As Election Day approaches in the Maryland, a candidate for Montgomery County Executive and one for Governor of Maryland join us for the Politics Hour.
All books (and even public radio programs), no matter how unique, adhere to some sort of formula. Phil Parker uses technology to compose original works on a scale far beyond the abilities of any human author or radio host. His computer programs have written hundreds of thousands of non-fiction books, created news programming in obscure languages, and even composed love poems. He joins Kojo to explore the increasingly sophisticated world of automated writing.
- Phil Parker Professor of Marketing and Management Science, INSEAD
MR. KOJO NNAMDIFrom WAMU 88.5 at American University in Washington, welcome to "The Kojo Nnamdi Show," connecting your neighborhood with the world. He sees himself as, well, maybe the Henry Ford of authors. His books, hundreds of thousands of them, are put together in sequence, according to a formula, an assembly line of words all done with computer programs. Using algorithms, he also creates a 24-hour radio news program, publishes the sources in obscure languages and even composes love poems.
MR. KOJO NNAMDIHis goal? A big one, to use computer automation to tackle problems in the developing world that stem from a lack of access to information. He's an economist by training, and he sees a frustrating imbalance. While in English, we enjoy the publication of around a half a million books a year, not to mention having billions of Web pages a click away, but the total number of books published in Laos or Mali barely reaches double digits. And there are hundreds of languages for which there is little incentive for publishers.
MR. KOJO NNAMDISo if a computer can cheaply compile and publish textbooks in local languages for schoolchildren or generate hyperlocal weather forecasts that help farmers, who are we to quibble about authorship? Joining us to discuss all of this is Phil Parker. He's a professor of management science at INSEAD, an international business school based in France, Singapore and Abu Dhabi. Phil Parker, thank you for joining us.
PROF. PHIL PARKERThank you very much, Kojo. Appreciate it.
NNAMDILet's start with poetry. Love poems in particular wouldn't seem to be the best candidate for a computer program, but, in fact, a computer can draw on thousands of years of love poetry. Can you, please, explain?
PARKERSure. Actually, poetry is one of my favorite genres of literature. It's one of the oldest as well, after accounting, on tablets. And some of the first things that humans wrote were small little love songs and poems, lyrics and the like. And as time goes on, different cultures and different places, languages, used different formula for creating poems. The sonnet being a 14-line iambic pentameter with a certain rhyming scheme is very, very formulaic.
NNAMDIWhich I was lousy at in high school...
NNAMDI...even though I thought I was great.
PARKERMe too. Me too. In fact, I think...
NNAMDII thought I was good.
PARKERAnd some of my programs do much better than I ever can. Certainly faster, that's for sure. And because of the formulaic nature, they lend themselves quite well to automation, using the algorithms perhaps that authors would do. And I've been inspired by those authors, of course, to do the poetry we've created.
NNAMDIBut how did you -- and I failed to do well in high school in areas which a computer can do well, say, on a sonnet. How do you teach a computer irony?
PARKERHow do you teach it irony? That's a great question.
PARKERActually, I think the mere fact that it's doing it is the irony, maybe.
PARKERMy favorite -- we have a site where I posted about a million poems. It's -- I'll do a little plug. It's called totopoetry.com, and it's for English-language learning. And if you go to a page called Truth, there's a sonnet there that's formulated on Shakespeare, "Sonnet No. 76." And in that sonnet, Shakespeare complains that his poetry is formulaic, that by reading that poem, you know who the author is, and that he's very averse to innovation.
PARKERSo I created a sonnet program where a computer admits that it can't write a sonnet well, basically. But, in doing so, it delivers it in sonnet form. And so the irony is that it's the computer -- it's metapoetry where the poem is really about the poet or the poem itself. So it's kind of one of these introspective genres of poetry. And so the irony is perhaps that the poem was created about how a computer can't write a poem.
NNAMDIYou call these edge poems not because of their themes but because of the mathematical formula that's used to create them.
PARKERThat's right, heavy reliance on graph theory, which is as old as mathematics almost, in a semantic web. We've recreated human, you know, categorizations, so to speak, in many, many languages. If you think of a graph as a circle with the word love at the middle, you'll notice that a lot of words are synonymous with love, perhaps adoration, affection, liking and zero. Why zero? Because zero is a 40-love in tennis.
PARKERYou'll also know that affection and adoration are linked together. Those links are called edges, and they have a certain value. A mathematician might call it a valence. The stronger the value, the more likely a love poem is going to have those words in it. Now, if the poem is about tennis, the love of tennis, then you might find a pun using the word zero showing up, just as a little tip-off to the reader that there's something going on here to give it the subtlety of what a human might write.
PARKERSo it's using a large semantic web. And because the computer doesn't understand English versus Chinese, the web can be created in any language. It's language independent, so to speak, 'cause it's all mathematical at that point.
NNAMDIWe're talking with Phil Parker. You can see where this conversation is going. If you'd like to join it, you can call us at 800-433-8850. What repetitive or maybe tedious writing tasks do you think a computer could replace? I wouldn't say that writing poetry is a tedious task. But, as Shakespeare pointed out, it can be formulaic. So what formulaic tasks do you think a computer could replace?
