WEBVTT - Why Silicon Valley Is Hiring Bird Experts 0:00:02.720 --> 0:00:05.320 A few years ago, around the time of Twitter's I 0:00:05.480 --> 0:00:08.640 p O, I noticed what I thought was an odd coincidence. 0:00:09.320 --> 0:00:13.720 It's chief engineer had studied birds. Specifically, he'd studied the 0:00:13.760 --> 0:00:17.360 auditory cortex of zebra finches. I thought that was pretty 0:00:17.360 --> 0:00:21.599 funny for an engineer, especially because Twitter's mascot is that 0:00:21.680 --> 0:00:24.480 little blue bird. I forgot about it until a couple 0:00:24.480 --> 0:00:27.360 of years ago when I noticed another bird brain scholar 0:00:27.400 --> 0:00:30.680 in the top echelons of tech. This person had been 0:00:30.800 --> 0:00:34.600 hired by Elon Musk, the entrepreneur behind Tesla and SpaceX, 0:00:34.960 --> 0:00:38.880 to join his new company, Neuralink. Neuralink is a very 0:00:38.920 --> 0:00:43.280 secretive futuristic company which is trying to supercharge the human brain. 0:00:43.880 --> 0:00:47.440 This all sounds pretty obscure. You'd think studying bird brains 0:00:47.479 --> 0:00:50.800 wouldn't be too relevant to studying human brains or figuring 0:00:50.800 --> 0:00:54.200 out social media, right, That's why I remembered it. It 0:00:54.280 --> 0:00:57.279 just felt so random, But now here were two. So 0:00:57.360 --> 0:00:59.280 one day, when I was a little bored, I just 0:00:59.360 --> 0:01:02.320 tried type being zebra finch and the names of several 0:01:02.400 --> 0:01:05.120 big tech companies into Google. And what did you find? 0:01:05.680 --> 0:01:08.160 I found quite a few employees who knew a lot 0:01:08.240 --> 0:01:12.200 about zebra finches. That was that companies like Intelent, Apple 0:01:12.319 --> 0:01:15.320 and Google too. I was surprised. Okay, hang on, Sarah, 0:01:15.360 --> 0:01:18.319 I'm googling this. The zebra finch is the most common 0:01:18.400 --> 0:01:21.520 astrill did finch of central Australia and ranges over most 0:01:21.520 --> 0:01:24.440 of the continent, avoiding only the cool, moist south and 0:01:24.520 --> 0:01:28.920 tropical far North. Zebrafinches are loud and boisterous singers. So 0:01:28.959 --> 0:01:31.600 what's the connection here, Well, that's where I jumped into 0:01:31.600 --> 0:01:35.000 the reporting process. Our goal is to figure out why 0:01:35.080 --> 0:01:38.160 tech companies are hiring bird brain experts, and this little 0:01:38.240 --> 0:01:41.679 quest has taken us to college campuses around the country. 0:01:41.840 --> 0:01:45.440 We visited several university labs. Some smelled better than others. 0:01:46.040 --> 0:01:48.360 We heard a lot of birds on So this is 0:01:48.400 --> 0:01:53.160 our guy, and so they're they're very light, um, but 0:01:53.280 --> 0:01:58.560 super active. How big would you say that thing is? 0:01:58.600 --> 0:02:01.840 It's like, so they're about fifty in Graham's total weight. 0:02:01.960 --> 0:02:06.640 They're mostly feathers. That's Tim Achi, a neuroscientist and assistant 0:02:06.640 --> 0:02:10.280 professor at Boston University, showing us czebra finch in his 0:02:10.400 --> 0:02:13.880 bird lab and telling us about his research on bird brains. 0:02:14.360 --> 0:02:18.079 He's running some pretty extraordinary experiments. Oh no, but the 0:02:18.480 --> 0:02:22.280 the implant. So this is actually just a lens. And 0:02:22.320 --> 0:02:24.079 so if you look at the top, so you see 0:02:24.080 --> 0:02:26.079 that little black thing that's sticking up, So that's a lens. 0:02:26.760 --> 0:02:28.480 If you look down and you couldn't see the brain, 0:02:28.480 --> 0:02:30.360 but there's not probably not enough light getting down there 0:02:30.400 --> 0:02:33.519 that you could actually see it. And it turns out 0:02:33.560 --> 0:02:36.120 what Tim's learning about the brains of these tiny birds 0:02:36.200 --> 0:02:39.120 is of great interest to the world's largest tech companies. 