Good Trouble

Let's use AI to do journalism

By Ben Welsh and Scott Klein • • Columbia Scholastic Press Association in New York City

Slides

Show the extracted slide text

Slide 1

>MAKING GOOD
  TROUBLE
  WITH AI

Slide 2

>SHALL WE
PLAY A GAME?

Slide 3

>WORD
ASSOCIATION

Slide 4

>READY?

Slide 5

>WHAT’S THE
FIRST THING
  YOU THINK
WHEN WE SAY…

Slide 6

>“SCIENTIST”

Slide 7

>“TECHIE”

Slide 8

>“ENGINEER”

Slide 9

>“MATHLETE”

Slide 10

>“JOURNALIST”

Slide 12

>SCIENTIST

Slide 14

>TECHIE

Slide 16

>ENGINEER

Slide 18

>MATHLETE

Slide 20

>THEY’RE ALL
JOURNALISTS

Slide 22

1967

Slide 23

>JOURNALISM
 IS A STEM
   MAJOR

Slide 24

>WHAT IS
JOURNALISM
  ANYWAY?

Slide 25

>JOURNALISTS START WITH
A QUESTION NOT AN
ANSWER. JOURNALISTS GO
WHERE THE FACTS LEAD,
EVEN IF THEY DON’T LIKE
IT.JOURNALISM ISN’T JUST
WRITING. JOURNALISM IS A
MINDSET AND A SET OF
RULES.

Slide 27

>THE MOST
IMPORTANT
 SKILL IS
 GIVING A
   💩

Slide 28

>WHAT IS
GENERATIVE
    AI?

