WEBVTT - The Story of OpenAI 0:00:04.400 --> 0:00:07.800 Welcome to tech Stuff, a production from I Heart Radio. 0:00:11.880 --> 0:00:14.960 Hey there, and welcome to tech Stuff. I'm your host 0:00:15.200 --> 0:00:18.640 Jonathan Strickland. I'm an executive producer with I Heart Radio. 0:00:18.800 --> 0:00:21.960 And how the tech are Young. Well, since it's been 0:00:22.000 --> 0:00:24.320 in the news quite a bit so far this year, 0:00:24.600 --> 0:00:28.920 I thought today we would look into open ai, both 0:00:29.000 --> 0:00:35.120 the for profit company and it's parent not for profit organization. So, 0:00:35.240 --> 0:00:38.080 for those of y'all who have managed to dodge all 0:00:38.159 --> 0:00:42.760 the hubbub, open ai is the company behind chat gpt. 0:00:43.320 --> 0:00:46.199 That's the chat bot that's been making headlines for everything 0:00:46.280 --> 0:00:49.720 from offending the musician Nick Cave of Nick Cave and 0:00:49.760 --> 0:00:53.600 the Bad Seeds Fame, to worrying teachers that their students 0:00:53.600 --> 0:00:55.720 are just going to use a chat bot to cheat 0:00:55.760 --> 0:00:59.680 on assignments rather than actually bother to learn something. But 0:00:59.760 --> 0:01:02.560 what about the company that made this thing in the 0:01:02.600 --> 0:01:07.399 first place. Well, the history of open ai dates back 0:01:07.440 --> 0:01:13.399 to twenty when a bunch of very wealthy tech entrepreneurs 0:01:13.400 --> 0:01:16.880 got together and said, you know what, maybe we should 0:01:16.959 --> 0:01:21.480 create an organization that aims to make helpful artificial intelligence 0:01:21.520 --> 0:01:26.760 before someone opens Pandora's box and Unleasha's malevolent you know, 0:01:26.920 --> 0:01:31.560 or at least uncaring super intelligence upon us all or 0:01:31.800 --> 0:01:36.440 something to that effect. Essentially, the goal was to develop 0:01:36.520 --> 0:01:39.400 AI and AI applications in a way that would be 0:01:39.400 --> 0:01:43.520 beneficial to humanity and try to avoid all the scary 0:01:43.640 --> 0:01:49.480 sky net terminator kind of stuff. But to talk about 0:01:49.520 --> 0:01:54.400 this requires us to define some terms, like terms that 0:01:54.520 --> 0:01:57.400 you might think are obvious on the face of it, 0:01:57.440 --> 0:02:00.440 but I would argue are not. So. The big one 0:02:00.520 --> 0:02:04.560 here would be artificial intelligence. There are certain words and 0:02:04.560 --> 0:02:07.800 phrases out in the world that have lots of different meanings, 0:02:08.080 --> 0:02:11.519 and this can sometimes cause confusion and miscommunication. I would 0:02:11.639 --> 0:02:15.840 argue artificial intelligence is a real doozy among these. You 0:02:15.840 --> 0:02:19.200 hear about someone working in AI and you start immediately 0:02:19.240 --> 0:02:24.919 getting preconceived ideas of what that means, and you're probably wrong. Actually, 0:02:25.480 --> 0:02:27.680 now that we're just talking about even just the word 0:02:27.720 --> 0:02:31.840 intelligence has some ambiguity to it. So what do we 0:02:31.919 --> 0:02:35.600 mean when we say that something is intelligent. Well, let's 0:02:35.600 --> 0:02:39.079 take a look at what some dictionaries say. So Webster 0:02:39.320 --> 0:02:43.760 defines intelligence as the ability to learn or understand or 0:02:43.840 --> 0:02:47.399 to deal with new or trying situations, or the ability 0:02:47.480 --> 0:02:51.360 to apply knowledge to manipulate one's environment or to think 0:02:51.400 --> 0:02:56.440 abstractly as measured by objective criteria such as tests. Thanks 0:02:56.480 --> 0:03:02.160 Webster Oxford defines it as the ability to learn, understand, 0:03:02.240 --> 0:03:05.000 and think in a logical way about things. The ability 0:03:05.080 --> 0:03:08.080 to do this, well, it's a little more succinct. But 0:03:08.160 --> 0:03:09.880 then if we really want to boil it down, the 0:03:09.880 --> 0:03:14.919 American Heritage Dictionary defines it as the ability to acquire, understand, 0:03:15.040 --> 0:03:18.840 and use knowledge. That's what intelligence is, according to those. 