0:02:39.720 --> 0:02:43.800 I'm brad Stone and Sarah mcbrad that, I'm Ashley Vance, 0:02:44.120 --> 0:02:52.320 and you're listening to decrypt it. So, guys, tell me 0:02:52.360 --> 0:02:54.840 a little bit more about Tim Acchi and his I'm 0:02:54.880 --> 0:02:58.800 sure wonderfully smelling lab out in Boston. Yeah, well, Tim's 0:02:58.880 --> 0:03:00.919 lab was one of the key st ups on our trip. 0:03:01.280 --> 0:03:04.680 Tim took over a Boston University lab from Tim Gardner, 0:03:04.880 --> 0:03:07.560 the guy who left to work for Elon Musk and Neuralink. 0:03:07.919 --> 0:03:10.560 Tim Acchi is a forty year old guy with black 0:03:10.600 --> 0:03:14.280 frame glasses who throws in plenty of references to Portlandia 0:03:14.440 --> 0:03:17.360 and other TV shows when he check chats on his 0:03:17.440 --> 0:03:21.400 office wall. He's got a print of voltage chases called 0:03:21.639 --> 0:03:24.959 five Seconds of Donkey Kong. Now we're talking about something 0:03:25.000 --> 0:03:27.600 I'm familiar with, So what's the connection here. It turns 0:03:27.600 --> 0:03:30.520 out that that old eighties video game became the subject 0:03:30.639 --> 0:03:34.400 of a famous neuroscience research paper. There's one other thing 0:03:34.440 --> 0:03:37.200 you should know about Tim. He has a giant zebra 0:03:37.280 --> 0:03:41.440 finch tattoo on his right forearm. Brad, how familiar are 0:03:41.480 --> 0:03:43.839 you with zebra finch? Well, actually, other than my very 0:03:43.840 --> 0:03:47.000 brief Wikipedia search just now, other than small birds that 0:03:47.120 --> 0:03:49.440 sing a lot, I would say not very familiar. They 0:03:49.480 --> 0:03:53.080 are very cute little birds. Indeed, they're about four inches long. 0:03:53.280 --> 0:03:56.080 The males have orange cheeks and black and white striped 0:03:56.080 --> 0:03:59.320 feathers across their chests, hence the name. They're super easy 0:03:59.360 --> 0:04:02.760 to breed, and they chirp a lot. Here's Tim's interpretation. 0:04:04.760 --> 0:04:07.360 So how did Tim get into studying bird brains? He 0:04:07.440 --> 0:04:10.880 took a circuitous route into neuroscience. One thing I found 0:04:10.920 --> 0:04:13.720 interesting is he started as an engineer working at a 0:04:13.760 --> 0:04:17.440 company that helped factories to automate. His job was teaching 0:04:17.520 --> 0:04:20.640 robots how to sort stuff, everything from car parts to 0:04:20.680 --> 0:04:23.680 gives moos for circuit boards. It was just astounding to 0:04:23.680 --> 0:04:25.599 me how difficult it was to get things to do this, 0:04:25.680 --> 0:04:29.640 and these were tasks that you know, children do. It 0:04:29.680 --> 0:04:33.359 really put in my mind the idea that, you know, 0:04:33.480 --> 0:04:37.000 a lot of the things that that children can do 0:04:37.200 --> 0:04:43.080 almost effortlessly and without almost any training, are incredibly impossible 0:04:43.400 --> 0:04:46.720 to get artificial systems to do or take an enormous 0:04:46.720 --> 0:04:49.720 amount of thought. Okay, so I think I understand Tim 0:04:49.800 --> 0:04:52.599 was curious about why a certain task is so easy 0:04:52.680 --> 0:04:55.400 for a child but so difficult for a robot. Yeah, 0:04:55.480 --> 0:04:58.640 that's exactly it. So after a detour to Tim ended 0:04:58.720 --> 0:05:02.040 up studying neuroscience at Harvard, and that's where he discovered 0:05:02.120 --> 0:05:05.200 zebra finches. Now he's teaching it be you and doing 0:05:05.200 --> 0:05:08.440 his own research. One thing Tim focuses on is how 0:05:08.560 --> 0:05:11.680 zebra finches learned to sing. Tim described his study of 0:05:11.760 --> 0:05:14.280 zebra finches is more of a means to an end. 0:05:14.880 --> 0:05:17.560 I don't, you know, really think of myself or or 0:05:17.640 --> 0:05:24.080 care too much about, you know, songbird neuroscience specifically, I 0:05:24.200 --> 0:05:27.080 take it as a as a way to investigate general 0:05:27.080 --> 0:05:31.640 principles and mechanisms in neuroscience and how brains function generally. 0:05:31.880 --> 0:05:34.280 So Tim is saying that he studies these teeny tiny 0:05:34.360 --> 0:05:36.600 zebra finch brains because he thinks it will give us 0:05:36.640 --> 0:05:40.040 insights into the way human brains function. That's right. Researchers 0:05:40.040 --> 0:05:42.480 study all different kinds of animals for all different kinds 0:05:42.480 --> 0:05:44.640 of purposes, but in this case they're trying to learn 0:05:44.640 --> 0:05:47.599 more about the human brain. Seeing how birds learn to sing, 0:05:47.680 --> 0:05:51.280 for example, can provide insights into how we learn things. 