Slide 30

>LET’S USE AI
TO MAKE GOOD
   TROUBLE

Recording

Show the timestamped transcript
  1. Hi everybody. Hello. Was it just lunch? Yeah. Did you eat on campus or did you leave? There was free food. They ate here. Did anybody venture out into the big city? Yeah. Where'd you go?
  2. Were they New York sandwiches? Yes. That means like pastrami piled up or what? Um, well, I'm a vegetarian, so it wasn't that exciting. It just Okay. Okay. Sorry. There's a terrific dumpling place not far from here called Ali's. Okay. Um, Colia students, I'm sure, can can vouch for this. How was it? How was
  3. Ali's? It was It was very good. It was super good. It was great. Good. Did you have the Do you have the dumplings? That's what you got to have. I got the the wantton soup and then my friend Maya got the the dumplings. Dumplings. Oh, awesome. Awesome. So, we got a thumbs up on the dumplings. Did anyone go to my
  4. favorite place, the Hungarian pastry shop? Yeah. Heck yeah. It's it's it's purum, right? They must have the hamandashen. I like to get the poppy seed myself. Do we have any prune fans here? Poppy seed is the sort of classic hamashen.
  5. You like the prune hamashen? I like pun. Oh yes, we all like pun. So, okay. So, you are here for making good trouble with AI, guys. Get ready. That's right. This is going
  6. to be interactive. We're going to try not to get on the stage, but we're going to put one of you on stage. Um, so get yourself ready. Uh, prepare yourself to volunteer. But there will be candy. There's one. Whoa. Candy.
  7. We Yeah. No, we didn't realize we were going right after lunch, but when we found out, Ben went out and got candy. uh to keep everybody alert. Uh fall asleep at the next one. Coming in.
  8. We're going. You got to wait. It got to be a reward. You got to earn it. You got to earn it. Scott says you have to earn it. Scott, who are you? I am Scott Klene. Uh for many Oh, thank you. for many years. Uh I worked at ProPublica uh here in New York. Uh but I have since then left and
  9. I now work uh I'm trying the world of software out. So, uh, I now work at Automatic. Uh, who here has heard of WordPress. There we go. So, we are the company that makes WordPress, but I work on a team that, uh, works with many, many small newsrooms around the country.
  10. Um, so I'm sure that we have a a publisher who uses Newspack, um, at the city you guys live in, but we don't have to get into that because we're not here to talk about that. And who are you, Ben? Hi, my name is Ben Welsh. I live here in Manhattan with my wife, but I'm a relatively new New Yorker. I've only
  11. been here about two years. I work at Reuters News in Times Square, but I'm originally from Eastern Iowa. Do we have any Ians in the house? What? Where? Hell yeah. I grew up in Swisser. Right
  12. there off 380. You know where it is. Hell yeah. And I am a nerd in the news. And I don't know if Scott, do you self-describe as a nerd? Do you identify yourself? Abs self-describe others describe me. I'm just sort of generally
  13. it is a truism that I am a nerd. Yeah. And so as you can probably tell from our presentation which is about to begin Scott and I are the nerds in the newsroom, right? We are we have kind of specialized roles that we've car carved out over the last couple decades where yes, we are reporters and yes, we are
  14. editors, but we're also computer programmers and product designers and just tinkerers who are trying to help the news industry kind of make it on the new rough and tough digital frontier. And but I just I will uh interestingly I spent 25 years as the nerd in the
  15. newsroom and now I'm the journalist in the nerd room. Whoa. How's that? Uh, you know, it's pretty awesome. I like it. So, right. Um, all right. So, this is going to be interactive. We're not going to make people stand up, but we are going to run at you with a microphone um
  16. and ask you to kind of raise your hand um and answer some questions. So, uh who here is ready to play a game? Ready or not, we're doing it. I don't know why I asked. All right. Scott, call on somebody. All right. Hold on. Hold on. All right.
  17. Sorry about that. Uh, I got to be closer to do those things, so I can't go that far. There we go. Do you guys know this reference anyway? No. Hell yeah. War games. There you go. Here comes your candy. We're going to play a game of word word association. Look it up. War
  18. games was a movie from when I was a young person. Uh, I gota Why do I keep doing this? Word association. All right, you ready? Finally. Can we get there? All right.
  19. So, what is the first thing the first thing you think of when we say scientist? No, we raise hands. We're gonna We're going to put you on mics. You again. Nerd. Nerd. Wow, that's rough. How about
  20. you? Biology. Biology. Very good. Anybody else? Oh, sorry. Smart. Smart. Andy, also true. I like that. Hold on. Wait. Good. Your answer. Research. Awesome. Love it. Experiment.
  21. Experiment. Fantastic. This is great. These are Colombia kits. Yep. Big words. Terrific. You earned some candy on that one. Ready? Should we move on to Ben? We We have Yeah. What's the next word? These
  22. are good. These are excellent. Techie technologist technologist first word that popped into your head. Yeah. Computer. Computer. Computer. You Zuckerberg. No. Weird. Zuckerberg. Okay, I'll come over.
  23. Um, Bay Area. Bay Area. Whoa. Bay Area. It's true. Okay, we got some folks over here. Scott, I'm making my way. All right. Yes, you guys. Wait, we got some candy. We got some candy earners over here. Big mouse. Mouse. Okay. You know,
  24. my intern last summer had never used a mouse in her whole life. She was like, "That's for old people." Did you guys know that? Shocked me. You got a word? Yes. This guy right here. Him. Great word. The sequel to her.
  25. What else we got for Techie? We need one more. Okay. Yeah. How about you guy right here, dude? One more. Another. Wait. Okay. Wait. We got two more. Ben, we got we got candy earners up here. Way too much candy is what Scott's telling me. Coding. Coding. Coding. That's a good one. Perfect. Bro,
