WEBVTT - TechStuff Rerun: Could we make a sarcastic supercomputer? 0:00:04.400 --> 0:00:12.600 Welcome to tex Stuff production from I Heart Radio. Hey there, 0:00:12.640 --> 0:00:16.200 and welcome to tech Stuff. I'm your host, Jonathan Strickland. 0:00:16.239 --> 0:00:18.400 I'm an executive producer with I Heart Radio and a 0:00:18.480 --> 0:00:21.520 love of all things tech, and I'm going to bring 0:00:21.560 --> 0:00:24.800 you guys a little bit of a rerun today. I 0:00:24.840 --> 0:00:27.960 am trying to get some stuff put together for a 0:00:28.040 --> 0:00:32.199 special series of episodes as well as prepare for some 0:00:32.280 --> 0:00:35.199 other stuff. So great things right around the corner. I 0:00:35.200 --> 0:00:37.600 did not want to leave you without an episode at all, 0:00:37.680 --> 0:00:40.000 So we're going to listen to this one that originally 0:00:40.040 --> 0:00:43.280 published on October eighteenth, two thousand eighteen, and it kind 0:00:43.280 --> 0:00:45.559 of goes in line with some other stuff we've been 0:00:45.560 --> 0:00:49.040 covering in recent episodes of Tech Stuff. This episode was 0:00:49.080 --> 0:00:54.600 titled Could We Make a Sarcastic Supercomputer? And yeah, it 0:00:54.640 --> 0:00:59.920 really dives into the whole concept of artificial intelligence, natural lane, 0:01:00.040 --> 0:01:03.960 which and just kind of understanding the quirks of what 0:01:04.040 --> 0:01:07.120 it is to be human and the whole concept of sarcasm. 0:01:07.360 --> 0:01:10.680 I hope you guys enjoy it. I mean that without 0:01:10.720 --> 0:01:13.280 even a hint of sarcasm. And I'll chat with you 0:01:13.480 --> 0:01:18.200 after the episode. Today. I want to talk to you 0:01:18.319 --> 0:01:21.920 about an interesting topic that I got to explore a 0:01:21.959 --> 0:01:25.760 couple of years ago with Joe McCormick and Lauren fogobaum 0:01:26.200 --> 0:01:30.880 As we debated the possibilities of computers learning how to 0:01:31.280 --> 0:01:36.440 understand sarcasm. We did it for a podcast called Forward Thinking, 0:01:36.760 --> 0:01:38.760 which was around for a couple of years. It was 0:01:38.800 --> 0:01:40.760 a lot of fun to work on that that show 0:01:41.319 --> 0:01:43.840 is over, but I thought I would revisit the topic 0:01:44.200 --> 0:01:47.160 and talk about it for you guys and kind of 0:01:47.200 --> 0:01:50.360 go over what would it take to have a computer 0:01:50.440 --> 0:01:54.360 that could actually understand when someone's being sarcastic. Now to 0:01:54.840 --> 0:01:57.360 understand why this is a big deal, it helps to 0:01:57.440 --> 0:02:01.520 have a refresher course on how computers process information. And 0:02:01.560 --> 0:02:03.960 I know I talked about this a lot, but I 0:02:04.000 --> 0:02:07.000 still think it's important to cover the basics when you 0:02:07.000 --> 0:02:10.160 want to talk about something as advanced as being able 0:02:10.200 --> 0:02:16.200 to detect and understand sarcasm. So computers understand machine code 0:02:16.280 --> 0:02:19.960 or assembly language. This is a language that corresponds with 0:02:20.080 --> 0:02:25.400 the actual physical architecture of the computers. So the way 0:02:25.440 --> 0:02:28.639 the computer is built, that's how this language interacts. It's 0:02:28.680 --> 0:02:32.440 it's essentially how the physical components of the computer are 0:02:32.520 --> 0:02:38.480 able to handle electric current or voltage differences in order 0:02:38.520 --> 0:02:45.600 to process information, and computers can interpret this and execute 0:02:45.720 --> 0:02:49.919 upon this language very quickly. It is the basic language 0:02:49.960 --> 0:02:55.600 of those physical components. However, it is almost impossible for 0:02:55.960 --> 0:02:58.320 humans to work with this, at least on a way 0:02:58.400 --> 0:03:02.720 that is at all of shion, because it ultimately for 0:03:02.919 --> 0:03:08.520 most computers boils down to binary language, right, zeros and ones. 0:03:09.360 --> 0:03:13.079 So you see a huge block of zeros and ones, 0:03:13.120 --> 0:03:15.640 and unless you are neo from the matrix, it means 0:03:15.680 --> 0:03:20.080 