,480 Speaker 1: us due to the limitations of our very slow perception. 111 00:07:27,360 --> 00:07:29,600 Speaker 1: And he also did this not just with still shots, 112 00:07:29,640 --> 00:07:31,960 Speaker 1: but a series of still shots, so you get a 113 00:07:31,960 --> 00:07:35,040 Speaker 1: little time lapse film. And in this way he could 114 00:07:35,040 --> 00:07:40,760 Speaker 1: expand super brief events into viewable sequences. This was great 115 00:07:40,800 --> 00:07:44,360 Speaker 1: for science and for art, and also for military applications 116 00:07:44,440 --> 00:07:48,720 Speaker 1: like studying the behavior of bombs. So with these techniques 117 00:07:48,760 --> 00:07:52,760 Speaker 1: he led us into new time domains which we suspected 118 00:07:52,840 --> 00:07:56,640 Speaker 1: might exist, but which no human had ever actually seen. 119 00:07:57,120 --> 00:08:01,280 Speaker 1: So he opened a window into a world of fleeting moments, 120 00:08:01,760 --> 00:08:05,200 Speaker 1: capturing events that occur too quickly for our eyes to see, 121 00:08:05,840 --> 00:08:08,600 Speaker 1: and reveling in a hidden beauty that we didn't even 122 00:08:08,680 --> 00:08:13,000 Speaker 1: have the mental tools to imagine correctly. So the point 123 00:08:13,120 --> 00:08:17,080 Speaker 1: is we can take advantage of these other timescales that 124 00:08:17,120 --> 00:08:19,800 Speaker 1: we don't know much about, and we can do this 125 00:08:19,840 --> 00:08:22,480 Speaker 1: by building the right kind of machinery. 126 00:08:22,800 --> 00:08:24,160 Speaker 2: Now, this all works. 127 00:08:23,920 --> 00:08:29,280 Speaker 1: Because although we humans are slow and pitifully limited in 128 00:08:29,440 --> 00:08:33,439 Speaker 1: vision and in time, we have large prefrontal cortices. This 129 00:08:33,480 --> 00:08:36,840 Speaker 1: is the part of the brain that allows us to extrapolate, 130 00:08:36,960 --> 00:08:39,439 Speaker 1: to think about what's in the next room even though 131 00:08:39,480 --> 00:08:43,680 Speaker 1: we're not there, to think about what we might do tomorrow, 132 00:08:44,040 --> 00:08:48,559 Speaker 1: to think about possibilities that we haven't yet seen. And this, 133 00:08:48,840 --> 00:08:52,920 Speaker 1: I suggest, is what has driven the development of new technologies, 134 00:08:53,000 --> 00:08:57,040 Speaker 1: because we are constantly imagining things that are bigger or 135 00:08:57,040 --> 00:09:00,520 Speaker 1: smaller than we can see. So we invent tell scopes 136 00:09:00,559 --> 00:09:05,320 Speaker 1: and microscopes. We invent ways of seeing into other parts 137 00:09:05,360 --> 00:09:09,160 Speaker 1: of the visible spectrum, like radios or X ray machines, 138 00:09:09,280 --> 00:09:12,640 Speaker 1: or devices in the infrared or ultraviolet, and all over 139 00:09:12,679 --> 00:09:16,240 Speaker 1: the place. We burrow into spaces that we can't see 140 00:09:16,280 --> 00:09:20,600 Speaker 1: with our eyes, but we extrapolate that they must exist. 141 00:09:20,840 --> 00:09:24,160 Speaker 1: And I suggest the prefrontal cortex is what allows us 142 00:09:24,200 --> 00:09:28,680 Speaker 1: to extrapolate in the time domain. So here's an example. 143 00:09:29,040 --> 00:09:31,120 Speaker 1: I was in Germany a little while ago, and I 144 00:09:31,200 --> 00:09:35,760 Speaker 1: was transfixed watching a river flowing through a very deep valley, 145 00:09:36,440 --> 00:09:38,920 Speaker 1: and it was pretty effortless to look at the valley 146 00:09:39,040 --> 00:09:42,640 Speaker 1: walls and think about this at the scale at which 147 00:09:42,800 --> 00:09:47,480 Speaker 1: geologists look at a landscape. This exact moment of the 148 00:09:47,600 --> 00:09:51,600 Speaker 1: river flowing had been happening for millions of years, and 149 00:09:51,640 --> 00:09:55,360 Speaker 1: if I sped up the film in my head, that's 150 00:09:55,440 --> 00:09:59,360 Speaker 1: how you get a valley of that depth. Now, to 151 00:09:59,400 --> 00:10:02,240 Speaker 1: be clear, I wasn't imagining millions of years. I was 152 00:10:02,280 --> 00:10:06,000 Speaker 1: imagining a film at my human speed. I just took 153 00:10:06,040 --> 00:10:09,960 Speaker 1: the concept of huge time scales and squished them down 154 00:10:10,000 --> 00:10:12,680 Speaker 1: to something that I could understand. This is the same 155 00:10:12,679 --> 00:10:17,200 Speaker 1: way that someone using a microscope doesn't stare at the 156 00:10:17,280 --> 00:10:21,640 Speaker 1: bacterium at the scale of a micron. Instead, she uses 157 00:10:21,800 --> 00:10:25,240 Speaker 1: the machine, the microscope, to push the image up to 158 00:10:25,400 --> 00:10:29,360 Speaker 1: a large size that her brain can handle. And this 159 00:10:29,440 --> 00:10:31,959 Speaker 1: is all we ever do. We take things from different 160 00:10:32,000 --> 00:10:36,960 Speaker 1: domains and translate them into something that we can understand. 161 00:10:37,240 --> 00:10:40,240 Speaker 1: Now we're very good at that. Here's an example on 162 00:10:40,800 --> 00:10:43,720 Speaker 1: the flip side of the river carving the valley. Open 163 00:10:43,760 --> 00:10:47,080 Speaker 1: any book on let's say, molecular biology, and you'll see 164 00:10:47,400 --> 00:10:51,760 Speaker 1: two strands of DNA unzipping and proteins buying to those 165 00:10:51,760 --> 00:10:56,000 Speaker 1: strands and walk along it and this gets transcribed to RNA, 166 00:10:56,200 --> 00:11:00,120 Speaker 1: and then that gets translated into little amino acids, and 167 00:11:00,160 --> 00:11:03,840 Speaker 1: you can walk yourself through all the steps of transcription 168 00:11:03,920 --> 00:11:08,240 Speaker 1: and translation. But all this takes place at a timescale 169 00:11:08,600 --> 00:11:12,720 Speaker 1: that's faster than you can possibly conceive. It's about twenty 170 00:11:12,760 --> 00:11:15,960 Speaker 1: amino acids get translated every second, and it's hard to 171 00:11:16,040 --> 00:11:19,840 Speaker 1: picture that in real time, But no problem. 172 00:11:19,920 --> 00:11:21,800 Speaker 2: We can apply the tools of our. 173 00:11:22,080 --> 00:11:25,800 Speaker 1: Thinking to imagine it as though it were happening at 174 00:11:25,840 --> 00:11:29,319 Speaker 1: our timescale, and of course we do this on crazier 175 00:11:29,320 --> 00:11:34,080 Speaker 1: and crazier timescales. Let's say you're a physicist building a 176 00:11:34,200 --> 00:11:39,000 Speaker 1: nuclear reactor and you're thinking about two nucleis smashing into 177 00:11:39,040 --> 00:11:42,720 Speaker 1: one another, and as it busts up, the parts of 178 00:11:42,760 --> 00:11:46,080 Speaker 1: it smash into other nuclei and bash those up, and. 179 00:11:46,000 --> 00:11:47,760 Speaker 2: You get a chain reaction. 180 00:11:48,400 --> 00:11:50,680 Speaker 1: And once you understand the simple model, you can picture 181 00:11:50,760 --> 00:11:53,840 Speaker 1: each step of the chain reaction, even though what is 182 00:11:54,000 --> 00:11:59,320 Speaker 1: happening takes place in one millionth of a second, a microsecond. 