Latest / Think Fast Talk Smart: Communication Techniques / 341. Think Outside the Bot: Communicate More Effectively With AI
Transcript
- 0:03When it comes to working with AI and improving our communication,
- 0:05we make a fundamental mistake.
- 0:08AI is not a tool, it's a teammate.
- 0:11My name's Matt Abrahams, and I teach Strategic Communication at
- 0:14Stanford Graduate School of Business.
- 0:16Welcome to Think Fast, Talk Smart, the podcast.
- 0:20Today, I am super excited to chat once again with my Stanford colleague
- 0:23and good friend, Jeremy Utley.
- 0:25Jeremy is a three-time guest to Think Fast, Talk Smart.
- 0:28He's an adjunct professor specializing in creativity and entrepreneurship at
- 0:32Stanford, and the author of Ideaflow.
- 0:35His new book is The Human Advantage: What AI Can't Do Without You.
- 0:40Well, Jeremy, welcome back.
- 0:41I always enjoy our podcast conversations almost as much as I enjoy our
- 0:45walks around the Stanford campus.
- 0:46Thanks for being here.
- 0:47Likewise, my friend.
- 0:48Thanks for having me.
- 0:49Yeah.
- 0:50Shall we get started?
- 0:51Let's do it.
- 0:51In your first book, Idea Flow, which is a great book, by the way,
- 0:56you argue that creativity comes from generating lots of ideas.
- 1:00You found that AI-assisted teams can actually become less creative if they
- 1:04just accept AI's first obvious suggestion.
- 1:07How can we prevent AI from becoming an average idea generator and making
- 1:11us intellectually lazy and just relying on whatever we see first?
- 1:16What does it take to have good ideas?
- 1:18So you can start there because the same thing that it takes for a human brain
- 1:21to have good ideas is what it takes for an artificial brain to have good ideas.
- 1:24And by the way, the same biases that prevent humans from having
- 1:27good ideas are what prevent AIs from having good ideas, very simply.
- 1:31There's a couple of things worth calling out.
- 1:33One is the reason most of us don't have better ideas is because we
- 1:37won't let ourselves have worse ideas.
- 1:39What you're actually after is variability.
- 1:41You want a variation in the spectrum of ideas.
- 1:44Ideas like all naturally occurring phenomena are normally distributed, right?
- 1:48Some are stupid, some are great, et cetera.
- 1:51The challenge is if you want to get kind of on the far right end of the
- 1:54distribution, you have to be willing to go to the far left end because you
- 1:59can't have an asymmetric distribution.
- 2:01And I was just listening to a podcast with the CEO of Airbnb.
- 2:06He said, "I have 100,000 ideas a year. 99,000 of them are no good." And I
- 2:12think if you ask the lay person to describe the path to having good ideas,
- 2:15they wouldn't say 99% failure rate.
- 2:18They just wouldn't do it, right?
- 2:19They think, "I gotta be smart," or, "I gotta go to a good school," or, "I gotta
- 2:22wait for lightning to strike." But they don't think of expanding the variability.
- 2:27What's true for humans is also true for an AI, right?
- 2:30So instead of asking the AI for an idea, one simple lever is ask the AI for 10
- 2:35ideas, and instead of only using kind of volume as your only lever, ask the
- 2:40AI for 10 radically different ideas.
- 2:43And if you want to get really fun, ask the AI for terrible ideas, right?
- 2:47And your human brain will actually do kind of the fun part of
- 2:50figuring out what's good there.
- 2:51But I think a lot of the problem basically is just thinking about
- 2:55what I need is one good idea.
- 2:56That's actually the problem.
- 2:57So we need to open up our aperture and look for more ideas, and if we're
- 3:03gonna leverage AI, have it help us come up with those more ideas, and
- 3:06sometimes ask it to do the bad ones.
- 3:08Absolutely.
- 3:09And then just getting back to another statement that you made earlier, you said,
- 3:13"I found that teams settle." That's true.
- 3:16And the reason that it's true is because of a widely known
- 3:19cognitive bias called satisficing.
- 3:21The fact that we satisfice means we're satisfied by that
- 3:24which suffices, so we satisfice.
- 3:27And so what we do is we kind of settle for good enough, and that's fine.