0:05:51.360 --> 0:05:55.080 So I think that you know, the songbird is one 0:05:55.120 --> 0:05:59.480 of those systems that we probably understand the best in 0:05:59.560 --> 0:06:02.840 terms of different brain regions that are involved, in terms 0:06:02.839 --> 0:06:05.600 of the roles of those different brain regions, and so 0:06:05.720 --> 0:06:08.799 I think that we can ask very very precise questions 0:06:09.000 --> 0:06:14.840 in the songbird about the interaction between brain activity and behavior. Sarah, 0:06:14.880 --> 0:06:17.080 you started out by saying that studying the brains of 0:06:17.160 --> 0:06:21.480 zebra finches are somehow interesting for tech companies. So if 0:06:21.520 --> 0:06:24.200 Tim studies the birds because there are connections between bird 0:06:24.200 --> 0:06:27.440 brains and human brains, are there also connections between bird 0:06:27.440 --> 0:06:31.600 brains and computers. Absolutely, But to understand it, I need 0:06:31.640 --> 0:06:34.800 to explain a little bit more about Tim's research. When 0:06:34.800 --> 0:06:37.080 we met him in Boston, before we went into the 0:06:37.160 --> 0:06:40.160 room with the real life zebra finches, he played us 0:06:40.160 --> 0:06:43.680 one of his best songbird clips, and so that's what 0:06:43.760 --> 0:06:47.359 the song sounds like. Okay, so that's one. Yeah, so 0:06:47.440 --> 0:06:50.760 that's what the that's what this this particular zebra finch sings. 0:06:50.800 --> 0:06:53.440 All of them have slightly different songs. That sounded like 0:06:53.480 --> 0:06:56.800 a cartoon. Yeah, it sounded a little like woody woodpecker. Right, 0:06:57.080 --> 0:06:59.440 but that but that is actually, uh, that is actually 0:06:59.480 --> 0:07:04.920 what it sounds. Yeah. Yeah, we had listened to these 0:07:04.920 --> 0:07:08.800 all day long. Tim told us while male and female 0:07:08.880 --> 0:07:12.960 zebra finches can chirp, only male zebra finches sing, even 0:07:13.000 --> 0:07:15.320 if their song is so short it doesn't sound like 0:07:15.400 --> 0:07:18.040 much of a song to us. Tim studies the brains 0:07:18.080 --> 0:07:20.760 of the baby birds as they learn. Here's the baby 0:07:20.840 --> 0:07:24.400 zebra finch trying to imitate his dad's song. He makes 0:07:24.440 --> 0:07:27.960 an early effort, and then a better one a month later. Okay, 0:07:28.000 --> 0:07:39.280 so this is the father this one, well, okay, and 0:07:39.320 --> 0:07:41.080 then we can do the middle one where he's not 0:07:41.160 --> 0:07:54.080 quite good. A right, Okay, this is the one. Yeah 0:07:55.000 --> 0:07:58.480 that's pretty good. So Ashley, I'm kind of praying that 0:07:58.520 --> 0:08:03.640 to these experiments, he's not hurting these beautiful, teeny tiny birds. Basically, 0:08:03.800 --> 0:08:06.040 what he does is they take these birds that have 0:08:06.080 --> 0:08:09.520 been injected with a benign virus. The virus makes their 0:08:09.520 --> 0:08:13.120 brains produce a type of protein that causes individual neurons 0:08:13.160 --> 0:08:16.040 to light up when they fire, they glow green and red. 0:08:16.760 --> 0:08:19.000 To see them in action, Tim and the grad students 0:08:19.000 --> 0:08:22.120 in his lab performed very delicate surgery on the zebra 0:08:22.200 --> 0:08:26.320 finches to implant tiny, tiny microscopes in their brain. Tim 0:08:26.360 --> 0:08:28.640 showed us one bird who had gone through this procedure, 0:08:29.600 --> 0:08:32.800 male zebra finish. And so haven't we on his hiding 0:08:34.760 --> 0:08:37.800 He's got one of these windows that I was telling 0:08:37.840 --> 0:08:43.480 you about. And so this guy has already been through 0:08:43.480 --> 0:08:46.360 the procedure in which we inject the viruses into the 0:08:46.400 --> 0:08:50.319 brain and then we've put a window or a lens 0:08:50.400 --> 0:08:52.880 on top so that we can actually image through it. 0:08:53.080 --> 0:08:55.720 And it's going to get a microscope. He will eventually 0:08:55.760 --> 0:09:00.840 get a microscope. Auch Sarah, that sounds completely crazy. Give 0:09:00.880 --> 0:09:03.600 us a better sense for what this looks like. So 0:09:03.640 --> 0:09:07.199 there are all these little birds hopping around with a 0:09:07.240 --> 0:09:11.240 tiny bit of their skull missing, and instead they have 0:09:11.440 --> 0:09:14.240 a microscope there and a little hole where you can 0:09:14.240 --> 0:09:16.400 look in and see what's going on in their brain. 