nothing to you. So we speak in natural language to 0:03:20.120 --> 0:03:24.120 one another. Natural language, however, is filled with a lot 0:03:24.160 --> 0:03:28.600 of components that make it very very challenging for machines 0:03:28.639 --> 0:03:33.280 to interpret, like ambiguity, or there might be double meanings 0:03:33.320 --> 0:03:36.320 in a phrase and you may mean both meanings at 0:03:36.320 --> 0:03:40.640 the same time, and that is too complicated for most 0:03:40.680 --> 0:03:43.680 machines to be able to process. They just can't deal 0:03:43.760 --> 0:03:47.520 with that. So to bridge the gap between the way 0:03:47.600 --> 0:03:51.840 we humans communicate and the way that computers process language, 0:03:52.280 --> 0:03:58.000 we have created programming languages and compilers. Now, programming languages 0:03:58.080 --> 0:04:02.120 fall into two broad category worries. It's more like a spectrum, 0:04:02.440 --> 0:04:04.840 and you could be further on one end than the other, 0:04:05.280 --> 0:04:08.920 and we typically call them high level programming languages and 0:04:09.000 --> 0:04:13.920 low level programming languages. The lower the level of programming language, 0:04:14.000 --> 0:04:17.719 the closer it is to machine code, and the easier 0:04:17.760 --> 0:04:20.800 it is for a computer to understand, but the harder 0:04:20.839 --> 0:04:22.880 it is to work with. If you happen to be, 0:04:22.960 --> 0:04:27.240 you know, a human being. High level programming languages are 0:04:27.320 --> 0:04:30.640 easier for humans to understand. Now, if you have never 0:04:30.720 --> 0:04:33.919 taken any courses in programming and you're looking at a 0:04:34.000 --> 0:04:38.039 page of code, it could seem indecipherable to you. It 0:04:38.200 --> 0:04:44.160 is just meaningless strings of characters. But once you learn 0:04:44.200 --> 0:04:49.200 the rules of that programming language, how you construct an instruction, 0:04:49.640 --> 0:04:51.720 and a series of instructions, how you go from one 0:04:51.720 --> 0:04:55.080 instruction to the next. Once you understand the rules, it 0:04:55.120 --> 0:04:58.159 actually becomes quite easy to use in the grand scheme 0:04:58.160 --> 0:05:00.640 of things, much more easy than machine which would be. 0:05:01.720 --> 0:05:04.680 But again, the problem here is that computers don't understand 0:05:04.720 --> 0:05:09.880 programming languages, not natively. Even though this is not exactly 0:05:09.920 --> 0:05:12.320 the same as human natural language, it's also not the 0:05:12.400 --> 0:05:15.920 same as machine language. That's why you need compilers. A 0:05:15.920 --> 0:05:21.840 compiler is essentially a translator. It takes this high level 0:05:21.920 --> 0:05:26.240 programming language or higher level anyway and then converts it 0:05:26.320 --> 0:05:29.440 into a machine readable language for the computer to actually 0:05:29.440 --> 0:05:32.680 execute upon. And this is all in the design of 0:05:32.680 --> 0:05:37.320 the programming languages and the compilers. So this is the 0:05:37.400 --> 0:05:41.960 way that for decades we have interacted with computers, when 0:05:41.960 --> 0:05:43.960 you're talking about it on a on a direct level, 0:05:44.000 --> 0:05:48.080 not just executing a program, but creating code, creating programs 0:05:48.080 --> 0:05:53.160 for computers to run. Over the last few decades, we've 0:05:53.200 --> 0:05:58.880 had some very very smart people working on natural language 0:05:58.880 --> 0:06:05.760 systems for machines, which would allow a computer to interpret 0:06:06.360 --> 0:06:09.880 natural language in a way that would make some sort 0:06:09.920 --> 0:06:11.640 of sense, and for the computer to be able to 0:06:11.680 --> 0:06:15.280 act upon that language. And we've seen this in plenty 0:06:15.320 --> 0:06:20.640 of examples recently. Most smartphones have some sort of smart assistant. 0:06:21.120 --> 0:06:25.279 You have standalone products like Amazon's Echo, you have Google Home, 0:06:25.400 --> 0:06:30.560 You've got tons of devices that can interact with people. 