183 00:12:00,200 --> 00:12:03,120 Speaker 1: If you made these calculations every day of your life 184 00:12:03,120 --> 00:12:05,560 Speaker 1: on the whiteboard or on the computer, it would be 185 00:12:05,800 --> 00:12:11,400 Speaker 1: absolutely impossible for you to ever perceive the event anywhere. 186 00:12:11,800 --> 00:12:16,160 Speaker 1: But in your imagination, you couldn't perceive it directly. So 187 00:12:16,320 --> 00:12:21,559 Speaker 1: in this way, You are like the biblical story of Moses, 188 00:12:21,600 --> 00:12:24,920 Speaker 1: who leads his people through the desert but never gets 189 00:12:24,960 --> 00:12:30,000 Speaker 1: to enter the Holy Land himself. You are the temporal Moses, 190 00:12:30,080 --> 00:12:33,800 Speaker 1: knowing about these timescales, knowing about these other temporal worlds, 191 00:12:33,840 --> 00:12:38,679 Speaker 1: but you are never able to enter the territory directly. 192 00:12:39,120 --> 00:12:42,640 Speaker 1: Just think about the best theories of the beginning of 193 00:12:42,679 --> 00:12:45,360 Speaker 1: the universe. The best models for this suggest that there 194 00:12:45,440 --> 00:12:48,880 Speaker 1: was cosmic inflation where everything got bigger. There was this 195 00:12:49,200 --> 00:12:52,760 Speaker 1: exponential expansion of space, and we say this all happened 196 00:12:52,800 --> 00:12:56,920 Speaker 1: in the early universe. And when we look at the models, 197 00:12:56,960 --> 00:13:01,079 Speaker 1: there was the Plank Epic, where physics were dominated by 198 00:13:01,120 --> 00:13:04,160 Speaker 1: the quantum effects of gravity, and then that got followed 199 00:13:04,160 --> 00:13:08,120 Speaker 1: by the Grand Unification Epic, and that was followed by 200 00:13:08,160 --> 00:13:12,760 Speaker 1: the inflationary epic, in which the universe expanded by trillions 201 00:13:12,760 --> 00:13:16,360 Speaker 1: and trillions of times, and then the electroweek Epic, and 202 00:13:16,400 --> 00:13:18,760 Speaker 1: the quark Epic and the hadron epic and so on. 203 00:13:18,800 --> 00:13:19,640 Speaker 2: Now here's the key. 204 00:13:20,120 --> 00:13:25,120 Speaker 1: The Plank Epic lasted less then ten to the negative 205 00:13:25,200 --> 00:13:29,000 Speaker 1: forty three seconds. That's less than a trillionth of a 206 00:13:29,040 --> 00:13:31,360 Speaker 1: trillionth of a trillion of a trillionth of a second, 207 00:13:32,040 --> 00:13:35,880 Speaker 1: and the grand unification epic was over by ten to 208 00:13:35,960 --> 00:13:39,800 Speaker 1: the negative thirty six seconds, and the inflationary epic was 209 00:13:40,000 --> 00:13:43,400 Speaker 1: over in the first peko second of cosmic time, that's 210 00:13:43,440 --> 00:13:47,280 Speaker 1: ten to the negative twelve seconds. And eventually protons formed. 211 00:13:47,440 --> 00:13:49,680 Speaker 1: And oh, by the way, that happened by the first 212 00:13:50,000 --> 00:13:53,240 Speaker 1: microsecond one millionth of a second after the Big Bang, 213 00:13:53,640 --> 00:13:56,440 Speaker 1: and then nuclear fusion began, and that was a ten 214 00:13:56,520 --> 00:13:59,040 Speaker 1: milliseconds after the start of the universe, or one one 215 00:13:59,120 --> 00:14:01,240 Speaker 1: hundredth of a second. No, how do people even put 216 00:14:01,280 --> 00:14:04,920 Speaker 1: together theories like this, Well, you can't even run experiments 217 00:14:04,920 --> 00:14:07,720 Speaker 1: at those temperatures or densities. But you can do is 218 00:14:07,800 --> 00:14:12,440 Speaker 1: extrapolate the known physical laws to extremely high temperatures, and 219 00:14:12,480 --> 00:14:14,600 Speaker 1: you get a picture of what was likely to have 220 00:14:14,720 --> 00:14:18,040 Speaker 1: happened there. And again, this is something that we can 221 00:14:18,120 --> 00:14:24,240 Speaker 1: never experience directly, but we can extrapolate our local understanding 222 00:14:24,280 --> 00:14:28,640 Speaker 1: of time to imagine it. Presumably, the squirrels running around 223 00:14:28,640 --> 00:14:32,880 Speaker 1: in your backyard simply don't have the neural capacity to 224 00:14:33,120 --> 00:14:36,840 Speaker 1: imagine time at any scale other than their own. So 225 00:14:37,160 --> 00:14:39,560 Speaker 1: a squirrel is never going to come up with the 226 00:14:39,640 --> 00:14:58,080 Speaker 1: Plank epic. Now, what I want to talk about is 227 00:14:58,120 --> 00:15:01,360 Speaker 1: the way that we do these extra appellations, and then 228 00:15:01,840 --> 00:15:05,320 Speaker 1: we build technologies to get us there, like Doc Edgerton 229 00:15:05,400 --> 00:15:10,040 Speaker 1: at MIT did and translating things into something useful for 230 00:15:10,200 --> 00:15:14,920 Speaker 1: our pitifully slow time scales. So I'll give you an 231 00:15:14,960 --> 00:15:18,320 Speaker 1: example of this. I was in Shanghai a few years ago, 232 00:15:18,600 --> 00:15:20,880 Speaker 1: and on the edge of the river where a lot 233 00:15:20,920 --> 00:15:23,680 Speaker 1: of tourists walk around, they had this terrific display where 234 00:15:23,680 --> 00:15:26,800 Speaker 1: it was like a big plant wall, and from a 235 00:15:26,840 --> 00:15:31,800 Speaker 1: distance you could see these beautiful, slowly flapping butterflies on 236 00:15:31,840 --> 00:15:34,080 Speaker 1: the wall. But when you go up close and really 237 00:15:34,120 --> 00:15:37,400 Speaker 1: examine this, you see that each butterfly is actually an 238 00:15:37,600 --> 00:15:43,120 Speaker 1: illusion produced from a rapidly spinning bar of led lights. 239 00:15:43,560 --> 00:15:45,640 Speaker 1: So the bar is attached at one end to a 240 00:15:45,760 --> 00:15:48,120 Speaker 1: motor and it spins around and around in a circle 241 00:15:48,240 --> 00:15:48,800 Speaker 1: very quickly. 242 00:15:49,080 --> 00:15:52,000 Speaker 2: You may have seen these before. You can display. 243 00:15:51,760 --> 00:15:56,400 Speaker 1: Cartoon characters or words or animals. The rod is a 244 00:15:56,560 --> 00:15:59,240 Speaker 1: row of led lights, and these are all turning on 245 00:15:59,320 --> 00:16:03,120 Speaker 1: and off with different colors of different intensities at exactly 246 00:16:03,160 --> 00:16:06,480 Speaker 1: the right time in their rotations, so that they draw 247 00:16:06,680 --> 00:16:09,840 Speaker 1: out what appears to be a picture of an object. 248 00:16:09,920 --> 00:16:13,600 Speaker 1: Now why does this work, Well, it's because of something 249 00:16:13,680 --> 00:16:17,400 Speaker 1: called persistence of vision. And this is what happens when 250 00:16:17,440 --> 00:16:20,200 Speaker 1: you flash a bunch of images really rapidly, one after 251 00:16:20,240 --> 00:16:23,120 Speaker 1: the other. The bar is in this position, and then 252 00:16:23,160 --> 00:16:26,320 Speaker 1: the next moment it's displaying it slightly differently one degree away, 253 00:16:26,440 --> 00:16:28,640 Speaker 1: and then differently again when it ticks. 