  26. bro, I know plenty of Okay. of non bro techies. All right. What's our next word? Hold on, Ben. We didn't We are We owe candy. Yes. Okay, here you go. Like half of this area. Just
  27. don't selfidentify as candy earners. All right. Engineer. You got a word? Gears. Gears. Okay. Robots. Not the first time. I sense robot. Robots. I said gear. You're done.
  28. Okay, you're out. Your cash machine. Machine. Building. Building. We got building for engineer. I love it. Good. Robots. Robots. More robots. Lots of robots.
  29. school. School you know where we are now. Robotics. Robotics. More Rob. We got robots. This is like a This is terrific. All right. This is great. And we got one more word, Scott. Or we got one more
  30. word. How do you say that one? Okay. Okay. I was genuinely worried this this is like a thing from my era and didn't exist anymore. All right. All right. These still exist, right? Good.
  31. Olympiad. Olympiad. Okay, good. These still exist. Thank you. Derivative. Derivative. Yeah, that had three syllables. Hell yes. Mean Girls. Mean Girls. Wow.
  32. Really? I didn't know. Does that happen in Mean Girls? Is there a mathlete? The new one or the old one? Oh, okay. Oh, here we go. Another mathlete. Competition. competition. Okay. All right. Very good. Excellent.
  33. Who uh who is owed a candy? I just want to make sure we don't Did you get your candy? You guys Oh, she don't want to get She don't want candy. That's okay. All right. Well, let us know there we got way too much. So, you are all going to get candy even if you didn't. Science, techie, engineer, mathlete. We
  34. did those four. All right. We got one last one and we want you to dig deep and just say what's the first thing that comes to your head. Don't be afraid. of my feelings. Ben can speak to him for his own. All right. First word that came into your head. Let's be honest. Yep.
  35. Honest. Writer. Writer. The first one we got was writer. Okay. Interviewer. Interviewer. Love it. Got one. These are all terrific. Inquisitive. Inquisitive. Excellent word. Such vocabularies.
  36. Storyteller. Storyteller. Absolutely love it. Okay, I got one. I see one. I see one over here. Hero. Hero. Hell yeah. Yearbook. Yearbook.
  37. We got candy. Yep. Good. All right. And let's get one right here, Scott. All right. Oh, he's passing you by. Underpaid. I agree. That one is absolutely true. Underpaid. That was funny. One more reporter. My Yes,
  38. reporter. Very good. Oh, here's your candy. Okay. I saw I heard a bit of a contrast from the four before. Don't you think, Scott? Yes. I I noticed that immediately. The first four were very
  39. similar, right? But journalists seemed very different. So what the vibe I get from the room is that you all think that journalism is very different than science, technology, engineering, and math. Is that right? Is it sort of like
  40. different? Like these like one of these things doesn't That's definitely my era. One of these things doesn't belong. Uh yes. Okay. All right. Well, we think I'm not sure how to do this. Can
  41. you do it? There you go. Come on, Scott. You get some candy if you do it. Boom. No. No. Not quite. Not yet. I got it. You got it. I got it. It turns out that's wrong. Sorry, you're wrong. That is not true at
  42. all. We think and we are we are proof that science, technology, engineering, and math are all things that are useful in journalism. and they are kinds of people that journalists can be and kinds of
  43. backgrounds um that journalists can have. And we're going to talk a little bit about some people that we know in the news industry who come from science, technology, engineering, ba math backgrounds. Uh so this is a guy named Peter Aldis. Do you want to talk about
  44. Peter? Peter was a scientist and he got bored and he thought working at the bench and writing these papers that very few people read wasn't a very good time. and he decided to change it up and become a reporter. And he took all the skills that he developed as a scientist
  45. and used those same skills to find and tell stories, including stuff with AI, which we can get into, and um in this particular case, tracking uh government spy planes and then writing about what they're up to out there in the world. Meanwhile, who can can anybody anybody with really good vision can you read uh
  46. what what Peter did in college? He has his PhD in animal behavior. Um he worked for a long time at BuzzFeed News uh of Blessed Memory. Um and I'm not sure where he is now, but you as Ben says, he is an investigative reporter
  47. that uses quantitative methods that he learned studying animal behavior um to do journalism projects. Um and as I said, here's the the project on kind of uh he figured out how to use uh helicopter transponders. Every airplane
  48. has a transponderant to sort of figure out whether uh FBI, whether and why FBI helicopters were uh were hovering over places. So that's Peter Aldis, a scientist. A scientist and a journalist and a journalist. Not not one or the
  49. other. Scott, what's going on, dude? There we go. Hell yeah.
  50. Uh this is Jeff Cow. Uh Jeff uh works at Bloomberg News now, though I worked with him when he was at ProPublica. Um this is a project that Jeff did uh in which uh Bloomberg uh has a obviously who knows about Bloomberg News? Who's heard of Bloomberg News or the Bloomberg terminal or Mayor Mike Bloomberg? All
  51. sort of related to each other. Great. So uh Bloomberg News Oh, I said I wasn't going to go on the stage. Bloomberg News uh did a project where they took uh so who here's from New York City? You are awesome. I'm also from New York City. We um uh we just started this
  52. thing called congestion pricing where you pay a toll if you drive your car in Manhattan under 60th Street. So you're you're safe here, but if you drive your car south of 60th Street, you get you pay a toll. And critics of the program said that uh this was not making traffic
