Latest / HumanPrint: How to Use AI Without Losing Yourself / Episode 7: Who's Actually Driving This Thing

Transcript

Episode 7: Who's Actually Driving This Thing · HumanPrint: How to Use AI Without Losing Yourself

• 0:00
  Foreign.
• 0:05
  Welcome to Human Print. I am Christine Whitmarsh,
• 0:08
  and if you are open to collaborating with AI but you're
• 0:12
  not interested in disappearing into AI, you're in the right place.
• 0:16
  Each week, we will look at one place where your authentic human signal is
• 0:20
  at risk for drift and one way to shore it up.
• 0:23
  Oh, and Human Print is brought to you by Abby AI Brand
• 0:28
  bank intelligence that helps authors, thought leaders, and enterprise
• 0:32
  teams train AI on their voice and point of view so written
• 0:37
  content stays unmistakably human, consistent and on brand.
• 0:41
  Learn more@abbey.ink a b
• 0:45
  b I.ink like you write with.
• 0:48
  Let's begin.
• 0:52
  Hi, everybody, and welcome back to to Human print.
• 0:56
  So, in February 2024, a man named Jake Moffett
• 1:01
  was booking a flight to go to his grandmother's funeral.
• 1:04
  He went on Air Canada's website, he found the customer service chatbot
• 1:08
  and asked about bereavement fares. The bot told him he could apply for
• 1:12
  a discounted bereavement rate retroactively after he'd traveled
• 1:16
  as long as he submitted the request within 90 days. So he went
• 1:20
  ahead, he traveled, he submitted the claim. But then Air Canada was like,
• 1:23
  sorry, yeah, no. Turns out their actual policy didn't allow
• 1:27
  retroactive refund requests.
• 1:30
  Like so many other bots we're used to, the Air Canada chat
• 1:34
  bot was wrong. So Moffat took his case to British Columbia's
• 1:39
  civil resolution Tribunal, and here's where it gets interesting. Air Canada's
• 1:43
  defense was basically that the chatbot was its own thing.
• 1:47
  It was like a separate entity. And. And the airline could not be held responsible
• 1:52
  for what it said. The tribunal rejected that completely.
• 1:55
  Air Canada was responsible for everything on its
• 1:59
  website, whether it came from a static page or a chatbot.
• 2:03
  So Moffitt was awarded exactly $650.88.
• 2:08
  Now, obviously, $650 is not a huge amount of
• 2:12
  money in the grand scheme of airline litigation, but that ruling is a big deal,
• 2:16
  because what the tribunal was really saying is you don't
• 2:20
  get to outsource the judgment and keep the authority. Those two
• 2:24
  things travel together, so to speak. You built the system,
• 2:28
  you deployed it, you put your name on it. Whatever it says, you said.
• 2:32
  And I want to be clear, what happened to Air Canada isn't really a story
• 2:36
  about a company trying to cheat anyone. The story was more about
• 2:40
  what happens when a company delegates a judgment call. In this case,
• 2:43
  what fare policy applies to this specific customer
• 2:47
  in this specific situation. So the judgment call to
• 2:50
  the system that really isn't equipped to make it the chatbot couldn't
• 2:54
  generate an answer. It just. Well, it could generate an answer, but what it couldn't
• 2:59
  do was evaluate whether that answer was correct. And that gap,
• 3:02
  the gap between generating an answer and being able to evaluate whether
• 3:07
  it's correct, that's what we're going to talk about today.
• 3:10
  Because, you know, most AI overuse doesn't start with laziness or
• 3:15
  bad intentions on the part of the user. It starts with kind of what feel
• 3:19
  like completely reasonable trade offs. Like, I'll just let it draft this thing
• 3:22
  and then I'll fix it after. This is just for speed. I'll be careful more
• 3:26
  next time about, you know, checking sources for accuracy. I'm sure it's fine.
• 3:30
  Which is similar to the same mental gymnastics as I'll start
• 3:34
  exercising tomorrow or, or New Year's.