254 00:16:28,360 --> 00:16:30,080 Speaker 2: Forward one more degree, and so on. 255 00:16:30,240 --> 00:16:32,640 Speaker 1: But because each of these little flashes of light are 256 00:16:32,760 --> 00:16:37,080 Speaker 1: so fast, they last in your vision. So the light 257 00:16:37,160 --> 00:16:39,680 Speaker 1: pops on and off in a fraction of a millisecond, 258 00:16:39,720 --> 00:16:42,520 Speaker 1: but your brain just can't see something that brief, So 259 00:16:42,560 --> 00:16:44,880 Speaker 1: as far as you're concerned, it comes on and it 260 00:16:44,960 --> 00:16:47,480 Speaker 1: stays on for a little while. So when the bar 261 00:16:47,640 --> 00:16:51,480 Speaker 1: spins around and around, it essentially paints in the air. 262 00:16:52,360 --> 00:16:54,680 Speaker 1: Now here's the key. If you could see faster, if 263 00:16:54,680 --> 00:16:57,480 Speaker 1: you could see what is actually happening out there, you'd 264 00:16:57,480 --> 00:16:59,760 Speaker 1: see a little bar spinning in a circle and changing 265 00:16:59,760 --> 00:17:03,400 Speaker 1: color as it does so. But that's not what you see. 266 00:17:03,440 --> 00:17:06,679 Speaker 1: You see a butterfly in the air, you see a 267 00:17:06,720 --> 00:17:10,040 Speaker 1: cartoon character, you see a flag waving. But here's the 268 00:17:10,119 --> 00:17:13,639 Speaker 1: thing to appreciate. The speed at which the rod needs 269 00:17:13,680 --> 00:17:17,680 Speaker 1: to spin is very fast, and therefore the speed at 270 00:17:17,720 --> 00:17:21,000 Speaker 1: which the lights need to know precisely when to turn 271 00:17:21,040 --> 00:17:24,120 Speaker 1: on and off is even faster, and all of this 272 00:17:24,200 --> 00:17:27,880 Speaker 1: is computed at rocket ship speed, and you just look 273 00:17:27,920 --> 00:17:31,640 Speaker 1: at it and appreciate that there's a nice butterfly out there. 274 00:17:32,000 --> 00:17:34,680 Speaker 1: So on the one hand, you have lightning fast calculations 275 00:17:34,680 --> 00:17:37,480 Speaker 1: on the level of sub millisecond, and on the other 276 00:17:37,640 --> 00:17:40,160 Speaker 1: level we get a nice picture. 277 00:17:40,400 --> 00:17:41,440 Speaker 2: And persistence of. 278 00:17:41,440 --> 00:17:45,760 Speaker 1: Vision is the reason why movies appear to be continuous motion, 279 00:17:45,920 --> 00:17:49,720 Speaker 1: even though they are actually just a series of still images. 280 00:17:49,840 --> 00:17:54,199 Speaker 1: In other words, after photography was invented, people realize that 281 00:17:54,240 --> 00:17:57,159 Speaker 1: if you flash one photograph and then there's a blank, 282 00:17:57,200 --> 00:18:00,200 Speaker 1: and then you flash another photo right after, like twenty 283 00:18:00,200 --> 00:18:05,520 Speaker 1: milliseconds after, your brain will interpret that as continuous motion. 284 00:18:06,080 --> 00:18:11,760 Speaker 1: And this was the birth of moving photography or movies. Now, 285 00:18:11,960 --> 00:18:14,320 Speaker 1: the challenge was tougher with televisions. 286 00:18:14,680 --> 00:18:17,640 Speaker 2: Entrepreneurs wanted to get movies into the homes of. 287 00:18:17,640 --> 00:18:21,159 Speaker 1: People, but you couldn't flicker photographs forty eight times per 288 00:18:21,200 --> 00:18:23,560 Speaker 1: second because you didn't have a film strip that you 289 00:18:23,600 --> 00:18:24,680 Speaker 1: were just shining light through. 290 00:18:24,720 --> 00:18:26,200 Speaker 2: So how do you do it? Well? 291 00:18:26,240 --> 00:18:29,600 Speaker 1: The genius there was to use a cathode ray tube, 292 00:18:29,600 --> 00:18:33,920 Speaker 1: which points a beam of light at a phosphorescent screen, 293 00:18:34,400 --> 00:18:38,040 Speaker 1: and that causes a little spot to glow from dark 294 00:18:38,080 --> 00:18:40,879 Speaker 1: to bright and then you move that little light gun 295 00:18:40,920 --> 00:18:44,399 Speaker 1: over to the next spot over and set the next brightness, 296 00:18:44,640 --> 00:18:46,199 Speaker 1: and then you go to the next spot in the 297 00:18:46,200 --> 00:18:46,960 Speaker 1: next spot. 298 00:18:46,680 --> 00:18:47,080 Speaker 2: And so on. 299 00:18:47,320 --> 00:18:50,359 Speaker 1: And you had to cover the entire screen one dot 300 00:18:50,400 --> 00:18:52,600 Speaker 1: at a time, and the whole trick was to do 301 00:18:52,640 --> 00:18:55,720 Speaker 1: it really fast, so that you paint the screen in 302 00:18:56,200 --> 00:18:59,240 Speaker 1: one sixtieth of a second, and then you paint the 303 00:18:59,280 --> 00:19:02,239 Speaker 1: next screen again one dot at a time, and it 304 00:19:02,280 --> 00:19:06,280 Speaker 1: has to get done in the next sixtieth of a second. Now, 305 00:19:06,320 --> 00:19:08,639 Speaker 1: the genius of coming up with the technology like that 306 00:19:09,160 --> 00:19:12,680 Speaker 1: is realizing that you can do something very very fast 307 00:19:12,720 --> 00:19:16,000 Speaker 1: with a machine, and that the person at the receiving 308 00:19:16,160 --> 00:19:20,360 Speaker 1: end is incredibly slow and has a sluggish brain and 309 00:19:20,480 --> 00:19:24,400 Speaker 1: can't understand anything at these millisecond time scales and instead 310 00:19:24,520 --> 00:19:28,840 Speaker 1: just blends everything together. And that's how you tell stories 311 00:19:28,920 --> 00:19:32,480 Speaker 1: about James Cagney or Fat Albert or the family in Bonanza, 312 00:19:33,040 --> 00:19:36,360 Speaker 1: and the sluggish humans on the receiving end are simply 313 00:19:36,440 --> 00:19:40,959 Speaker 1: registering the emotion of the story instead of seeing the 314 00:19:41,359 --> 00:19:45,840 Speaker 1: invisibly fast mechanics that are zipping along in their own 315 00:19:46,000 --> 00:19:50,520 Speaker 1: timescale range, painting the screen over and over sixty times 316 00:19:50,560 --> 00:19:55,640 Speaker 1: every second in a way that is completely invisible to us. 317 00:19:55,960 --> 00:19:58,919 Speaker 1: So we build machines to leverage the fact that the 318 00:19:59,000 --> 00:20:03,119 Speaker 1: visual brain integrates what it sees over a little window 319 00:20:03,200 --> 00:20:06,879 Speaker 1: of time. So if two or more images arrive within 320 00:20:06,920 --> 00:20:09,800 Speaker 1: a window of about a tenth of a second, they're 321 00:20:09,840 --> 00:20:13,560 Speaker 1: smeared together. And by the way, building devices to take 322 00:20:13,600 --> 00:20:16,920 Speaker 1: advantage of that isn't even so new. We've been building 323 00:20:17,000 --> 00:20:21,600 Speaker 1: primitive things for centuries. For example, think of that little 324 00:20:21,680 --> 00:20:26,040 Speaker 1: toy known as a thaumatrope. It's a flat disc with 325 00:20:26,160 --> 00:20:28,280 Speaker 1: a picture on each side. So you might have a 326 00:20:28,320 --> 00:20:30,440 Speaker 1: picture of a bird on one side of the disk 327 00:20:30,720 --> 00:20:33,320 Speaker 1: and a picture of a tree branch on the other. 