  53. any better. And Bloomberg wanted to see if that was true. Um and because we they have nerds, uh Jeff and a band of his colleagues aimed a high resolution camera out the window of their building and then wrote
  54. a computer vision piece of computer vision software that counted the number of cars that passed by. and they were able to find out because they're they're under 60th Street. Um they were able to determine that thanks to congestion pricing that actually the number of cars
  55. in Manhattan did go down. Um and in fact the computer vision they wrote was able to determine the difference between a civilian car, a cab, and other kind of you know uh service vehicles. Um so this was a program that they had to write to
  56. kind of answer this question. Um and where did Jeff go to school? right here, right here, right on this very campus. Um, he went to the University of Wateroo in Canada where he was
  57. a he was an engineering major. Um, and then, uh, he got a law degree here at Colombia and then decided not to study law, not to practice law. Um, but yeah, so and Jeff um, so Jeff is a techie. So, we have science and we
  58. have technology. Where are we gonna go next? Ben, you want to talk about Arena? Yes. So, Arena is a journalist who has worked at NPR and um ProPublica and it's
  59. now at the New York Times, I think, as of a couple weeks ago. and she studied actually engineering at Stanford University and has since become a journalist who like us works with data to find and tell stories. Um, and without those specialized skills,
  60. without that nerdiness, she wouldn't be able to do what she does. That's right. These projects, so they she did a project where she did not just a whole lot of math to uh to do a story about uh who here's from the Pacific Northwest.
  61. Oh, we got one. Yay. Uh, so there are, uh, salmon fisheries. The salmon industry in the Pacific Northwest has really been, uh, you know, getting smaller because of over damning, I guess you would call it. Um, and so Inreina
  62. did a big mathematical project to try to figure out why that was happening and whether the federal government's uh, processes for fixing this were working. Narrator, they're not. Uh, and so, uh, I Ina used her quantitative skills and her
  63. more traditional journalism skills, whoever said interviewing and things like that, to do this story, uh, working with a reporter, um, out in Oregon. Um, she also did a project, um, if you look at the map, she did a project where she looked to see, um, how many state legislatures um, had fewer than 20 less
  64. than 20% women. Um, and she found that actually all of the ones apparently in the southeast, almost all the southeast had legislatores that have less than 20% women. So, uh, lots of math, lots of engineering and and kind of technology
  65. to be able to kind of find stories that would otherwise not be findable. Keep that in mind. We're going to be finding stories in about a minute. Um, so Inrea is actually probably both a scientist and an engineer. What does that leave?
  66. And finally, we have Amanda Cox. Does anyone know Amanda? She is now the data editor at Bloomberg News, where she runs a large team of people, including the one Jeff's on, I suppose, and has been a pioneer in uh digital data journalism now for more
  67. than 20 years or so, and did that after studying statistics in college. That's right. She got her bachelor's degree in economics from St. Olaf. We have any any St. Olafs? Man, that would have been fun. Uh, she got her bachelor's degree uh
  68. from St. Olaf's and then she got in economics and then got her master's degree in statistics. So, what does that make her? There we go. She's our math. She's our mathlete. And so, all of these people,
  69. what do they all have in common? They're nerds. Scott. And oh, oh, by the way, this is what this is. Uh, do people know this? Amanda invented the needle at the New York Times election night. Yeah. Check it
  70. out, guys. That's what that's needed. Yeah. Yeah. We don't want to skip that slide. That's That's Amanda's claim to fame. They're all also journalists. They're nerds and journalists. They're are nerds and they're journalists. They are not intention at all. um you can do
  71. all of the things that a journalist does and use your biology skills, your engineering skills, your math skills. You can bring all of yourself um to journalists and we need all of that in journalism and in newsrooms to tell the stories that can be told now that have to be told
  72. now. So if we leave you with something in addition to kind of showing you how to use AI to do stuff, um it's that we believe I believe Ben do you believe this? With my full heart, Scott, I believe with my full heart that journalism is a STEM major. Salaries haven't caught up
  73. yet, but the reality is that we need biologists and chemists and animal behaviorists and electrical engineers um and all of those things in journalism. Um, which raises a
  74. question, seeing everything that you've seen and it's kind of thinking about the words that you describe journalist before we showed you all of those slides. Knowing what you know now, what is journalism? Can someone define it for the class? Oh,
  75. hell yeah. I love this. You're gonna get if you answer. Break it down. I'm giving you the mic. What's journalism? Journalism is the ability to take a story and distill it down into a
  76. consumable format for the general public. Okay. He's taking the production angel, Scott. I kind of like that. That's not bad. So, that was the packaging and storytelling side. I love it. I love it. Anyone else want to try to define journalism? What's missing from that definition?
  77. Hand the mic. Let's hear this guy. I think you might be trying to get at the journalism side. That is the scientific method. Maybe a little, but there's the there's also the gathering part is what I'm thinking. I heard the publishing part. Yeah. You got to go get
  78. that, right? Yeah. The gathering of information. Yeah, man. This is what we think journalism is. It is all of those things as