0:06:30.839 --> 0:06:35.600 It can be activated by typically an alert phrase, which 0:06:35.600 --> 0:06:37.279 I'm not going to say because I don't want any 0:06:37.320 --> 0:06:39.120 of you guys to have to deal with that. I 0:06:39.200 --> 0:06:41.559 know how irritating it is when I'm watching a video 0:06:41.680 --> 0:06:48.120 and someone activates their specific system and then mine begins 0:06:48.160 --> 0:06:50.200 to respond, and all my lights started going on and 0:06:50.279 --> 0:06:53.760 off because the people on YouTube we're talking funny. I 0:06:53.800 --> 0:06:56.640 know how irritating that is, but use that it activates 0:06:56.920 --> 0:07:00.839 and then you can speak. And typically you can say 0:07:00.880 --> 0:07:06.279 the same thing several different ways and the device appears 0:07:06.279 --> 0:07:09.240 to understand you no matter how you word it. And 0:07:09.320 --> 0:07:11.960 this is a real challenge because we human beings can 0:07:12.000 --> 0:07:14.760 find lots of different ways to say the same thing. 0:07:15.120 --> 0:07:17.800 For example, if I say what is the weather today, 0:07:18.640 --> 0:07:20.640 it could be very similar to if I if I 0:07:20.680 --> 0:07:24.040 ask a question is it going to rain today? Both 0:07:24.080 --> 0:07:27.280 of those are asking for information about the weather, but 0:07:27.400 --> 0:07:31.160 are very different ways of saying that. A good natural 0:07:31.280 --> 0:07:34.560 language recognition program will be able to parse that information 0:07:35.320 --> 0:07:40.760 and then return the appropriate response. This is not an 0:07:40.760 --> 0:07:45.040 easy thing to do. Typically it involves creating a neural 0:07:45.080 --> 0:07:49.000 network structure, and I've talked about artificial neural networks recently. 0:07:49.640 --> 0:07:56.520 That's a typically a network that can accept multiple binary inputs, 0:07:56.560 --> 0:07:59.600 so either a zero or a one input that represents 0:07:59.680 --> 0:08:03.760 some thing uh, some sort of yes, no or on 0:08:03.920 --> 0:08:09.000 off kind of feature. It can accept multiple multiple inputs 0:08:09.160 --> 0:08:12.360 of that nature, so multiple zeros or ones that all 0:08:12.600 --> 0:08:15.520 factor into making a decision, and then it has a 0:08:15.560 --> 0:08:19.800 waiting for each of those components, and then it produces 0:08:19.880 --> 0:08:22.960 a single output that's also binary in nature, either is 0:08:23.040 --> 0:08:26.320 zero one, and it passes that on to other artificial 0:08:26.360 --> 0:08:30.040 neurons further down the chain. Sometimes that will come back 0:08:30.080 --> 0:08:33.760 around and you have a recursive artificial neural network. The 0:08:33.880 --> 0:08:40.080 goal here is for this process two ultimately result in 0:08:40.640 --> 0:08:46.000 a response that is reasonably certain to meet the requirements 0:08:46.040 --> 0:08:49.199 of the person asking the question. This tends to be 0:08:49.720 --> 0:08:53.240 talked about in the realm of probabilities. We we talked 0:08:53.240 --> 0:08:56.680 about how certain the machine is that the response is 0:08:56.720 --> 0:09:00.440 the appropriate one, and if it falls below a certain toushold, 0:09:01.120 --> 0:09:04.079 then the machine would typically respond with I'm sorry, I 0:09:04.160 --> 0:09:06.600 don't know what you're asking for, or something similar to that. 0:09:08.000 --> 0:09:10.520 There are cases where you just get misinterpreted and you'll 0:09:10.520 --> 0:09:13.240 get a response that does not reflect whatever you ask. 0:09:13.400 --> 0:09:16.080 That's a little different. That's where the machine has drawn 0:09:16.120 --> 0:09:19.320 a conclusion, has been reasonably certain that it came to 0:09:19.320 --> 0:09:21.320 the right conclusion. It turns out it was wrong the 0:09:21.320 --> 0:09:26.000 whole way, but that's the process. Now, when it comes 0:09:26.040 --> 0:09:34.000 to sarcasm, that adds yet another layer of difficulty, because 0:09:34.400 --> 0:09:38.360 now a machine isn't just parsing what you are saying. 