WEBVTT 00:00:00.080 --> 00:00:02.120 Hey there! I’m Jabril. 00:00:02.120 --> 00:00:04.759 John Green Bot: And I am John Green Bot 00:00:04.760 --> 00:00:08.160 and welcome to Crash Course Artificial Intelligence. 00:00:08.160 --> 00:00:10.780 Now, I want to make sure we’re starting on the same page. 00:00:10.780 --> 00:00:12.740 Artificial intelligence is everywhere. 00:00:12.750 --> 00:00:17.180 It’s helping banks make loan decisions, and helping doctors diagnose patients, it’s 00:00:17.180 --> 00:00:21.450 on our cell phones, autocompleting texts, it’s the algorithm recommending YouTube 00:00:21.450 --> 00:00:23.199 videos to watch after this one! 00:00:23.199 --> 00:00:26.710 AI already has a pretty huge impact on all of our lives. 00:00:26.710 --> 00:00:30.610 So people, understandably, have some polarized feelings about it. 00:00:30.610 --> 00:00:35.410 Some of us imagine that AI will change the world in positive ways, it could end car accidents 00:00:35.410 --> 00:00:40.120 because we have self-driving cars, or it could give the elderly great, personalized care. 00:00:40.120 --> 00:00:44.080 Others worry that AI will lead to constant surveillance by a Big Brother government. 00:00:44.080 --> 00:00:46.310 Some say that automation will take all our jobs. 00:00:46.310 --> 00:00:50.200 Or the robots might try and kill us all. 00:00:50.200 --> 00:00:52.380 No, we’re not worried about you John Green Bot. 00:00:52.380 --> 00:00:55.860 But when we interact with AI that’s currently available like Siri... 00:00:55.860 --> 00:00:58.020 Hey Siri. 00:00:58.020 --> 00:01:00.300 Is AI going to kill us all?” 00:01:00.300 --> 00:01:03.500 Siri: “I don’t understand ‘Is AI going to kill us all.’” 00:01:03.500 --> 00:01:06.680 … it’s clear that those are still distant futures. 00:01:06.680 --> 00:01:11.940 Now to understand where artificial intelligence might be headed, and our role in the AI revolution, 00:01:11.940 --> 00:01:15.340 we have to understand how we got to where we are today. 00:01:15.340 --> 00:01:24.340 [INTRO] 00:01:24.340 --> 00:01:29.120 If you know about artificial intelligence mostly from movies or books, AI probably seems 00:01:29.120 --> 00:01:32.640 like this vague label for any machine that can think like a human. 00:01:32.640 --> 00:01:37.500 Fiction writers like to imagine a more generalized AI, one that can answer any question we might 00:01:37.500 --> 00:01:39.760 have, and do anything a human can do. 00:01:39.760 --> 00:01:45.800 But that’s a pretty rigid way to think about AI and it’s not super realistic. 00:01:45.800 --> 00:01:49.880 Sorry John Green-bot, you can’t do all that yet. 00:01:49.880 --> 00:01:54.880 A machine is said to have artificial intelligence if it can interpret data, potentially learn 00:01:54.880 --> 00:01:59.540 from the data, and use that knowledge to adapt and achieve specific goals. 00:01:59.540 --> 00:02:04.480 Now, the idea of “learning from the data” is kind of a new approach. 00:02:04.480 --> 00:02:05.950 But we’ll get into that more in episode 4. 00:02:05.960 --> 00:02:10.480 So let’s say we load up a new program in John Green-bot. 00:02:18.280 --> 00:02:22.950 This program looks at a bunch of photos, some of me and some of not of me, and then learns 00:02:22.950 --> 00:02:24.120 from those data. 00:02:24.120 --> 00:02:28.170 Then, we can show him a new photo, like this selfie of me here in the studio filming this 00:02:28.170 --> 00:02:32.360 Crash Course video, and we’ll see if he can recognize that the photo is me. 00:02:32.360 --> 00:02:38.160 John Green Bot: You are Jabril. 00:02:38.160 --> 00:02:42.340 If he can correctly classify that new photo, we could say that John Green-bot has some 00:02:42.340 --> 00:02:43.970 artificial intelligence! 00:02:43.970 --> 00:02:49.270 Of course, that’s a very specific input of photos, and a very specific task of classifying 00:02:49.270 --> 00:02:53.120 a photo that’s either me or not me. 00:02:53.120 --> 00:02:57.390 With just that program John Green-bot can’t recognize or name anyone who /isn’t/ me… 00:02:57.390 --> 00:03:01.450 John Green Bot: You are not Jabril. 00:03:01.450 --> 00:03:12.980 He can’t navigate to places. 