  79. well. But people are doing journalism right now who don't call themselves journalists, who don't realize they're doing journalism. Journalism, it's a way of thinking. It's a set of rules that keep us honest um
  80. that hopefully make us credible in the world. When we say something important, people believe us because we apply we impose rules upon ourselves. Um what are the rules, Ben? What's a rule? Tell the truth. Tell the truth. That's right.
  81. Don't give money to political parties. Yeah. Right. We sort of there's a little bit of a monkishness to it about what we can do and what we can't do. We don't sign political uh pamph.
  82. Yes. Uh we kind of hold ourselves out of we don't go to protests, right? There's some there are some rules to keep us honest. We don't pay sources. Do people know about that? Like if you ever hear see a source, if you see a person talking on TV, it is absolutely forboden, forbidden to pay that person for being on your air or for being for
  83. stalking for your story there. So, it's a set of rules that keep us honest, but it's also a way of thinking. Um, and it's already you've already seen it on the board, so I don't have to go too deeply into it, but one of those things is um that we start with a question. Um, we are all biased. We all have opinions.
  84. Um, some of our opinions are really deeply held. Some of those opinions are political. Um, and we can't turn those off no matter how many rules we make for ourselves. But the important part is that we start with a question, not an answer. We don't we don't take an answer and then find the proof that we need.
  85. Um, the scientists you can call that uh inductive, right? Inductive is when you start with a answer and then and then uh and then find the question. Um, we start with the question um and find the answer. And if we don't like where that goes and if we if it goes against our
  86. side or if it makes us feel really bad and we wish it weren't true, it is too bad. If if Ben if I was doing something bad, which I'm not, and Ben figured it out and wrote a story about me, even though we're actually close friends, I would understand. He would have to do
  87. that, right? And so to to return to our examples, Peter started with the question of what are all these we where are the spy planes? Can I find them? I know the government has them. Where are they? How frequently do they fly? What are they up to? Right. And then our
  88. techie example, Jeff started with, is congestion pricing actually reducing traffic in Manhattan, which is what it claims it's going to do? And Arena started with the question of are we approaching anything like gender parody in our state legislatures, right? And Amanda started with the question of can
  89. we from the early election results as they begin to emerge in the night give you some sense of where they're going to end up or a better sense of where they're going to end up based on what we know about the political science of it all. So each of them started with a question in a sense like any reporter
  90. might but they used their science technology engineering and math skills to find the answers right and then even some of those same skills to publish something interesting on the internet as well.
  91. You can do it Scott. Yeah. And in that way, but journalism is not one thing. Journalism is the intersection of things. Yes. There's writing or photography or illustration or coding, but it also is
  92. an intersection with the what you bring to it, the expertise that you have. My whole career uh I have been a journalist and an engineer. I have managed people who are journalists and illustrators, journalists and photographers.
  93. Journalist and something is very much a thing. Um, and frankly more so as technology improves, as as the internet happens, and as traditional newsrooms are kind of competing with and working with things like influencers and on
  94. social media, um, and kind of building, you know, helicopter trackers, um, stuff that, you know, 40 years ago would not have been a thing that news
  95. did. There we go. Um, really the most important thing. Can I swear? Ben, should I swear? You should swear, Jennifer. Can I swear? If you've never heard this before, tell your parents. I'm sorry. But, uh, the
  96. most important thing, the most important skill of a journalist is giving enough of a to use the stuff you're good at to tell the truth. Should we give Scott some candy for cursing? Oh, thank you. Well done. Well done. That was not easy for him.
  97. Um, so all right, I'm having one myself. Okay. Is this the last candy question? No, this is not the last candy question. What is generative AI? Only wrong answers. No, I'm joking.
  98. You can give a right answer. It's like you're putting in your ideas into it and from what you put in it generates something. Sure. Yeah. The generative. Exactly. Generative generates. Exactly right. You put in you put in some inputs and it gives you some outputs. Anyone else?
  99. I got one here, Scott. It's like uh kind of amalgamation of code that uses a large data set of information and then kind of recreates it and replicates it in a unique and original way. Pretty good, dude. That's pretty good. Pretty good. That's definitely worth the candy. What about you? Can you top it? Top guy. Sorry,
  100. Aaron. Um, I'd say something similar like it's it has a lot of information and it's instead of just like like a regular search engine where it's just pulling up an answer from something else that already exists, it's going off of other information and from that creating
  101. something new. Pretty good. You got something to add to that? We got one more. One more. It uses statistics to figure out what it's you're most likely looking for and to use what the most like most likely correct answer is based on a massive set of data it already has.
  102. Hell yeah. That was very good. Baze himself would be proud. Here here's a question. Now, I'm not talking about your homework or essays or cheating in any way because no one who gets themselves invited to this conference would ever do that. But who here has used uh a conversational AI like chat