NNAMDI800-433-8850, you can go to our website, kojoshow.org, join the conversation there. Send us a tweet, @kojoshow, or email to firstname.lastname@example.org. We're talking with Phil Parker. He's a professor of management science at INSEAD, which is an international business school based in France, Singapore and Abu Dhabi. There's the old line about how a chimp in front of a typewriter would eventually write "Hamlet," but this process really isn't random, is it?
PARKEROh, far from that. That would be horrible, actually. It would take too long for the chimps to come up with anything useful, especially if it was a weather report in a local language in some country. So, yeah, the idea is to avoid having chimps, in fact, or any costly labor that might be involved in the feeding all these chimps, et cetera. So, no, the idea is to look for genres of literature or writing that is going to be -- that's very formulaic but also, at the same point, impactful.
PARKERYou can easily use such authoring engines for nefarious purposes, but what I've attempted to do is find areas where, really, there's no hope that anyone will probably author in a certain area, so try to replace it with automation if it's possible.
NNAMDIWell, following the line of fiction for a second, I'm not sure if...
NNAMDI...you're a fan of, say, the Stieg Larsson books, the Swedish thrillers, "The Girl with the Dragon Tattoo" and the others. But as a challenge, would you consider that kind of genre, a thriller, and the subgenre, dark Scandinavian thriller, to be replicable?
PARKERThe minute someone writes an idiot's guide to doing something, it's probably possible with a computer. So if there's an idiot's guide to writing dark Scandinavian fiction...
PARKERThere's a great famous Canadian company, Harlequin, that has on their website for romance authors actual instructions on how to follow the formula, what needs to happen in chapter one, what needs to happen in chapter two, et cetera. And one of the most fascinating things I've seen in fiction is that, typically, in some other subgenres of fiction, chapter nine, especially in romance novels, would have to have where the heroine and the hero consummate their relationship.
PARKERChapter 10 is where they wonder if it was real or not. In chapter 11, which is an epilogue, it's happily ever after. If you take that chapter nine, the consummation chapter, and put it as chapter one, and it's Charles and Cindy who are those in chapters, and you add one last sentence, where Charles takes out a knife and slices Cindy's neck, and she bleeds to death on the floor, then start chapter one of another romance novel -- keep Charles, but change the heroine -- it now becomes a thriller.
PARKERSo by rearranging content in a certain way, authors use certain conventions. And so if professors of literature speak of conventions and how one writes in a certain genre, they're basically telling you about a formula that can be followed. Truly original genres are probably not formulaic. They're just so different that no one has ever seen anything like it, and that's something a computer would have a very tough time coming up with.
NNAMDISo if I understand this, you look for what is formulaic about any particular format, whether we're talking about a sonnet or a novel. And, in fact, you've found most formats out there to be formulaic. Let's start with something trivial, like radio. What's formulaic about radio programs?
PARKERWell, one of the things we're working on with Farm Radio International, which does radio programming in Africa -- and we do Farmer Voice Radio --is we're finding that a lot of community radio stations simply don't have back production crews to help do some research to prepare a radio host to do an interview. So you can type into a computer program, I'm going to interview a carpenter, I'm going to interview an inventor, I'm going to interview X-person. There's a standard way of approaching those types of interviews once you see enough of those interviews.
PARKERYou have to look and audit many, many interviews, and you start noticing a pattern. It typically starts with the introduction of who the person is. With me today, I have X, Y and Z. This person does this, a small bio, and then it goes on to other things. So what we're developing for these folks, as a part of the project I'm working on, we call them issue packs. It's a short little issue package, usually a piece of paper or two that outlines exactly what might be covered inside.
PARKERSo we're using the automation programs not to replace the host, per se, but to save time because a lot of these rural radio stations simply don't have the staff to prepare notes for the people that are going to be on the air, the talent on the air. We've taken it to a bigger extreme, however, with actual radio reports where we've pre-recorded the voices of radio hosts for every known weather pattern that can possibly happen. We then take a satellite feed, concatenate the little phrases together in the local languages, and then it gets broadcast over the air that way.
NNAMDIYeah. I want to talk about a little bit more about that, and we'll play a clip about that. But I'm glad you said the part about not eliminating the radio host...
NNAMDI...because my next...
NNAMDI...question was not really going to be a question. It was going to be out, out, out...
PARKERNo, have no fear. Where there's talent, there's always a market, you know?
NNAMDI800-433-8850 is the number you can call if you'd like to join the conversation. Here is Elizabeth in Westminster, Md. Elizabeth, you're on the air. Go ahead, please.
ELIZABETHI'm fascinated by this idea, and I'm wondering if your guest has ever asked the computer to try to write a play.
PARKERActually, when we created the fiction genre -- we're not releasing that to the public, by the way, for fear that people might use it for slanderous reasons. But we created a little fiction writing thing. It turns out that many novels with a certain formula can be converted into a screenplay. It can be converted into a movie script. It can be converted into a variety of formats. So the answer is yes.