0:03:19.840 --> 0:03:24.120 Dr dah lyoel Lee, and I apologize Dr Lee for 0:03:24.400 --> 0:03:28.360 butchering your name is a professor of neuroscience and author 0:03:28.560 --> 0:03:32.120 of Birth of Intelligence, and he defines intelligence as the 0:03:32.160 --> 0:03:36.560 ability to solve complex problems or make decisions with outcomes 0:03:36.600 --> 0:03:41.280 that benefit the actor. Dr Lee also acknowledges that intelligence 0:03:41.320 --> 0:03:44.200 is actually pretty hard to define, and that there are 0:03:44.240 --> 0:03:47.520 many different definitions, which you know, we've just seen. Like 0:03:47.600 --> 0:03:52.120 even though all the definitions I mentioned have significant overlap 0:03:52.200 --> 0:03:55.280 between them and they all seem to be dancing around 0:03:55.320 --> 0:03:59.480 the same kind of concept, you might feel like none 0:03:59.520 --> 0:04:02.320 of them quite get it right. And that's where some 0:04:02.440 --> 0:04:06.800 of these challenges come from. Is that just defining intelligence 0:04:07.040 --> 0:04:11.600 before we even get to artificial intelligence is hard. All right, Well, 0:04:11.680 --> 0:04:15.640 let's let's say that intelligence generally is the ability to 0:04:15.760 --> 0:04:18.640 learn and to acquire knowledge and then to use that 0:04:18.720 --> 0:04:21.960 knowledge in new situations. Let's just use it by that 0:04:22.000 --> 0:04:23.680 and say that, you know, it's got an element of 0:04:23.760 --> 0:04:26.560 problem solving that goes with that, which I think is 0:04:26.600 --> 0:04:32.360 pretty much implied. So artificial intelligence, then will artificial suggests 0:04:32.400 --> 0:04:35.119 that it's something that's created by humans rather than found 0:04:35.160 --> 0:04:39.520 in nature. Oxford Languages defines artificial intelligence as the theory 0:04:39.640 --> 0:04:43.320 and development of computer systems able to perform tasks that 0:04:43.480 --> 0:04:49.039 normally require human intelligence, such as visual perception, speech recognition, 0:04:49.400 --> 0:04:54.760 decision making, and translation between languages. So that's a fairly 0:04:54.920 --> 0:04:58.479 decent definition. Uh, But here's where we run into some 0:04:58.480 --> 0:05:03.000 more ambiguity. When we talk about artificial intelligence. We're not 0:05:03.040 --> 0:05:06.880 necessarily using the word intelligence to mean the exact same 0:05:06.920 --> 0:05:10.440 thing when we apply it to a human context. You know, 0:05:10.480 --> 0:05:14.960 a person working in artificial intelligence isn't necessarily trying to 0:05:15.000 --> 0:05:18.880 make a machine think or appear to think like a 0:05:19.000 --> 0:05:24.240 human does. In fact, they're probably not doing anything of 0:05:24.320 --> 0:05:26.479 the sort. They might be working on something that, when 0:05:26.560 --> 0:05:30.640 collected with the work of countless others, ends up contributing 0:05:30.680 --> 0:05:35.360 to that kind of machine but that's different, so AI 0:05:35.440 --> 0:05:40.880 involves a lot of different disciplines and technologies. Facial recognition 0:05:41.600 --> 0:05:45.640 is a type of AI. Speech recognition is a type 0:05:45.640 --> 0:05:49.840 of AI. Text to speech is related to artificial intelligence. 0:05:50.320 --> 0:05:54.640 Robotics share a lot of features with AI, although you 0:05:54.680 --> 0:05:57.640 could also have robots that are fully programmed to complete 0:05:57.640 --> 0:06:01.200 precise routines and and that cases they're just following a 0:06:01.240 --> 0:06:04.440 list of instructions and there's no decision making component right there, 0:06:04.480 --> 0:06:08.680 just literally following step one, step two, step three, step four, repeat. 0:06:09.360 --> 0:06:12.839 So those kinds of robots aren't really in the artificial 0:06:12.880 --> 0:06:16.360 intelligence realm, but there are other robots that are now 0:06:16.400 --> 0:06:20.440 frequently I find that the general public associates the concept 0:06:20.480 --> 0:06:23.960 of artificial intelligence with a machine that appears to have 0:06:24.200 --> 0:06:29.440 knowledge gathering and problems solving capabilities, usually paired with some 0:06:29.520 --> 0:06:32.880 method to put solutions into action, so often in the 0:06:32.920 --> 0:06:35.960 form of a robot or a computer system that's connected 0:06:36.000 --> 0:06:38.719 to stuff that can actually get crap done. I almost 0:06:38.760 --> 0:06:41.039 said the other phrase, but this is a family show, 0:06:42.200 --> 0:06:45.039 so they're thinking about what is often referred to as 0:06:45.240 --> 0:06:48.320 strong AI. These are machines that have a form of 0:06:48.360 --> 0:06:54.080 intelligence that is to all practical purposes. Indistinguishable from human intelligence. 