  103. GPT or C-Pilot or Gemini? Everybody love it. That's fantastic to do what kind of stuff to do. Again, you didn't do homework and you're being recorded.
  104. Uh what did you do? What did you use it for? Um, when it first like came out, I tried it out and I just like held a conversation with it to see how useful it was and how like responsive it was to my questions. And was it was it responsive to your questions? Yes. That's awesome. That's great. Who else who else is has an example of a time
  105. they used genai for not homework and you're being recorded? Yeah. Um, I wrote a a program that uses chatbt to automatically email professors and ask for internships. That is awesome. Um, and did it get you an internship? Um,
  106. no. We only tested it on my mother and she was not happy that we tested. Okay. And she was already going to give you the internship, so you didn't even need to. Yeah. Yeah. So, all right. One more and then we're going to start using generative AI. For art class, we were using generative AI to try to compare to see if it would
  107. replicate human art. And did it? No. No. It was pretty bad. Okay. One last one, Scott. All right. Okay. Um, so for yearbook class, um, we always have trouble like with headlines and everything, so might as well just ask AI to help us come up with a
  108. headline for for like a page. How to do? So, it did pretty good. Yeah. And it's really nice, especially around like final deadline time. Like I remember like the final day like we were just like we had like chat GPT up and we were like typing like um aggressively to try
  109. to find a headline and you know what it worked. So that's awesome. That's great. Um, so we're not going to get we only have 15 minutes and we want to use a genai so we're not going to go too deeply into it but we've had some excellent answers here somewhere in that you're getting getting at the answer.
  110. Um, who here but people have heard right that there are some anxiety that Geni is going to start taking jobs. I assume everybody has heard that. Sure. Sure. Um, I think very differently. I think that generative AI is like the autopilot of an airplane. I don't think that any
  111. pilot jobs were lost when they introduced autopilot. But flying got safer. Um pilots started needing new skills and they started using the autopilot as much as flying the plane themselves. So they still need to know how to fly the plane themselves. And I think it will be an incredible tool u
  112. that will transform a lot of fields especially journalism right and not just in the English language stuff. We kind of heard here we heard some English class answers but also in those in the STEM part of our work as well right in the science technology engineering and math side. So who said when we were
  113. talking about journalism who said uh interviewer? Awesome. That was a great one. Thank you for that. I get extra candy just extra credit. So yes, interviewing is an incredibly important part of of being a journalist
  114. and you interview people. But one of the sayings we have in data journalism is interviewing the data. So the data just like a person can also be interviewed. Uh we also talk about how you trust it exactly as much as you trust a person.
  115. Uh which is not at all. Uh, but we are going to use some AI to ask questions about data uh to hopefully make some good trouble. Um, who wants to be at the Who has to type on my computer? Need to
  116. volunteer. Red shirt. Do you want to come do it? Awesome. Thank you. There's like so much candy in this for you. Yeah. Just don't look at the browser history. Okay. No, no, no.
  117. All right. What's your name? Uh, Enzo. Enzo. Wonderful. Where are you from? Uh, New York. Oh, from what part? Uh, Harlem. You're from Oh, right around here. Oh, that's awesome. That's great. I am from Brooklyn. Um, okay. So, you stand here. No more slides. So, I'm like off the hook now. Um, so who flew here? Who took
  118. an airplane? Yeah. All right. Be prepared to tell us the airport you flew out of. This is going to be airplane airport related information. Who here knows what a what a runway incursion is? A runway incursion, a government word,
  119. if there ever was one. It's when there's either a vehicle or like some sort of fuel spilled on a runway or some some reason why a plane can't land on a runway. That is like three pieces of candy right there. That is exactly right. It is when something blows onto
  120. the runway at a time when they don't expect it to or something drives on the runway or something is stuck on the runway. Every time this happens, it turns out that there's an FAA regulation that says that the airport has to report that to the federal government. This can be trivial. It can be little nothing. It
  121. could be uh you know, a person ran onto it and was incredible dangerous. So, it can be a lot of things. And the government actually publishes all of the data. So every runway incursion in any airport in the country uh is reported to the federal
  122. government is actually put in this data set um for uh a while. So let's So this is it's been happening for a while. So like 15 years I think. Um so I just need an airport.
  123. Wait, what's that? MSY. Oh MSY in uh in in Minneapolis, right? Or New Orleans. Sorry. All right. MSY. So, let's see what we find from MSY in New Orleans. Excellent. By the way, airport code. Yes, airport code knowledge is you're going to be a data journalist. There we go. So, we can look
  124. Louisiana MSY. And if you click on search RWRWS on the lower right, you can very quickly see Oh, my session expired. Hit okay. Try it. Yeah. Yeah. Just do it again.
  125. MSY. That's fine. Right. There we go. So, we can see all of the runway incursions that happened at New Orleans uh airport um since a while now, since 2017. And you can click
  126. on each of these. Uh but Enzo, go back to the page before. How might we find out what airport in the country has the most runway incursions or who had the mo how many