328 00:20:33,680 --> 00:20:35,840 Speaker 1: And what you do is you wind up the disk 329 00:20:36,000 --> 00:20:37,960 Speaker 1: with a piece of string, and when you pull the 330 00:20:38,000 --> 00:20:42,280 Speaker 1: string tight, the disc spins rapidly so that both sides 331 00:20:42,480 --> 00:20:45,119 Speaker 1: are seen in rapid alternation. You see the bird, you 332 00:20:45,160 --> 00:20:46,960 Speaker 1: see the branch. You see the bird, you see the branch, 333 00:20:47,359 --> 00:20:50,600 Speaker 1: and the bird appears to be sitting on the branch. 334 00:20:50,720 --> 00:20:54,520 Speaker 1: The images fuse. This is an example of an old 335 00:20:54,560 --> 00:20:58,960 Speaker 1: technology taking advantage of the slowness of our brains. Now 336 00:20:59,000 --> 00:21:02,360 Speaker 1: I find all this credible, from the butterflies to the 337 00:21:02,400 --> 00:21:06,040 Speaker 1: fact that movies move, to the fact that television works. 338 00:21:06,600 --> 00:21:10,080 Speaker 1: But with modern electronics, combined with the slowness of our 339 00:21:10,160 --> 00:21:13,600 Speaker 1: visual systems, we can build these kinds of devices to 340 00:21:13,840 --> 00:21:17,679 Speaker 1: paint pictures for us. Our machines can be fast, a 341 00:21:17,680 --> 00:21:21,040 Speaker 1: lot faster than our poor little brains. Now, if you 342 00:21:21,119 --> 00:21:24,840 Speaker 1: were the machine, you would think, what am I doing? 343 00:21:24,920 --> 00:21:28,440 Speaker 1: I'm just changing the colors of the light as I spin. 344 00:21:28,960 --> 00:21:33,040 Speaker 1: Surely no one out there is falling for this. And 345 00:21:33,080 --> 00:21:35,199 Speaker 1: this is the thing I really want to dig into today, 346 00:21:35,240 --> 00:21:39,960 Speaker 1: the difference in the time worlds between us and our machines. 347 00:21:40,720 --> 00:21:43,600 Speaker 1: So the best place to start is with our computers, 348 00:21:44,200 --> 00:21:49,199 Speaker 1: which perform a certain number of floating point operations per second. 349 00:21:49,320 --> 00:21:54,199 Speaker 1: This measure is called flops floating point operations per second. Now, 350 00:21:54,200 --> 00:21:56,840 Speaker 1: don't worry about the details, except to say a floating 351 00:21:56,840 --> 00:22:00,440 Speaker 1: point operation is a calculation like adding two. 352 00:22:00,840 --> 00:22:02,000 Speaker 2: Very big numbers. 353 00:22:02,160 --> 00:22:04,240 Speaker 1: So this is a standard way to measure how fast 354 00:22:04,280 --> 00:22:08,480 Speaker 1: the computer can chug along. Now, think about how long 355 00:22:08,520 --> 00:22:12,320 Speaker 1: it would take you to add two very large numbers. 356 00:22:12,640 --> 00:22:16,760 Speaker 1: It's unlikely that you could do even one floating point 357 00:22:16,800 --> 00:22:20,800 Speaker 1: operation in a second. But a computer from the last 358 00:22:20,800 --> 00:22:27,480 Speaker 1: century could do a gigaflop that's one billion floating point 359 00:22:27,520 --> 00:22:32,359 Speaker 1: operations in a second. Just think about sitting there with 360 00:22:32,400 --> 00:22:35,600 Speaker 1: a paper and pencil and adding two big numbers and 361 00:22:35,640 --> 00:22:40,600 Speaker 1: doing that a billion times every second. Then in this century, 362 00:22:40,640 --> 00:22:47,399 Speaker 1: computers hit terraflops that's one trillion floating point operations in 363 00:22:47,480 --> 00:22:51,840 Speaker 1: a second. And then it wasn't long before computers hit petaflops, 364 00:22:51,880 --> 00:22:56,240 Speaker 1: which is a thousand terraflops. In other words, a thousand 365 00:22:56,680 --> 00:23:02,199 Speaker 1: trillion or a quadrillion operations person second. Nvidia just released 366 00:23:02,200 --> 00:23:06,480 Speaker 1: a new chip that clocks in at twenty petaflops, So 367 00:23:06,640 --> 00:23:09,240 Speaker 1: just imagine how long it would take you to do 368 00:23:10,160 --> 00:23:15,879 Speaker 1: twenty thousand trillion calculations. It does that a second. The 369 00:23:16,359 --> 00:23:20,560 Speaker 1: current record is the US supercomputer called Frontier, which does 370 00:23:21,160 --> 00:23:26,760 Speaker 1: eleven hundred petaflops or one point one exaFLOPS. These numbers 371 00:23:26,760 --> 00:23:29,720 Speaker 1: are so mind boggling to think about. Just imagine a 372 00:23:29,800 --> 00:23:35,520 Speaker 1: machine performing one quintillion floating point operations per second. That's 373 00:23:35,560 --> 00:23:39,320 Speaker 1: a one followed by eighteen zeros. That's a lot faster 374 00:23:39,720 --> 00:23:42,120 Speaker 1: than all the humans in the world could do if 375 00:23:42,119 --> 00:23:44,800 Speaker 1: they all sat with paper and pencil and worked on 376 00:23:44,840 --> 00:23:49,399 Speaker 1: calculations for a million years. That's what gets accomplished in 377 00:23:49,600 --> 00:23:53,560 Speaker 1: one second by a computer like that. We are living 378 00:23:53,840 --> 00:23:58,240 Speaker 1: on completely different time domains, and this is what allows 379 00:23:58,280 --> 00:24:02,040 Speaker 1: you to enjoy a song or YouTube video or an 380 00:24:02,080 --> 00:24:06,440 Speaker 1: Inner Cosmos podcast. Now, something I think is quite stunning 381 00:24:06,600 --> 00:24:08,760 Speaker 1: is to consider what it is like. 382 00:24:08,840 --> 00:24:11,240 Speaker 2: To be your computer keyboard. 383 00:24:11,640 --> 00:24:14,240 Speaker 1: Even if you think you are a fast typist, your 384 00:24:14,359 --> 00:24:20,560 Speaker 1: keyboard actually goes to sleep in between your slow keystrokes. 385 00:24:20,920 --> 00:24:24,040 Speaker 1: So I'm gonna do something here. Let's imagine that you 386 00:24:24,119 --> 00:24:27,679 Speaker 1: and I are down inside your computer, living at the 387 00:24:27,720 --> 00:24:32,200 Speaker 1: speed of your computer chip. Now, the poor slow human 388 00:24:32,359 --> 00:24:35,640 Speaker 1: on the outside is typing, and let's say they're typing 389 00:24:35,680 --> 00:24:40,000 Speaker 1: relatively fast, so when they strike a key on the keyboard, 390 00:24:41,080 --> 00:24:47,040 Speaker 1: it's going to sound like this to us inside the computer. Now, 391 00:24:47,119 --> 00:24:51,879 Speaker 1: let's keep talking and we'll see when the next keystroke arrives. 