0:09:38.760 --> 0:09:42.720 It has to understand what you mean, the meaning of 0:09:42.760 --> 0:09:47.679 your words and the meaning of the way you deliver them. 0:09:47.720 --> 0:09:50.320 It could be different. So if I were to just 0:09:50.480 --> 0:09:55.480 write out a phrase with no tone, nobody language, uh, 0:09:55.840 --> 0:10:00.200 not emphasizing any one word over another, it might be 0:10:00.320 --> 0:10:04.520 very difficult to detect what my intent was. It may 0:10:04.559 --> 0:10:07.760 seem like I'm being sincere, when in fact I'm being insincere. 0:10:08.080 --> 0:10:12.520 For example, Uh, if I were to say that guy 0:10:12.600 --> 0:10:18.240 is super tall, but I'm being sarcastic, then just in 0:10:18.280 --> 0:10:21.640 that phrase the way I write it out, you would think, oh, well, 0:10:21.679 --> 0:10:26.160 that person he's looking at must be super tall. How 0:10:26.200 --> 0:10:30.319 do you recognize sarcasm? How can you detect that this 0:10:30.400 --> 0:10:33.520 is in place and then understand what the meaning underneath 0:10:33.520 --> 0:10:37.960 it is. One of the approaches that has been put 0:10:38.000 --> 0:10:44.720 forward relates to IBM's Watson platform. Now. Watson first made 0:10:44.720 --> 0:10:48.680 headlines back when it was a contestant on Jeopardy. It 0:10:48.960 --> 0:10:53.120 went up against two former champions, including Ken Jennings, who 0:10:53.200 --> 0:10:56.440 shows up on a house Stuff Works podcast. Anyway, Watson 0:10:56.480 --> 0:11:00.120 went up against these two former champions, and it is 0:11:00.160 --> 0:11:03.400 able to interpret natural language. It had to in order 0:11:03.400 --> 0:11:05.400 to play the game of Jeopardy, And for those who 0:11:05.440 --> 0:11:08.120 do not know what jeopardy is or they're not familiar 0:11:08.160 --> 0:11:11.319 with the game show, Jeopardy is a game where you 0:11:11.360 --> 0:11:17.320 are presented with categories of trivia and each category has 0:11:17.400 --> 0:11:23.920 multiple uh questions or multiple entries in it, and they 0:11:24.040 --> 0:11:29.600 range in dollar value, and the lower dollar value ones 0:11:29.640 --> 0:11:33.240 are easier to answer than the higher dollar value ones, 0:11:34.320 --> 0:11:37.920 and UH, you're Typically the way jeopardy works is that 0:11:37.960 --> 0:11:40.840 you're you're given quote unquote the answer and you have 0:11:40.920 --> 0:11:46.199 to provide the question. So, uh, if the answer were 0:11:47.600 --> 0:11:53.640 this film that detailed the adventures of a young playwright 0:11:53.880 --> 0:11:57.560 in sixteenth century England, one best picture, you would say, 0:11:57.880 --> 0:12:02.160 what was Shakespeare in Love? So this computer is playing 0:12:02.160 --> 0:12:04.480 against these two former champions. This was sort of an 0:12:04.520 --> 0:12:09.800 exhibition series of games. It wasn't meant for, uh, a 0:12:09.840 --> 0:12:12.600 competition in the way the typical Jeopardy games were there 0:12:12.640 --> 0:12:16.440 was money on the line. Was an exhibition and Watson won. 0:12:16.880 --> 0:12:19.480 It beat both of the champions, and it did what 0:12:19.559 --> 0:12:23.160 I was telling you. It It would analyze the clue 0:12:23.600 --> 0:12:27.200 that was given, the answer that was given, It would 0:12:27.240 --> 0:12:30.199 try and generate a question to correspond with that answer, 0:12:30.559 --> 0:12:33.720 and only if the question met a certain threshold of 0:12:33.760 --> 0:12:37.120 confidence with Watson buzz in. If it did not meet 0:12:37.160 --> 0:12:41.280 that level of confidence, Watson would remain quiet. And most importantly, 0:12:41.520 --> 0:12:44.160 Watson was not at all connected to the Internet. All 0:12:44.240 --> 0:12:49.880 the information was contained within a massive series of servers 0:12:50.760 --> 0:12:53.320 more than gosh I can't even remember. There's a ton 0:12:53.400 --> 0:12:58.680 of processors attached to it. Um so a very powerful machine, 0:12:59.720 --> 0:13:05.880 but it still wasn't exactly able to detect sarcasm. It 0:13:05.960 --> 0:13:10.240 could work with word play and it could work with riddles, 0:13:10.280 --> 0:13:13.200 so that was really impressive. But what it really did 0:13:13.240 --> 0:13:15.800 was it gave IBM the opportunity to say, we have 0:13:15.960 --> 0:13:20.600 