WEBVTT 00:00:29.359 --> 00:00:30.719 It is 1893. 00:00:30.719 --> 00:00:35.119 A damp dark night on a lake in northern Michigan. 00:00:35.119 --> 00:00:39.920 Two men sit in a rowboat drifting along the reedy edge. 00:00:39.920 --> 00:00:46.560 In the stern, John Hammer holds a wooden paddle and keeps their progress quiet. 00:00:46.560 --> 00:00:53.520 At the bow, a wooden frame balances two heavy box cameras side by side. 00:00:53.520 --> 00:00:59.119 Each holding a single sheet of glass coated in light-sensitive emulsion. 00:00:59.119 --> 00:01:01.200 One exposure. 00:01:01.200 --> 00:01:08.799 Above them, a kerosene lamp with a reflector throws a single sharp spear of light across the water. 00:01:08.799 --> 00:01:17.280 Behind that beam, hidden in the shadows, George Shiras holds a small pan of magnesium powder. 00:01:17.280 --> 00:01:22.640 They watch the shoreline until a pair of eyes catches the light. 00:01:22.640 --> 00:01:25.599 Twin emerald points in the brush. 00:01:25.599 --> 00:01:30.319 A deer stands frozen, transfixed by the lamp. 00:01:30.319 --> 00:01:32.640 The boat creeps closer. 00:01:32.640 --> 00:01:37.280 Fifty feet, forty feet, thirty. 00:01:37.280 --> 00:01:40.719 Shiras squeezes the trigger. 00:01:40.719 --> 00:01:43.359 There is a crack, a white hot hiss. 00:01:43.359 --> 00:01:50.079 For a fraction of a second, the deer, the hills, the lake all stand out in dark daylight. 00:01:50.079 --> 00:01:55.599 Then the smoke rolls over the water and the blackness rushes back in. 00:01:55.599 --> 00:02:04.239 When they develop the plate, the deer is still there, stunned at the water's edge, alive and fixed in glass. 00:02:04.239 --> 00:02:10.000 But the only creatures they can capture this way are the ones that come to the shoreline. 00:02:10.000 --> 00:02:13.439 Shiras wants to see the rest of the forest. 00:02:13.439 --> 00:02:16.800 And he is at his core a hunter. 00:02:16.800 --> 00:02:20.800 He knows the woods by habit, by patience. 00:02:20.800 --> 00:02:22.960 He knows where the deer walk. 00:02:22.960 --> 00:02:33.439 So he carries the heavy camera into the woods, stretches a single black thread across an animal path, and connects it to the shutter and the flash. 00:02:33.439 --> 00:02:35.439 Then he leaves. 00:02:35.439 --> 00:02:39.759 Sometime after dark, a body crosses the trail. 00:02:39.759 --> 00:02:45.360 The thread tightens, and the darkness turns white. 00:02:45.360 --> 00:02:49.919 In the morning, the deer is there on the glass. 00:02:49.919 --> 00:02:54.479 Shiras has built a camera that waits. 00:02:54.479 --> 00:02:58.960 His thread has given it a sense of touch. 00:02:58.960 --> 00:03:01.919 But the thread sags in the rain. 00:03:01.919 --> 00:03:04.000 It tangles. And it cannot catch a bat or a bird. 00:03:04.000 --> 00:03:11.520 Or anything that steps over it or under it. 00:03:11.520 --> 00:03:16.159 To remove the thread, the camera needed to feel more than pressure. 00:03:16.159 --> 00:03:18.960 It needed to feel heat. 00:03:18.960 --> 00:03:24.879 I'm Daina Bouquin, and this is Found in the Machine. 00:03:24.879 --> 00:03:33.599 Every warm thing gives off invisible light. 00:03:33.599 --> 00:03:36.159 Infrared radiation. 00:03:36.159 --> 00:03:41.680 A deer. 00:03:41.680 --> 00:03:44.879 The raccoon beside the trash cans. 00:03:44.879 --> 00:03:51.680 You, right now, you are glowing, broadcasting your presence wherever you are. 00:03:51.680 --> 00:03:57.199 But if you warm up a piece of tourmaline, it doesn't just glow. 00:03:57.199 --> 00:04:01.280 It begins to pull at the world around it. 00:04:01.280 --> 00:04:07.680 Dust, ash, and tiny pieces of straw cling to its surface. 