  127. are in runway incursions increasing or decreasing? How would we figure that out from this interface? Uh I no candy there is no way to figure that out in the interface. That's something that you would have to do like a data journalist like me or Ben
  128. or like Ireina or Peter would download the data or figure out how to get the data from the government and run that analysis themselves. But, and this is an awesome trick that works with way too many websites. If we uh if you actually reload so that we don't have MSY in
  129. anymore. Did that do it? Yep, it did. So, if we put nothing in the search form, try this. It works shockingly a lot. Now, if you just with a blank search, if you click on search
  130. RWS, it will actually just give you all the data. So, this is if you go to the bottom, it'll tell you how many rows. 33,000 rows of data. So this I now have we have as a as a group all 34,000 runway incursions that have happened in United States since the teens. And if you go to the
  131. bottom you can see CSV download. So I can take it and have it all to myself. Three sweetest letters in the English language. CSV. Anyone know what it stands for? Hold on. Enzo knows. Enzo sorry. Hell yeah, dude. Comma separated values. Commaepparated values. And that
  132. just means a spreadsheet. It's just like a dumb acronym for means this is data in a spreadsheet. Thank you. And so now we can download it. And now should I Oh, wait. What happened to my chat JPT? Oh no. What' I do? I got this here. Hold on
  133. one second. Hold that for me. Yeah. Thanks. Now you're all seeing my uh my browser history. Okay. Here we go. So, this is chat GPT. People have
  134. seen chat GPT. So, with this uh this LLM with this model, the 03 mini high is really good at this, but you can actually use it with a bunch of models. If you click on plus on the left of that thing, you can upload a file. So, upload from
  135. computer. And I have already downloaded the let me just go to the desktop. I have already downloaded this. There we
  136. go. So, here is the runway incursions. And I'm going to hit open. Yeah. And it's going to think about it a second. And it's going to load all of this data into its terrifying hive mind. Uh, and it's going to be we're going to be able to ask it's already there. So, now we can interview the data. We can ask it
  137. questions that are very interesting to us. So, I'm going to start with one and then I want to hear from others. Um, uh, what airport has the most runway incursions? Had the it'll figure it
  138. out. Reasoning considering airport runway incursions. This this model is the creepiest because it's going to tell you how it's thinking about it. So, it asks the user asked which runway incursion. There's an uploaded file.
  139. It's decided it has to load the pi the file through some Python code. Let's try that approach. What's Python code? Who knows? Anybody? Anybody know what Python is? Enzo knows. Yeah, it's I mean it's just like the
  140. coding language that it's going to use to try to interpret the data and like list it all in into values. That's right. It's a programming language and you can use it to analyze data files. And 10 years ago, you had to learn how to write the code. Now, Enzo is the perfect person to bring
  141. up here. Enzo, you are awesome. Enzo, what has Chat GPD told us? It's Chicago O'Hare Airport. Um, anybody fly in from Chicago O'Hare?
  142. You were just there. So, yeah. Careful the runway incursions. I guess it's like it is like a Titanic airport. Like I you know, who knows, right? I I wouldn't worry. Um and and Enzo just did this, but if you actually click on the little icon to the right, uh we can actually
  143. see a little bit of the Python code that it used uh to interpret this information. And Ben, is that good Python code? It's okay. It's just okay. Better. Yeah. But it works, which hurts me. And so you can imagine a reporter who wants to interview the data can use
  144. chat GPT for this. And if they have enough Python, Enzo, do you think that's good Python code? It's okay. It's fine. It's fine. It's not like, you know, we're not writing uh Microsoft Word or something, right? It's, you know, it's pretty simple. But um and so if you are
  145. a programmer rather than writing all of the code to do this yourself, you can interview the data using check GPT almost literally. But then you're a programmer so you can look at the data and understand like did it mess up? Like did it do something that was wrong? Did it do something that might lead to a wrong answer. So who has other questions
  146. that we would like to know about runway incursions? We can ask and this is where STEM hold on meets journalist runway incursion. What is the most recent runway incursion?
  147. Okay, while it's thinking, let's put yourself in the mind of a journalist and tell me what you would do with the knowledge that you gained by inter by interviewing this data. Would you just immediately write a story that said the Chicago O'Hare has the most runway incursions? You wouldn't, right? Why
  148. not? It doesn't really like it's just like it's an answer. It's not like a story about like what's why does it have the most? Excellent. Excellent. It's not a story. It's just a fact. So, what would you do to add add knowledge to
  149. that so you could write it into a story? What are the things that you might do? Maybe interviewing someone who experienced a runway encouragement at Excellent. Excellent. Right. Sorry for cutting you off. No. Right. interview someone who has experienced a runway incursion. What
  150. else? Take Well, first of all, sorry. First of all, I would figure out what airport has the most runway incursions per like amount of flights. Hell yeah. Not just the most amount. And then I would probably explore that more than the first
  151. question. Um, but I think like right like going you like getting in contact with people in the airport and figuring out like what happens when there's a runway incursion. What does that mean for the plane and what um and why does it happen and why is why is it happening