- 3:30The challenge is that satisficing, while it helps us in the majority of our life
- 3:34get through the otherwise overwhelming amount of information and decisions that
- 3:39we could make, where it kind of nips us in the rear, so to speak, is when we're
- 3:44trying to generate a new idea, right?
- 3:47Or a new product or a new service.
- 3:49And so what we know is that this human cognitive bias actually plagues
- 3:53creativity and it plagues innovation.
- 3:56And what Kian, my research partner and I discovered, when we studied teams that
- 3:59we gave access to AI, is the following.
- 4:03Unassisted teams actually worked hard, workshopping, spitballing, going back
- 4:08and forth, devil's advocating, et cetera.
- 4:10Assisted teams often, not always, but often plugged a superficial staccato
- 4:17prompt into an AI, got a pretty mediocre response, and because it's better than
- 4:22the response they could have come up with in that 30 seconds, they declare victory.
- 4:27And they basically satisfice.
- 4:29Of course, because they're not grading the ideas, they don't realize
- 4:31what's theoretically possible.
- 4:33They just realize that the AI did something they couldn't have done in
- 4:36that amount of time, and they assume that there's nothing left to do.
- 4:39Which is to say that left without intervention, AI merely amplifies the
- 4:46underlying existing human cognitive bias.
- 4:49So AI actually isn't helping us in that case, it's actually making things worse.
- 4:53You know, you tell lots of great stories, Jeremy, and I always love when you
- 4:56regale me with some of your stories.
- 4:57Would you mind sharing your story about your daughter and the broken window?
- 5:04Yeah.
- 5:04Why I like this story is because most people, when they think about AI, they're
- 5:08thinking about business and efficiency and productivity and that sort of thing.
- 5:12And I often say I'm not embarrassed to say that AI has helped me be a better parent.
- 5:17So I tell them the story that I live in a really old house, 130-year-old home, and
- 5:22one of my favorite windows in the house is in the front bathroom of the house.
- 5:26It's got this beautiful glass.
- 5:28And as you know, I've got four girls.
- 5:30And the other morning, I hear a very loud bang in the middle of making breakfast.
- 5:34And then my daughter, she yells.
- 5:35She's like, "Dad, there's glass everywhere. Get up here. Come up here."
- 5:38So I go to the front of the house, and lo and behold, there's glass everywhere.
- 5:42And this beautiful Victorian window that I quite prize and adore is
- 5:47shattered all over the floor.
- 5:48And thankfully, nobody's too hurt, but everybody's looking at me going,
- 5:53"How is Dad gonna respond?" Because basically what had happened, one of the
- 5:57girls needed a little personal space.
- 5:59And how did she accomplish that?
- 6:00She decided to slam the door so hard in her sister's face that the window
- 6:04shattered, which is obviously a violation of a few household rules.
- 6:08So the question is, how do I respond?
- 6:09And we all have these broken window moments, and that feeling
- 6:13of, I wish I knew what to do.
- 6:16And so I am not embarrassed to say that AI actually helped me be a better
- 6:21parent to my daughters in that moment.
- 6:23And this is actually part of the core premise of my new book, which
- 6:26is called The Human Advantage.
- 6:28But the basic premise is that the most special thing about AI is
- 6:32actually you, the expertise that you bring to your collaboration.
- 6:37And the great irony I find in this AI era is we have this amazing ability
- 6:42to basically interact with machines.
- 6:43And us humans are the robotic ones.
- 6:47We aren't capable of actually bringing our humanity to the interaction.
- 6:51And so this, to me, it's so important to realize that AI's gonna give you
- 6:56the average of the internet, except if you bring your own expertise,
- 7:02knowledge, wisdom, discernment, taste to the conversation.
- 7:04And as we just discussed, the value of bad ideas, as a creativity expert, I
- 7:09happen to know that in that moment, if I want a good idea, what do I need, Matt?
- 7:14I need a lot of ideas, and I need permission to have bad ideas.
- 7:18And so when I came to the front of the house, I just said to
- 7:21my AI, "I need 20 terrible ways to respond to this situation."
- 7:25And it started advocating for certain torture tactics and
- 7:29involving the authorities.
- 7:31It mentioned expelling her from school, which is funny because
- 7:34we homeschool, as you know.
- 7:35And what's funny is actually the amazing thing about the human brain
- 7:39is you go, "Ah, that's a bad idea," but it's kind of interesting.