0:06:54.720 --> 0:06:58.400 Now that's not to say that it's processing information the 0:06:58.400 --> 0:07:02.359 exact same way that we peep process information, but that 0:07:02.440 --> 0:07:04.800 the outcome is the same that at the end of 0:07:04.800 --> 0:07:08.080 the day, if the machine and the person were to 0:07:08.120 --> 0:07:12.000 come to the same conclusion, doesn't really matter what steps 0:07:12.080 --> 0:07:14.920 in the middle were taken. Now, if such a thing 0:07:14.960 --> 0:07:18.680 as possible, we're not there yet. We aren't at the 0:07:18.760 --> 0:07:21.520 point where we have this. But the work done in 0:07:21.600 --> 0:07:24.760 AI right now, which is really in the field of 0:07:24.840 --> 0:07:29.480 weak AI, that is, artificial intelligent solutions designed for specific purposes, 0:07:30.280 --> 0:07:36.160 is contributing toward the creation of strong AI. Now there's 0:07:36.160 --> 0:07:38.560 another phrase for strong AI that we need to talk about, 0:07:38.640 --> 0:07:44.120 which is artificial general intelligence or a g I. And 0:07:44.160 --> 0:07:46.720 I know there are a lot of initialisms, that's always 0:07:46.760 --> 0:07:49.480 the case when we talk about tech. But a g 0:07:49.760 --> 0:07:53.120 I general intelligence that kind of tells you, Okay, this 0:07:53.240 --> 0:07:56.520 is an AI that's meant to do lots of different stuff. Right, 0:07:56.560 --> 0:08:00.400 It's not designed to do a specific task and just 0:08:00.480 --> 0:08:03.320 get better and better and better at doing that task. 0:08:03.920 --> 0:08:08.400 It's meant to handle lots of different things, maybe any thing. 0:08:09.160 --> 0:08:11.440 And it's just like if you put a human and 0:08:11.520 --> 0:08:14.560 you have that human go into a situation they've never 0:08:14.600 --> 0:08:19.000 experienced before, how do they cope? Well, it's the goal 0:08:19.080 --> 0:08:22.040 is to create an artificial intelligence that would be able 0:08:22.120 --> 0:08:25.120 to handle new situations in a similar way to the 0:08:25.160 --> 0:08:29.520 way humans do. That's the artificial general intelligence. Again, no 0:08:29.560 --> 0:08:32.280 one has made one of these yet, but that would 0:08:32.360 --> 0:08:37.160 become open a eyes. Primary goal is to create an 0:08:37.240 --> 0:08:40.800 a g I, the first to create an a g I. Now, 0:08:40.880 --> 0:08:44.240 week AI does not mean that artificial intelligence is bad 0:08:44.280 --> 0:08:47.760 at its job or its inferior in some way. In fact, 0:08:47.760 --> 0:08:51.640 week AI might be much better at doing its specific 0:08:51.720 --> 0:08:56.160 task than humans are at completing that specific task. It's 0:08:56.160 --> 0:08:58.400 just that this is all the week AI can do. 0:08:58.600 --> 0:09:02.520 It can't do other things things it's operating under constraints. 0:09:02.520 --> 0:09:05.360 So as an example, let's just think of something that's 0:09:05.400 --> 0:09:08.280 really simple that you wouldn't even think of as being intelligent, 0:09:08.840 --> 0:09:13.360 like a basic calculator, not even a scientific calculator, a 0:09:13.440 --> 0:09:17.400 basic calculator like one that might be handed out by 0:09:17.559 --> 0:09:23.280 a bank, and you can enter a pretty tough mathematical 0:09:23.360 --> 0:09:26.360 problem into the calculator and it will provide a solution 0:09:26.400 --> 0:09:28.480 in a fraction of the time it would take your 0:09:28.520 --> 0:09:31.680 average human to do the same work, but that same 0:09:31.760 --> 0:09:34.480 human could do other stuff like maybe that human can 0:09:34.480 --> 0:09:38.720 play the guitar or juggle or paint or play a 0:09:38.840 --> 0:09:42.960 video game or any of an endless number of other tasks. 0:09:43.000 --> 0:09:45.719 But the calculator can't do that. It can just calculate. 