this platform here and we're welcoming developers to create applications 0:13:20.600 --> 0:13:24.360 that tap into this platform and make use of this 0:13:25.120 --> 0:13:28.800 in order to do interesting stuff with it. And IBM 0:13:28.960 --> 0:13:31.559 was largely working with the medical industry at that point 0:13:31.600 --> 0:13:37.839 to try and help doctors treat and diagnose patients, and 0:13:37.920 --> 0:13:40.000 it was sort of computer guidance. It wasn't that you 0:13:40.040 --> 0:13:44.200 had an automatic doctor, but rather the doctor had what 0:13:44.559 --> 0:13:49.720 equates to a medical expert to confer with when trying 0:13:49.760 --> 0:13:53.000 to determine why's the best course of action for a patient. 0:13:54.000 --> 0:13:58.000 IBM put up an application Program Interface or API, and 0:13:58.080 --> 0:14:02.720 let developers create their own cognitive computing applications built on 0:14:02.840 --> 0:14:08.000 top of Watson. One of those was called the tone analyzer. 0:14:08.240 --> 0:14:11.439 It still exists back when we were doing this episode 0:14:11.440 --> 0:14:15.679 for Forward Thinking. It was in the form of analyzing 0:14:15.720 --> 0:14:18.720 some text and telling you whether or not that text 0:14:18.720 --> 0:14:23.760 would come across as agreeable or argumentative, or positive or negative, 0:14:24.240 --> 0:14:28.880 and it would assign tone to those pieces. I'll explain 0:14:28.960 --> 0:14:32.160 more about how it did and what it did in 0:14:32.240 --> 0:14:34.480 just a minute, but first let's take a quick break 0:14:34.680 --> 0:14:46.040 to thank our sponsor. So how did this tone analyzer work. 0:14:46.520 --> 0:14:52.960 It would search for cues in any written text, social cues, 0:14:53.040 --> 0:14:56.960 written cues, emotional cues in order to determine the overall 0:14:57.040 --> 0:14:59.960 tone of a piece, which actually meant that the analy 0:15:00.040 --> 0:15:05.200 lizer would tag individual words within a text, words that 0:15:05.280 --> 0:15:11.000 it recognized and had already pre labeled as falling into 0:15:11.080 --> 0:15:14.560 various categories. So words that might have a positive meaning 0:15:14.960 --> 0:15:20.600 like happy, glad, joy, things like that, those would get 0:15:20.680 --> 0:15:24.600 tagged as cheerful. But then it would then assign all 0:15:24.640 --> 0:15:28.560 the individual words tags and then tally everything up. So 0:15:28.840 --> 0:15:31.000 let's say you've got a bunch of sentences and it 0:15:31.080 --> 0:15:36.960 starts individually labeling certain words as being cheerful or sad, 0:15:37.120 --> 0:15:41.000 or angry or helpful, and then it adds it all 0:15:41.080 --> 0:15:43.520 up and then would give you a percentage. So a 0:15:43.560 --> 0:15:49.520 message might be agreeable or thirty conscientious, you would actually 0:15:49.560 --> 0:15:52.640 get multiples of these, and that would just really indicate 0:15:52.680 --> 0:15:58.240 the density of those types of words within the message itself. Now, 0:15:58.920 --> 0:16:02.000 in an ideal world, if language were very simple to 0:16:02.800 --> 0:16:07.160 understand and interpret by machines, this would help you gauge 0:16:07.360 --> 0:16:11.200 how people would respond to your work. Right, So, you 0:16:11.200 --> 0:16:15.240 could write a message. Before you send it, you put 0:16:15.280 --> 0:16:18.240 it through the tone analyzer and it tells you what 0:16:18.440 --> 0:16:22.480 sort of a tone you are setting. So if you 0:16:22.520 --> 0:16:25.680 wanted to create a business letter, you could send it 0:16:25.680 --> 0:16:27.560 through this tone analyzer and if it came back as 0:16:27.600 --> 0:16:32.200 saying it's coming across as as a indecisive, you might 0:16:32.240 --> 0:16:35.360 want to go back in and edit that message so 0:16:35.400 --> 0:16:40.200 that you can make a more straightforward and uh decisive 0:16:40.400 --> 0:16:43.800 message and not give the wrong impression before you send 0:16:43.880 --> 0:16:47.160 the message out to your actual human recipient and come 0:16:47.240 --> 0:16:49.760 up with alternate word choices in order to make sure 0:16:49.800 --> 