0:09:17.040 --> 0:09:20.480 And the microscope sits there for days, weeks, months, at 0:09:20.480 --> 0:09:24.400 a time, looking at their neurons fire in in real time, 0:09:24.440 --> 0:09:26.320 and the birds are in their little cages, and then 0:09:26.400 --> 0:09:28.520 when you go into their lab, it's actually it's pretty cool. 0:09:28.640 --> 0:09:32.200 There's all these wires going off these cages straight into 0:09:32.240 --> 0:09:35.199 like a data center where they store all of this information. 0:09:35.240 --> 0:09:37.440 And so basically you just get to just get to 0:09:37.480 --> 0:09:41.360 watch this bird's brain behave in real time and then 0:09:41.400 --> 0:09:43.600 go back and look through all the information. So, Sarah, 0:09:43.640 --> 0:09:46.920 should we feel sorry for these birds? I didn't. They 0:09:46.960 --> 0:09:50.359 seemed really happy. They were hopping around, they were chirping, 0:09:50.600 --> 0:09:53.640 they were acting normal as far as they could tell 0:09:53.720 --> 0:09:56.959 based on the ones they saw without microscopes in their head. 0:09:57.000 --> 0:09:59.280 They really didn't seem to be too much difference. Okay, 0:09:59.280 --> 0:10:01.920 so how is all this useful for the neuroscientists. They 0:10:01.920 --> 0:10:04.440 look deep inside each bird's brain and they look to 0:10:04.480 --> 0:10:07.240 see which neurons are firing and for how long, when 0:10:07.320 --> 0:10:09.760 say the bird is learning to sing, and they can 0:10:09.800 --> 0:10:14.080 make hypothesis on the relationship between different neurons. So that's 0:10:14.120 --> 0:10:17.199 actually a technique used pretty widely in science now and 0:10:17.320 --> 0:10:20.360 all kinds of animals. While we were visiting Tim in Boston, 0:10:20.400 --> 0:10:23.240 we also stopped by a mouse lab that does something similar, 0:10:23.320 --> 0:10:25.360 and a fish lab. In each of the labs, the 0:10:25.400 --> 0:10:28.480 neuroscientists there are studying other things as well, like how 0:10:28.600 --> 0:10:31.960 animals move, what happens in their brains when they make decisions. 0:10:32.679 --> 0:10:35.480 For example, at the Roland Institute in Cambridge, we saw 0:10:35.520 --> 0:10:38.400 a couple of video games that mice play using tiny 0:10:38.480 --> 0:10:41.480 joysticks size for a mouse paw Wow. So how do 0:10:41.520 --> 0:10:43.559 they how do they mice get the quarters in the 0:10:43.679 --> 0:10:48.400 video game machine. They're highly trained and and the video 0:10:48.440 --> 0:10:50.880 games tells us what it tells us how they're making 0:10:50.880 --> 0:10:54.600 decisions exactly. So the mice are a lot like the birds. 0:10:54.640 --> 0:10:57.559 They have bits of their their skull have been removed, 0:10:57.600 --> 0:11:01.000 and we're watching their brains and real time again and 0:11:01.160 --> 0:11:03.080 and so you watch them play these video games, and 0:11:03.120 --> 0:11:05.920 you see how they adapt. Sometimes they get different rules, 0:11:06.200 --> 0:11:08.760 and you see how the mouse learns the rules of 0:11:08.760 --> 0:11:11.120 the game, and they move this choice stick around and 0:11:11.160 --> 0:11:13.600 one of them to find the edges of a box. 0:11:14.080 --> 0:11:16.720 And if the mouse is successful, it gets a little 0:11:16.760 --> 0:11:19.520 bit of sugar water. And and all this time the 0:11:19.559 --> 0:11:21.760 scientists are sitting in there seeing which parts of the 0:11:21.800 --> 0:11:24.480 brain light up and how the mouse reacts to these 0:11:24.520 --> 0:11:27.280 different situations. So what is it about this research that 0:11:27.400 --> 0:11:31.040 tech companies find so interesting? One reason is that these 0:11:31.040 --> 