• 3:38
  Now, it sounds believable in the moment, but if you do it enough times,
• 3:41
  it becomes a habit that you didn't mean to create. We already talked
• 3:45
  in episode three about governance and the handoff line where support,
• 3:49
  where support ends and where substitution begins. And I have a
• 3:53
  ghostwriting analogy, as usual, that I think nails this. So that's where
• 3:57
  we're gonna go today. But before we get to the ghost writing,
• 4:01
  I wanna make a quick detour into filmmaking because I got to play there in
• 4:04
  the indie film world back in la. And there's a saying in filmmaking that also
• 4:08
  seems to fit in this context of AI. And that saying is, that's okay,
• 4:12
  we'll fix it in post. Which means whatever scene isn't working now that
• 4:17
  we're shooting, we'll work it out in the editing bay. Post production will save the
• 4:20
  day. And the editors of the world are like, yeah, no,
• 4:24
  we're not magicians. We can work with what exists that was actually
• 4:28
  filmed, but we cannot shoot entire scenes that were never filmed.
• 4:32
  We cannot build a foundation that wasn't laid.
• 4:34
  So. So if you half assed it on the set and you were hoping that
• 4:38
  editing is going to fix it, well, you're about to spend a lot of time
• 4:41
  in a dark room being very disappointed. So there's a lot of chatbot
• 4:45
  conversations right now, obviously. People handing rough, underdeveloped prompts
• 4:49
  to generic AI tools, assuming it'll produce something they can quote, unquote,
• 4:53
  fix after fix it in post and end up staring at this,
• 4:57
  you know, generic, mind numbing, rambling kind of not X, but why?
• 5:01
  Crappy metaphor. EM dash filled. I love EM dashes,
• 5:05
  by the way, but they just put way too many of them in the chatbots
• 5:08
  so this kind of like this generic output thinking, so you know what,
• 5:13
  maybe this actually can't be fixed in post. Maybe I should have put in the
• 5:16
  work at the beginning. Maybe I should have built the foundation of the house
• 5:20
  before building the frame. And here's the thing about foundations.
• 5:24
  You can't add them later. You can't renovate your way to a foundation. You either
• 5:28
  laid one and it's good or you didn't. With my book clients,
• 5:32
  the foundation is the work that we do before a single chapter gets written.
• 5:36
  My book blueprint process, which is about figuring out what the book is
• 5:40
  actually about, what the author genuinely believes, what story they
• 5:44
  are uniquely positioned to tell, and why their target readers should care.
• 5:49
  That process takes time and it cannot be skipped. Authors who
• 5:53
  try to skip it end up rewriting their book from scratch later. Believe me,
• 5:57
  that takes longer than just doing it right the first time. Same principle
• 6:01
  with AI, the input work. Your thinking, your point of view,
• 6:04
  your stories, your actual judgment about what belongs,
• 6:08
  that is the foundation of the house. If you skip it and you ask AI
• 6:12
  to build the frame anyway, well, you get a very polished frame with nothing
• 6:16
  to hold it up. So back to my ghost writing analogy. You know, there are
• 6:20
  two different ways to ghost write a book. There's probably like more than that,
• 6:24
  but there's two majorly different ways. The first way, pretty common in the industry,
• 6:29
  is that the ghostwriter meets with the author maybe a handful of times,
• 6:32
  puts a recorder on the table, records some interviews, goes, and then they go,
• 6:36
  go off to write the whole thing and they come back months later with the
• 6:39
  manuscript. Maybe just a few, you know, read throughs in between, but the
• 6:43
  author reads it, they give some notes, maybe a couple rounds of revisions, and the
• 6:47
  book gets published under their name. This obviously works. And there are tons of successful
• 6:51
  ghost written books out there as proof, so I'm not knocking it. But my
• 6:55
  process dating back to 2001, it's always been completely
• 6:59
  different. I work very collaboratively with my clients over many
• 7:04
  months with weekly interviews, back and forth conversations,
• 7:07
  brainstorming, more like co authoring, digging into ideas together,
• 7:10
  figuring out what they actually think and believe through their
• 7:14
  process of trying to put it into words. Because that is,
• 7:17
  you know, thinking through words. Sometimes that's what it takes. You have to talk
• 7:21
  something out to figure out what you think about it. The writing emerges
• 7:25
  from the collaboration, so the book emerges from that collaboration. It's not
• 7:30
  me going off to produce something and the client signing off on it.