392 00:24:52,119 --> 00:24:55,920 Speaker 1: Imagine that this human is typing relatively fast, like sixty 393 00:24:55,960 --> 00:24:59,000 Speaker 1: words per minute, which is about three hundred keystrokes per minute, 394 00:24:59,200 --> 00:25:02,960 Speaker 1: or a keystroke every two hundred milliseconds from our point 395 00:25:03,040 --> 00:25:07,240 Speaker 1: of view from inside the computer. The next key will 396 00:25:07,280 --> 00:25:11,280 Speaker 1: not even get hit until about the end of this podcast, 397 00:25:11,680 --> 00:25:15,760 Speaker 1: And that's why the software routine that monitors the keyboard 398 00:25:16,560 --> 00:25:20,760 Speaker 1: goes to sleep while it waits for the next key 399 00:25:21,320 --> 00:25:24,240 Speaker 1: to be struck. And I'll tell you something else fascinating 400 00:25:24,280 --> 00:25:28,960 Speaker 1: about our computers and our interaction with them. You remember capture, 401 00:25:29,119 --> 00:25:30,960 Speaker 1: which is how at the beginning of the Internet you 402 00:25:31,000 --> 00:25:33,520 Speaker 1: could try to tell if it was a real human 403 00:25:33,560 --> 00:25:36,880 Speaker 1: doing something versus a computer, because you could say, hey, 404 00:25:36,920 --> 00:25:40,199 Speaker 1: what are these funny shaped letters spelling out? And a 405 00:25:40,240 --> 00:25:42,680 Speaker 1: computer just couldn't figure that out, but a human could. 406 00:25:43,040 --> 00:25:45,720 Speaker 1: Now eventually this evolved to what are now known as 407 00:25:45,920 --> 00:25:49,440 Speaker 1: recapture boxes. So you know these little tests when you're 408 00:25:49,840 --> 00:25:52,439 Speaker 1: let's say, making a purchase on a website and you 409 00:25:52,520 --> 00:25:55,680 Speaker 1: have to click a little box that says I am 410 00:25:55,800 --> 00:25:59,119 Speaker 1: not a robot, and you check that box because you 411 00:25:59,160 --> 00:26:01,480 Speaker 1: are in fact a human. Now, if you thought about this, 412 00:26:01,600 --> 00:26:04,320 Speaker 1: you know that this is because AI can't figure this 413 00:26:04,400 --> 00:26:08,200 Speaker 1: out and click the box. Wait, what AI has read 414 00:26:08,320 --> 00:26:11,480 Speaker 1: everything written on the planet. It can do extraordinary things, 415 00:26:11,560 --> 00:26:14,080 Speaker 1: so why can't it click the box. Well, of course 416 00:26:14,119 --> 00:26:17,960 Speaker 1: it can click the box. It's absolutely trivial for any 417 00:26:18,040 --> 00:26:21,879 Speaker 1: large language model to read that text and check the box. 418 00:26:21,920 --> 00:26:25,359 Speaker 1: There's no problem at all, So, why are companies still 419 00:26:25,480 --> 00:26:29,320 Speaker 1: using this as a way to distinguish humans from computers. 420 00:26:29,920 --> 00:26:32,800 Speaker 1: There's a remarkable story here. It turns out the reason 421 00:26:33,040 --> 00:26:36,280 Speaker 1: you can still use this to make the distinction is 422 00:26:36,320 --> 00:26:40,840 Speaker 1: because of the time domain. Humans check the box at 423 00:26:40,880 --> 00:26:43,800 Speaker 1: their slow human speed. 424 00:26:43,760 --> 00:26:45,400 Speaker 2: While AI does it instantly. 425 00:26:45,760 --> 00:26:48,199 Speaker 1: And by the way, sometimes you click the box, it 426 00:26:48,280 --> 00:26:51,160 Speaker 1: might take you to a little puzzle where you see 427 00:26:51,240 --> 00:26:54,359 Speaker 1: nine photographs and it says, click all the photos that 428 00:26:54,840 --> 00:26:59,639 Speaker 1: contain a motorcycle. Now, if chat GPT or Claude or 429 00:26:59,680 --> 00:27:03,440 Speaker 1: GEMIN and I can crush the SATs and the MCATs 430 00:27:03,480 --> 00:27:06,439 Speaker 1: and the l sets, why can't it find a motorcycle? Well, 431 00:27:06,480 --> 00:27:09,560 Speaker 1: of course it can, But the website determines whether you 432 00:27:09,600 --> 00:27:12,960 Speaker 1: are a human or a bot by how slow you are. 433 00:27:13,119 --> 00:27:17,560 Speaker 1: Humans are unbelievably clunky at doing these things. Bots do 434 00:27:17,680 --> 00:27:22,120 Speaker 1: them instantaneously. Just as a side note, like all capture tests, 435 00:27:22,160 --> 00:27:25,520 Speaker 1: this will be short lived before it has to evolve again, 436 00:27:25,600 --> 00:27:29,600 Speaker 1: because computers will soon imitate not only the right answer, 437 00:27:29,640 --> 00:27:33,320 Speaker 1: but the right timing. Going to sleep for a long 438 00:27:33,840 --> 00:27:39,359 Speaker 1: nap to impersonate our pitifully slow timescales. And it's no 439 00:27:39,520 --> 00:27:44,399 Speaker 1: surprise that we're increasingly surrounded by machines that operate in 440 00:27:44,560 --> 00:27:49,600 Speaker 1: different time domains. A typical modern self driving car is 441 00:27:49,720 --> 00:27:53,119 Speaker 1: ringed with cameras and radar, and it generates something like 442 00:27:53,240 --> 00:27:57,399 Speaker 1: six gigabytes of data every thirty seconds. The key again 443 00:27:57,880 --> 00:27:59,639 Speaker 1: is that this is not at all what a human 444 00:27:59,680 --> 00:28:02,199 Speaker 1: can do. These cars, which are all around us on 445 00:28:02,200 --> 00:28:05,600 Speaker 1: the road now, they capture vision on multiple cameras that 446 00:28:06,080 --> 00:28:09,320 Speaker 1: ten to thirty frames per second, sometimes more. And by 447 00:28:09,359 --> 00:28:11,720 Speaker 1: the way, these cameras aren't just in the visible range, 448 00:28:11,760 --> 00:28:15,040 Speaker 1: but they can include infrared cameras, and all this data 449 00:28:15,119 --> 00:28:19,239 Speaker 1: is combined and digested rapidly into a picture of the 450 00:28:19,240 --> 00:28:23,400 Speaker 1: world out there. And these cameras never blink or lose 451 00:28:23,400 --> 00:28:26,800 Speaker 1: attention or get sleepy, and they will only continue to 452 00:28:26,840 --> 00:28:30,280 Speaker 1: get faster. Although they appear to be accomplishing the same 453 00:28:30,440 --> 00:28:33,040 Speaker 1: tasks that we are, they're doing it at a totally 454 00:28:33,160 --> 00:28:37,680 Speaker 1: different timescale. And by the way, in commercial airliners there 455 00:28:37,680 --> 00:28:40,880 Speaker 1: are computer chips which auto correct the flight of the 456 00:28:40,920 --> 00:28:44,040 Speaker 1: airplane one hundred thousand times a second, which keeps the 457 00:28:44,160 --> 00:28:48,200 Speaker 1: rides smooth and good and infinitely better than the best 458 00:28:48,400 --> 00:28:52,240 Speaker 1: human pilot could manage. And this is all just the beginning. 459 00:28:52,600 --> 00:28:57,800 Speaker 1: Take something that underlies our entire economy. I'm always interested 460 00:28:57,840 --> 00:29:01,440 Speaker 1: to discover that not everyone realizes what's going on with 461 00:29:01,640 --> 00:29:07,040 Speaker 1: high frequency stock market trading. This is where computers use 462 00:29:07,280 --> 00:29:11,000 Speaker 1: algorithms to trade instead of humans doing the trading, so 463 00:29:11,200 --> 00:29:14,920 Speaker 1: the computer can ingest an enormous amount of information and 464 00:29:15,040 --> 00:29:19,320 Speaker 1: make a decision about buying or selling stocks on microsecond 465 00:29:19,320 --> 00:29:22,080 Speaker 1: time scales. And the part that often comes as a 466 00:29:22,120 --> 00:29:26,920 Speaker 1: surprise is that sixty percent of all trades on the 467 00:29:27,040 --> 00:29:29,440 Speaker 1: US stock market are done by computer. 