0:11:33.839 scientists are working with tons and tons of data, and 0:11:33.880 --> 0:11:37.640 that's something that every tech company needs to do. And 0:11:37.800 --> 0:11:42.240 also it's to do with artificial intelligence, the field that 0:11:42.360 --> 0:11:46.959 has computer systems mastering tasks that require human traits like 0:11:47.200 --> 0:11:51.800 visual perception or decision making. Maybe how to identify a cat. 0:11:52.480 --> 0:11:55.800 That sounds simple, but it's actually way more nuanced than 0:11:55.840 --> 0:11:59.520 a traditional computer task and very hard for a computer 0:11:59.640 --> 0:12:02.600 to masse term. So there's this one school of thought 0:12:02.760 --> 0:12:06.679 that AI moving forward should loosely be modeled after the 0:12:06.760 --> 0:12:11.160 human brain. The AI systems we have today are still 0:12:11.320 --> 0:12:15.480 basically number crunching systems. They're doing tons of statistical calculations. 0:12:15.720 --> 0:12:18.000 And if we want to get to this this future 0:12:18.040 --> 0:12:21.120 that Sarah's talking about where you actually have decision making 0:12:21.640 --> 0:12:25.280 and much more sophisticated thought, the ideas that we could 0:12:25.280 --> 0:12:29.360 borrow from the human brain and and maybe I have 0:12:29.520 --> 0:12:32.000 something that's way more flexible than what a computer could do. 0:12:32.160 --> 0:12:35.360 And so presumably, since we can't cut open human skulls 0:12:35.360 --> 0:12:37.920 and make people play video games against their will, the 0:12:38.280 --> 0:12:42.080 zebra finch brains are first Yeah, exactly, human brains are 0:12:42.080 --> 0:12:44.720 a little too big and complicated to study in the 0:12:44.800 --> 0:12:47.560 kind of detail we can get from animals right now, 0:12:48.000 --> 0:12:52.280 they're smaller, easier to study, and obviously the ethics of 0:12:52.400 --> 0:12:56.199 studying a living human brain very tricky. So I'm so 0:12:56.240 --> 0:12:58.680 curious about why zebra finches have become the bird to 0:12:58.720 --> 0:13:01.920 study and what exactly tech companies are doing with all this. 0:13:02.200 --> 0:13:17.680 Let's get to that after the break, Okay, Sarah n Ashley, 0:13:17.800 --> 0:13:21.560 So before the break, you explain how neuroscience is starting 0:13:21.559 --> 0:13:25.400 to inform the way tech companies design AI systems. So 0:13:25.440 --> 0:13:27.480 I guess this means it's a very good time to 0:13:27.520 --> 0:13:30.120 be a neuroscientist. Yeah. So these people that used to 0:13:30.160 --> 0:13:33.160 be in academia or working at universities for their whole 0:13:33.200 --> 0:13:37.360 career are now finding tons of job opportunities in Silicon Valley. 0:13:37.520 --> 0:13:41.480 Companies like Google and Apple and Amazon are all snatching 0:13:41.520 --> 0:13:44.520 them up, including a lot of these zebra finch experts 0:13:44.520 --> 0:13:46.960 that we've been talking about. It's how the company that 0:13:47.080 --> 0:13:50.040 probably makes the chips in your laptop won't say exactly 0:13:50.040 --> 0:13:52.520 how many people it has working in AI, but it's 0:13:52.559 --> 0:13:55.720 a lot, and many of them have this expertise. Intel 0:13:55.760 --> 0:13:58.199 helps its customers soup up machines to get them to 0:13:58.280 --> 0:14:01.800 behave smarter and in more uman lake ways. That could 0:14:01.880 --> 0:14:05.120 mean maybe working with a self driving carmaker, and that 0:14:05.240 --> 0:14:08.800 carmaker needs its vehicles to make lightning fast decisions on 0:14:08.840 --> 0:14:11.800 the road, and they have to be good decisions. A 0:14:11.840 --> 0:14:14.400 promising way to do that is to build computer systems 0:14:14.440 --> 0:14:17.640 that imitate how the human brain works. In a human brain, 0:14:17.880 --> 0:14:21.480 the systems are called signapses and pathways. In a machine, 0:14:21.760 --> 0:14:25.720 they're called neural networks. We talked to a mirror because Shahi, 0:14:26.000 --> 0:14:30.960 the chief technology officer for Intel's AI products divisions of 0:14:31.040 --> 0:14:35.040 architectures having wildly successful and uh sends around two thousand 0:14:35.120 --> 0:14:45.440 