- 7:41But when the AI said, "You could expel her from school," I go, "School?
- 7:45Oh.
- 7:47Wait, what if we give her the assignment that her job today is to prepare a
- 7:51lesson for her sisters in the schoolroom about why our household rules matter?"
- 7:57She's uniquely qualified to deliver that lesson.
- 8:00And as a consequence for slamming the door in her sister's face and
- 8:04shattering the window, her consequence was she had to teach a lesson to her
- 8:08sisters about why our rules matter.
- 8:10Far more inventive and effective as a technique for helping our girls be
- 8:17productive and effective and harmonious members of the home than take away
- 8:22the Kindle Fire for three days.
- 8:24Love it.
- 8:25Love it.
- 8:26Thank you.
- 8:27Your story about the broken glass is not only really helpful to
- 8:32understand the idea of how AI can help us and help us by giving us
- 8:38ideas that we would never implement, but get us thinking more creatively.
- 8:43In your book, The Human Advantage, you talk a lot about unassisted
- 8:47living and handcrafted artisanal work.
- 8:51Share more about what that means.
- 8:54Right now, most of us, even though we have access to kind of call it
- 8:58Jetstones age technology, we're still living with Flintstones age habits.
- 9:04So very simply, we have the cognitive equivalent now of what Winston
- 9:09Churchill had a generation ago, right?
- 9:11Everybody knows that one of the things that made Winston Churchill's
- 9:14speeches great was the fact that he had an exceptional secretary or
- 9:18assistant who knew his voice and his audience and his objectives and what
- 9:23he'd said in the last speech and what Neville Chamberlain was saying, right?
- 9:27But most people don't have that.
- 9:28We don't come from means.
- 9:30We don't have the ability to have that kind of cognitive uplift, right?
- 9:33And technology is always democratizing what's been available to the elites.
- 9:37But the problem is we have grown accustomed to unassisted living.
- 9:42Assistance is available to us, but we have habitually learned to not
- 9:47only lean on our own abilities, but also to value doing it all ourselves.
- 9:52That's why I joke about handmade artisanal labor.
- 9:55And for whatever reason, we think that our particular work
- 9:58product, there's some virtue in it being handmade artisanal, right?
- 10:01If I'm a financial analyst, I go, "This is handmade artisanal financial analysis." If
- 10:06I'm a marketer, I think there's some value to handmade artisanal marketing copy.
- 10:10And this is patently absurd the moment you think about what's possible.
- 10:15You would never go to a cardiologist who practices
- 10:18handcrafted artisanal cardiology.
- 10:20But why?
- 10:21Because you want the best medical process.
- 10:25You want the best healthcare.
- 10:27And the doctor who prides herself that she puts her ear on your chest, you
- 10:31go, "What are you doing, doctor?" She says, "Oh, we practice only handcrafted
- 10:35artisanal cardiology here." You go, "I need a real doctor." Because for
- 10:39a lot of professionals today, they don't see there's now, call it a new
- 10:43standard of care available to them.
- 10:45And to me, that's unassisted living, is the best I can do on
- 10:49my own, is unassisted living.
- 10:51Whereas people who have learned to avail themselves of assistance,
- 10:54who have learned, as I said, to reflexively invite AI in, that's Tobi
- 10:58Lütke's phrase, not mine, but those people are becoming more AI native.
- 11:03And rather than it being a last resort, it's the first phone call they make.
- 11:08Got it.
- 11:08So it's attitude and approach that matters.
- 11:12It's taking advantage of what you've got.
- 11:14Yeah.
- 11:14And the thing is, why don't we do it?
- 11:16I mean, there's beliefs below it, right?
- 11:18I mean, if you hand your boss the latest report or piece of analysis
- 11:21or whatever you ought to do, and your boss asks you, "Did AI help with
- 11:24this?" It's almost like an accusation.
- 11:27And what we should do is say, "Absolutely, I availed myself of every possible means
- 11:32of delivering the best possible work." But we don't, and that belies this belief
- 11:37that the thing that I do on my own is more valuable, and that's simply not true.
- 11:42We hesitate to delegate.
- 11:43We have a resistance to assistance, and that actually prevents us from getting
- 11:48the kind of boost that's now available.
- 11:50So you've become poetic.