PARKERWe have played with that. Our first application is writing radio dramas, oddly enough, for a farmer who visits a village and wants to give agricultural tips on improved crops inside -- in a village. And a daughter walks -- a young woman walks up and starts saying: What are you doing on my father's land? And then the father sees him at a distance, and there's a little bit of interesting drama that takes place.
PARKERAnd what we noticed is that that kind of sequence of events can happen across many different crops, many different cultures. What changes is the soil types, the crops involved. And so we have a computer program where it can select the type of crop, the type of -- the country. We'll know the elevation of the country, the longitude and latitude. We'll know what kind of pests and diseases are present in those countries. And it automatically generates those kind of plays, if you would. So it's a radio play as opposed to a theatrical play.
NNAMDIYet another example, Elizabeth, of how art and technology imitates life. I got into radio as a result of being in radio plays for children.
PARKERYeah, it's interesting. Yeah.
NNAMDIElizabeth, thank you so much for your call. You've always loved words, and you were a collector of dictionaries, it's my understanding, since you were a child. What fascinated you about them?
PARKERI think it was more a fact that I was a very, very slow reader. I was diagnosed with a kind of a form of dyslexia, I guess, when I was a kid. And I never really enjoyed reading books because I'd always forget the first paragraph by the time I got to the bottom of the page. And I've only read half dozen books in my life and -- except for I found reference books to be very useful. You can go straight to the subject. You get exactly what you need with not too much effort.
PARKERSo I started -- I have probably a few hundred dictionaries at this point. My favorite of the dictionaries from the 1700s where they weren't very politically correct and they'd say exactly what they had on their mind, you know?
PARKEROats being -- by Samuel Johnson, you know? -- the food of horses and Scotsmen, you know, that kind of thing. So I very much enjoy those, "The Devil's Dictionary" and others. And the more I got involved in dictionary collecting, the more fun I had in it. And so, I guess, that was the driving force, just a fun hobby.
NNAMDIWell, later in the broadcast, we'll talk about how that led to INSEAD. Phil Parker is a professor of management science at INSEAD, which is an international business school based in France, Singapore and Abu Dhabi. He's our guest this hour. We're taking your calls at 800-433-8850. Would you mind if your weather report, sports articles and even novels were computer generated? 800-433-8850. Send us a tweet, @kojoshow, or join the conversation at our website, kojoshow.org. I'm Kojo Nnamdi.
NNAMDIWelcome back. We're talking about when the author is an algorithm. Our guest is Phil Parker. He's a professor of management science at INSEAD, which is an international business that's school based in France, Singapore and Abu Dhabi. We're taking your calls at 800-433-8850. Would you mind if your weather report, sports articles and even novels were computer generated?
NNAMDIWe raise this because radio programs in local languages in developing countries is one of the bigger projects that Phil Parker has been working with. Let's listen to a weather report from Malawi that aired in May of this year.
UNIDENTIFIED WOMAN #1(speaks foreign language)
NNAMDIWeather report in Malawi. This is obviously very useful for farmers in developing countries, for example, and what's especially important is that it's in the local language. What was the response to it?
PARKEROh, that was an interesting thing. I was very humbled by the whole experience. The reactions were mixed at the beginning, in particular because the concepts of certain weather topics were new to the population. They had never heard of weather report in their local language before. They were asking about, what do you mean by visibility? What do you mean by kilometer? What's a degree? Who is Celsius? And all these kind of basic questions.
PARKERSo we -- and we also learned that their scale of temperature was slightly different in the local vernacular, so you have four temperatures: very hot, hot, warm and cool. And part of the finesse of agriculture is often within a degree or two or three for plant germination and the like, so we had to have additional program that explains, you know, what is Celsius -- this Swedish inventor who came up with a thermometer -- and what is visibility? And, you know, very basic concepts. But then, as the time goes on, there's little celebrities forming in some of the communities.
PARKEROne fellow in particular named Phillip has gone down as being the weather man, in essence. And everyone's waiting for Phillip's, you know, latest temperature report, and so people are walking around, talking about 17 degrees hot or cold or whatever. It's kind of interesting, so it takes a while. I think we're at the beginning of a lifecycle curve, if you want, for some of these communities. And as time goes on, they'll learn to calibrate and understand that a forecast isn't as accurate maybe as a crystal ball, so to speak.
NNAMDIWe know that.
PARKERYeah. And we have to be patient with the technology, and it takes a while. But if...
NNAMDIPhillip is a computer generated voice.
PARKERWell, Phillip's the actual announcer.
NNAMDIOh, he is.
PARKERThat person does rip and read. What you just played is -- I'm sorry, which is where he reads a script.
PARKERSo we print the script for him, and he just reads it. What you just played was one where it's not Phillip. This is the automated one where we're using an audio engine. And that -- the response is very good. People actually will say, oh, the voice is so beautiful, et cetera. The -- but I was saying one person in particular has become the local celebrity because of his reading of some of the scripts in the local language.
NNAMDINow, you got to explain how you use this audio engine.