eleven UM, and while wider and excreted vision, m navigation, 0:14:45.680 --> 0:14:50.520 reforcement learning things that are car related to neuroscience becauses 0:14:50.520 --> 0:14:54.880 are also tacit humans doing pretty well. A mirror is 0:14:54.880 --> 0:14:58.480 a computational neuroscientists by training. He got his PhD at 0:14:58.560 --> 0:15:02.280 UC Berkeley, and he ended up hiring another Berkeley PhD grad, 0:15:02.520 --> 0:15:06.320 Tyler Lee, to work it into and helping virtual assistants 0:15:06.440 --> 0:15:09.840 understand human speech. Tyler spends a lot of time thinking 0:15:09.880 --> 0:15:14.400 about how speech works in different environments like cars. Knowing 0:15:14.480 --> 0:15:17.960 context simple things like if the speakers the driver, or 0:15:18.000 --> 0:15:20.760 the passenger makes it a lot easier to understand what 0:15:20.800 --> 0:15:23.800 they're saying. For example, a driver is more likely to 0:15:23.840 --> 0:15:26.600 ask about directions. Some of the work on context he 0:15:26.640 --> 0:15:29.000 does actually reminds him a lot of his PhD work 0:15:29.200 --> 0:15:34.360 studying you guessed it zebra fitches. The bird has to 0:15:34.400 --> 0:15:38.360 recognize what type of call is being being omitted by 0:15:38.480 --> 0:15:41.600 by the one it's hearing, uh, and then then it 0:15:41.640 --> 0:15:43.960 can go and recognize who that bird is. Is that 0:15:44.400 --> 0:15:46.600 is that a family member of mate? You know? Is 0:15:46.600 --> 0:15:48.600 it another zuber fringe or a different type of bird? 0:15:48.960 --> 0:15:52.120 Should I be concerned? The vocal identification is one thing 0:15:52.120 --> 0:15:54.800 where it's it's context specific, and recognizing the context would 0:15:54.920 --> 0:15:59.160 let you better understand the vocal signature. Mostly, Tyler says 0:15:59.160 --> 0:16:03.000 his studies help him big picture stuff. Neuroscience teaches you 0:16:03.120 --> 0:16:06.960 how to think about complex problems of signal processing. Where 0:16:07.000 --> 0:16:10.360 I take some something from the world that comes in. 0:16:10.440 --> 0:16:15.040 It's an image, it's uh, it's sound, and it's it's noisy, 0:16:15.320 --> 0:16:18.320 and it's high very very high dimensional, and I have 0:16:18.400 --> 0:16:21.400 to like break it down into into features that I 0:16:21.440 --> 0:16:25.560 can then use to to like do something with, solve 0:16:25.600 --> 0:16:27.680 a task with. That's what the brain does all the time, 0:16:27.720 --> 0:16:30.880 and that's sort of the abstract level. Tyler's boss a 0:16:30.960 --> 0:16:34.160 Mirror says he's no zebra finch chauvinist. He's got people 0:16:34.200 --> 0:16:38.640 on his team who studied flies, rats, locus, even worms. 0:16:38.680 --> 0:16:42.600 These very simple organisms exhibit really complex behaviors that are 0:16:42.600 --> 0:16:46.680 still a challenge for us to to simulate and silicon 0:16:46.800 --> 0:16:49.520 using our neual networks and machine and learning. So even 0:16:49.640 --> 0:16:55.040 a simple worm inchworm is a really complicated robotic machine 0:16:55.080 --> 0:17:00.000 that's really miraculous. So we look for inspiration from simple 0:17:00.280 --> 0:17:03.880 to humans. So I guess Intel and all these other 0:17:03.920 --> 0:17:07.240 companies must be developing products which incorporate AI developed with 0:17:07.280 --> 0:17:10.520 the help of these zebra finch experts. Besides commands to 0:17:10.560 --> 0:17:13.520 self driving cars, is there any area where knowing a 0:17:13.520 --> 0:17:16.520 lot about sound itself is helpful? And I guess when 0:17:16.560 --> 0:17:19.280 when do all these years of studying zebra finches finally 0:17:19.359 --> 0:17:21.880 pay off? You can think about features on your phone 0:17:21.960 --> 0:17:24.639 or computer that let you unlock the device with your voice, 0:17:25.320 --> 0:17:28.000 or stuff like noise reduction and phone calls and on 0:17:28.119 --> 0:17:30.119 video calls. I can't wait till I can sing a 0:17:30.119 --> 0:17:33.680 little Zebra Finch song and my phone unlocks. But anything 0:17:33.680 --> 0:17:37.520 beyond gadgets, yeah, absolutely. Animal neuroscience connects to a lot 0:17:37.600 --> 0:17:41.600 of fields, especially health. Somewhere could be relevant for Parkinson's 0:17:41.640 --> 0:17:45.120 research because animals help researchers figure out how to stop tremors. 