0:09:45.760 --> 0:09:48.160 That's all it can do, and it can do it 0:09:48.320 --> 0:09:52.160 really well, but it's unable to extend this capability to 0:09:52.400 --> 0:09:56.480 anything beyond that purpose. Now, sometimes when we encounter a 0:09:56.679 --> 0:10:00.880 really good week AI, we can fool ourselves into thinking 0:10:00.880 --> 0:10:04.200 that the AI is doing something really magical, or that 0:10:04.280 --> 0:10:08.000 it's matching our own capabilities to think. It can actually 0:10:08.000 --> 0:10:11.000 be pretty easy to fall into this trap. A sufficiently 0:10:11.040 --> 0:10:14.320 sophisticated chatbot might fool listen to thinking that the machine 0:10:14.360 --> 0:10:18.280 we're chatting with is actually thinking itself. But it's not, 0:10:18.559 --> 0:10:21.920 at least not in the same way that people do. Now, 0:10:21.920 --> 0:10:24.280 why did I go through all of that trouble to 0:10:24.360 --> 0:10:28.120 define all these things? Well? The founding principle of open 0:10:28.160 --> 0:10:32.880 AI is to create artificial general intelligence and AI applications 0:10:32.880 --> 0:10:38.960 and technologies through a responsible, thoughtful approach, and that implies 0:10:39.400 --> 0:10:42.600 that there's an irresponsible way to do this, and that 0:10:42.720 --> 0:10:45.720 following such an irresponsible way could lead to disaster. And 0:10:45.760 --> 0:10:48.640 that's where we get to our science fiction stories, and 0:10:48.679 --> 0:10:51.440 that certainly tracks. You know, I'm not here to tell 0:10:51.440 --> 0:10:56.079 you that that's an unreasonable fear. That fear is totally reasonable. 0:10:56.640 --> 0:11:00.760 In fact, we've been seeing how weak AI can and 0:11:00.840 --> 0:11:04.640 does cause problems, or maybe how I should say, are 0:11:04.640 --> 0:11:09.280 our reliance upon week AI can cause problems. The AI 0:11:09.440 --> 0:11:12.000 on its own may not be able to cause a 0:11:12.040 --> 0:11:15.760 problem by itself, but because we rely on it, then 0:11:15.840 --> 0:11:17.880 we go and we create these problems. So let's go 0:11:17.920 --> 0:11:21.400 with facial recognition for this one. It has been shown 0:11:21.960 --> 0:11:25.439 time and again that many of the facial recognition technologies 0:11:25.480 --> 0:11:29.120 that are actively deployed in the world today have bias 0:11:29.160 --> 0:11:32.840 built into them. They are fairly reliable at identifying people 0:11:32.840 --> 0:11:38.080 within certain populations, like white people primarily, but then with 0:11:38.120 --> 0:11:41.760 people of color, these systems aren't nearly as accurate. So 0:11:41.920 --> 0:11:45.720 what happens is that these facial recognition systems can generate 0:11:45.760 --> 0:11:50.360 false positives more frequently for say, black people. And because 0:11:50.400 --> 0:11:53.440 we have law enforcement agencies that are making active use 0:11:53.559 --> 0:11:57.600 of facial recognition technologies when looking for suspects, this means 0:11:57.640 --> 0:12:01.839 that police can and do end up harassing innocent people, 0:12:02.000 --> 0:12:06.640 all based off of this misidentification. So imagine one day 0:12:06.880 --> 0:12:09.840 you're just going about your business and then suddenly law 0:12:09.920 --> 0:12:13.240 enforcement swoops in and arrests you for a crime not 0:12:13.320 --> 0:12:16.959 only you didn't commit, but you also have no knowledge 0:12:17.400 --> 0:12:20.480 of this crime. And it's all because a machine somewhere 0:12:20.520 --> 0:12:23.920 said this is the person you want. Now, imagine how 0:12:23.960 --> 0:12:27.280 your life would be affected. What if it happened while 0:12:27.400 --> 0:12:30.040 you were at work or at school. How do you 0:12:30.080 --> 0:12:32.640 think the people around you would react when police come 0:12:32.640 --> 0:12:35.440 in and arrest you. How many of those people would 0:12:35.440 --> 0:12:38.880 treat you differently even after hearing that the whole thing 0:12:38.960 --> 0:12:42.560 was just a mistake. What kind of stress would that 0:12:42.600 --> 0:12:46.120 put on you and the people in your life? Now? 0:12:46.120 --> 0:12:48.440 The reason I'm really nailing this home is because this 0:12:48.520 --> 0:12:52.360 stuff is happening right. This problem is a real problem. 