- 11:52Hesitate to delegate.
- 11:54I like it.
- 11:55I think the core belief issue is that people are simply outsourcing
- 11:59the work to AI rather than doing what you're advocating for.
- 12:01You're not saying outsource it, you are saying intensely engage with it.
- 12:06And I think the stereotype that we carry in our head is that if you leverage AI,
- 12:10you're just letting it do the work, and that's not what you're saying at all.
- 12:13I'm not advocating superficial random acts of AI.
- 12:17I'm advocating proficient fluency.
- 12:21Very few of us are thinking, "I have at my disposal a highly capable
- 12:26teammate who's going to thrive under exceptional leadership."
- 12:29Okay, I want to get really practical and tactical.
- 12:31You talk about reverse prompting, where you ask AI to interrogate your assumptions
- 12:36and identify the missing information.
- 12:38So one of the ways we can get more value out of AI and our relationship
- 12:43with it is to do something like that.
- 12:44Break it down for us.
- 12:46Simple.
- 12:46Add a single sentence to any prompt.
- 12:50Before you perform this task, please ask me three questions to ensure
- 12:56you have all the context you need so that you can do a great job.
- 13:00Excellent.
- 13:00So you're asking the tool to help you give it the information it needs to help you.
- 13:06Yeah.
- 13:07And we realize this when we're working with people, right?
- 13:08I mean, you and I both, we have longtime partners.
- 13:11What percentage of the time does your partner perfectly understand what you say?
- 13:15I mean, if you're humble, you say it's maybe 10%, right?
- 13:18Why in the world would we think that machines are any different?
- 13:22We give an AI a crappy prompt, and we assume it's dumb.
- 13:26It is not a technology problem.
- 13:28It's a management problem, and it's incumbent upon us to be
- 13:32good managers, good leaders.
- 13:33If AI is a thought partner, we got to be the thought leader.
- 13:36And so what you have to do is you can employ all of the techniques you have
- 13:39learned, especially the more experience you have, the better, as a manager,
- 13:43as a coach, as a teacher, as a mentor.
- 13:45And everything you've learned about getting exceptional work product
- 13:49out of this weird intelligence called humans is relevant to AI.
- 13:53And all the tips and tricks and things that you do to get great work out of
- 13:56people, you can now get out of AI, provided you bring those same tricks.
- 14:00And one of the most basic ones is what we just talked about, which is
- 14:03with a human, what do you always say?
- 14:05My door's always open.
- 14:07If anything is unclear, feel free to ask me any questions, right?
- 14:09But I would challenge every listener to the show, when's the last time
- 14:12you gave AI permission to ask you questions if anything was unclear?
- 14:17Most people go, "Oh, I didn't know it could ask me questions." That's why
- 14:20you're getting mediocre, suboptimal output because you assume you're
- 14:24perfectly clear in your prompt.
- 14:26And we don't assume that with humans.
- 14:28Why would we assume it with AI?
- 14:29Excellent.
- 14:30I think that analogy is so powerful to help us really think through
- 14:35how we have to work with the AI and have it help us work with it.
- 14:40You also suggest working effectively with AI as fundamentally a communication skill.
- 14:46Unpack that for us and help us understand how we can better communicate with AI.
- 14:51The ability to bring good management to the table is kind of the ballgame.
- 14:56Most people will take an AI output and they will refine it themselves,
- 15:02and that leads to underperformance.
- 15:03And it's not much of a lift, right?
- 15:04Because you get something that's eighty percent of the way there.
- 15:06You still got to do a lot of work.
- 15:07Exceptional collaborators to AI say, "Instead of refining output, let me
- 15:12refine the input here. What kind of context do I need to give to help the
- 15:16AI do a better job next time?" And very simply, giving yourself that challenge
- 15:21almost always leads to better results.
- 15:23And here's where this is rooted in the research.
- 15:25There's a concept called theory of mind.
- 15:27Are you familiar with that?
- 15:28Yes, absolutely.
- 15:29It's your awareness of what you're doing and how you think about things yourself.
- 15:33Yeah.
- 15:33Specifically, what the other person is thinking.
- 15:37Someone who has a well-developed theory of mind is thinking about,
- 15:41"How is this person perceiving me?" Or, "What are they thinking?