00:03:12.980 --> 00:03:16.880 Or hold a meaningful conversation. 00:03:16.880 --> 00:03:18.400 No. 00:03:18.400 --> 00:03:20.440 I just don’t get it. 00:03:20.440 --> 00:03:26.440 Why would anyone choose a bagel when you have a perfectly good donut right here? 00:03:26.440 --> 00:03:32.440 John Green Bot: You are Jabril 00:03:32.440 --> 00:03:39.120 Thanks John Green Bot. 00:03:39.160 --> 00:03:44.440 He can’t do most things that humans do, which is pretty standard for AI these days. 00:03:44.440 --> 00:03:49.000 But even with this much more limited definition of artificial intelligence, AI still plays 00:03:49.000 --> 00:03:51.520 a huge role in our everyday lives. 00:03:51.520 --> 00:03:57.569 There are some more obvious uses of AI, like Alexa or Roomba, which is kind of like the 00:03:57.569 --> 00:03:59.890 AI from science fiction I guess. 00:03:59.890 --> 00:04:02.660 But there are a ton of less obvious examples! 00:04:02.660 --> 00:04:07.790 When we buy something in a big store or online, we have one type of AI deciding which and 00:04:07.790 --> 00:04:09.950 how many items to stock. 00:04:09.950 --> 00:04:14.680 And as we scroll through Instagram, a different type of AI picks ads to show us. 00:04:14.680 --> 00:04:20.809 AI helps determine how expensive our car insurance is, or whether we get approved for a loan. 00:04:20.809 --> 00:04:23.479 And AI even affects big life decisions. 00:04:23.479 --> 00:04:28.099 Like when you submit your college (or job) application AI might be screening it before 00:04:28.099 --> 00:04:29.300 a human even sees it. 00:04:29.300 --> 00:04:32.990 The way AI and automation is changing everything, from commerce to jobs, is sort of like the 00:04:32.990 --> 00:04:35.830 Industrial Revolution in the 18th century. 00:04:35.830 --> 00:04:41.550 This change is global, some people are excited about it, and others are afraid of it. 00:04:41.550 --> 00:04:46.159 But either way, we all have the responsibility to understand AI and figure out what role 00:04:46.159 --> 00:04:48.229 AI will play in our lives. 00:04:48.240 --> 00:04:51.439 The AI revolution itself isn’t even that old. 00:04:51.440 --> 00:04:55.360 The term artificial intelligence didn’t even exist a century ago. 00:04:55.360 --> 00:04:59.200 It was coined in 1956 by a computer scientist named John McCarthy. 00:04:59.210 --> 00:05:03.759 He used it to name the “Dartmouth Summer Research Project on Artificial Intelligence.” 00:05:03.759 --> 00:05:06.150 Most people call it the “Dartmouth Conference” for short. 00:05:06.150 --> 00:05:10.289 Now, this was way more than a weekend where you listen to a few talks, and maybe go to 00:05:10.289 --> 00:05:11.740 a networking dinner. 00:05:11.740 --> 00:05:15.189 Back in the day, academics just got together to think for a while. 00:05:15.189 --> 00:05:19.729 The Dartmouth Conference lasted eight weeks and got a bunch of computer scientists, cognitive 00:05:19.729 --> 00:05:22.779 psychologists, and mathematicians to join forces. 00:05:22.779 --> 00:05:26.689 Many of the concepts that we’ll talk about in Crash Course AI, like artificial neural 00:05:26.689 --> 00:05:31.009 networks, were dreamed up and developed during this conference and in the few years that 00:05:31.009 --> 00:05:32.009 followed. 00:05:32.009 --> 00:05:36.809 But because these excited academics were really optimistic about artificial intelligence, 00:05:36.809 --> 00:05:38.389 they may have oversold it a bit. 00:05:38.389 --> 00:05:43.419 For example, Marvin Minsky was a talented cognitive scientist who was part of the Dartmouth 00:05:43.419 --> 00:05:44.499 Conference. 00:05:44.499 --> 00:05:49.689 But he also had some ridiculously wrong predictions about technology, and specifically AI. 00:05:49.689 --> 00:05:54.449 In 1970, he claimed that in "three to eight years we will have a machine with the general 00:05:54.449 --> 00:05:57.000 intelligence of an average human being." 00:05:57.000 --> 00:05:58.729 And, uh, sorry Marvin. 00:05:58.729 --> 00:06:01.289 We’re not even close to that now. 00:06:01.289 --> 00:06:05.059 Scientists at the Dartmouth Conference seriously underestimated how much data and computing 00:06:05.059 --> 00:06:08.680 power an AI would need to solve complex, real world problems. 