00:04:07.680 --> 00:04:17.199 Inside the stone, a change in temperature shifts its electrical balance and creates a faint charge at the surface. 00:04:17.199 --> 00:04:21.040 Not heat in electricity out exactly. 00:04:21.040 --> 00:04:23.920 A change in temperature in. 00:04:23.920 --> 00:04:26.720 A fleeting electrical signal out. 00:04:26.720 --> 00:04:30.079 That signal can be amplified. 00:04:30.079 --> 00:04:34.000 It can be used to close a circuit. 00:04:34.000 --> 00:04:36.079 It can wake a camera. 00:04:36.079 --> 00:04:39.040 This is pyroelectricity. 00:04:39.040 --> 00:04:45.680 But a single detector cannot tell what caused the temperature to change. 00:04:45.680 --> 00:04:49.519 To a stone, warmer is simply warmer. 00:04:49.519 --> 00:04:53.759 The sun rising over a clearing warms the air. 00:04:53.759 --> 00:04:56.399 The detector warms. 00:04:56.399 --> 00:05:02.240 Wire that detector to a camera and it will confidently photograph the empty field. 00:05:02.240 --> 00:05:12.480 To use pyroelectricity, to make the signal useful, engineers needed the machine to also notice movement. 00:05:12.480 --> 00:05:22.000 So they paired two pyroelectric elements together and balanced them against each other like two sides of a scale. 00:05:22.000 --> 00:05:32.399 When a slow even shift in temperature, like sunlight warming a room, reaches both elements at once, the scale stays balanced. 00:05:32.399 --> 00:05:35.519 That broad change produces silence. 00:05:35.519 --> 00:05:42.480 But if a body moves through the room, it warms one element and then the other. 00:05:42.480 --> 00:05:46.160 For a fraction of a second, the scale tips. 00:05:46.160 --> 00:05:51.920 That disagreement becomes a trigger, telling a camera to wake up. 00:05:51.920 --> 00:05:56.079 This is what happens with the floodlight above your driveway. 00:05:56.079 --> 00:05:59.680 With the lights that turn on when you enter a room. 00:05:59.680 --> 00:06:02.720 It's what happens inside a camera trap. 00:06:02.720 --> 00:06:07.600 The sensors in those little machines do not see you. 00:06:07.600 --> 00:06:16.639 They contain pyroelectric elements that just notice when the infrared patterns in front of them briefly stop agreeing with each other. 00:06:16.639 --> 00:06:23.360 It is called a passive infrared sensor because it sends nothing into the world. 00:06:23.360 --> 00:06:26.560 No beam, no searching pulse. 00:06:26.560 --> 00:06:32.160 It simply waits for the invisible light already being emitted around it to move in a certain way. 00:06:32.160 --> 00:06:40.720 But even with two pyroelectric elements, the sensor still needs help to focus all that infrared light. 00:06:40.720 --> 00:06:46.639 And it turns out the cheapest solution came from the sea. 00:06:46.639 --> 00:06:52.800 From the work of an early 19th-century French physicist named Augustin Fresnel. 00:06:52.800 --> 00:07:06.000 Fresnel designed a new kind of glass for lighthouses, made of step-cut concentric rings that could gather the scattered, wasted light that flew towards the land instead of the water. 00:07:06.000 --> 00:07:16.879 His lens bent the lighthouse's brightness into a single intense beam aimed at the horizon so the ships could see the light from further away. 00:07:16.879 --> 00:07:30.560 Now if you look at the milky plastic window on a motion detecting light, or a burglar alarm, or a trail camera, you will see a pattern of tiny molded ridges. 00:07:30.560 --> 00:07:36.160 That is a Fresnel lens, molded in plastic. 00:07:36.160 --> 00:07:38.480 Cheap and easy to make. 00:07:38.480 --> 00:07:49.600 But instead of throwing visible light out into the dark to warn ships, it reaches out and gathers up the scattered invisible light and bends it inward toward the sensor. 