PARKEROh, audio engine is very much similar to Xbox 360, if you play some of these video games, where you have a -- let's say it was in French and -- that's FIFA. FIFA is a sport thing. They'll say, (speaks foreign language). You know, and then he'll name his name, right? The second one will said, (speaks foreign language) or whatever, they say (speaks foreign language), so that would be a pronoun. And then the third one will say, (speaks foreign language) or whatever. What a score. One is a proper noun. One is...
NNAMDISee, when you get confused like I do with English and French and you say (speaks foreign language), we don't think of score. We think of head butt.
PARKEREspecially against Italy.
NNAMDIExactly right. (unintelligible)
PARKERYeah, of course. It's just infamy. So mixing out -- mixing those three types with pronoun, with proper noun, with neither, in a certain order, it actually sounds like natural flowing languages. So we take the satellite feed codes, which are in bits and bytes, we then say the appropriate phrases to be concatenated together would be the following. And we make it so it's not boring by having the same words repeated too often because that doesn't sound very natural. So that's what an audio engine is all about.
NNAMDII like the fact that there's a video that goes along with that weather report that we just heard showing a cloud and a man in a rain hat.
PARKERYeah, yeah. That's for a low literacy environment, so that people -- this is with the Grameen Foundation, where the people in the villages have -- one villager will have a smartphone, and they then can show the weather report to the local villagers. And it'll be in their local language, but it can also show visually what the weather looks like. So they can see that, you know, there's high visibility, so you may have noticed there's a tower with some big binoculars that blink and things like that. And if you -- and if the tower can't see, low visibility, then it's clouded with fog.
NNAMDIHow do you navigate the cultural issues, the cultural sensitivities, when you're creating content in a situation like that? Isn't there always the possibility of offending?
PARKERI imagine there is a possibility of it. We haven't run across that yet, although we did run across people not being familiar with the terms used.
PARKERBut so far so good. I think one thing that helps us -- we work with local partners who themselves know the sensitivities. They wouldn't broadcast something I don't think that would not pass through that filter. Some of the sensitivities that we've discovered are more political than they are cultural in that we're producing forecast for villages that are very, very remote that the national government themselves don't broadcast in the local language.
NNAMDIBecause a lot of times, people in the national government don't understand that language themselves.
PARKERThat can happen. Or they just don't have the budget to do so, and there's just a lack of resources. So these are all generated in San Diego, Calif. and then beamed across the world and then played in the local language, so it's kind of interesting.
NNAMDIYou have created 24-hour radio channels. What else is broadcast?
PARKEROh, that's where we -- for example, let's say you wanted to have a classic radio station in Swahili. The person -- the speaker then introduces the piece. We've fully have bought out the rights for worldwide redistribution of a lot of different genres of music. So it's recreating the music industry, so to speak, in a local language because a lot of the local radio stations don't have the content themselves. So we have a classic radio station. We have a hip-hop station. We've got a jazz station. We've got a classic music station and an African fusion station.
PARKERSo there'll be a little voice that introduces the piece, and then at the end it'll say, you've listened to, you know, Mozart's "Magic Flute." And then it'll be interrupted with a weather program and then maybe an agricultural tip about the season, et cetera -- some public service announcement, things like that.
NNAMDIHere's Ernie in Gaithersburg, Md. Ernie, you're on the air. Go ahead, please.
ERNIEGood afternoon. I'm very curious about Phil's involvement with the Khan Academy videos and how can he distribute those throughout the globe for the lesser-developed countries as well as people?
PARKERYeah, that's a very good question. The Khan Academy is a wonderful medium.
PARKERYes, thank you for the question. And it's a wonderful medium. It's all in English, of course, and it's actually oriented toward what I'd call North American educational systems. There's very little content with respect to subsistence farming and the topics they're interested in, in these rural areas that I'm focusing on right now, in the projects I'm involved in. So, from a basic arithmetic point of view, I think it's a wonderful format at this point.
PARKERWe've actually created some templates and some other publications for basic learning for, like, for example, coloring books for local languages, videos that explain certain local tips that people can use to improve their crop yield and therefore get higher incomes and things like this. So we're really focused on health and education, but -- and education, specifically rural education in Africa, things you needed to learn about crops and pests and diseases.
PARKERThe Khan Academy is a very big project that handles what I'd call a mathematics language learning in sciences for people that needed a brush up and/or a path towards a better education in those subject areas. So we're thinking about working in that area, but, for now, we're mostly focused on agriculture and health at this point as opposed to arithmetic, math and the basic scientist -- sciences.
NNAMDIIf you have called -- and thank you for your call, Ernie -- stay on the line. We will get to your call. I just wanted to pursue this for a second with you, Phil. Talk a little bit about how you got involved in all of these and exactly what INSEAD is and does?
PARKEROK. Well, INSEAD was founded around the time of the Treaty of Rome in the late 1950s as Europe's first business school by former Harvard MBAs who thought Europe needed such education. And it's a -- or MBA student who are trilingual. And they've -- can be in all of our campuses and do work and study in our different campuses. So it's a very multi-cultural organization to begin with.