00:06:08.680 --> 00:06:13.729 See, an artificial intelligence doesn’t really “know” anything when it’s first 00:06:13.729 --> 00:06:15.479 created, kind of like a human baby. 00:06:15.479 --> 00:06:20.319 Babies use their senses to perceive the world and their bodies to interact with it, and 00:06:20.319 --> 00:06:23.099 they learn from the consequences of their actions. 00:06:23.099 --> 00:06:27.350 My baby niece might put a strawberry in her mouth and decide that it’s tasty. 00:06:27.350 --> 00:06:30.889 And then she might put play-doh in her mouth and decide that it’s gross. 00:06:30.889 --> 00:06:36.379 Babies experience millions of these data-gathering events as they learn to speak, walk, think, 00:06:36.379 --> 00:06:37.470 and not eat play-doh. 00:06:37.470 --> 00:06:42.080 Now, most kinds of artificial intelligence don’t have things like senses, a body, or 00:06:42.080 --> 00:06:46.729 a brain that can automatically judge a lot of different things like a human baby does. 00:06:46.729 --> 00:06:49.699 Modern AI systems are just programs in machines. 00:06:49.699 --> 00:06:52.599 So we need to give AI a lot of data. 00:06:52.599 --> 00:06:57.550 Plus, we have to label the data with whatever information the AI is trying to learn, like 00:06:57.550 --> 00:06:59.869 whether food tastes good to humans. 00:06:59.869 --> 00:07:04.309 And then, the AI needs a powerful enough computer to make sense of all the data. 00:07:04.320 --> 00:07:07.120 All of this just wasn’t available in 1956. 00:07:07.120 --> 00:07:12.569 Back then, an AI could maybe tell the difference between a triangle and a circle, but it definitely 00:07:12.569 --> 00:07:16.039 couldn’t recognize my face in a photo like John Green-bot did earlier! 00:07:16.039 --> 00:07:22.099 So until about 2010 or so, the field was basically frozen in what’s called the AI Winter. 00:07:22.099 --> 00:07:27.690 Still there were a lot of changes in the last half a century that led us to the AI Revolution. 00:07:27.690 --> 00:07:32.399 As a friend once said: “History reminds us that revolutions are not so much events 00:07:32.399 --> 00:07:33.960 as they are processes.” 00:07:33.960 --> 00:07:38.880 The AI Revolution didn’t begin with a single event, idea, or invention. 00:07:38.880 --> 00:07:43.379 We got to where we are today because of lots of small decisions, and two big developments 00:07:43.379 --> 00:07:44.400 in computing. 00:07:44.400 --> 00:07:48.830 The first development was a huge increase in computing power and how fast computers 00:07:48.840 --> 00:07:50.200 could process data. 00:07:50.200 --> 00:07:53.760 To see just how huge, let’s go to the Thought Bubble. 00:07:53.760 --> 00:07:59.589 During the Dartmouth Conference in 1956, the most advanced computer was the IBM 7090. 00:07:59.589 --> 00:08:04.789 It filled a whole room, stored data on basically giant cassette tapes, and took instructions 00:08:04.789 --> 00:08:06.860 using paper punch cards. 00:08:06.860 --> 00:08:12.360 Every second, the IBM 7090 could do about 200,000 operations. 00:08:12.360 --> 00:08:17.400 But if you tried to do that it would take you 55 and a half hours! 00:08:17.409 --> 00:08:21.479 Assuming you did one operation per second, and took no breaks. 00:08:21.479 --> 00:08:22.479 That’s right. 00:08:22.480 --> 00:08:25.200 Not. Even. For. Snacks. 00:08:25.240 --> 00:08:28.960 At the time, that was enough computing power to help with the U.S. Air Force's Ballistic 00:08:28.960 --> 00:08:30.120 Missile Warning System. 00:08:30.120 --> 00:08:34.960 But AI needs to do a lot more computations with a lot more data. 00:08:34.960 --> 00:08:39.720 The speed of a computer is linked to the number of transistors it has to do operations. 00:08:39.720 --> 00:08:44.600 Every two years or so since 1956, engineers have doubled the number of transistors that 00:08:44.600 --> 00:08:46.760 can fit in the same amount of space. 00:08:46.760 --> 00:08:49.190 So computers have gotten much faster. 00:08:49.200 --> 00:08:54.720 When the first iPhone was released in 2007, it could do about 400 million operations per second. 