00:07:49.600 --> 00:08:00.160 Now take that sensor with its paired elements and its inside-out lighthouse lens and strap it to a tree on the coast of British Columbia. 00:08:00.160 --> 00:08:07.199 In Heiltsuk territory, an invasion was quietly taking hold under the water. 00:08:07.199 --> 00:08:10.319 European green crabs were moving in. 00:08:10.319 --> 00:08:15.839 They threatened the native clams, the salmon, the eelgrass. 00:08:15.839 --> 00:08:19.439 So the Nation's Environmental Guardians fought back. 00:08:19.439 --> 00:08:25.439 They sunk traps into the deep water, baiting them with herring and sea lion meat. 00:08:25.439 --> 00:08:29.279 But the traps started coming back broken. 00:08:29.279 --> 00:08:39.519 The Guardians would find the cages dragged out of the water, netting shredded, the little orange bait cups mangled or completely missing. 00:08:39.519 --> 00:08:43.039 Sometimes a trap vanished entirely. 00:08:43.039 --> 00:08:50.799 At first they guessed it was sea lions or maybe otters, but the damage didn't quite make sense. 00:08:50.799 --> 00:08:54.960 They needed to know what was happening beneath the surface. 00:08:54.960 --> 00:08:58.480 So they pointed a camera at the surf to watch. 00:08:58.480 --> 00:09:06.960 Then, one day in May 2024, the sensor registered a change in temperature on the shore. 00:09:06.960 --> 00:09:09.120 The camera woke up. 00:09:09.120 --> 00:09:13.279 It recorded a female coastal wolf. 00:09:13.279 --> 00:09:16.720 In this territory, the wolves are different. 00:09:16.720 --> 00:09:24.080 They cross open water between islands with their backs submerged, only eyes and ears and snouts above the surface. 00:09:24.080 --> 00:09:28.240 They bite the heads off spawning salmon and eat only the brains. 00:09:28.240 --> 00:09:30.960 Crack muscles with their jaws. 00:09:30.960 --> 00:09:34.879 But this wolf wasn't hunting in the tide pools. 00:09:34.879 --> 00:09:39.200 The camera caught her wading into the cold salt water. 00:09:39.200 --> 00:09:44.240 She swam out until she reached a floating buoy attached to a crab trap. 00:09:44.240 --> 00:09:50.480 She grabbed it in her jaws and swam back, dropping it on the rocks. 00:09:50.480 --> 00:09:56.080 Then she took the rope in her mouth and she pulled. 00:09:56.080 --> 00:10:07.200 She stood on the shore, pulling the line over and over, understanding that this rope connected to something that she could not see, heavy and hidden in the depth. 00:10:07.200 --> 00:10:13.919 She dragged the awkward contraption all the way onto the beach and tore through the mesh. 00:10:13.919 --> 00:10:20.720 She extracted that little orange bait cup, licked the sea lion meat clean, and trotted away. 00:10:20.720 --> 00:10:25.039 In less than three minutes, the camera solved the mystery. 00:10:25.039 --> 00:10:29.200 It recorded a wolf executing a deliberate multi-step heist. 00:10:29.200 --> 00:10:37.039 It captured the animal understanding a cause and effect relationship that scientists didn't know wild wolves were capable of grasping. 00:10:37.039 --> 00:10:45.360 But the camera that caught the wolves hauling in crab traps also caught many, many, many things that were not wolves. 00:10:45.360 --> 00:10:52.399 A branch warmed by the sun swinging across the field of view, the camera wakes. 00:10:52.399 --> 00:10:57.440 A patch of light lands suddenly on wet leaves and the camera wakes. 00:10:57.440 --> 00:11:03.600 For every photograph of a wolf on a beach, there may be dozens with nothing in them at all. 00:11:03.600 --> 00:11:06.240 Nothing, nothing, nothing. 