PARKERAnd my interest in this subject area started, actually, when I was working in Washington, D.C., oddly enough, with a very -- a group of very smart economists for a firm called Nathan Associates. And I worked on some projects in the telecommunications area with a very smart fellow, Andrew Roscoe, and some other people on development projects. And it was very clear early on that a lot of micro-enterprises out there -- for example, in Haiti, I did one project on export promotion from Haiti during the Caribbean Basin Initiative days.
PARKERAnd there were simply no data available to these enterprises on what the export market was, how big the market was, et cetera, because most enterprises make very, very narrow things. They don't make broad things like automobiles. They make very narrow things like copper coils casings or something. And so I was thinking way back then about how can one create such reports and data without hiring hundreds of GWU economists, so to speak...
PARKER...and going bankrupt at the same time. So automation was one avenue that I pursued and finally cracked the dilemma, so to speak.
NNAMDIBecause it's not just weather forecasts that are essential in the developing world. It turns out there's an almost total lack of information on some of the most important issues for a farmer, for example.
PARKERYou know, that's right. It's interesting, when you go to a Barnes & Noble bookstore, you go to the self-help section. You also notice it's a very huge section, very large. You go to France, and the section is much smaller. And you might think, well, it's because Americans are flipped out, and the French are centered. But it turns out there are simply less people who speak French, so the economies of publishing in the French language are not so great. As you go farther and farther, you'll notice that most languages in the world don't have biology books.
PARKERThey don't have basic math books, no algebra books, no algebra content, no video content. There's a content shortage, so to speak. The Internet is just lacking content, and it's not just...
NNAMDIAnd we're sitting around here thinking that there's too much content in the world...
PARKERYeah, yeah, probably in the English language, that might be true. I don't want to pass judgment on anyone's...
PARKER...creativity or their production, but in a lot of the smaller languages, that's definitely the case. And it turns out that people who speak those languages as their primary languages are the lowest income people of the world.
NNAMDII was about to say, because as an economist by training, there's -- it's one thing to notice that there are languages spoken by relatively few people, but when you create resource materials in those languages, because you're an economist by training, you see that there's a basic economics problem here.
PARKERYeah, it's a market failure problem for sure. There's just -- the returns on investment on traditional authorship or radio content creation or whatever the format might be -- mobile phone applications, et cetera -- it's just not profitable to do so. And so we're -- automation is one way, not the only way, of course, but to handle at least the formulated types of information.
NNAMDISo that at least people in those cultures can get information that they can understand in their language to deal with specific problems that they have, having to do with how they're going to earn a living.
PARKERYeah, basically that's...
NNAMDIHere is George in Frederick, Md. George, you're on the air. Go ahead, please.
GEORGEThank you, Kojo. Hi. I think -- I bought an e-book once that was a compilation of articles about a certain medical subject that my daughter was born with. I change (unintelligible) much information about it, so you can't find the book on the subject. And I think I found a book compiled by your guest, but upon going through it, there was something all about it. There was no central stress. There was a quality of shallowness to the like.
GEORGENow, there was no understanding of -- or guidance on the subject, and, I think, years later, I realized that we have been -- (word?) had been compiled by an algorithm, and that's a dark side of this. The algorithm has no real model that we've -- in understanding and it's our understanding of the subject math, so it was pretty disappointing. It was expensive, and it wasn't useful. So that's a danger with -- when we're dealing with algorithms, that they don't really model the complexity of a human thought, that (unintelligible) keywords.
NNAMDIWell, you know, George, one of the questions we have planned on asking Phil, and you gave us the opportunity to ask, is that there are issues when a computer does research without a human editor. There can be, I guess, a problem with errors, a problem with not having a central theme around which George can wrap his head, so to speak.
PARKERThat's an interesting point that he raises there. What's interesting about that genre, it was created about the year 2000, 2002 when we had a request from an organization called OCLC, which is a big distributor to libraries, and was designed for how to research using the Internet, diseases. That's what the genre was. That's what it was created for. And so the genre was to -- where are the official sources of information for medical diseases? It wasn't about the disease itself. It's, where do you research if you want to take on patient research yourself, if you want to do it yourself?
PARKERAnd what's odd about that is we really didn't use computer algorithms to write the text. We had, actually, editors. We hired a staff. Well, all we used the editors for was -- all we used computers for was to create the index in the back of the book and the table of contents formatting and a few things like that. And so we used human-authored materials. Now, it was a compilation of those human-authored materials. In some disease areas, there's simply not a lot available.
PARKERSo it was designed to actually help research librarians inside medical libraries -- there's a few thousand of those -- and distribute it through OCLC. Amazon contacted us and asked us to list these books that were originally designed for research librarians in medical schools to distribute those. And the results are mixed. Some trade associations -- like all books, I suppose some trade associations actually recommended it to their patients. Others do not.
PARKERAnd like all types of books, some are rated well and others less so well, and, you know, that's just the nature of the beast of publishing content, I think, in general.
NNAMDIIf I can pursue this for a second, George, what were you looking for in the book?
