WEBVTT - The State of AI 0:00:04.120 --> 0:00:07.160 Get in touch with technology with tech Stuff from how 0:00:07.200 --> 0:00:13.880 stuff works dot com. Hey there, and welcome to tech Stuff. 0:00:13.920 --> 0:00:17.040 I'm your host, Jonathan Strickland. I'm an executive producer at 0:00:17.040 --> 0:00:20.000 how Stuff Works in I Love all Things tech, and 0:00:20.040 --> 0:00:25.800 welcome to episode nine hundred ninety nine of tech Stuff. Yep, 0:00:25.880 --> 0:00:28.040 we're gonna have a really big shin dig for the 0:00:28.080 --> 0:00:30.159 next one. And by shin dig, I mean I'm going 0:00:30.200 --> 0:00:33.559 to record another show. But today we're going to talk 0:00:33.600 --> 0:00:39.159 about the state of artificial intelligence because in late June 0:00:40.000 --> 0:00:45.159 Dr Nathan Bane and Ian Hogarth presented a report titled 0:00:45.280 --> 0:00:48.080 the State of a I, giving an update on the 0:00:48.080 --> 0:00:51.560 advancements that have happened in the field of artificial intelligence 0:00:51.640 --> 0:00:55.360 over the course of the last twelve months, so end 0:00:55.400 --> 0:00:59.000 of well summer of seventeen to the summer of eighteen essentially. 0:00:59.120 --> 0:01:00.640 So I thought it might be in using to talk 0:01:00.640 --> 0:01:04.039 about what they found as they researched the topic. But first, 0:01:04.120 --> 0:01:08.240 let's define a few terms, because artificial intelligence is one 0:01:08.240 --> 0:01:13.920 of those categories of topics that tends to encompass a 0:01:13.920 --> 0:01:16.479 lot of different ideas and unless you define what you're 0:01:16.520 --> 0:01:19.959 talking about early on, uh, you could have two people 0:01:20.000 --> 0:01:24.000 talking about different aspects of AI, and they don't realize 0:01:24.000 --> 0:01:26.800 they're talking about different aspects and they think they're disagreeing, 0:01:26.840 --> 0:01:30.080 but in reality they're actually agreeing. It's just they haven't 0:01:30.080 --> 0:01:32.880 defined their terms. So we're gonna do that first. So 0:01:33.440 --> 0:01:39.440 the term artificial intelligence dates to nineteen fifty five. John McCarthy, 0:01:39.560 --> 0:01:42.240 who was was a computer scientist, had coined the phrase 0:01:42.240 --> 0:01:45.640 while developing a plan for a conference that took place 0:01:45.720 --> 0:01:50.000 at Dartmouth, and it was supposed to happen in nineteen 0:01:50.040 --> 0:01:52.160 fifty six. So he came up with the term artificial 0:01:52.160 --> 0:01:54.880 intelligence about a year before the conference was to take place. 0:01:55.560 --> 0:01:59.280 And his was a definition of general intelligence, which means 0:01:59.680 --> 0:02:02.560 it would be a way to allow machines to reason, 0:02:03.000 --> 0:02:07.040 engage an abstract thought, do some problem solving, and also 0:02:07.080 --> 0:02:10.080 to pursue self improvement. And I think a lot of 0:02:10.120 --> 0:02:13.960 people still think of this when they are thinking of 0:02:13.960 --> 0:02:17.320 the phrase artificial intelligence. That means a computer that can 0:02:17.440 --> 0:02:20.320 process information in a way that is at least comparable, 0:02:20.520 --> 0:02:24.959 if not identical, to the way we humans process information. 0:02:25.400 --> 0:02:29.000 With the concept of self improvement included, there's at least 0:02:29.000 --> 0:02:32.240 an implication that such a machine would at least possess 0:02:32.320 --> 0:02:34.680 some degree of self awareness. It would have to know 0:02:35.240 --> 0:02:40.080 at least that it needed to improve upon something. It 0:02:40.160 --> 0:02:42.600 may not really have a full sense of self. However, 0:02:42.680 --> 0:02:47.200 that's not necessarily required in order to have self improvement, 0:02:47.760 --> 0:02:52.120 so uh, there'll just be a degree of it. Um. However, 0:02:52.160 --> 0:02:54.520 it could go as far as a machine that actually 0:02:54.560 --> 0:02:58.519 knows that it exists within a world it has a 0:02:58.560 --> 0:03:05.680 sense of self. That could be a potential uh way 0:03:05.680 --> 0:03:09.760 to interpret this concept. But as time has gone on, 0:03:09.840 --> 0:03:14.359 we have