00:11:06.240 --> 00:11:07.440 Wolf. 00:11:07.440 --> 00:11:08.559 Nothing. 00:11:08.559 --> 00:11:09.440 Again. 00:11:09.440 --> 00:11:13.440 But a false alarm only wastes time. 00:11:13.440 --> 00:11:17.519 A missed animal can rewrite the record. 00:11:17.519 --> 00:11:27.440 It can say "Nothing was here." For 30 years, engineers tried to make the signal itself smarter. 00:11:27.440 --> 00:11:31.519 They filtered waveforms and analyzed pulse shapes. 00:11:31.519 --> 00:11:40.320 But the size and signal depends on how warm the body is, how fast it moves, how far away it stands. 00:11:40.320 --> 00:11:43.679 And none of that tells you what it is. 00:11:43.679 --> 00:11:46.960 So people sit at screens. 00:11:46.960 --> 00:11:54.639 Students and tens of thousands of volunteers working on citizen science projects from home. 00:11:54.639 --> 00:11:58.960 They move through images one at a time. 00:11:58.960 --> 00:12:00.639 Nothing. 00:12:00.639 --> 00:12:01.519 Nothing. Nothing. 00:12:02.399 --> 00:12:04.240 Maybe a bird. 00:12:04.240 --> 00:12:05.440 Nothing. Nothing. 00:12:06.240 --> 00:12:07.360 Wolf. 00:12:07.360 --> 00:12:13.360 Each label turns the photograph into an example. 00:12:13.360 --> 00:12:17.600 Collect enough examples and they can train a new machine. 00:12:17.600 --> 00:12:21.440 The first machine answered, did something change? 00:12:21.440 --> 00:12:26.399 But the second one tries to answer, what was it? 00:12:26.399 --> 00:12:30.159 This is supervised machine learning. 00:12:30.159 --> 00:12:35.919 But the machine learning classifier does not know what a wolf is. 00:12:35.919 --> 00:12:41.360 It learns statistical relationships from images that people have labeled wolf. 00:12:41.360 --> 00:12:46.159 Textures, contours, arrangements of pixels. 00:12:46.159 --> 00:12:51.600 When a new photograph arrives, it estimates which of its categories fits best. 00:12:51.600 --> 00:12:52.639 Nothing. Nothing. 00:12:53.519 --> 00:13:03.679 Something to label "wolf." The sensor detects, the camera records, the classifier proposes a name. 00:13:03.679 --> 00:13:06.960 Then people decide what that name means. 00:13:06.960 --> 00:13:15.039 For most of a century, there were almost no wolves left in the American West. 00:13:15.039 --> 00:13:18.879 We had shot and poisoned them nearly to nothing. 00:13:18.879 --> 00:13:23.759 As the few survivors began to wander back, the reports came in. 00:13:23.759 --> 00:13:27.279 A coyote seen from a moving car. 00:13:27.279 --> 00:13:29.679 A shape on a ridge. 00:13:29.679 --> 00:13:32.240 A big dog at dusk. 00:13:32.240 --> 00:13:37.840 Wildlife offices filled with sightings of wolves that were not wolves. 00:13:37.840 --> 00:13:38.879 Nothing. Nothing. 00:13:39.600 --> 00:13:40.480 Nothing. 00:13:40.480 --> 00:13:47.759 But in 2023, Colorado began reintroducing gray wolves to the state's western slope. 00:13:47.759 --> 00:13:58.000 Animals were captured in Oregon and British Columbia, fitted with GPS collars that logged their position every four hours, and released into the mountains. 00:13:58.000 --> 00:14:01.039 Trail cameras watched den sites. 00:14:01.039 --> 00:14:04.399 Aerial surveys tracked pack movements. 00:14:04.399 --> 00:14:08.240 Biologists collected scat for genetic analysis. 00:14:08.240 --> 00:14:13.840 It became one of the most closely monitored wolf populations in history. 00:14:13.840 --> 00:14:22.799 In June 2025, a trail camera outside a den in Root County captured the first images of pups born to the King Mountain Pack. 00:14:22.799 --> 00:14:26.720 No person was there to see them, but the camera was. 00:14:26.720 --> 00:14:32.799 It recorded the tumbling, clumsy little bodies before anyone else even knew they had arrived. 