GEORGEWell, it doesn't have to be -- I was looking for what you -- whether you want to understand the advances, the latest, the cutting edge researching in a subject for some in the medical field. You expect somebody that edits -- there is some clearance that goes through the articles presented. There is some guidance or some understanding of the added value. This was pretty much a collection like I could assemble of PDF (unintelligible)...
NNAMDIYeah, but I'm trying to figure out what was the -- what was your purpose where -- well, George, I'm trying to figure out. Were you doing this for purposes of...
GEORGE...running a Google search.
NNAMDIWere you doing it for purposes of research yourself?
GEORGENevertheless, you can find that because, like that with a Google search and makes a combination by yourself. And chances are your (unintelligible)...
NNAMDII get the impression that George was not looking for an -- that George was looking for some kind of expository work rather than...
PARKERThere's -- yeah. In the area of health care, there's, like, four or five subgenres. One is the patient testimonial book, which a computer can never write. Another one is the doctor's strong opinion on an area, and they will write something that's very towards their particular point of view. We found one genre which was a lot of things float around the Internet. Which ones can you rely on? Which can you not rely on? And this was an e-book that was sent out to libraries, originally.
PARKERAnd then there was a print-on-demand request for these books, and that's when they got listed on to Amazon for what sites can you rely on. We did a focus group with nurses, and we asked them what are the -- what sources would rely on the most? And during one of those focus groups, we said well, would you recommend the NIH, for example? And one of the nurses said, what's the NIH? And at that point, we said, OK. So we'll just do a simple little guide book on here are the ones that are official.
PARKERThese are the ones that are less official for both alternative medicine and other types of health care treatments. And it's possible that, in George's case, he's already an intensive Internet researcher when, in fact, it was never designed for people like that. It was for the people who were more intimidated by the Internet, so and that -- we haven't re-entered that genre for many years now because people now are very, very comfortable with the Internet. It was created back in the days based on the request from the librarians at that point.
NNAMDIGeorge, thank you very much for your call. And, speaking of people who are not that comfortable with the Internet, here is Jim who is on the Beltway somewhere. Jim, you're on the air. Go ahead, please.
JIMThank you very much for taking my call. I'm driving -- I-95. Phil...
JIM...let's say, next year, Kojo decides at the beginning of each of his show, he'll give a 30-second reporting of the best play the day before in the Washington Nationals' baseball game. So...
JIM….I want to -- I'm very illiterate about computers. I'm in my 60s. I can do email, but I don't know how computers really work. Other than that, they're, you know, zeroes and ones. So let's say, Ryan Zimmerman hits a homerun, I mean a triple, and the guy who was on first, Jesus Flores, gets caught in a rundown between third and home, so only a two-run score. And Kojo gets sick. And so he says, you know, I want to give the listeners that exciting reporting about that play.
JIMHow would that get translated to a robot or a computer? What do you do to get from, A, what happened, through B to C where it gets reported robotically or however without Kojo in the show?
NNAMDIHere's Phil parker, Jim.
PARKERActually, that's rather trivial from a computer programming point of view. Basically, there's a data feed that comes in right off of the sports, you know, networks. And it comes in as basic data about what happens when. And the tricky part is not actually concatenating the right voice of Kojo. We would put him in a studio for a week, and he would basically say, oh, my gosh and need to come up with all these crazy, like, phrases, every possible one that he could ever think of. And those are stored on a hard disk.
PARKERNow, the question is, how do you play them in the right order? That's the question. So the art of it is mathematics, actually. If you look at a zero-zero score, right, in any sport, you can actually say something or, you know, six-zero whatever, somebody got blanked or got wiped out. But the adjective you're going to use depends on that score. If it's zero-one, you could say a highly defensive game, whatever. If it was -- this is true for football or anything.
PARKERAnd if it goes to 0-12, if it goes at 0-50 in a baseball game, he's going to use quite different phrases. So what we do is we subcategorize every possible outcome of every possible play. We have him pre-record every possible player's name in a certain order, very much like Xbox 360. I'd encourage anyone to just pick up a, you know, Madden or ESPN football and -- or baseball or soccer and just play the games and just listen the way it's done. So we'd have Kojo sit in the studio.
PARKERHis voice would be permanently recorded forever, and he'd be on the air until humanity, you know, is extinguished, basically. So he would always be calling baseball games or football games.
NNAMDIJim, thank you very much for your call. And we're going to take a short break. When we come back, we'll continue this conversation. You can still call us though, 800-433-8850. What repetitive or tedious writing task do you think a computer could replace? And we got someone who called and left a message saying, "What are other human activities that Phil thinks can be replicated in a similar way by computers?" What do you think? 800-433-8850 or go to our website, kojoshow.org. Join the conversation there. I'm Kojo Nnamdi.
NNAMDIWelcome back to our conversation with Phil Parker. We're talking about when the author is an algorithm. Phil Parker is a professor of management science at INSEAD, an international business school based in France, Singapore and Abu Dhabi. We're taking your calls at 800-433-8850. How would you feel if you knew your favorite radio program was created by an automated program? 800-733-8850. You often create resource materials like dictionaries and languages with very few speakers even left.