needed to narrow down various definitions of artificial 0:03:14.400 --> 0:03:19.600 intelligence because intelligence itself human intelligence is a very broad term. 0:03:19.680 --> 0:03:23.200 It encompasses many different things. It takes more than creating 0:03:23.240 --> 0:03:26.720 a machine that can process information at a faster rate 0:03:26.720 --> 0:03:30.320 than humans. We wouldn't say that a calculator is more 0:03:30.360 --> 0:03:33.680 intelligent than a person just because a calculator can, uh 0:03:33.919 --> 0:03:37.960 can perform a complex mathematical calculation in a fraction of 0:03:38.000 --> 0:03:41.000 the amount of time your average person could, right, because 0:03:41.040 --> 0:03:43.160 your average person can still do tons of stuff that 0:03:43.200 --> 0:03:46.280 a calculator can't do, so we wouldn't call the calculator intelligent. 0:03:46.560 --> 0:03:49.760 It takes more than just processing information at a very 0:03:49.840 --> 0:03:54.920 fast rate, so we have to go beyond just efficiency. 0:03:55.080 --> 0:03:59.240 If we want to talk about intelligence, the Encyclopedia Britannica 0:03:59.360 --> 0:04:02.720 defines AI as the ability of a digital computer or 0:04:02.800 --> 0:04:09.560 computer controlled robot to perform tasks commonly associated with intelligent beings. Uh. 0:04:09.560 --> 0:04:12.440 This isn't a bad definition, but it does require you 0:04:12.480 --> 0:04:15.480 to take another step back to consider what sort of 0:04:15.520 --> 0:04:21.160 actions would be considered intelligent versus those that are more instinctive. So, 0:04:21.320 --> 0:04:25.800 for example, if a fly sees that there is a 0:04:25.880 --> 0:04:28.880 hand coming toward it to squish it and it flies away, 0:04:29.120 --> 0:04:32.200 is that instinctual or is that intelligent? While you would 0:04:32.240 --> 0:04:36.680 argue it's more instinct, it's not a sign of intelligence. Uh. 0:04:36.720 --> 0:04:39.320 There have been a lot of experiments with insects in 0:04:39.400 --> 0:04:43.440 general to display that behaviors they have that appear to 0:04:43.560 --> 0:04:48.600 be intelligent because they involve complex behaviors, Uh, turn out 0:04:48.640 --> 0:04:51.840 to be instinctual because if you start messing with them, 0:04:51.880 --> 0:04:54.440 they will just repeat the exact same sequence over and 0:04:54.480 --> 0:04:57.080 over again. They don't have a way of retaining the 0:04:57.120 --> 0:05:00.120 information that just happen in order to build upon it, 0:05:00.400 --> 0:05:03.279 which is all part of learning. Right as human beings, 0:05:03.560 --> 0:05:07.239 when we encounter new information, we can incorporate that into 0:05:07.240 --> 0:05:10.760 our knowledge and then we can build upon that in 0:05:10.960 --> 0:05:16.039 future encounters with similar scenarios. We can even generalize from 0:05:16.080 --> 0:05:22.560 that situation and apply that knowledge to similar but different scenarios. 0:05:23.279 --> 0:05:29.200 This is something that differentiates intelligence from just instinct. AI 0:05:29.320 --> 0:05:33.000 is also a multidisciplinary field. It's not just one area 0:05:33.000 --> 0:05:36.640 of study. It's in fact, lots of different areas of study, 0:05:36.680 --> 0:05:41.560 everything from computer science to biology, to neuroscience, to psychology, 0:05:41.600 --> 0:05:45.800 to engineering and more. It's tons of different areas of 0:05:45.800 --> 0:05:49.279 study that all go into AI, and they're also a 0:05:49.320 --> 0:05:53.159 lot of different ways that we can categorize AI. One 0:05:53.160 --> 0:05:56.080 way is that we could just divide AI into two 0:05:56.200 --> 0:06:01.599 very large categories week AI and strong long AI. John 0:06:01.720 --> 0:06:06.320 Searle proposed this way of defining AI back in In fact, 0:06:06.400 --> 0:06:09.360 he argued that week AI is probably the best will 0:06:09.360 --> 0:06:12.800 ever do, will never have strong AI. Week AI refers 0:06:12.839 --> 0:06:16.359 to a simulation of human thought. That you could have 0:06:16.400 --> 0:06:19.359 a machine that appears to think like a person, but 0:06:19.480 --> 0:06:23.080 that's just an appearance. In reality, once you strip away 0:06:23.120 --> 0:06:26.520 all the different layers, it turns out it's just a simulation. 