0:12:52.440 --> 0:12:55.600 This is not a theoretical it's not a hypothetical. Real 0:12:55.840 --> 0:13:00.160 people have had their lives up ended because police have 0:13:00.280 --> 0:13:06.320 relied upon faulty facial recognition technology and saying, oops, it 0:13:06.400 --> 0:13:09.000 was our mistake doesn't fix your life when it's been 0:13:09.000 --> 0:13:13.199 turned upside down. Or as Matthew Grissinger of the Institute 0:13:13.240 --> 0:13:17.400 for Safe medication practices has put it quote. The tendency 0:13:17.480 --> 0:13:21.160 to favor or give greater credence to information supplied by 0:13:21.160 --> 0:13:24.680 technology e g. And a d C display, and to 0:13:24.760 --> 0:13:29.600 ignore a manual source of information that provides contradictory information 0:13:30.000 --> 0:13:33.400 e g. Handwritten entry on the computer generated m a R, 0:13:34.360 --> 0:13:39.360 even if it is correct, illustrates the phenomenon of automation bias. 0:13:39.880 --> 0:13:44.240 Automation complacency is a closely linked, overlapping concept that refers 0:13:44.280 --> 0:13:48.400 to the monitoring of technology with less frequency or vigilance 0:13:48.679 --> 0:13:51.600 because of a lower suspicion of error and a stronger 0:13:51.640 --> 0:13:54.920 belief in its accuracy end quote. So in other words, 0:13:55.800 --> 0:13:58.959 we have a tendency to trust the output of machines, 0:13:59.800 --> 0:14:03.200 and that trust is not always warranted. This can get 0:14:03.280 --> 0:14:05.600 us into trouble. We can trust that the machines know 0:14:05.640 --> 0:14:09.040 what they're doing and that the way they process information 0:14:09.559 --> 0:14:15.400 is reliable and even infallible, and by acting upon that 0:14:15.480 --> 0:14:20.480 we can create terrible consequences. Mr griss Singer's context was 0:14:20.560 --> 0:14:24.760 within the field of medication prescriptions, which, obviously, if you 0:14:24.800 --> 0:14:27.920 were to rely solely upon automated output and that automated 0:14:27.960 --> 0:14:32.160 output was wrong, could result in terrible consequences. But I'm 0:14:32.200 --> 0:14:34.840 sure you can imagine countless other scenarios in which an 0:14:34.880 --> 0:14:39.240 over reliance on technology could lead to disaster. We'll talk 0:14:39.240 --> 0:14:43.160 about another one when we come back from this quick break. 0:14:52.720 --> 0:14:55.200 We're back, and before the break, I was talking about 0:14:55.200 --> 0:14:59.000 how we have a tendency to put too much trust 0:14:59.560 --> 0:15:03.400 inte knowlogy in general and AI in particular, and how 0:15:03.440 --> 0:15:06.680 this can come back to haunt us. So an example 0:15:06.680 --> 0:15:10.280 that leaps to my mind is autonomous cars. And I'm 0:15:10.320 --> 0:15:12.240 going to be the first to admit I jumped on 0:15:12.320 --> 0:15:16.480 the autonomous car bandwagon without applying nearly enough critical thinking. 0:15:17.440 --> 0:15:21.560 I was really considering just the surface level of what 0:15:21.640 --> 0:15:24.440 it would mean to have autonomous cars. So here's how 0:15:24.480 --> 0:15:27.640 my flawed logic went. This is why I was so 0:15:27.760 --> 0:15:32.400 like Gung Ho on autonomous cars several years ago now 0:15:32.840 --> 0:15:36.800 and have subsequently changed my my thinking. So the way 0:15:36.880 --> 0:15:40.520 I originally thought was computer processors are wicked fast, right 0:15:40.640 --> 0:15:46.240 like a CPU in your computer can complete calculations so quickly, 0:15:46.360 --> 0:15:50.440 millions of them every second, billions in fact, depending upon 0:15:50.520 --> 0:15:55.080 the the sophistication of the of the operations. And then 0:15:55.200 --> 0:15:58.560 you have parallel processing, right like if you have a 0:15:58.640 --> 0:16:03.480 multi core processor could have lots of functions all being 0:16:03.520 --> 0:16:07.920 performed simultaneously by this processor. Then, on top of that, 0:16:08.000 --> 0:16:10.600 you could have sensors on your car that cover three 0:16:11.040 --> 0:16:14.520 sixty degrees of view around the vehicle, so you would 0:16:14.560 --> 0:16:18.160 be able to have the system pay