- 15:43What are they wondering?" And it turns out, this is very cool and
- 15:46very unexpected, folks who have a well-developed theory of mind do
- 15:51better with AI than folks who don't, adjusted for all other variables, okay?
- 15:55To say nothing about their ability to collaborate even with other people.
- 15:58But why would that be?
- 15:59Because you're thinking, "What is this other intelligence thinking?" And you
- 16:03could say it's a proxy for empathy.
- 16:05It's not exactly the same thing, but the people who go, "Why would
- 16:09it have responded that way?" That is a profoundly communication-centric
- 16:14way of thinking and understanding what was it about what I said that
- 16:18made the AI go in that direction.
- 16:19And so, for example, my dad is an attorney, and sometimes he'll send
- 16:23me a text message saying, "Hey, can Claude do XYZ?" And one day, he's got,
- 16:26like, all of these jury testimonies or depositions, and he's got them in a
- 16:30CSV file, and he needs to convert it to some other format, and he said, "Hey,
- 16:34can, you know, can Claude do this?"
- 16:36So I go to Claude, "Hey, can you please convert this?" Instantaneously,
- 16:39Claude's like, "Here you go.
- 16:40Download the file here." But then I thought, "You know what?
- 16:42I should open this file," right?
- 16:43Open it.
- 16:43It's like a single line with like a thousand columns.
- 16:47But here's the thing.
- 16:48What do I do there?
- 16:49What you do when AI disappoints you is hugely important.
- 16:54I said, "Hang on.
- 16:55What about the way I communicated with this AI led it to do that?
- 17:00And how might I change how I communicate?
- 17:02And what are the words I'm using?" But I went back to the AI and I used
- 17:06a simple trick called chain-of-thought prompting, which basically, tell
- 17:09me how you'll think about this problem before you solve it, right?
- 17:11This is well known in cognitive science to help humans make better decisions.
- 17:16So I literally went back into the same chat with Claude.
- 17:19I edited my prompt and I said, "Hey, before you make this file conversion,
- 17:23tell me how you think about problems like this." And then instead of saying, "I've
- 17:27done it," it answers my question, which is now not do the thing, but explain
- 17:31how you do the thing before you do it.
- 17:33And so to comply with my request, Claude says, "I would look and see
- 17:36what kind of file type this is.
- 17:38I would make sure that I have the conversion tool required to do that.
- 17:41I would look for these kinds of headings.
- 17:43I would then execute with this rubric.
- 17:46Here's your file." And I opened it, and it's a three hundred line
- 17:50CSV with all of the data in it.
- 17:52And all I did was use a communication tactic called tell me how you
- 17:57think about this before you do it.
- 17:59What I love about this, Jeremy, is it really does highlight
- 18:03effective communication skills and how they can be applied to AI.
- 18:07Thinking to yourself, in this moment, this communication, what can I say
- 18:11that will help our future interactions be better, is a tremendously powerful
- 18:16tool for when we communicate.
- 18:18This idea of theory of mind, where you're thinking about how would this person
- 18:22think of this, incredibly valuable.
- 18:24In fact, it's the most frequently cited advice that comes from this show.
- 18:28And the other thing that you said that fits right in is think about how does
- 18:32that person tend to process information and ask them, you know, how do you do it?
- 18:37Well, Jeremy, I knew this would be a fantastic conversation
- 18:40'cause they always are.
- 18:41I want to end like we always do.
- 18:43I got three questions.
- 18:44One I'm gonna make up just for you.
- 18:46Two you've answered before, but I'll see if they've changed over time.
- 18:49You ready for that?
- 18:50Yeah.
- 18:50Go ahead.
- 18:50So every chapter in The Human Advantage, your new book, ends with
- 18:53a try this now, and I love that.
- 18:55My book does try this now as well.
- 18:57But you end with a try this now prompt to make the next step less intimidating.
- 19:01If you could give our audience one immediate, surprisingly fun prompt
- 19:05or exercise to try to write with AI that might fundamentally change
- 19:09how they view it, what would it be?
- 19:11So one of my favorite ones, and I just put this in the newsletter,
- 19:14so it's top of my mind, is consult with three characters: an alien, a
- 19:18toddler, and a retired sea captain.
- 19:21And you basically tell the AI, "For whatever you're doing, I'd like for
- 19:24you to embody three distinct roles: an alien, a toddler, and a sea captain.