NNAMDIYou created a thesaurus for aboriginal languages in Australia, some of which had only a handful of native speakers left. Are these publications meant to preserve languages? Are they meant to teach? What are they intended to do?
PARKERActually, there's -- they're not really there to preserve languages. That's not something I'm involved in. It's more that there's a study that was done and published in the American Economic Review that said that people who can acquire a second language, their incomes, per capita, generally are substantially higher. The more you can acquire new languages -- and one language you might want to acquire is English or French or Arabic or whatever.
PARKERAnd so we tried to think of, like, what is it that's required minimally to acquire new languages and basic dictionaries to and from languages, but also with the sources for SAT preparations so you can get into certain schools. These preparation guides just don't exist in these minority languages so that you can help study. Even the notion of synonyms doesn't necessarily exist in many languages. So bridging between English words and what are non-English words to a list of English synonyms is kind of something that's new for many languages.
PARKERSo we weren't trying to preserve the languages, actually, helping people acquire commercial languages, which are the languages of traders locally. So it could be in (word?) Africa be acquiring French or acquiring English in English-speaking countries. So it's more -- becoming more bilingual, actually. It can help people, especially -- the stakes are pretty clear on that regard, especially for the people at the very lowest levels of income. Their ability to learn materials outside of their language gives them access to many things they wouldn't have seen otherwise.
NNAMDIAnd you have patented a process of automated authoring. Tell us about that.
PARKERThe patent covers all forms of authorship, be it video or audio or books or PC software or even websites. So a computer is just generating from the fly, you know, websites and videos. The -- a good example might be -- the movie "Shrek" is programmed, the animation, using a software package called Maya. Underlying Maya is a computer language called MEL or MEL scripting. And so we have compute programs that write MEL scripts directly, therefore, there's no animators involved.
PARKERThere's no production crew, no director, no artist, no -- all of those things are kind of reverse engineered, so to speak. And then we write computer algorithms to mimic those behaviors. And that's how you saw that weather video you mentioned before. That was fully generated on the fly, using algorithms that automate video creation pixel by pixel, so to speak.
NNAMDIHere's Jamie in Ellicott City, Md. Jamie, you're on the air. Go ahead, please.
JAMIEHi. How are you guys doing?
JAMIEI had -- I just had a question about the -- back to what he was talking about at the beginning, how they used it for creating, like, poetry and stuff. Couldn't you use that same, I guess, algorithm to -- in even, like, robotics or -- I know you said that already you use them in gaming, but, like, make it more real-time, like in games like "Call of Duty," like the military games, like, you could have -- you could actually ask the computer a question or something and have it respond in real-time then?
PARKERYeah. We've created a -- thank you for the question. We created a character named EVE, like the mother of all languages, and she's actually an artificial bot. You can ask her questions, and she answers. You can say, write a poem for me in a certain -- write an -- you know, and she actually responds as a friend would respond in creating things. We're using -- we're calling her a bot, and we're using her right now in Africa as an SMS bot.
PARKERYou can SMS to EVE a question, like prices of coco, and she'll answer you. Or you can say banana wilt, and she'll describe banana wilt and how to cure it. You can ask her a variety of questions. She's endowed with billions of pieces of information, and it's an SMS functionality because a lot of the phones that are out there in the rural areas is, in fact -- are, in fact, SMS-compatible.
PARKERSo people can actually SMS her, and she gets an answer back, and very heavily used for weather, tips, mathematics, one plus one equals what, you know, these kind of questions that are highly formulaic in their nature. So she's read billions of lines of text, very similar to IBM's Watson, so to speak, but heavily focused on agriculture, health care and things that would be relevant to people in the rural areas.
NNAMDIJamie, thank you very much for your call. I don't know how large the leap is from information about specific aspects of agriculture to writing, well, a college term paper. This email we got from Will in Silver Spring: "How well could computer algorithms write papers for high school and college assignments? Computers can now identify when someone has plagiarized a paper that has already been..."
NNAMDI"...posted online or in a database. But could we take a paper that has been published online and use algorithms to modify it enough so that paper plagiarism algorithms wouldn't identify it and, at the same time, still making sense and getting a good grade?" This kid has larceny on his heart.
PARKERYeah, yeah, that's the nefarious side. In fact, I could write computer programs that could torpedo any candidate for political office by just spreading a rumor mill, literally a mill for rumors. Yeah, of course, that could be done. I think what's more interesting is actually creating a Ph.D. thesis, go a little bit -- step farther because many doctoral theses are very formulaic in nature.
PARKERFirst, you state the problem. You state where the literature on that problem. Then you state the model you're interested in doing. You state what are the data, what is the model you're using and then the conclusion of the model and then a summary with some references in the end. That is very formulaic. I'm working on one project right now, as a matter of fact, that will write academic papers that'll be the effective X on Y: A Z Perspective.