attention in every 0:16:18.200 --> 0:16:22.520 single direction simultaneously, whereas a human driver can only pay 0:16:22.520 --> 0:16:25.520 attention within their field of view and then with the 0:16:25.520 --> 0:16:28.440 help of some mirrors, get a little extra you know, 0:16:28.640 --> 0:16:33.240 awareness around them. You could have mechanical systems that could 0:16:33.280 --> 0:16:36.920 react immediately upon receiving a command from the processors with 0:16:36.960 --> 0:16:40.000 no delay, so you don't have that delay of action 0:16:40.080 --> 0:16:43.160 between when you sense something happening and when you are 0:16:43.200 --> 0:16:47.680 able to act on that. So, surely such a system 0:16:47.720 --> 0:16:52.160 with incredible processing power, with three sixty degrees of awareness, 0:16:52.200 --> 0:16:56.120 with this immediate ability to react, would be able to 0:16:56.160 --> 0:16:59.960 engage in defensive driving faster, more effectively, and safer than 0:17:00.200 --> 0:17:04.879 in a human ever could. Clearly, machines are superior. We 0:17:04.920 --> 0:17:07.600 should all be in autonomous cars. This is where I 0:17:07.640 --> 0:17:10.680 ran into the problem of overreliance on technology. Sure, in 0:17:10.880 --> 0:17:14.640 isolated cases, everything I was thinking might be at least 0:17:14.680 --> 0:17:17.399 partly true, but when you take it together and you 0:17:17.400 --> 0:17:20.280 start to apply it in the field in a vehicle, 0:17:20.760 --> 0:17:23.800 things are far more complicated than I ever gave it 0:17:23.840 --> 0:17:27.160 credit for. And as we have seen with advanced driver 0:17:27.240 --> 0:17:31.720 assist features, if we rely too much on this technology, 0:17:31.760 --> 0:17:36.240 it can and does lead to tragedy. So we've seen 0:17:36.320 --> 0:17:40.640 this play out where people have depended too heavily upon 0:17:40.760 --> 0:17:43.280 this tech and have paid for it with their lives. 0:17:43.760 --> 0:17:47.159 So we know that this is more complex than what 0:17:47.400 --> 0:17:50.280 I initially thought of back in my naive days of 0:17:50.359 --> 0:17:55.080 being so, you know, flag bearing for the whole autonomous 0:17:55.320 --> 0:17:58.959 car our movement, and I still believe in autonomous cars 0:17:59.359 --> 0:18:04.080 and how they could contribute to greater safety, but I 0:18:04.160 --> 0:18:08.200 also recognize that it's a far more complex problem than 0:18:08.280 --> 0:18:13.960 what I originally imagined. All right, so we have thoroughly 0:18:14.520 --> 0:18:18.280 defined the problem at this point. Right. Artificial intelligence has 0:18:18.359 --> 0:18:22.240 the potential to help us do amazing things, but only 0:18:22.520 --> 0:18:26.800 if we develop and deploy it properly. Otherwise it could 0:18:26.840 --> 0:18:31.720 exacerbate existing problems or even create all new problems. So 0:18:31.800 --> 0:18:36.000 there's a need to be thoughtful about design and application 0:18:36.160 --> 0:18:42.000 and deployment and distribution. So who decided to codify this 0:18:42.119 --> 0:18:47.560 philosophy of being careful about AI and create an organization 0:18:47.680 --> 0:18:51.160 dedicated to doing that. Well, the two people who are 0:18:51.160 --> 0:18:55.520 frequently cited as the co founders for open ai are 0:18:56.160 --> 0:19:00.080 Elon Musk and Sam Altman, though I would hate and 0:19:00.200 --> 0:19:03.040 add there were many other people who are really co 0:19:03.200 --> 0:19:07.080 founders as well, but these are the two that, you know, 0:19:07.280 --> 0:19:10.639 everyone says, these are the guys who started talking and 0:19:10.680 --> 0:19:13.480 kind of generated the initial idea that became open AI. 0:19:14.440 --> 0:19:18.679 So let's start with Musk. So years before he decided 0:19:18.720 --> 0:19:21.880 to drop billions of dollars in an effort to troll 0:19:21.960 --> 0:19:26.040 the Internet whenever he wanted to, Mr Musk was something 0:19:26.080 --> 0:19:29.639 of an AI doomsayer. You know, he was warning that 0:19:29.760 --> 0:19:34.840 artificial intelligence could potentially pose an existential threat to humans. 