0:06:26.600 --> 0:06:31.360 The computer is not quote unquote actually thinking. Uh, it's 0:06:31.680 --> 0:06:34.960 mimicking the way we think, and it tends to be 0:06:35.000 --> 0:06:38.760 in a relatively narrow band of applications. It may perform 0:06:38.800 --> 0:06:41.880 those applications very well, it might even do so better 0:06:41.920 --> 0:06:45.880 than a human can, but it cannot act outside of 0:06:45.920 --> 0:06:48.760 that narrow band, or if it does attempt to act 0:06:48.800 --> 0:06:51.240 outside of it, it doesn't do so very well. So 0:06:51.520 --> 0:06:55.320 an example of this might be IBM's Deep Blue, which 0:06:55.520 --> 0:07:00.520 was the computer they designed to play against chess grand masters, 0:07:00.560 --> 0:07:03.520 and it plays chess really, really well, well enough to 0:07:03.560 --> 0:07:07.520 beat grand masters at chess, But you couldn't then tell 0:07:07.560 --> 0:07:12.920 it to sort a complex series of problems so that 0:07:12.960 --> 0:07:16.320 you could tackle them properly, or ask it about what 0:07:16.400 --> 0:07:19.160 the weather is going to be like three days from now. 0:07:19.400 --> 0:07:22.040 You couldn't set it to other tasks. It was programmed 0:07:22.040 --> 0:07:25.360 for a very specific application, and you could not just 0:07:25.440 --> 0:07:29.160 leverage that quote unquote intelligence to something else. Whereas with 0:07:29.200 --> 0:07:33.280 a person, you can have a person tackle all sorts 0:07:33.320 --> 0:07:35.960 of different problems, and even if the person has never 0:07:36.040 --> 0:07:40.320 encountered that problem before, he or she can apply the information, 0:07:40.360 --> 0:07:44.200 the knowledge that they have accumulated throughout their experiences an 0:07:44.200 --> 0:07:47.360 attempt to apply it as best they are able to 0:07:47.480 --> 0:07:49.760 the new task. They might not be very good at 0:07:49.760 --> 0:07:52.360 the new task, but they can at least try to 0:07:52.400 --> 0:07:54.720 do that based upon the information and the knowledge that 0:07:54.760 --> 0:07:58.239 they have gained in their past experiences. It doesn't matter 0:07:58.240 --> 0:08:01.480 how many games of chess Deep Blue plays, it's still 0:08:01.520 --> 0:08:04.640 not going to be good at doing other tasks. Strong 0:08:04.760 --> 0:08:08.200 AI would refer to an artificial intelligence that could actually 0:08:08.280 --> 0:08:10.960 think in a way that's at least analogous to the 0:08:10.960 --> 0:08:14.080 way we humans think, even if it uses a different 0:08:14.120 --> 0:08:17.640 methodology to do so, and it could apply that intelligence 0:08:17.640 --> 0:08:21.600 to any situation, not just a narrow set of situations. 0:08:21.640 --> 0:08:23.840 And that does not mean that it would immediately be 0:08:23.920 --> 0:08:27.640 great at everything. This isn't a definition of a super 0:08:27.680 --> 0:08:32.520 intelligent computer. It might also be pretty crappy at brand 0:08:32.520 --> 0:08:36.200 new tasks, but it can learn over time and improve 0:08:36.360 --> 0:08:41.319 over time, self improvement being an important concept in intelligence. 0:08:42.120 --> 0:08:44.920 So I can learn from mistakes and even generalize what 0:08:45.040 --> 0:08:48.640 it's learned into new, unrelated situations, so it would not 0:08:48.720 --> 0:08:51.760 be bound by those narrow set of applications like a 0:08:51.800 --> 0:08:54.760 week AI would. Cirl, By the way, is also the 0:08:54.760 --> 0:08:58.360 philosopher who proposed the Chinese room thought experiment to argue 0:08:58.400 --> 0:09:01.960 against strong AI. I've talked about the Chinese room thought 0:09:02.000 --> 0:09:07.520 experiment and other episodes. Basically, it's this this thought experiment 0:09:07.600 --> 0:09:10.840 where you imagine that you are in a room and 0:09:10.880 --> 0:09:13.080 the room only has a door, and the door has 0:09:13.120 --> 0:09:16.760 a little slot in it, and occasionally a piece of 0:09:16.800 --> 0:09:19.320 paper is shoved through the slot. You get the piece 0:09:19.320 --> 0:09:23.040 of paper, something is written in an alphabet that you 0:09:23.120 --> 0:09:25.920 don't you don't understand, you have you have no knowledge 0:09:25.960 --> 0:09:28.600 of this alphabet. It's just it looks like squiggles to you. 0:09:29.280 --> 0:09:32.439 But you have an enormous book, and that enormous book 0:09:32.520 --> 0:09:36.040 has a list of all these different pages of squiggles. 