0:17:45.600 --> 0:17:48.560 They also help in areas like how to handle prosthetic limbs. 0:17:48.920 --> 0:17:52.040 Outside of medicine, the work might grow even more futuristic. 0:17:52.280 --> 0:17:55.040 This is where we get into dystopian scenarios. I suspect 0:17:55.080 --> 0:17:58.960 what what else exactly like Neuralink. I mentioned that company 0:17:59.000 --> 0:18:01.840 earlier because Ellen Musk get hired a Zebra Finch scholar, 0:18:01.920 --> 0:18:04.080 and it's one of the companies that we believe is 0:18:04.119 --> 0:18:08.320 working on very futuristic technology. Ellen Musk keeps dropping hints 0:18:08.400 --> 0:18:10.560 on Twitter that the company is about to announce a 0:18:10.600 --> 0:18:14.320 big breakthrough. Nobody knows exactly what, but it's going to 0:18:14.359 --> 0:18:17.200 have something to do with brain machine interfaces. So I'm 0:18:17.240 --> 0:18:19.760 trying hard not to think about some Star Trek episodes 0:18:19.760 --> 0:18:22.680 on this topic, which all ended quite badly. But help 0:18:22.720 --> 0:18:26.320 me understand actually what a brain machine interfaces about. I mean, 0:18:26.480 --> 0:18:28.720 at its most basic level, is this idea that you 0:18:28.800 --> 0:18:33.040 have a two way interplay between humans and computers where 0:18:33.040 --> 0:18:36.639 you could actually funnel information back and forth. We already 0:18:36.640 --> 0:18:38.960 have examples of stuff like this with the implants that 0:18:39.000 --> 0:18:43.160 help people here or stop Parkinson's tremors in this case, 0:18:43.359 --> 0:18:46.760 I think people are looking at much more futuristic applications 0:18:46.800 --> 0:18:50.120 where you might even have like a mesh that's attached 0:18:50.160 --> 0:18:53.440 to your brain and you could full on download your 0:18:53.440 --> 0:18:57.040 brain to a machine or learn Japanese in five seconds. 0:18:57.720 --> 0:19:00.639 There's another company called Colonel that's in this same field, 0:19:01.000 --> 0:19:05.080 and like Neuralalink, it's also very mysterious, and in some 0:19:05.119 --> 0:19:07.640 ways that's kind of the best part. When we don't 0:19:07.760 --> 0:19:10.920 know exactly what they're doing, we can imagine all kinds 0:19:10.920 --> 0:19:15.080 of crazy stuff. Going back to Elon Musk, he's been 0:19:15.080 --> 0:19:19.040 talking about where neurally could go, maybe allowing people to 0:19:19.119 --> 0:19:22.920 have this kind of superhuman cognition where you could you 0:19:22.960 --> 0:19:25.280 could think on par with a machine, are certainly much 0:19:25.320 --> 0:19:28.919 better than we do today. That means basically stuff like 0:19:28.960 --> 0:19:32.439 you could download an entire FOURGN language directly into your brain, 0:19:32.840 --> 0:19:37.399 or maybe instantly grabbing encyclopedia. People like Timachi have been 0:19:37.400 --> 0:19:40.840 thinking about exactly these scenarios for years and can really 0:19:40.880 --> 0:19:44.960 nerd out on the possibilities. Trivia would be over, Jeopardy 0:19:45.000 --> 0:19:48.160 would not be a thing anymore. Um, Alex Trebek would 0:19:48.160 --> 0:19:50.280 be out of a job. He has some more serious 0:19:50.280 --> 0:19:53.439 thoughts on the topic too. I find the idea that 0:19:53.480 --> 0:19:57.359 we could, you know, pretend one day in the maybe 0:19:57.400 --> 0:20:01.000 just in future, you know, really right information directly into 0:20:01.000 --> 0:20:05.320 the brain, that we could actually have a high bandwidth way, 0:20:05.800 --> 0:20:08.359 uh to get really sci fi about it, kind of 0:20:08.400 --> 0:20:11.880 a matrix E like, Um, I think that would be amazing. 0:20:11.960 --> 0:20:15.320 We are nowhere near knowing anything about how to get there. 