• 7:33
  We find our way to the answers together. So in the
• 7:37
  first more frequent ghostwriting model, the author is approving someone else's
• 7:41
  version of their thinking, but in mine, the author is developing their thinking through
• 7:46
  the process. Same book on the outside, very different ownership
• 7:50
  of the meaning on the inside. The way that most people work with
• 7:54
  AI right now, it's more like the first model. They hand things over, they get
• 7:58
  something back, they approve it or they don't, they edit it a little bit,
• 8:00
  they move on. And sure, for some tasks that works fine, maybe like summarizing
• 8:05
  a transcript or cleaning up formatting, generating headline options,
• 8:08
  you can just kind of judge what you like best. But for the things that
• 8:12
  require your actual thinking, your point of view, your meaning, your judgment,
• 8:16
  that handed over model, that's where substitution starts
• 8:20
  creeping in. Because at some point you're not developing ideas with AI,
• 8:25
  you are approving ideas from AI,
• 8:28
  and those are not the same thing. You have one
• 8:32
  big shot to get your story and your message out there in a way that
• 8:35
  actually helps people. With books anyway, one. So when authors defer to what I
• 8:39
  think of as the peanut gallery making edits based on what everybody else in the
• 8:43
  world thinks that they should write versus trusting themselves, that I
• 8:47
  see that as a wasted opportunity. AI, when used on
• 8:50
  autopilot without real input, is a version of deferring
• 8:54
  to the peanut gallery. You're not making the calls, you're letting a system do
• 8:59
  it for you and putting your name on whatever comes out. And again, the outputs
• 9:02
  might be fine, but fine isn't the same thing as one, what you actually set
• 9:07
  out to do. That book that could have changed someone's life,
• 9:10
  the article that could have made someone feel less alone,
• 9:14
  the point of view that could have shifted how someone thinks about their own situation.
• 9:18
  That stuff requires you to take the lead, even if it makes you nervous,
• 9:22
  even if it makes you sweat a little bit. Assistance supports your
• 9:26
  thinking. Substitution replaces the moment where
• 9:30
  you choose. And that's the distinction. It sounds simple until you're
• 9:35
  in the middle of a deadline and the AI version is like right there,
• 9:37
  and it's pretty good and you're tired. And that's where the line gets blurry.
• 9:41
  Air Canada's chatbot didn't choose to mislead
• 9:45
  Jake Moffat. It generated a plausible answer based
• 9:49
  on patterns in its training. Nobody checked whether that answer was
• 9:53
  accurate for this specific situation. The judgment call got skipped and
• 9:58
  someone paid for it. First Moffett, then Air Canada.
• 10:01
  The stakes in your own AI use are probably lower than an airline's legal
• 10:05
  liability, but the mechanism is really the same when the machine makes the
• 10:09
  call and you don't check it. Whatever you said. Whatever it said is what you
• 10:13
  said. Whatever it decided, you decided. But your name's on
• 10:17
  the COVID So, two things this week to try. First, before you use
• 10:21
  AI for something that really matters, write this one sentence down and fill in the
• 10:25
  blanks. AI can help with blank,
• 10:28
  but I am in charge of blank.
• 10:31
  Fill in both of those blanks. And second, for one piece of
• 10:35
  work this week, mark every single place where AI Helped you, and then mark
• 10:39
  every place where you made the final call. If that second list is
• 10:43
  a little thin, or if you can't remember where the final calls were, you might
• 10:46
  want to look at your process. And that's maybe where your authority went. Use the
• 10:51
  help. Keep the authority. All right,
• 10:54
  that's it for today's episode of Human Print. Thanks for tuning in. And until next
• 10:58
  time, use the tool. If it helps, just don't hand
• 11:02
  it the pen.
• 11:11
  And that's it for this episode of Human Print. If you want a
• 11:15
  practical framework for protecting your voice amid the AI Noise,
• 11:19
  please follow the show. And before you go, take a minute and ask yourself,
• 11:22
  as the AI noise grows louder and louder, which areas of
• 11:27
  yourself do you need to protect the most? I'll see you next time.