0:09:36.040 --> 0:09:38.240 So what you do is you consult the enormous book, 0:09:38.280 --> 0:09:40.959 you look for a page of squiggles that's identical to 0:09:41.040 --> 0:09:43.120 the one that was sent to you, and then you 0:09:43.160 --> 0:09:46.600 follow the instructions of what to do. When you get 0:09:46.640 --> 0:09:50.040 a page that has the squiggles, you follow the instructions, 0:09:50.360 --> 0:09:52.560 you put the output through the slot in the door, 0:09:52.800 --> 0:09:56.280 and things continue. Searl argued, this is essentially the same 0:09:56.320 --> 0:10:00.600 way that computers process information. They don't understand the information 0:10:00.640 --> 0:10:04.160 that's coming in. They look at the information, they look 0:10:04.320 --> 0:10:08.160 for the match of that information against what is programmed 0:10:08.160 --> 0:10:12.079 to do, and then they respond with the appropriate response. 0:10:12.559 --> 0:10:16.200 But they don't understand the process. Uh. You would say 0:10:16.240 --> 0:10:18.520 that the same thing is true of the Chinese room 0:10:18.559 --> 0:10:21.800 thought experiment. That if you did a piece of paper 0:10:21.880 --> 0:10:25.880 written in Chinese, and the person inside does not understand Chinese, 0:10:26.320 --> 0:10:28.760 they see the piece of paper, they have the book, 0:10:29.160 --> 0:10:32.439 they write the response, they send it back through from 0:10:32.480 --> 0:10:35.720 an external observer, it would appear that whoever is inside 0:10:35.720 --> 0:10:39.800 the room understands Chinese. But the truth of the matter 0:10:39.920 --> 0:10:43.840 is you don't understand Chinese. You're just following the instructions. Now, 0:10:43.880 --> 0:10:46.239 there have been a lot of responses to this particular 0:10:46.280 --> 0:10:49.200 thought experiment, but that's another episode, so I'm not gonna 0:10:49.200 --> 0:10:53.160 go into it here, but it's a very interesting philosophical notion. 0:10:53.880 --> 0:10:57.160 There was an assistant professor of integrative biology and computer 0:10:57.240 --> 0:11:00.680 science and engineering at Michigan State University named Erin Hints 0:11:00.720 --> 0:11:03.840 who lays out four types of AI uh in an 0:11:03.920 --> 0:11:07.520 article that he wrote for the Conversation dot com. His 0:11:07.640 --> 0:11:10.600 four types of AI start with type one, which are 0:11:10.679 --> 0:11:15.160 reactive machines. These base all operations on the current state 0:11:15.440 --> 0:11:19.720 of any given situation, but it cannot refer back to 0:11:20.080 --> 0:11:24.520 past events. Uh, As Hints points out in this piece 0:11:24.559 --> 0:11:27.760 Deep Blue, When I was mentioning earlier. It falls into 0:11:27.800 --> 0:11:31.720 this type one category. When Deep Blue played chess, it 0:11:31.800 --> 0:11:35.200 wasn't tracking moves. It wasn't saying, all right, well, the 0:11:35.280 --> 0:11:38.640 last five moves, my opponent did such and such, So 0:11:38.760 --> 0:11:42.239 I suspect based on that that I'm learning more about 0:11:42.320 --> 0:11:45.600 his style of play. Deep Blue would look at the 0:11:45.600 --> 0:11:48.400 state of the board, where all the pieces were, and 0:11:48.440 --> 0:11:50.840 it would do this as if it was a the 0:11:50.920 --> 0:11:54.040 first time Deep Blue had ever seen the chessboard. It's 0:11:54.040 --> 0:11:57.200 not referring to the previous moves. It's just looking at 0:11:57.280 --> 0:11:59.920 what is on the board at that moment and then 0:12:00.120 --> 0:12:02.920 starts to evaluate all the potential moves it can make, 0:12:03.320 --> 0:12:06.680 all the potential moves its opponent can make, and then 0:12:06.920 --> 0:12:10.120 chooses a move among that set. The next time it's 0:12:10.160 --> 0:12:12.360 Deep Blues turn, it