0:16:52.000 that your message is received the way you intended it 0:16:52.440 --> 0:16:55.240 and anyone who has communicated over the Internet can think 0:16:55.320 --> 0:16:58.040 of ways that this might have been helpful in the past, 0:16:58.160 --> 0:17:03.600 because again, language depends on so many different elements to 0:17:03.760 --> 0:17:06.840 get your meaning across, and when you reduce it to 0:17:07.920 --> 0:17:11.040 the written form, especially the written form online, where we 0:17:11.119 --> 0:17:15.639 tend to be very short with our our communication, it 0:17:15.840 --> 0:17:19.200 comes in very quick bursts, a couple of sentences here 0:17:19.280 --> 0:17:22.400 or there. We lack all that body language, we lack 0:17:22.480 --> 0:17:25.720 that tone. It's very easy to misinterpret. I'm sure there's 0:17:25.760 --> 0:17:28.880 been an example in your life where either you got 0:17:28.880 --> 0:17:31.520 offended from receiving something that was meant in a way 0:17:31.760 --> 0:17:34.520 that was different from the way you you interpreted it, 0:17:34.640 --> 0:17:37.080 or the reverse happened where you sent a message and 0:17:37.160 --> 0:17:41.440 somebody had a reaction you did not anticipate because they 0:17:41.520 --> 0:17:44.400 could not tell what tone you were using just from 0:17:44.440 --> 0:17:48.119 the words you were using. Machines have that same problem. 0:17:48.359 --> 0:17:51.879 In the future, an analyzer like this tone analyzer, it 0:17:51.920 --> 0:17:56.560 could be incorporated into word processors, or email servers, or 0:17:56.920 --> 0:18:00.520 email services, i should say, or social media platform. So 0:18:00.560 --> 0:18:03.320 you start typing in your message and before you hit 0:18:03.760 --> 0:18:07.320 published or post or send, you could analyze that text. 0:18:07.840 --> 0:18:09.719 It could tell you what the tone is and then 0:18:09.760 --> 0:18:12.600 you could say, oh, no, that's gonna come across totally 0:18:12.760 --> 0:18:15.000 the wrong way, and you could actually fix it before 0:18:15.080 --> 0:18:17.160 you posted it or sent it, and then you wouldn't 0:18:17.200 --> 0:18:20.840 have that awkward decision of whether or not to edit something, or, 0:18:20.840 --> 0:18:23.800 in the case of Twitter, which continues to refuse to 0:18:23.840 --> 0:18:27.080 allow you to edit tweets, to delete a tweet. I 0:18:27.160 --> 0:18:30.119 deleted a tweet the other day when I posted a 0:18:30.200 --> 0:18:32.840 link to a news story, and I had done a 0:18:32.920 --> 0:18:36.240 rookie mistake, one that I try to avoid, but I 0:18:36.800 --> 0:18:39.960 did it this pastime, which is that I didn't think 0:18:40.000 --> 0:18:42.199 to look at the date when the news item had 0:18:42.240 --> 0:18:45.400 been published, and had been published a full year earlier, 0:18:45.760 --> 0:18:48.080 so it was not new news, it was old news. 0:18:48.600 --> 0:18:51.399 And uh then deleted the tweet and it wasn't up 0:18:51.440 --> 0:18:53.680 for long, but I still felt dumb about it. It 0:18:53.680 --> 0:18:55.399 would have been nice to have been able to check that. 0:18:55.600 --> 0:18:58.280 Although that's not tone obviously, that's but similar in the 0:18:59.000 --> 0:19:02.360 and the idea that you want to check before you 0:19:03.080 --> 0:19:06.400 end up offending someone, unless you're one of those jerk 0:19:06.480 --> 0:19:08.880 faces that just sets out to offend people, in which 0:19:08.880 --> 0:19:12.760 case rethink your strategy. There are better things to do. 0:19:12.920 --> 0:19:15.080 It's just as you can make just as big an 0:19:15.080 --> 0:19:17.520 impact being a positive person as you can being a 0:19:17.560 --> 0:19:20.000 jerk face. I know it can seem like it's more work, 0:19:20.080 --> 0:19:22.359 but it's also more rewarding in the long run. Okay, 0:19:22.400 --> 0:19:25.720 soapbox done. So. There is a demo of the tone 0:19:25.760 --> 0:19:30.119 analyzer that's available online, and back when we were recording 0:19:30.400 --> 0:19:33.840 Forward Thinking, the demo worked in a way where it 0:19:33.840 --> 0:19:38.080 would tell you about emotional tone and break it down 