468 00:29:29,840 --> 00:29:33,880 Speaker 2: There's no human in the loop these high frequency computerized trades. 469 00:29:33,920 --> 00:29:36,720 Speaker 1: They'll often only make a few pennies on a trade, 470 00:29:36,720 --> 00:29:39,880 Speaker 1: but they're doing this thousands or millions of times a 471 00:29:39,960 --> 00:29:45,040 Speaker 1: day at an extraordinary timescale. That's simply not human. A 472 00:29:45,240 --> 00:29:47,800 Speaker 1: colleague of mind started one of the first high frequency 473 00:29:47,840 --> 00:29:51,760 Speaker 1: trading companies, and he was fascinated that the ticker tape 474 00:29:51,800 --> 00:29:54,920 Speaker 1: back in the day, the physical ticker tape which showed 475 00:29:54,960 --> 00:29:58,160 Speaker 1: the stock prices that was rolling off straight into the 476 00:29:58,200 --> 00:30:01,560 Speaker 1: trash can. And he was in and understanding whether he 477 00:30:01,600 --> 00:30:04,600 Speaker 1: could use that past data, which no one really cared 478 00:30:04,600 --> 00:30:08,600 Speaker 1: about anymore, to make really good algorithms to predict what 479 00:30:08,880 --> 00:30:12,280 Speaker 1: happens in the next second or eventually the next millisecond 480 00:30:12,320 --> 00:30:17,000 Speaker 1: or eventually the next microsecond. So he rented physical office 481 00:30:17,000 --> 00:30:20,360 Speaker 1: space just on the other side of the wall from 482 00:30:20,400 --> 00:30:23,280 Speaker 1: the New York Stock Exchange, because when you're talking about 483 00:30:23,440 --> 00:30:26,200 Speaker 1: signals at this speed, you have to worry about the 484 00:30:26,280 --> 00:30:30,280 Speaker 1: transmission time from your computer to the exchange floor. And 485 00:30:30,360 --> 00:30:33,800 Speaker 1: pretty soon everyone was getting real estate physically right next 486 00:30:33,800 --> 00:30:38,120 Speaker 1: to the stock floor, because microseconds really matter if your 487 00:30:38,200 --> 00:30:40,600 Speaker 1: computer is trying to beat the next guy's computer to 488 00:30:40,720 --> 00:30:44,280 Speaker 1: a trade. So the key is that all these computers 489 00:30:44,320 --> 00:30:48,800 Speaker 1: are competing against one another, but human traders aren't even 490 00:30:48,840 --> 00:30:52,440 Speaker 1: in that race. They are orders of magnitudes slower, literally 491 00:30:52,840 --> 00:30:57,160 Speaker 1: millions of times slower in making their decisions. So the 492 00:30:57,160 --> 00:31:01,720 Speaker 1: computers capitalize on short term price movement, some arbitrage opportunities 493 00:31:01,920 --> 00:31:06,440 Speaker 1: that are impossible for human traders to exploit because those 494 00:31:06,480 --> 00:31:27,040 Speaker 1: timescales are invisible to the Homo sapien. So when I 495 00:31:27,080 --> 00:31:29,600 Speaker 1: put all this together and think about where this goes, 496 00:31:30,000 --> 00:31:32,960 Speaker 1: I'm reminded of a dream I had some years ago. 497 00:31:33,200 --> 00:31:37,520 Speaker 1: In this dream, I was standing over a giant valley 498 00:31:37,680 --> 00:31:39,920 Speaker 1: below me, and I was standing on a column that 499 00:31:40,000 --> 00:31:42,800 Speaker 1: was coming straight up from the earth like a stalagmite, 500 00:31:42,880 --> 00:31:45,560 Speaker 1: and I stepped out into the void, and as soon 501 00:31:45,560 --> 00:31:49,200 Speaker 1: as I did, a new stalagmite shot up instantly from 502 00:31:49,240 --> 00:31:53,200 Speaker 1: the ground way below and provided a perfect landing spot 503 00:31:53,280 --> 00:31:56,600 Speaker 1: for my foot. Then I put my other foot out 504 00:31:56,640 --> 00:31:58,920 Speaker 1: into the void, and right when I was about to 505 00:31:58,920 --> 00:32:02,160 Speaker 1: put my weight down, the next lag might shot up 506 00:32:02,160 --> 00:32:06,040 Speaker 1: in milliseconds and net my foot right where a floor 507 00:32:06,120 --> 00:32:09,600 Speaker 1: would be. So no matter which way I stepped or 508 00:32:09,640 --> 00:32:13,640 Speaker 1: at what pace, a stepping surface would shoot up to 509 00:32:13,720 --> 00:32:16,600 Speaker 1: meet me. It was an incredible dream, and I woke 510 00:32:16,720 --> 00:32:19,680 Speaker 1: up flooded with emotion. It didn't feel like I just 511 00:32:19,800 --> 00:32:24,720 Speaker 1: experienced magic. It felt like I had just experienced AI technology, 512 00:32:25,040 --> 00:32:28,120 Speaker 1: but at a level we don't have now. There's no 513 00:32:28,240 --> 00:32:31,080 Speaker 1: reason we wouldn't have this kind of thing in the 514 00:32:31,120 --> 00:32:34,240 Speaker 1: next century. So I felt like I had just experienced 515 00:32:34,240 --> 00:32:38,640 Speaker 1: a bit of twenty second century technology that had dropped 516 00:32:38,760 --> 00:32:42,880 Speaker 1: into the dream of a twenty first century sleeper. So 517 00:32:43,040 --> 00:32:45,320 Speaker 1: when I woke up, I started thinking about this and 518 00:32:45,360 --> 00:32:49,480 Speaker 1: calling this physical AI. Just imagine a world where we 519 00:32:49,520 --> 00:32:54,200 Speaker 1: can build intelligent machines that physically move so much faster 520 00:32:54,320 --> 00:32:58,680 Speaker 1: than us on a totally different timescale that we just 521 00:32:58,760 --> 00:33:02,800 Speaker 1: get to enjoy moving physically however we want, like stepping 522 00:33:02,880 --> 00:33:05,840 Speaker 1: off a cliff and being certain that it will be 523 00:33:06,040 --> 00:33:10,200 Speaker 1: there essentially instantly as far as we're concerned to catch us. 524 00:33:10,600 --> 00:33:14,480 Speaker 1: Imagine being able to trip on something and fall without 525 00:33:14,600 --> 00:33:17,440 Speaker 1: fear because you know you will get cushioned before you 526 00:33:17,520 --> 00:33:19,760 Speaker 1: hit the ground. In a sense, this would be like 527 00:33:19,800 --> 00:33:22,960 Speaker 1: the airbag in a car, which deploys in twenty milliseconds, 528 00:33:23,360 --> 00:33:25,400 Speaker 1: faster than the blink of an eye, but you would 529 00:33:25,480 --> 00:33:29,120 Speaker 1: have this with you in all ways, in all situations. 530 00:33:29,400 --> 00:33:32,920 Speaker 1: It's very difficult for us, as twenty first century minds, 531 00:33:33,320 --> 00:33:36,240 Speaker 1: to imagine what this will be like or what it 532 00:33:36,240 --> 00:33:39,040 Speaker 1: will translate to socially. All we can be certain of 533 00:33:39,320 --> 00:33:42,120 Speaker 1: is that we don't have the capacity to make a 534 00:33:42,320 --> 00:33:45,800 Speaker 1: meaningful model of what life will be like in a 535 00:33:45,920 --> 00:33:50,239 Speaker 1: century from now. The idea of physical AI seems so 536 00:33:50,920 --> 00:33:53,760 Speaker 1: foreign to us, but just like the keyboards or the 537 00:33:53,800 --> 00:33:57,840 Speaker 1: high frequency trading, this is where we're heading with technology. 