does it all over again. It 0:12:12.480 --> 0:12:16.280 is not referring to its past experience. It's only looking 0:12:16.320 --> 0:12:19.839 at the current state of the board. So again, it 0:12:20.240 --> 0:12:23.520 can't analyze player behavior leading up to the turn and 0:12:23.520 --> 0:12:26.319 then base the decision off of that. So Deep Blue 0:12:26.400 --> 0:12:29.480 blade chess as if on every single turn it was 0:12:29.559 --> 0:12:33.120 the first time it ever scenes ever I had a 0:12:33.160 --> 0:12:36.559 chance to look at that board. Um Ronnie Brooks and 0:12:36.640 --> 0:12:40.240 AI researcher argued that this is really the only type 0:12:40.280 --> 0:12:43.040 of AI we should try to build, because any other 0:12:43.160 --> 0:12:47.720 AI would require some sort of internalized concept of the world. 0:12:48.080 --> 0:12:52.120 It would need to have some form of representation of 0:12:52.120 --> 0:12:55.880 the world in its for lack of a better word, mind, 0:12:56.559 --> 0:12:58.480 in order for it to be able to react off 0:12:58.520 --> 0:13:01.679 of that and be us. We humans are the people 0:13:01.840 --> 0:13:05.840 programming these machines, we would have to program that representation 0:13:05.880 --> 0:13:09.040 of the world, and no matter how carefully we do, 0:13:09.080 --> 0:13:12.160 that is never going to be as good a representation 0:13:12.320 --> 0:13:14.439 as the world actually is. Right, It's only going to 0:13:14.520 --> 0:13:18.080 be a weak simulation of what we see the world 0:13:18.160 --> 0:13:22.000 to be, and therefore any decisions such a machine makes 0:13:22.200 --> 0:13:26.000 based off of this imperfect representation of the world will 0:13:26.040 --> 0:13:31.000 in turn be imperfect. Moreover, if I were to program 0:13:31.000 --> 0:13:33.680 such a computer, I'm doing so based off my own 0:13:34.440 --> 0:13:37.160 concept of what the world is. But my concept of 0:13:37.200 --> 0:13:40.120 the world is different from someone else's concept of the world. 0:13:40.160 --> 0:13:42.920 Someone who comes from a very different background, with a 0:13:43.000 --> 0:13:47.240 very different set of experiences might have a very very 0:13:47.400 --> 0:13:50.920 radically different perception of what the world is. And if 0:13:50.920 --> 0:13:53.720 I were to design a machine off my perceptions, it 0:13:53.800 --> 0:13:56.280 might behave in a way that is completely alien to 0:13:56.360 --> 0:13:59.080 this other person, perhaps in a way that is harmful 0:13:59.120 --> 0:14:01.520 to this other person, and we'll get more into that later. 0:14:01.880 --> 0:14:05.240 That falls into the realm of bias. So there are 0:14:05.280 --> 0:14:07.560 some who argue that we shouldn't even try to go 0:14:07.640 --> 0:14:11.240 beyond type one because of the potential problems we could encounter. 0:14:11.600 --> 0:14:17.000 Type two. AI possesses some limited ability to remember. This 0:14:17.040 --> 0:14:19.960 is sort of like short term memory for humans, though 0:14:20.000 --> 0:14:23.800 perhaps it's more transient than short term memory. The memories 0:14:23.840 --> 0:14:27.520 never get converted into long term storage for these machines, 0:14:27.840 --> 0:14:31.000 but rather they serve to help a machine make immediate decisions. 0:14:31.520 --> 0:14:35.200 Hence points to self driving cars as an example of 0:14:35.240 --> 0:14:40.240 this type of AI. Hence says that, well, they have 0:14:40.360 --> 0:14:43.040 to identify and monitor other elements on the road that 0:14:43.120 --> 0:14:46.240 are constantly changing, such as other vehicles. They have to 0:14:46.280 --> 0:14:50.240 be able to tell how fast is another vehicle traveling, 0:14:50.640 --> 0:14:53.680 How close to your vehicle is this other vehicle, what 0:14:53.720 --> 0:14:56.960 direction are they traveling in? Uh Like, if you're going 0:14:57.040 --> 0:14:59.680 down the highway and there are other cars on the highway, 0:15:00.400 --> 0:15:03.040 your car needs to know how many there are, where 0:15:03.040 --> 0:15:05.400 they are in relation to you, how fast they're traveling. 