0:19:38.080 --> 0:19:41.120 by percentage. It's a little different now, but I want 0:19:41.119 --> 0:19:45.000 to tell you the what words and the results we 0:19:45.080 --> 0:19:48.680 got in the past because they were so much fun. 0:19:49.280 --> 0:19:51.359 Granted you would get a different result now because the 0:19:51.359 --> 0:19:55.399 tone analyzer has been tweaked since we recorded that episode. So, 0:19:55.440 --> 0:19:58.760 when we recorded that episode, one of my co hosts 0:19:59.119 --> 0:20:02.600 decided to put a a sentence that is somewhat known 0:20:02.720 --> 0:20:06.000 in literary circles into this tone analyzer and find out 0:20:06.000 --> 0:20:09.280 what it said. And the sentence used was it is 0:20:09.320 --> 0:20:13.040 a truth universally acknowledged that a single man in possession 0:20:13.040 --> 0:20:16.680 of a good fortune must be in want of a wife. Now, 0:20:16.680 --> 0:20:21.879 the analyzer said that this emotional tone was cheerful, the 0:20:21.960 --> 0:20:26.720 social tone was seventy six percent open and fifty one agreeable, 0:20:26.960 --> 0:20:31.360 and the writing tone was analytical. You can also view 0:20:31.400 --> 0:20:34.200 the sentence in terms of word count as opposed to 0:20:34.240 --> 0:20:37.359 the weighted value of individual words, and using that view, 0:20:37.440 --> 0:20:41.720 five percent of the sentence sentences were in an emotional tone, 0:20:41.760 --> 0:20:44.399 eighty nine percent in a social tone, and five percent 0:20:44.440 --> 0:20:48.159 in a writing tone. Now, the analyzer highlights each word 0:20:48.200 --> 0:20:52.600 according to how it classifies them, so emotional words would 0:20:52.600 --> 0:20:55.879 be highlighted in red or pink in that older version 0:20:55.880 --> 0:20:58.919 of the Tone Analyzer, social words would show up in blue, 0:20:59.359 --> 0:21:02.960 and writing tones would be in green. And you could 0:21:02.960 --> 0:21:05.800 click on any word and the analyzer would offer alternative 0:21:05.880 --> 0:21:08.320 words that you might want to use and classify those 0:21:08.320 --> 0:21:11.520 words in the tones that they are associated with, so 0:21:11.560 --> 0:21:13.520 that you could shape your message to meet the tone 0:21:13.520 --> 0:21:17.679 you wish to convey. Also, the Tone Analyzer demo used 0:21:17.840 --> 0:21:22.560 the business letter format as the means of comparison, so, 0:21:23.040 --> 0:21:27.920 in other words, we compared Jane Austen to a business letter. Presumably, 0:21:27.960 --> 0:21:30.680 if you were to use a full version of the analyzer, 0:21:30.760 --> 0:21:33.360 not just the demo version, you would have other options 0:21:33.359 --> 0:21:37.520 so you could compare it with other models, not just 0:21:37.560 --> 0:21:42.960 a business letter. Joe McCormick. He included an excerpt from 0:21:43.560 --> 0:21:47.840 Dostoyevsky's Notes from Underground. That excerpt was, I could not 0:21:47.960 --> 0:21:51.960 become anything, neither good nor bad, neither a scoundrel nor 0:21:52.000 --> 0:21:55.520 an honest man, neither a hero nor an insect. And 0:21:55.560 --> 0:21:58.400 now I am eking out my days in my corner, 0:21:58.760 --> 0:22:03.480 taunting myself with the bitter and entirely useless constellation that 0:22:03.600 --> 0:22:07.959 an intelligent man cannot seriously become anything, that only a 0:22:08.000 --> 0:22:12.320 fool can become something. The feedback was that the emotional 0:22:12.359 --> 0:22:19.120 tone had anger at cheerfulness at so happy anger negative at. 0:22:20.119 --> 0:22:25.840 The social tone was agreeable zero percent conscientious, zero percent open. 0:22:26.320 --> 0:22:31.320 The writing tone was analytical, zero percent confident and tentative. 