0:19:35.320 --> 0:19:37.960 Kind of this idea of we create a human level 0:19:38.080 --> 0:19:43.240 or even superhuman level strong AI, and then it turns 0:19:43.280 --> 0:19:46.919 on us and wipes us out. And certainly bad AI 0:19:47.200 --> 0:19:49.720 can be a huge issue. We just talked about how 0:19:49.760 --> 0:19:53.960 even weak AI can be a really big problem. Now, 0:19:54.000 --> 0:19:57.040 I don't think we're close to having a human, let 0:19:57.040 --> 0:20:02.640 alone superhuman intelligence determined to wipe out humanity emerge, but 0:20:02.720 --> 0:20:04.960 you know, you can definitely have bad AI contribute to 0:20:05.040 --> 0:20:09.760 human suffering. See also Tesla, one of Mr Musk's companies. 0:20:09.840 --> 0:20:12.359 One might even argue that Elon Musk knows the danger 0:20:12.640 --> 0:20:16.000 that artificial intelligence poses to humanity because one of his 0:20:16.080 --> 0:20:19.080 companies is leading the charge in that field in the 0:20:19.119 --> 0:20:24.000 form of Tesla autopilot and full self driving modes. Now again, 0:20:24.240 --> 0:20:27.119 you could say that I'm being unkind, because we do 0:20:27.200 --> 0:20:30.919 need to remember that Tesla, despite the languages it uses 0:20:30.960 --> 0:20:35.159 for marketing purposes, does alert drivers that they are not 0:20:35.200 --> 0:20:37.720 supposed to take their hands off the wheel or stop 0:20:37.720 --> 0:20:41.000 paying attention to the road, and that at least in 0:20:41.200 --> 0:20:44.960 all the accounts I have read about terrible accidents involving 0:20:45.000 --> 0:20:48.560 Tesla vehicles that were in driver assists mode, it sounds 0:20:48.600 --> 0:20:52.600 like the driver wasn't following those directions. So you could 0:20:52.640 --> 0:20:56.119 argue that, you know, the driver ultimately is at fault 0:20:56.119 --> 0:21:00.399 because they're failing to adhere to the instructions that Tesla gives. 0:21:00.960 --> 0:21:03.600 The flip side of that is that Tesla markets these 0:21:03.640 --> 0:21:08.600 features as if they are more than you know, sophisticated 0:21:08.640 --> 0:21:13.440 driver assist features. The other co founder of open Ai 0:21:13.520 --> 0:21:17.639 that's frequently mentioned is Sam Altman, the current CEO of 0:21:17.720 --> 0:21:22.920 open Ai. Sam Altman was previously president of y Combinator. 0:21:23.160 --> 0:21:26.000 He became president of y Combinator in fourteen, which was 0:21:26.040 --> 0:21:29.119 the year before he co founded open Ai with Elon Musk. 0:21:29.880 --> 0:21:32.360 And you might say, well, what is why Combinator. It's 0:21:32.400 --> 0:21:36.800 a startup accelerator, which doesn't really mean anything either, right, Well, 0:21:37.240 --> 0:21:40.560 that's a company that helps people who have startup business 0:21:40.600 --> 0:21:44.719 ideas get the support they need in order to launch 0:21:44.800 --> 0:21:47.440 their idea and make it a reality. So that can 0:21:47.440 --> 0:21:50.720 include stuff like mentoring the startup leaders so that they 0:21:50.760 --> 0:21:55.240 can build a good business model and create the right 0:21:55.680 --> 0:21:58.119 corporate structure that they're going to need in order to 0:21:58.119 --> 0:22:02.160 do business, all the way up to prepping them and 0:22:02.240 --> 0:22:05.679 connecting them with people that they can pitch their idea 0:22:05.760 --> 0:22:09.680 to in order to get investment into their startup. So 0:22:09.760 --> 0:22:13.240 one of the big valuable services that companies like y 0:22:13.320 --> 0:22:18.200 Combinator provide is access to the investor community that you 0:22:18.280 --> 0:22:22.199 might not otherwise be able to get to without that 0:22:22.280 --> 0:22:25.520 kind of support. Now, Altman would continue to serve as 0:22:25.680 --> 0:22:28.840 y Combinator president until twenty nineteen. At that point he 0:22:28.920 --> 0:22:32.640 stepped down from that position to focus on open Ai. 0:22:33.200 --> 0:22:36.280 H Elon Musk would sit on the board of directors 0:22:36.280 --> 0:22:39.359 for open Ai until ten. We'll talk about that in 0:22:39.440 --> 0:22:42.240 just a bit. Now. I mentioned that there were also 0:22:42.320 --> 0:22:45.920 other co founders, So in addition to these two entrepreneurs, 0:22:46.320 --> 0:22:50.520 early founders in the open Ai initiative included Greg