0:15:05.640 --> 0:15:10.120 This is all ongoing information. So the alternative would be 0:15:10.160 --> 0:15:12.200 for your AI to look at the world as a 0:15:12.240 --> 0:15:16.680 series of snapshots. Right, But snapshots don't tell you things 0:15:16.760 --> 0:15:20.280 like speed. They would tell you the thing was here, 0:15:20.600 --> 0:15:23.600 now it's here. You could interpret it as speed if 0:15:23.600 --> 0:15:27.040 you knew how long it was between snapshots. But that's 0:15:27.040 --> 0:15:29.960 a lot of unnecessary processing power. It makes more sense 0:15:30.400 --> 0:15:34.640 to design AI that has the ability to at least 0:15:34.680 --> 0:15:41.000 hold information in short term to understand things like velocity. 0:15:41.080 --> 0:15:47.200 So uh, slightly different version of AI, slightly more sophisticated 0:15:47.240 --> 0:15:49.280 than type one. But you don't have to worry about 0:15:49.320 --> 0:15:51.240 other stuff, right, You don't have to worry about anything 0:15:51.240 --> 0:15:56.840 outside of whatever the AI's purposes. So a self driving car, 0:15:56.880 --> 0:15:59.520 for example, doesn't need to know how expensive a jug 0:15:59.640 --> 0:16:01.840 of milk is, right. It doesn't need to know any 0:16:01.880 --> 0:16:04.680 of that. It just needs to know rules of the road, 0:16:04.920 --> 0:16:08.960 needs to know how to identify things that it's places 0:16:09.000 --> 0:16:11.840 where a car can go versus where a car should 0:16:11.840 --> 0:16:16.000 not go. It has to be able to identify pedestrians, uh, 0:16:16.040 --> 0:16:18.560 these sort of things. But outside of that, the car 0:16:18.720 --> 0:16:20.640 doesn't need to worry about it, so it still has 0:16:20.640 --> 0:16:24.280 a fairly narrow set of parameters that it follows. Its worldview, 0:16:24.320 --> 0:16:27.400 in other words, is constrained, so you don't have to 0:16:27.400 --> 0:16:30.560 worry about creating a perfect representation of the world. You 0:16:30.640 --> 0:16:34.000 just have to create as perfect a representation of the 0:16:34.040 --> 0:16:37.560 specific part of the world the AI will inhabit as 0:16:37.600 --> 0:16:40.720 you possibly can to make sure it behaves properly. Type 0:16:40.720 --> 0:16:44.280 three AI should not only have an internal representation of 0:16:44.320 --> 0:16:48.040 the world, but also a concept of other entities that 0:16:48.120 --> 0:16:51.160 are within that world as and it doesn't just note 0:16:51.200 --> 0:16:54.640 the presence of other things within its environment, but recognizes 0:16:54.880 --> 0:16:59.520 which of those things have agency. Humans possess agency. We 0:16:59.600 --> 0:17:04.680 understand our own faculties. We can recognize that others possess 0:17:04.840 --> 0:17:07.879 similar abilities. If you and I were to have a conversation, 0:17:08.680 --> 0:17:11.960 we would do so knowing that the other person possesses 0:17:12.040 --> 0:17:15.160 at least some of the same abilities that we ourselves do. Right, 0:17:15.840 --> 0:17:20.640 so we know like you would know that I have motivations, 0:17:20.720 --> 0:17:23.840 that I have needs and wants. You would know this, 0:17:23.960 --> 0:17:25.720 and I would know the same about you. We might 0:17:25.760 --> 0:17:29.159 not know what they all are, but we recognize that 0:17:29.200 --> 0:17:32.920 the other person has them. A Type three AI would 0:17:32.920 --> 0:17:36.960 be able to recognize this and other entities It would 0:17:37.000 --> 0:17:42.080 not itself necessarily possess needs, wants, anything like that, but 0:17:42.160 --> 0:17:46.080 it would recognize that other entities do have those things, 0:17:46.119 --> 0:17:50.360 So it is not self aware, but is aware of others. 0:17:50.920 --> 0:17:54.080 So if we just assume that we are the only 0:17:54.080 --> 0:17:58.359 ones who possess these faculties, then any conversation we would 0:17:58.359 --> 0:18:01.040 ever have with anyone else would be akin to speaking 0:18:01.119 --> 0:18:05.160 to ourselves, because we would assume other people don't have 0:18:05.200 --> 0:18:08.199 those faculties, they don't possess the intelligence that we have. 0:18:08.760 --> 0:18:11.439 That would mean that every single episode I did of 0:18:11.440 --> 0:18:13.960 tech stuff would essentially just turn into Tacos and Lord 0:18:14.000 --> 0:18:15.480 of the Rings, because that's all I would care to 0:18:15.520 --> 0:18:17.880 talk about. But I assume you guys want to hear 0:18:17.880 --> 0:18:21.239 more than that, Sorry, wants to hear about Tacos and 0:18:21.320 --> 0:18:23.240 Lord of the Rings, or at least one or the other. 