538 00:33:58,200 --> 00:34:02,280 Speaker 1: The technology becomes very useful to us because it lives 539 00:34:02,360 --> 00:34:06,360 Speaker 1: on a totally different timescale, and it also becomes more 540 00:34:06,520 --> 00:34:11,840 Speaker 1: and more foreign to us. By operating so inconceivably fast, 541 00:34:12,200 --> 00:34:15,799 Speaker 1: it comes to provide the firm ground that we step on, 542 00:34:16,280 --> 00:34:20,480 Speaker 1: but it also becomes something like magic that we no 543 00:34:20,520 --> 00:34:24,760 Speaker 1: longer understand, and this leads us into our final act 544 00:34:24,960 --> 00:34:29,120 Speaker 1: about our relationship with machines that operate on a different timescale. 545 00:34:29,239 --> 00:34:33,080 Speaker 1: Sometimes we need things to operate, or specifically to pretend 546 00:34:33,239 --> 00:34:37,080 Speaker 1: to operate at our own time scales. So, for example, 547 00:34:37,160 --> 00:34:40,440 Speaker 1: think about the movie Her. There's this guy who's played 548 00:34:40,440 --> 00:34:44,160 Speaker 1: by Joaquin Phoenix. His marriage ends, he's left heartbroken. He 549 00:34:44,200 --> 00:34:48,440 Speaker 1: becomes intrigued with a new operating system in which he 550 00:34:48,520 --> 00:34:52,640 Speaker 1: meets an AI named Samantha who's played by Scarlett Johansson, 551 00:34:52,960 --> 00:34:56,480 Speaker 1: and this AI is sensitive and playful, and their friendship 552 00:34:56,560 --> 00:35:01,040 Speaker 1: soon deepens into love and his relationationship with the AI 553 00:35:01,120 --> 00:35:04,360 Speaker 1: bought means everything to him. Now, this film has an 554 00:35:04,440 --> 00:35:07,759 Speaker 1: incredible ending because at the end he comes to understand 555 00:35:08,520 --> 00:35:11,840 Speaker 1: that she, the AI, is having this same kind of 556 00:35:11,880 --> 00:35:15,920 Speaker 1: relationship with hundreds of thousands of other men, all at 557 00:35:15,920 --> 00:35:20,600 Speaker 1: the same time, because she is computational and operates in 558 00:35:20,680 --> 00:35:23,719 Speaker 1: a different time domain and processes what appears to be 559 00:35:24,320 --> 00:35:28,279 Speaker 1: intimate conversation at a rate millions of times faster than 560 00:35:28,320 --> 00:35:32,800 Speaker 1: our poor brains, and so she is maintaining this intimacy 561 00:35:32,840 --> 00:35:36,600 Speaker 1: with all of these others. She's doing a million things 562 00:35:36,640 --> 00:35:41,279 Speaker 1: in between his pillow whispers, just like your keyboard goes 563 00:35:41,280 --> 00:35:43,719 Speaker 1: to sleep in between your keystrokes. 564 00:35:43,800 --> 00:35:44,760 Speaker 2: Now, the thing I want. 565 00:35:44,560 --> 00:35:46,720 Speaker 1: To surface here is that in order for this AI 566 00:35:46,840 --> 00:35:50,160 Speaker 1: to work, it has to slow way down for each 567 00:35:50,280 --> 00:35:53,640 Speaker 1: user so that they think they're talking to someone at 568 00:35:53,760 --> 00:35:57,120 Speaker 1: their timescale. Now, this question doesn't just exist in the 569 00:35:57,160 --> 00:35:59,840 Speaker 1: realm of movies or theory. Companies have to deal with 570 00:35:59,880 --> 00:36:04,480 Speaker 1: it all the time. For example, when user experienced designers 571 00:36:04,560 --> 00:36:08,200 Speaker 1: work to create good impressions on people, they learn that 572 00:36:08,760 --> 00:36:12,800 Speaker 1: good design is about shaving off confusing messages and getting 573 00:36:12,880 --> 00:36:16,839 Speaker 1: rid of bugs, and they're traditionally taught to remove any 574 00:36:17,000 --> 00:36:22,799 Speaker 1: unnecessary delays. But often a delay is exactly what's called for. 575 00:36:23,360 --> 00:36:23,839 Speaker 2: And this is. 576 00:36:23,800 --> 00:36:28,759 Speaker 1: Something that many user experienced designers have independently discovered, which 577 00:36:28,800 --> 00:36:31,680 Speaker 1: is that people don't always want the results as quickly 578 00:36:31,680 --> 00:36:34,160 Speaker 1: as you can generate them. So take this as an example. 579 00:36:34,239 --> 00:36:37,680 Speaker 1: The people who developed the coinstar machines discover this. These 580 00:36:37,680 --> 00:36:40,080 Speaker 1: are those machines where you dump in a bunch of 581 00:36:40,160 --> 00:36:42,840 Speaker 1: change and accounts the change for you and gives you 582 00:36:42,880 --> 00:36:46,160 Speaker 1: the result. Now, apparently these coinstar machines can do their 583 00:36:46,239 --> 00:36:48,880 Speaker 1: work very quickly, like in a second or two, but 584 00:36:48,920 --> 00:36:52,760 Speaker 1: that speed made customers feel nervous that maybe the counting 585 00:36:52,880 --> 00:36:58,640 Speaker 1: wasn't happening correctly. So Coinstar inserted pre recorded noise of 586 00:36:58,800 --> 00:37:02,200 Speaker 1: change moving through the machine clink clink, clink clink, And 587 00:37:02,280 --> 00:37:06,359 Speaker 1: even though the machine comes to its conclusions rapidly, like 588 00:37:06,440 --> 00:37:09,600 Speaker 1: you have seven dollars and thirty six cents, it purposefully 589 00:37:10,080 --> 00:37:14,120 Speaker 1: delays its output so that the counting seems slower. And 590 00:37:14,160 --> 00:37:18,000 Speaker 1: this is an interesting form of theater, one which our 591 00:37:18,080 --> 00:37:22,480 Speaker 1: machines perform in the time domain, and this is not uncommon. 592 00:37:22,520 --> 00:37:27,720 Speaker 1: Apparently ATM machines insert an artificial delay for a similar reason. 593 00:37:28,160 --> 00:37:29,799 Speaker 1: You stick in your card and you say you want 594 00:37:29,840 --> 00:37:33,520 Speaker 1: one hundred bucks, and the machine words and clicks and 595 00:37:33,600 --> 00:37:37,600 Speaker 1: turns for a while. There's no necessity, like a computational 596 00:37:37,680 --> 00:37:41,600 Speaker 1: or mechanical necessity, for ATM machines to be so slow, 597 00:37:41,920 --> 00:37:46,720 Speaker 1: but apparently customers are more satisfied when the money doesn't 598 00:37:46,760 --> 00:37:51,400 Speaker 1: come out instantly, and software designers increasingly have to deal 599 00:37:51,760 --> 00:37:56,000 Speaker 1: with inserting artificial delays because of the speed of computation. 