Brockman, 0:22:50.680 --> 0:22:54.159 who's still there. I believe he's a former chief technology 0:22:54.200 --> 0:22:58.560 officer of Stripe, the payment processing company. The PayPal co 0:22:58.680 --> 0:23:02.040 founder Peter Thiel was also one of the early investors 0:23:02.119 --> 0:23:06.639 in open Ai. LinkedIn co founder read Garrett Hoffman, another 0:23:06.680 --> 0:23:11.679 one one of Altman's y Combinator colleagues. Jessica Livingston was another, 0:23:11.840 --> 0:23:15.920 and there were a few more. Now collectively, the founders 0:23:15.920 --> 0:23:21.000 and partners all pledged one billion dollars to fund open ai, 0:23:21.080 --> 0:23:24.360 which again was meant to be a nonprofit organization dedicated 0:23:24.400 --> 0:23:28.640 to developing productive, friendly AI and not the scary pew 0:23:28.720 --> 0:23:32.600 pew lasers kind of AI. But then there's also the 0:23:32.760 --> 0:23:38.080 open part of open ai. So during the brainstorming that 0:23:38.119 --> 0:23:40.720 would lead to the founding of this organization, the co 0:23:40.880 --> 0:23:44.719 founders talked about how big tech companies typically do all 0:23:44.760 --> 0:23:49.840 their AI development behind closed doors with no transparency, and 0:23:49.960 --> 0:23:53.879 that their version of AI was meant to benefit the 0:23:54.320 --> 0:23:58.439 parent company, not humanity as a whole. The open Ai 0:23:58.640 --> 0:24:02.359 organization is going to take a different approach. The idea 0:24:02.480 --> 0:24:05.879 was to share the benefits of AI research with the 0:24:05.920 --> 0:24:09.639 world and do that as much as possible on an 0:24:09.720 --> 0:24:12.399 effort to evolve AI in a way that helps but 0:24:12.520 --> 0:24:16.359 doesn't harm. Researchers would be encouraged to publish their work 0:24:16.440 --> 0:24:20.680 in various formats as frequently as they could, and any 0:24:20.760 --> 0:24:24.800 patents that open ai would secure would similarly be shared 0:24:24.880 --> 0:24:27.760 with the world. The message appeared to be the goal 0:24:28.000 --> 0:24:32.800 is more important than the organization, that friendly AI is 0:24:32.880 --> 0:24:37.520 the chief important goal here, and that open ai only 0:24:37.560 --> 0:24:41.920 exists to see that become reality, and that open ai 0:24:42.040 --> 0:24:45.199 was really kind of more of a a shepherd of 0:24:45.320 --> 0:24:48.639 pushing AI into this direction rather than brazenly forging a 0:24:48.720 --> 0:24:51.880 path into the wilderness, although that's not how things would 0:24:51.880 --> 0:24:55.480 turn out now. Early on, the organization grew mostly through 0:24:55.600 --> 0:24:59.919 connections in the AI research community, with Luminaries and x 0:25:00.040 --> 0:25:04.159 birds joining the organization, but the organization itself kind of 0:25:04.240 --> 0:25:08.320 lacked a real sense of leadership or direction. There was 0:25:08.359 --> 0:25:11.640 this noble goal, right, Everyone knew that they were trying 0:25:11.680 --> 0:25:17.400 to make reliable, safe, friendly, beneficial AI, but how there 0:25:17.440 --> 0:25:20.560 wasn't really any plan for how to get to where 0:25:20.560 --> 0:25:26.160 they wanted to be. Google researcher Dariomday visited open ai 0:25:26.280 --> 0:25:29.880 in mid and he came away thinking that no one 0:25:29.920 --> 0:25:32.280 at the organization really had any idea of what they 0:25:32.280 --> 0:25:36.200 were doing. Despite that, or maybe because of it, i'm 0:25:36.200 --> 0:25:39.280 a Day would join the organization a couple of months 0:25:39.400 --> 0:25:43.000 later and became head of research there. Now. One of 0:25:43.040 --> 0:25:45.840 the first things to emerge from open ai was in 0:25:45.920 --> 0:25:50.080 ten like it was founded in late and in twenty 0:25:50.200 --> 0:25:54.720 sixteen they were already producing some interesting stuff. And the 0:25:54.800 --> 0:25:58.160 first up was a testing environment that the organization called 0:25:58.480 --> 0:26:03.080 Jim Jim as a gymnasium, not as in Jimmy Jim 0:26:03.200 --> 0:26:08.359 Jim Jim Hawkins. So what was being tested, well, they 0:26:08.400 --> 0:26:12.879 were testing learning agents. This brings us to a discipline 0:26:12.920 --> 0:26:18.080