PARKERSo it could be the Effect of Affect for Books on Literacy: A Cross-National Perspective, for example. And we're using Internet queries where we have people answering questions online, and that data is automatically analyzed, calibrated, and then an academic paper comes out at the other end. So that way, many under-researched topics can actually be researched by the computer.
PARKEROur first major application was in agriculture on biodiversity, where a lot of plants are so rare no one had time to research them in detail. There's 450,000 plant species, but there's not enough budget to cover people researching all of these plants. So the question is, for example, what are the chromosome counts of those plants? If a certain pest affects them, what would be a good abatement strategy?
PARKERAnd because there's so little funding, we have computer algorithms that mimic -- we call it a virtual laboratory -- that mimic what a scientist would speculate would be the appropriate chromosome count or the appropriate, let's say, disease abatement strategy. So we're using it for kind of knowledge discovery, if you want. So that's a little bit more than just a term paper for a student. But, at the master's level or the Ph.D. level, it certainly could be used to advance knowledge as opposed to get out of writing a paper.
PARKERSo I would discourage you looking for such an algorithm to fool the online people. I'd probably spend my time, myself personally, on actually creating something that really does generate something that's, you know, original.
NNAMDIIn other words, are we talking about here replacing not just, well, radio hosts, but also scientists and other academics? You're working on a project you're calling virtual labs, and we're talking about getting computers to work on Ph.D.-level analysis.
PARKEROh, yeah, absolutely. That's probably the biggest area that you can probably use the technology for, is literally, like, what do scientists do? They literally are author of things. I mean, the cure for malaria will be a sentence. The cure for malaria is, and they'll say something. The question is, what is that sentence? What will it be? It'll be based on a lot of research and scientific speculation, et cetera. And, often, scientific speculation leads to hard research. So we're creating a speculation engine, so to speak.
NNAMDIYet human knowledge moves forward only when someone goes beyond what we already know and asks the question that no one else has as yet: Copernicus, Newton, Einstein. Could a computer ever move ahead of the knowledge that we already have?
PARKERThat's kind of -- I think it can if it's couched in the right context, of course, but it's not going beyond the knowledge that one knows one can achieve. It's just speeding the process up, so we know we can achieve a certain knowledge if we invested so much human effort doing research in an area. The question is, can the computer make it faster by replicating what graduate students might do, by reading a lot of text and drawing lines between the dots, so to speak, and then ultimately authoring something that's digestible?
NNAMDIWhat's the distinction between what you're working on and AI, artificial intelligence?
PARKERWell, I think AI is -- when you really get into it, is less mysterious than one would think 'cause a lot of what we use in our brains is a very narrow area. When I worked as an economist doing econometrics, I realized I really didn't use that much of my mind to come up with a -- an academic paper, so to speak, and has a very small part. So the algorithms we work on are actually very small slices. We're not trying to mimic the human brain. We're just trying to get the outcome desired, like a sonnet. And a sonnet, you only need to know that it's, you know, so many lines, and it follows a certain pattern.
PARKERSo rather than trying to invent an entire human intelligence and then ask it to write a sonnet, instead write programs that write sonnets. And as you ask it to -- as you program more and more, ultimately, in the long run, someone might come up with something that's very all-encompassing, another human mind. But that's not really what I'm working on. I was working on more goal-based things, you know, weather for people, and that's pretty much where we started.
NNAMDIWell, since you're working with the Grameen Bank, I think Maria in Washington's question might be relevant. Maria, you're on the air. Go ahead, please.
MARIAThank you. I work for the nonprofit microfinance opportunities, and my question to Phil was, are they planning on adding financial education, tips on the radio or the SMS for rural areas in Africa such as savings tips for the month or how to manage my loan for my agricultural production?
PARKEROK, great question. Thank you. By the way, we're working with the Grameen Foundation. As you know, the Grameen name is used in different organizations.
PARKERAnd we're in Uganda, and we're creating actually pricing, exactly what you described, pricing tips. We have to be careful, by the way, because in many African countries, the pricing information isn't available. So even a radio host, if we try to mimic a radio host, there'd be -- we'd have a very difficult time because the data are not available. So we're creating these little SMS systems that query people, ask them what the prices are, then it comes back, does a statistical analysis, throwing out the outliers and coming up with means and then rebroadcasting it back out to people and other types of tips.
PARKEROn the microfinance in particular, this is something we'll probably pursue more. But, right now, we're really focusing on basic agricultural knowledge. So that would be prices and weather and tips on agriculture.
NNAMDIMaria, thank you very much for your call. I'm afraid that's all the time we have. Phil Parker, thank you so much for joining us.
NNAMDIPhil Parker is a professor of management science at INSEAD, an international business school that's based in France, Singapore and Abu Dhabi. And thank you all for listening. I'm Kojo Nnamdi.
Most Recent Shows
The "Bug Guy" is back, and answering all your questions about summer insects.
DCist was resurrected this week, acquired by WAMU after being abruptly shut down seven months ago. We explore the state of local media here in our region, and how the re-launched site fits into that landscape.
What do voters want to know? Kojo continues the conversation on D.C.'s contentious proposal with WAMU's Ally Schweitzer and an undecided business owner.