0:18:23.280 --> 0:18:25.600 I don't know which. I'm gonna guess Tacos if I 0:18:25.640 --> 0:18:29.280 have to guess. So this comes to the theory of 0:18:29.280 --> 0:18:32.359 the mind, which is also pretty close to what Alan 0:18:32.400 --> 0:18:35.199 Touring was talking about during the Turing Test. If we 0:18:35.200 --> 0:18:38.720 were to create a machine that could reliably mimic a 0:18:38.920 --> 0:18:41.760 human well enough so that your average person couldn't tell 0:18:42.320 --> 0:18:44.800 if the responses it was getting the human was getting 0:18:44.840 --> 0:18:47.520 were from a person or from a computer, You would 0:18:47.560 --> 0:18:50.760 say that computer passes the Turing test, and Touring would say, 0:18:50.880 --> 0:18:54.359 you might as well grant that the computer possesses intelligence, 0:18:54.800 --> 0:18:57.200 because you would do the same thing to another human being. 0:18:57.400 --> 0:18:59.280 Right if you talk to another human being, the human 0:18:59.320 --> 0:19:01.800 talks to you like a human being, you say, oh, 0:19:02.000 --> 0:19:06.200 this person has some of the same basic aspects of 0:19:06.280 --> 0:19:09.359 humanity that I have, like intelligence and motivations and needs 0:19:09.359 --> 0:19:11.919 and wants. Turing said, you might as well extend that 0:19:11.960 --> 0:19:15.560 to computers if they're able to mimic human interaction close 0:19:15.680 --> 0:19:17.199 enough so that you could not tell if it was 0:19:17.200 --> 0:19:20.120 a human or a computer. Even if the computer doesn't 0:19:20.119 --> 0:19:22.840 possess those things, we might as well assume it does, 0:19:23.119 --> 0:19:26.400 because we give that same consideration to other people. However, 0:19:26.960 --> 0:19:30.400 Hints would say that would really only apply to type 0:19:30.400 --> 0:19:34.240 four AI. Those are the types of artificial intelligence that 0:19:34.359 --> 0:19:37.600 have self awareness. This is an AI that not only 0:19:37.640 --> 0:19:40.639 recognizes there are other entities out there, but understands that 0:19:40.720 --> 0:19:45.679 it itself is an entity possessing intelligence, and AI in 0:19:45.760 --> 0:19:49.240 type three would recognize that humans have thoughts, feelings, and motivations, 0:19:49.440 --> 0:19:53.280 but an AI in type four would have those of 0:19:53.320 --> 0:19:55.960 its own. It would have its own motivations, its own needs, 0:19:55.960 --> 0:19:59.560 its own wants, its own loves and hates all the 0:19:59.560 --> 0:20:02.280 way in main not defined it in such terms. It 0:20:02.280 --> 0:20:04.720 would not only be able to recognize motivations, but also 0:20:04.880 --> 0:20:09.840 understand motivations. It could put itself in the place of 0:20:09.880 --> 0:20:14.239 another entity and say, this other entity wants to do 0:20:14.760 --> 0:20:19.960 X because entity is why. So it might be Jonathan 0:20:20.040 --> 0:20:23.639 wants to kick open the door because he's hungry for tacos, 0:20:24.200 --> 0:20:26.520 and the computer would be able to understand this concept, 0:20:26.760 --> 0:20:29.320 although it may not ever want to eat a taco. 0:20:30.480 --> 0:20:35.120 I consider that imperfect machine. So that is where we are. 0:20:35.240 --> 0:20:38.120 That's the definitions of artificial intelligence, so that you kind 0:20:38.119 --> 0:20:42.600 of understand, uh, the philosophical approach to what we consider 0:20:42.680 --> 0:20:45.680 a I. When we come back, we'll talk more about 0:20:45.720 --> 0:20:49.120 the actual report and what the two researchers found as 0:20:49.160 --> 0:20:51.680 they were looking into the advances that have been made 0:20:51.680 --> 0:20:54.560 in AI over the past twelve months. But first let's 0:20:54.560 --> 0:21:04.760 take a quick break and thank our sponsor. Much of 0:21:04.800 --> 0:21:07.440 