00:08:54.760 --> 00:08:58.560 But ten years later, Apple says the iPhone X’s processor can 00:08:58.560 --> 00:09:01.320 do about 600 billion operations per second. 00:09:01.320 --> 00:09:05.480 That’s like having the computing power of over a thousand original iPhones in your pocket. 00:09:05.490 --> 00:09:10.370 (For all the nerds out there, listen you’re right, it’s not quite that simple - we’re 00:09:10.370 --> 00:09:11.990 just talking about FLOPS here) 00:09:11.990 --> 00:09:17.550 And a modern supercomputer, which does computational functions like the IBM 7090 did, can do over 00:09:17.550 --> 00:09:20.540 30 quadrillion operations per second. 00:09:20.540 --> 00:09:25.710 To put it another way, a program that would take a modern supercomputer one second to 00:09:25.710 --> 00:09:31.790 compute, would have taken the IBM 7090 4,753 years. 00:09:31.790 --> 00:09:33.290 Thanks Thought Bubble! 00:09:33.290 --> 00:09:37.200 So computers started to have enough computing power to mimic certain brain functions with 00:09:37.200 --> 00:09:43.760 artificial intelligence around 2005, and that’s when the AI winter started to show signs of thawing. 00:09:43.760 --> 00:09:48.440 But it doesn’t really matter if you have a powerful computer unless you also have 00:09:48.440 --> 00:09:50.360 a lot of data for it to munch on. 00:09:50.360 --> 00:09:54.390 The second development that kicked off the AI revolution is something that you’re using 00:09:54.390 --> 00:09:57.250 right now: the Internet and social media. 00:09:57.250 --> 00:10:01.410 In the past 20 years, our world has become much more interconnected. 00:10:01.410 --> 00:10:05.510 Whether you livestream from your phone, or just use a credit card, we’re all participating 00:10:05.510 --> 00:10:06.650 in the modern world. 00:10:06.650 --> 00:10:11.460 Every time we upload a photo, click a link, tweet a hashtag, tweet without a hashtag, 00:10:11.460 --> 00:10:17.280 like a YouTube video, tag a friend on Facebook, argue on Reddit, post on TikTok [R.I.P. 00:10:17.280 --> 00:10:22.660 Vine], support a Kickstarter campaign, buy snacks on Amazon, call an Uber from a party, 00:10:22.660 --> 00:10:25.720 and basically ANYTHING, that generates data. 00:10:25.720 --> 00:10:29.510 Even when we do something that /seems/ like it’s offline, like applying for a loan to 00:10:29.510 --> 00:10:35.320 buy a new car or using a passport at the airport those datasets end up in a bigger system. 00:10:35.320 --> 00:10:40.260 The AI revolution is happening now, because we have this wealth of data and the computing 00:10:40.260 --> 00:10:42.130 power to make sense of it. 00:10:42.130 --> 00:10:43.180 And I get it. 00:10:43.180 --> 00:10:47.370 The idea that we’re generating a bunch of data but don’t always know how, why, or 00:10:47.370 --> 00:10:51.700 if it’s being used by computer programs can be kind of overwhelming. 00:10:51.700 --> 00:10:55.760 But through Crash Course AI, we want to learn how artificial intelligence works because 00:10:55.760 --> 00:10:58.320 it’s impacting our lives in huge ways. 00:10:58.320 --> 00:11:00.650 And that impact will only continue to grow. 00:11:00.650 --> 00:11:05.000 With knowledge, we can make small decisions that will help guide the AI revolution, instead 00:11:05.000 --> 00:11:08.320 of feeling like we’re riding a rollercoaster we didn’t sign up for. 00:11:08.320 --> 00:11:13.550 We’re creating the future of artificial intelligence together, every single day. 00:11:13.550 --> 00:11:15.360 Which I think is pretty cool. 00:11:15.360 --> 00:11:20.440 Next time, we’ll start to dive into technical ideas like supervised, unsupervised, and reinforcement 00:11:20.440 --> 00:11:21.440 learning. 00:11:21.440 --> 00:11:24.590 And we’ll discuss what makes a Machine Learning algorithm good. 00:11:24.590 --> 00:11:25.590 See you then! 00:11:25.590 --> 00:11:28.360 Thanks to PBS for sponsoring Crash Course AI! 00:11:28.360 --> 00:11:31.610 If you want to help keep all Crash Course free for everybody, forever, you can join 00:11:31.610 --> 00:11:33.880 our community on Patreon. 00:11:33.880 --> 00:11:41.960 And if you want to learn more about how computers got so fast, check out our video on Moore’s Law.