0:22:31.920 --> 0:22:34.479 Joe would actually end up highlighting some of the words 0:22:34.960 --> 0:22:37.960 to find out which words were the ones that ended 0:22:38.040 --> 0:22:44.080 up giving that cheerfulness result. Those four words were good, honest, hero, 0:22:44.320 --> 0:22:51.200 and intelligent and that kind of are that That's important 0:22:51.440 --> 0:22:55.600 because those words, the way they are used, uh in 0:22:55.640 --> 0:22:59.800 that passage are not used in a positive sense. They 0:23:00.000 --> 0:23:05.159 are positive words, but they're meant to show kind of 0:23:05.200 --> 0:23:11.440 a negation there not and not an assertion. So that 0:23:11.520 --> 0:23:14.880 really highlights a big problem in this tone analyzer, which 0:23:14.920 --> 0:23:20.879 is that it's tagging these words individually without context. So 0:23:20.920 --> 0:23:24.840 if I wrote the phrase I am not glad, it 0:23:24.880 --> 0:23:27.680 would tag the word glad and say that's a cheerful word. 0:23:28.359 --> 0:23:32.040 But I said I am not glad. You if I 0:23:32.119 --> 0:23:35.120 told you I am not glad, you would not think, oh, well, 0:23:35.119 --> 0:23:37.080 that's a cheerful thing to say or a positive thing 0:23:37.119 --> 0:23:40.720 to say. But according to the tone analyzer, it would 0:23:40.760 --> 0:23:44.040 come across as a cheerful statement because it had tagged 0:23:44.080 --> 0:23:46.280 that word as as being cheerful. In the other words 0:23:46.520 --> 0:23:50.040 are not that strong, They don't They don't warrant being 0:23:50.040 --> 0:23:54.439 tagged in a way like that. Now, over time, we 0:23:54.520 --> 0:23:57.520 might have a tone analyzer that can actually take context 0:23:57.760 --> 0:24:02.040 into account, and then you would learn a lot more 0:24:02.080 --> 0:24:05.840 about the actual meaning behind a phrase. It would be 0:24:05.880 --> 0:24:08.679 more than just tone. So if you were trying to 0:24:08.680 --> 0:24:13.960 get across tone by using more complicated and subtle word 0:24:14.040 --> 0:24:19.160 choice where you're sort of being kind of uh poetic 0:24:19.560 --> 0:24:23.680 in your expression, you're trying to get across a feeling 0:24:24.160 --> 0:24:29.439 by using irony or sarcasm, then a tone analyzer like 0:24:29.480 --> 0:24:31.879 this would totally miss it because it would just be 0:24:31.920 --> 0:24:36.000 counting the hits and not understanding the usage. There the 0:24:36.080 --> 0:24:40.480 hidden meeting the word play, so that is going to 0:24:40.520 --> 0:24:45.760 be a real challenge. So it's kind of another interesting 0:24:45.880 --> 0:24:48.120 use of IBMS Watson. There are a lot of other 0:24:48.119 --> 0:24:50.600 ones that we could talk about, like Chef Watson, which 0:24:50.680 --> 0:24:54.840 was my favorite. Chef Watson would generate new recipes based 0:24:54.920 --> 0:24:57.280 upon ingredients that you would tell it that you had 0:24:57.320 --> 0:25:01.600 on hand, and it wouldn't it wouldn't go and reference 0:25:01.840 --> 0:25:05.520 old recipes and pull one up for you. Instead, it 0:25:05.520 --> 0:25:09.200 would make flavor profiles based upon all the different combinations 0:25:09.240 --> 0:25:11.800 of food that were found in various recipe books and 0:25:11.840 --> 0:25:14.520 generate a brand new recipe for you right there on 0:25:14.560 --> 0:25:18.280 the spot. And sometimes they were whacka doodle crazy, y'all. 0:25:18.760 --> 0:25:21.959 So in a way, you could say that Chef Watson 0:25:22.000 --> 0:25:25.320 was another another way of seeing how I b M 0:25:25.359 --> 0:25:28.119 S Watson has a lot of promise, but it requires 0:25:28.200 --> 0:25:32.240 a ton of work on the app level in order 0:25:32.280 --> 0:25:35.760 to leverage it and make actual practical use out of it. 0:25:36.160 --> 0:25:40.560 I have more to say about computers detecting sarcasm, but 0:25:40.680 --> 0:25:52.960 first let's take a quick word from our sponsor. So 0:25:53.400 --> 0:25:57.720 back in two there were some researchers at the Hebrew 0:25:57.840 --> 0:26:02.760 University in Israel who designed a system called the Semi 0:26:02.800 --> 0:26:10.080 Supervised Algorithm for Sarcasm Identification or SAZI, and they used 0:26:10.160 --> 0:26:15.239 SAZI to analyze collections of nearly six million tweets and 0:26:15.359 --> 0:26:20.320 also around sixty six thousand product reviews from Amazon. They 0:26:20.400 --> 0:26:25.879 wanted to find rich treasure troves of sarcasm that turns 0:26:25.880 --> 0:26:30.159 out reviews and tweets. They fit the bill sarcasm is. 0:26:30.760 --> 0:26:35.600