the work and artificial intelligence that have been following has 0:21:07.480 --> 0:21:10.800 fallen firmly in the type one category I mentioned before 0:21:10.840 --> 0:21:13.680 the break. And that's not to say that the work 0:21:13.760 --> 0:21:17.000 is boring or it's not useful. The technology needed for 0:21:17.040 --> 0:21:20.520 type one AI is incredibly sophisticated, so it involves not 0:21:20.560 --> 0:21:23.119 just developing sensors for a machine to be able to 0:21:23.240 --> 0:21:27.760 observe its environment, but also the various programs algorithms necessary 0:21:27.800 --> 0:21:30.840 to process information in a meaningful way so the machine 0:21:30.880 --> 0:21:33.600 can react in the way that we wanted to react. 0:21:34.040 --> 0:21:38.920 Stuff like image recognition, voice recognition, depths, sen saying, all 0:21:38.960 --> 0:21:41.440 that kind of stuff sort of fall into that category, 0:21:42.240 --> 0:21:45.320 although they can also be incorporated into higher categories of 0:21:45.359 --> 0:21:49.800 artificial intelligence, but they are sort of their building blocks essentially. 0:21:50.320 --> 0:21:53.120 One of the first topics from the report focuses on 0:21:53.760 --> 0:21:57.080 machine learning and a concept called transfer learning, so we 0:21:57.119 --> 0:22:00.200 get to talk about what that means. Machine learning is 0:22:00.240 --> 0:22:04.800 an approach that involves a computer examining data, learning from 0:22:04.880 --> 0:22:07.639 that data, and then using what has been learned for 0:22:07.760 --> 0:22:12.640 future decisions. So, for example, let's take Amazon's shopping suggestions. 0:22:12.680 --> 0:22:15.880 When you buy something off Amazon, you'll see a recommendation 0:22:15.960 --> 0:22:18.400 for stuff that other people have bought when they were 0:22:18.440 --> 0:22:21.600 purchasing the same thing you just bought. So Amazon is 0:22:21.680 --> 0:22:24.679 using machine learning to try and up sell you more stuff. 0:22:24.720 --> 0:22:26.879 It's like saying, hey, when other people bought that thing 0:22:26.920 --> 0:22:29.760 of a jig, they also gotta do hicky. You probably 0:22:29.800 --> 0:22:31.720 also want to do hicky because you just buy that 0:22:31.760 --> 0:22:34.720 thing of a jig. That's a simple example of this. 0:22:35.480 --> 0:22:38.800 Then you have deep learning that's kind of a a 0:22:38.880 --> 0:22:42.080 subset of machine learning. Deep learning is sort of a 0:22:42.119 --> 0:22:45.880 self correcting branch in machine learning. You train a computer 0:22:45.960 --> 0:22:48.600 on sets of data, and occasionally you have to step 0:22:48.640 --> 0:22:51.720 in to make corrections and adjustments to make certain the 0:22:51.720 --> 0:22:54.600 computer is on the right track, that the computer is 0:22:54.640 --> 0:22:59.640 not making bad suggestions, which will happen just because it's 0:22:59.680 --> 0:23:02.439 work from large amounts of data, and sometimes it is 0:23:02.480 --> 0:23:05.679 making choices that to the computer seem logical but to 0:23:05.720 --> 0:23:08.919 an outside observer seem wackadoodle crazy, So you have to 0:23:08.920 --> 0:23:11.760 go in and tweak things. You might train an algorithm, 0:23:11.760 --> 0:23:14.480 for example, to recognize pictures of coffee mugs, and then 0:23:14.480 --> 0:23:17.320 occasionally you have to pop in and you see something 0:23:17.359 --> 0:23:20.520 that isn't a coffee mug that has been mistakenly identified 0:23:20.520 --> 0:23:23.080 as one. You have to tell the computer, no, computer, 0:23:23.200 --> 0:23:25.800 that is not a coffee mug, and then it learns 0:23:25.840 --> 0:23:30.240 from there. Deep learning depends upon artificial neural networks. These 0:23:30.280 --> 0:23:35.119 are neural networks that mimic the way brains process information, 0:23:35.760 --> 0:23:39.840 and they have algorithms that behave like neurons. Each algorithm 0:23:40.000 --> 0:23:44.640 processes some information, then assigns a weight to how correct 0:23:44.680 --> 0:23:48.200 it believes its conclusion to be, like I'm pretty sure 0:23:48.240 --> 0:23:50.920 this is right, all the way down to this might 0:23:50.960 --> 0:23:53.320