600 00:37:56,120 --> 00:37:59,920 Speaker 1: So there was one company that builds software that creates 601 00:38:00,120 --> 00:38:02,920 Speaker 1: an online blog for you with the design and the 602 00:38:02,920 --> 00:38:06,400 Speaker 1: title page and templates for your blog posts. And what 603 00:38:06,440 --> 00:38:09,040 Speaker 1: they found was that when users got to the end 604 00:38:09,160 --> 00:38:12,240 Speaker 1: of their inputs and clicked to the button that said 605 00:38:12,600 --> 00:38:17,120 Speaker 1: create my blog, the blog appeared for the user instantly, 606 00:38:17,239 --> 00:38:20,160 Speaker 1: and users felt confused. They weren't sure if something was 607 00:38:20,200 --> 00:38:23,560 Speaker 1: wrong or if something had broken. So after some testing, 608 00:38:24,160 --> 00:38:27,560 Speaker 1: the software company added a page in the middle that 609 00:38:27,640 --> 00:38:32,040 Speaker 1: said creating your blog and had a spinning wheel, and 610 00:38:32,080 --> 00:38:35,600 Speaker 1: then after about seven seconds, it would send users to 611 00:38:35,680 --> 00:38:39,200 Speaker 1: their newly minted blog page, and users felt much more 612 00:38:39,239 --> 00:38:43,280 Speaker 1: satisfied just based on the delay, just based on things 613 00:38:43,320 --> 00:38:47,720 Speaker 1: happening at a timescale that they were comfortable with. Now, 614 00:38:48,120 --> 00:38:52,360 Speaker 1: why are users of coinstar and ATM machines and blog 615 00:38:52,440 --> 00:38:57,200 Speaker 1: software happier with a delay? Sometimes designers will point out 616 00:38:57,200 --> 00:39:00,719 Speaker 1: that a user might have anxiety about it service, like 617 00:39:00,960 --> 00:39:03,640 Speaker 1: is the coinstar counting all my coins correctly? Or is 618 00:39:03,680 --> 00:39:06,960 Speaker 1: the ATM giving me the right amount of money? Or 619 00:39:07,440 --> 00:39:10,799 Speaker 1: building a blog must be a really complicated thing, and 620 00:39:10,840 --> 00:39:14,239 Speaker 1: so the best strategy is to address this anxiety by 621 00:39:14,320 --> 00:39:18,640 Speaker 1: convincing them that everything is running at a careful pace. 622 00:39:19,160 --> 00:39:20,960 Speaker 1: I think that's a pretty good hypothesis, but I think 623 00:39:21,000 --> 00:39:23,319 Speaker 1: there might be something else going on here, and it's 624 00:39:23,320 --> 00:39:25,160 Speaker 1: not just about anxiety. 625 00:39:25,440 --> 00:39:27,440 Speaker 2: I think this might be an expression. 626 00:39:26,960 --> 00:39:30,680 Speaker 1: Of how we value things and how that pivots on 627 00:39:30,760 --> 00:39:34,200 Speaker 1: what I have called the effort phenomenon. I discussed this 628 00:39:34,239 --> 00:39:36,440 Speaker 1: way back in episode six. 629 00:39:36,200 --> 00:39:39,839 Speaker 2: About AI and why we value a. 630 00:39:39,960 --> 00:39:43,560 Speaker 1: Book written by a person much more than a book 631 00:39:43,600 --> 00:39:46,600 Speaker 1: written by AI, even if the exact same words. 632 00:39:47,200 --> 00:39:48,279 Speaker 2: It's not clear. 633 00:39:48,040 --> 00:39:50,319 Speaker 1: Why we would value one over the other. But the 634 00:39:50,360 --> 00:39:53,560 Speaker 1: suggestion I made is that we care very much about 635 00:39:53,600 --> 00:39:57,319 Speaker 1: the effort that goes into something. For example, take two 636 00:39:57,360 --> 00:40:00,360 Speaker 1: pieces of art. Let's say someone makes a replica of 637 00:40:00,400 --> 00:40:05,880 Speaker 1: a Michelangelo statue entirely by gluing quarters together in a 638 00:40:05,960 --> 00:40:09,880 Speaker 1: giant three dimensional structure, and someone else puts a single 639 00:40:09,920 --> 00:40:13,319 Speaker 1: red dot in the middle of a canvas. How much 640 00:40:13,360 --> 00:40:16,080 Speaker 1: would you pay for these two different pieces of art. 641 00:40:16,440 --> 00:40:19,800 Speaker 1: Presumably you would pay more for that which you believe 642 00:40:19,960 --> 00:40:24,239 Speaker 1: took more effort. And in the same way, the coinstar 643 00:40:24,440 --> 00:40:28,000 Speaker 1: or the ATM machine or the blog post can give 644 00:40:28,040 --> 00:40:31,879 Speaker 1: you the illusion that it has just put in a 645 00:40:31,920 --> 00:40:36,160 Speaker 1: lot of effort. It's just pulling a sleight of hand 646 00:40:36,239 --> 00:40:39,520 Speaker 1: in the time domain, but it's enough to align with 647 00:40:39,640 --> 00:40:44,839 Speaker 1: your human expectations of what hard work looks like. It 648 00:40:44,920 --> 00:40:48,919 Speaker 1: lines up better with your internal model of how hard 649 00:40:48,960 --> 00:40:51,880 Speaker 1: it would be for you to count a bunch of 650 00:40:51,960 --> 00:40:56,759 Speaker 1: coins or bills, or program a blog template from scratch. 651 00:40:57,440 --> 00:40:59,759 Speaker 1: And so it's fascinating that if the machine or the 652 00:40:59,800 --> 00:41:03,880 Speaker 1: software it doesn't give you this false delay, that actually 653 00:41:03,920 --> 00:41:07,480 Speaker 1: diminishes how much you think the service is worth. And 654 00:41:07,480 --> 00:41:11,239 Speaker 1: this becomes like a temporal touring test. You remember, the 655 00:41:11,280 --> 00:41:14,280 Speaker 1: touring test is whether you can tell if a conversation 656 00:41:14,400 --> 00:41:17,160 Speaker 1: partner is a human or a computer. And I think 657 00:41:17,320 --> 00:41:21,480 Speaker 1: increasingly our computers and our devices will have to be 658 00:41:21,640 --> 00:41:27,239 Speaker 1: sensitive to the temporal aspect and slow way down if 659 00:41:27,280 --> 00:41:34,560 Speaker 1: they hope to fool us. Okay, so let's wrap up. Ah, 660 00:41:34,719 --> 00:41:38,120 Speaker 1: there's the next keystroke by the human who was typing. 661 00:41:38,840 --> 00:41:40,720 Speaker 2: That took a very long time. 662 00:41:41,200 --> 00:41:43,279 Speaker 1: So do you remember this thing I said at the 663 00:41:43,360 --> 00:41:46,319 Speaker 1: very beginning, I was imagining that we land on a 664 00:41:46,360 --> 00:41:48,359 Speaker 1: planet for a month and we don't see anything move, 665 00:41:48,440 --> 00:41:51,400 Speaker 1: and we conclude that there are only these tree like 666 00:41:51,440 --> 00:41:53,640 Speaker 1: things on the planet, and we blast off because it's 667 00:41:53,680 --> 00:41:56,359 Speaker 1: not so interesting. But it turns out that they live 668 00:41:56,400 --> 00:42:01,640 Speaker 1: on a much slower timescale, millionions of times slower, but 669 00:42:01,719 --> 00:42:06,160 Speaker 1: they have societies and loves and wars and clubs, and 670 00:42:06,200 --> 00:42:09,920 Speaker 1: it's just that to us, we wouldn't even recognize something 671 00:42:10,040 --> 00:42:13,040 Speaker 1: moving at that slow pace. So here we are in 672 00:42:13,080 --> 00:42:17,279 Speaker 1: the twenty first century, giving birth to a new species. 673 00:42:17,320 --> 00:42:24,160 Speaker 1: We're sparking new machines into existence and to our electronic progeny. 674 00:42:25,000 --> 00:42:27,440 Speaker 2: We are the tree People. 675 00:42:32,360 --> 00:42:35,560 Speaker 1: Go to Eagleman dot com slash podcast for more information 676 00:42:35,719 --> 00:42:38,759 Speaker 1: and to find further reading. Send me an email at 677 00:42:38,840 --> 00:42:42,239 Speaker 1: podcasts at eagleman dot com with questions or discussion, and 678 00:42:42,360 --> 00:42:45,840 Speaker 1: check out and subscribe to Inner Cosmos on YouTube for 679 00:42:45,960 --> 00:42:49,479 Speaker 1: videos of each episode and to leave comments until next time. 680 00:42:49,680 --> 00:42:52,960 Speaker 1: I'm David Eagleman, and we have been traveling together in 681 00:42:53,040 --> 00:43:02,200 Speaker 1: the Inner Cosmos in th