00:14:32.799 --> 00:14:36.639 Then came January 2026. 00:14:36.639 --> 00:14:43.600 A licensed hunter in Northwest Colorado saw an animal. 00:14:43.600 --> 00:14:47.360 And to him it looked like a coyote. 00:14:47.360 --> 00:14:52.000 Coyotes are legal to hunt year-round in Colorado. 00:14:52.000 --> 00:14:55.200 So he aimed. 00:14:55.200 --> 00:14:57.600 He fired. 00:14:57.600 --> 00:15:05.120 She was a wolf. 00:15:05.120 --> 00:15:12.000 She had been translocated from British Columbia a year earlier to support repopulation efforts. 00:15:12.000 --> 00:15:16.879 When the hunter saw her GPS collar, he reported the mistake immediately. 00:15:16.879 --> 00:15:22.080 That collar had been logging her position every four hours. 00:15:22.080 --> 00:15:25.120 The trail cameras were recording. 00:15:25.120 --> 00:15:28.320 The data was being collected. 00:15:28.320 --> 00:15:32.240 None of the technology failed. 00:15:32.240 --> 00:15:42.080 We have spent more than a century building machines that wait in our place because our own attention falters. 00:15:42.080 --> 00:15:44.960 We need them to sit in the snow. 00:15:44.960 --> 00:15:47.120 To wait in the dark. 00:15:47.120 --> 00:15:50.799 To feel the heat we cannot feel. 00:15:50.799 --> 00:15:55.360 We need them to do the mundane, tireless watching. 00:15:55.360 --> 00:16:02.879 But when George Shiras drifted across that lake in a rowboat in 1893, he didn't just want data. 00:16:02.879 --> 00:16:07.039 He used the camera to show other people what he had seen. 00:16:07.039 --> 00:16:16.879 To make them understand that the deer at the water's edge was not just a shape, not just a target, but something worth paying attention to. 00:16:16.879 --> 00:16:20.720 His images were meant to create a sense of presence. 00:16:20.720 --> 00:16:23.440 To make people really look. 00:16:23.440 --> 00:16:26.720 Shiras gave the machine a thread. 00:16:26.720 --> 00:16:30.000 Engineers gave it a way to feel heat. 00:16:30.000 --> 00:16:36.559 Lenses allowed it to focus, and people trained algorithms to name what it sees. 00:16:36.559 --> 00:16:40.480 The machine can document what moves through the dark. 00:16:40.480 --> 00:16:43.519 It can show us what we missed. 00:16:43.519 --> 00:16:46.240 But the machine does not think. 00:16:46.240 --> 00:16:49.039 It does not believe. 00:16:49.039 --> 00:16:51.039 It does not doubt. 00:16:51.039 --> 00:16:57.039 It feels the heat, throws the switch, and leaves the rest to us. 00:16:57.039 --> 00:17:02.879 I'm Daina Bouquin and this is Found in the Machine. 00:17:02.879 --> 00:17:09.119 Before I go, I want to tell you about something new. 00:17:09.119 --> 00:17:19.599 I've just launched a Patreon for the show, and members will get access to a bonus mini episode series called Found in the Margins. 00:17:19.599 --> 00:17:24.480 I'll be putting it out every other week, alternating with the main feed. 00:17:24.480 --> 00:17:36.559 These are short stories about the strange and wonderful stuff I stumble across while researching the main show, and just I'm not quite sure how to ever turn it into a full episode. 00:17:36.559 --> 00:17:42.000 And I'll be honest with you, I was laid off from my day job this week. 00:17:42.000 --> 00:17:48.880 So if you've ever thought about supporting the show, now it would mean a lot. 00:17:48.880 --> 00:17:56.559 You can find the Patreon link along with other ways to support Found in the Machine at foundinthemachine.com/support. 00:17:56.559 --> 00:18:02.400 And thank you as always for listening.