  152. more in the one the airport that we're looking at? Like what makes it different? And especially for Chicago, like if we're looking at per like flights, like O'Hare is just a really big airport. But like if we look at per like flights, then what's the difference between like O'Hare and Midway, right?
  153. And like why? You're exactly right. You know what I mean? I think normalizing by a denominator for a rate is almost always the right answer. Some airports are bigger than others. A bigger airport should have more incursions. Your logic's right. I myself worked on a story about helicopter accidents and we
  154. we to the Robinson R44 helicopter according to analysis of a similar database had the most accidents but guess what it's the most common helicopter right and so we wanted to calculate an accident rate before we said it was the deadliest right so that was uh crashes per flight hour in this
  155. case it might be what incursions per takeoff or something takeoff per landing right you want it per something so that you can tell is this a big deal or not? In journalism, sometimes we say, "Is this a man bites dog story or a dog
  156. bites man story?" And if your editor says to you, "This is a dog bites man story," that means you're done. That's it. Like, you're not allowed to publish your story. Uh, it means it's not newsworthy and not surprising. So, what is Have you heard of that? Have you heard the dog bites man versus man bite dog? It makes sense. It's like an old
  157. person thing to say. So, but still um you will definitely hear it if you're an industrial. Um all right, I think we have time for one more query of the data and then we be then we need to wrap up. Uh Enzo, do you want to take typers privilege? Um sure. What's your question? Yeah, I
  158. say it out loud. Let's think. Well, I mean we can ask just ask the question of how what is the airport that has the highest density of incursions per takeoff? Let's ask. Absolutely. Let's do it. Yeah. I don't think the data is in
  159. there. This is where it's going to get created. It might go for it, dude. Right. What Ben is saying, one of the things we didn't do when we asked the first question is we didn't tell it to limit its knowledge to what's in the data. So, you're normally, and you're writing a prompt like this, there's this concept called prompt engineering where you kind of know how to write these
  160. prompt. And we normally would say, don't go outside the data set to figure things out. Um, and so, right, I think Ben is right. I think if we put that in, it's going to see if it can figure out how many takeoffs every runway in the country has.
  161. But let's see if reasoning says that, right? Clarifying airport incursion data. While we wait, I mean, what's truly incredible about this as I stuff my face with chocolate is to do this type of interviewing of a data set as a journalist, you used to have to spend
  162. weeks or months learning how to be a computer programmer. And now these tools are just making it much more accessible. They are imperfect, but they're a great start. We could try clicking search and letting it use the internet. But but what's important here is that it actually identified and we used to talk
  163. about like a you know ancient history like 10 months ago we used to talk about the problem of hallucination. Who here has heard of the problem of AI hallucination? So AI hallucination is the tendency for AI because AI is really doing a lot of like very very sort of
  164. higher math um kind of prediction. Um and it sometimes will give you a prediction answer which is actually not true because it's not really thinking and it's not about truth and fiction. It's about kind of prediction. So sometimes the prediction will be wrong. But AI is getting much better. And so
  165. it's done the right thing ultimately which is to tell you that it can't figure out out from this data set. Right? Um like you know a question I would ask is what's the total number of incursions by year. Right. Right. This is the type of questions especially like to me when I get a new data set these
  166. first questions I ask I kind of call the first date questions. You know I'm just trying to get to know the data set and kind of see what's in the different columns and just like basic counts and filters um are just a way to kind of begin to see what's there. And then exactly as that uh person in
  167. the back said, um you got to start making phone calls, right? This is where the interviewer and where the non- STEM, I guess you might say, parts of the job kick in. You need to go find an expert who knows a lot about runway incursions. Maybe call O'Hare and say, "Hey, I think you got
  168. the most runway incursions last year. What's up with that?" Right? Hey, hey, Chat GBT, can you make that into a bar chart? So, uh, we chose a data set that we knew we could download. Um, but there you can do this with just about any data set
  169. with mixed results. You got to be careful. You should also kind of probably check it to make sure that it's right before you start calling and accusing O'Hare airport of stuff. Uh, but you can use this on almost any data set. Uh, we tried before we got here with some New York City data sets. It
  170. works shockingly well. Um, and these are great ways to get yourselves uh into stories. So I think we have run out of time. Uh right.
  171. So it makes me almost sad to see this because it took me years of learning how to write Python code to be able to take a data set I had downloaded 30 seconds before and begin to do these first day questions and get to know it and start to figure out what's there. And now thanks to these tools, what we just were farting around and did it in 10 minutes,
  172. guys. It's just wild. So go out, make good trouble, download uh more troublesome data sets uh and uh please thanks everybody.

Downloads

Slides PDF · Recording video · Extracted slide text · Timestamped transcript