- 19:28Would you please consider my request from each of those perspectives and
- 19:32stay in character to deliver your assessment, output, analysis, et cetera?"
- 19:37Almost always, it will unlock a perspective or an angle that is
- 19:43profound, and it just perfectly illustrates how much words matter.
- 19:47Because all of these techniques with language models are based on things
- 19:51we know about humans and the small language model on our heads, right?
- 19:54So when you invoke certain personas, it's called the attention mechanism
- 19:59of the model, but it's very similar to our own neural networks, which is it
- 20:03influences the kind of information that gets brought to bear on the problem,
- 20:08and it's incredibly valuable, and the word that got me to say that was fun.
- 20:12It's incredibly fun.
- 20:13I love the toddler.
- 20:14I think that's fantastic.
- 20:16Question number two: Who's a communicator that you admire and why?
- 20:20A communicator I admire, Nate Bargatze.
- 20:24I admire him because his comedy is clean, and it's an incredibly difficult thing to
- 20:30do to be funny without resorting to crass or sexual or racist kinds of commentary.
- 20:40And I also, I love the process by which he has arrived at his now, you know, kind of
- 20:47lovely destination, which is he had to opt out of a lot of the conventional pathways.
- 20:52He knew, based on his values and what the kind of work that he wanted to put in the
- 20:57world, he could not accept all offers.
- 21:00There were things he refused to yes and.
- 21:02And to me, it's an incredibly principled stand.
- 21:05It made it way harder for himself, but I think the proof is in the
- 21:09pudding, and his craft and his success are a testament to the value
- 21:15of choosing the road less traveled.
- 21:17I really like his work, and I see exactly what you're saying and the way
- 21:21he had to get there Final question.
- 21:24What are the first three ingredients that go into a successful communication recipe?
- 21:29Successful communication.
- 21:31I think care.
- 21:32I think being loose, I would say.
- 21:35I've never communicated in a way that I'm proud of when I'm really uptight.
- 21:38That's just my own experience.
- 21:40I think you do a great job of cultivating an environment where
- 21:43people can be loose here on this show.
- 21:44And then the third, I mean, obviously listening.
- 21:48What you're saying is probably less than half of the communication.
- 21:51We have two ears and one mouth for a reason, right?
- 21:54And I think it kind of highlights some of what we discussed today
- 21:56about considering the audience.
- 21:58And you learn a lot about what's landing or resonating when you
- 22:02shut your mouth and open your ears.
- 22:04Care, looseness, and listening.
- 22:06And I really do like the loose idea.
- 22:08You have to be in the moment, able to be agile.
- 22:12When you're tense and locked into something, it gets in the way.
- 22:14And thank you for the compliment.
- 22:15I do try to make this a place where people feel comfortable.
- 22:17Jeremy, tons of valuable information, tons of passion.
- 22:21We all need to be pushing ourselves to be more creative in our work and leveraging
- 22:28AI as a teammate to help us do that.
- 22:30Thank you so much for your revisiting us, and best of luck
- 22:34on your book, The Human Advantage.
- 22:35Thanks, my friend.
- 22:36Appreciate you having me.
- 22:39Thank you for joining us for another episode of Think
- 22:40Fast, Talk Smart, The Podcast.
- 22:42It was awesome to talk once again with Jeremy.
- 22:45One of the big takeaways I had is that getting more proficient and better at
- 22:49using AI is just like getting better and more proficient in our communication.
- 22:54It's about thinking about the other person, in this case, the AI.
- 22:57It's about thinking about how would you like the next interactions to go.
- 23:01That empathy, that thinking about the other perspective
- 23:05really makes a difference.
- 23:06To hear more from Jeremy, listen to episode 134, where he's joined by
- 23:10his research colleague, Kian Gohar.
- 23:12This episode was produced by Katherine Reed, Ryan Campos, Alex
- 23:16McCarthy, and me, Matt Abrahams.
- 23:18Our music is from Floyd Wonder.
- 23:20Please find us on YouTube and wherever you get your podcasts.
- 23:23Be sure to subscribe and rate us.
- 23:25Also, check out our enhanced English language learning content for those of you
- 23:29interested in learning more about English.
- 23:31We have ELL content for each and every episode.
- 23:34Go to ell.fastersmarter.io.