0:20:15.840 --> 0:20:19.119 We can barely even scratch the surface of what that 0:20:19.119 --> 0:20:22.480 would be like. But you know, in terms of fantasy, 0:20:24.119 --> 0:20:26.200 what would I like to do one day? I would 0:20:26.240 --> 0:20:29.080 love to be able to contribute even a small way 0:20:29.200 --> 0:20:32.800 to figuring out how we can have this sort of 0:20:33.400 --> 0:20:37.240 bi directional interface with the brain. Oh my god. So yeah, 0:20:37.280 --> 0:20:39.760 this is While the evocation of the matrix does not 0:20:39.880 --> 0:20:42.199 make me feel more comfortable about this, what do the 0:20:42.200 --> 0:20:45.240 skeptics say about about it all? Well, there are plenty 0:20:45.280 --> 0:20:48.880 of skeptics out there. I checked in with one scholar 0:20:48.960 --> 0:20:53.040 at the University of Chicago, Dan Margolias. The idea that 0:20:53.080 --> 0:20:59.440 we're going to reverse engineers, not reverse engineer, uh, forward 0:20:59.520 --> 0:21:03.800 engineer or the human brain, so we can download tons 0:21:03.840 --> 0:21:07.479 of material into it very rapidly. And I don't know 0:21:07.520 --> 0:21:12.400 what pick up a language overnight or something. I it's 0:21:12.440 --> 0:21:15.879 it's the way people make progresses to dream and so 0:21:16.000 --> 0:21:21.919 I'm I'm a scientist. I'm for that. But that really 0:21:21.960 --> 0:21:30.159 sounds more fantastical than realistic. It ignores it ignores the 0:21:30.240 --> 0:21:35.520 remarkable ways we learn, and it ignores our evolutionary history. Um, 0:21:35.840 --> 0:21:39.399 so I would, I would. Uh, it'll be interesting to 0:21:39.440 --> 0:21:46.000 see what progress they make for sure. So guys, the 0:21:46.040 --> 0:21:50.840 possibility of super human cognition sounds appealing. Um, you know, 0:21:50.880 --> 0:21:53.679 if we're up to you two and and somebody was 0:21:53.720 --> 0:21:56.240 offering to put a chip or an apparatus and in 0:21:56.280 --> 0:21:58.359 the year brains like they're doing to the to the 0:21:58.400 --> 0:22:00.720 poor little zebra finch, would you do it? I mean, 0:22:00.880 --> 0:22:03.800 you know, in some sence, we're already doing this stuff today. 0:22:03.920 --> 0:22:07.159 If you have an implant to help you here, or 0:22:07.480 --> 0:22:10.520 things to stop Parkinson's tremors, yeah, if I had one 0:22:10.520 --> 0:22:13.560 of those conditions, I would absolutely get one of these 0:22:13.560 --> 0:22:16.919 implants and supercharge myself. When you start going into this 0:22:17.080 --> 0:22:19.640 this next wave of stuff. It gets I think, far 0:22:19.680 --> 0:22:25.040 more philosophical and complicated, because you're talking about changing humans 0:22:25.040 --> 0:22:27.639 from what they are, some sort of weird next step 0:22:27.720 --> 0:22:32.320 of evolution where we're kind of half man half machine. Um. 0:22:32.359 --> 0:22:34.719 You know, in some ways there's people I talked to, 0:22:35.359 --> 0:22:38.399 like the guys at Colonel, who argue that this is 0:22:38.440 --> 0:22:40.280 the only way humans will be able to keep up 0:22:40.359 --> 0:22:44.280 with machines. And we always hear about losing jobs artificial 0:22:44.280 --> 0:22:48.240 intelligence and and seeing what humans can do going away. 0:22:48.320 --> 0:22:50.240 And so you know, if you're half and half, you 0:22:50.280 --> 0:22:52.280 can you can keep up, but maybe keep some of 0:22:52.320 --> 0:22:56.320 your humanness as well. Sarah, what about cyborg? Sarah McBride, 0:22:56.400 --> 0:22:58.679 will we ever see? Will we ever see that? I 0:22:58.720 --> 0:23:03.320 already hate myself, so I doubt it. But um, what 0:23:03.359 --> 0:23:07.080 about this idea of keeping up with robots? And and yeah, 0:23:07.359 --> 0:23:09.879 there are people who think that AI could help us 0:23:09.960 --> 0:23:14.640 solve problems like climate change. I think that's too optimistic. 0:23:15.119 --> 0:23:17.199 The funny part about doing the story for me was 0:23:17.280 --> 0:23:21.080 that that in the AI camp you tend to have 0:23:21.280 --> 0:23:23.640 I feel like people who think the technology is really 0:23:23.640 --> 0:23:26.199 far along. You know, if you're talking about specifically computer 0:23:26.280 --> 0:23:30.240 scientists and the Silicon Valley kind of crew. Um, they're 0:23:30.359 --> 0:23:32.720 very impressed what they've come up with over the last 0:23:32.760 --> 0:23:34.360