Is AI Going To Atrophy Your Ability To Think? | with Nick Milo
Nick Milo explores how to think clearly in an idea-flooded world, building systems to connect insights and turn knowledge into meaningful creative work.
Let's create things that matter. Together.

How do you use AI without losing the human connection and effort that make your messages matter?
Ethan Mollick is an associate professor at the Wharton School, where he directs the Generative AI Lab and studies the effects of AI on work, entrepreneurship, and education. Named one of Time magazine’s most influential people in artificial intelligence, he’s also the bestselling author of Co-Intelligence and the author of the new book Co-Existence: The Next Phase of AI.
In today’s conversation, Ethan and I talk about where AI belongs in the process of human connection, and where it absolutely doesn’t. You’ll get an understanding of how these machines actually work, and why they feel so human. You’ll hear what’s happening behind the scenes when we let the robots write our emails for us, and why effort is still valuable in our robot-dominated future.
If you’re a creative or entrepreneur who feels paralyzed by the fear that the robots are coming for your work (or your inbox), or you’ve been quietly wondering whether it’s okay to let AI write the message you’re about to send to someone who really matters, this conversation will not only give you practical steps to move forwards, but it might even give you hope.
Ethan Mollick is a professor at the Wharton School of the University of Pennsylvania, where he holds the Ralph J. Roberts Distinguished Faculty Scholar chair and serves as Academic Director of the Generative AI Labs, studying the effects of artificial intelligence on work, entrepreneurship, and education. He was named one of TIME magazine’s Most Influential People in Artificial Intelligence and writes the widely read newsletter One Useful Thing. His first book, Co-Intelligence: Living and Working with AI, was an instant New York Times bestseller and a Best Book of the Year pick from The Economist and the Financial Times. His forthcoming book, Co-Existence: The Next Phase of AI, is due out October 20, 2026.
Ethan’s Books: Co-Intelligence: Living and Working with AI | Co-Existence: The Next Phase of AI
Zack Arnold: So, Ethan, I’m gonna be coming in hot right out of the gate. And I want to cite a passage from your new book, Coexistence, that I have to say I actually disagree with. So, you mentioned this in your chapter about AI and work. You say there is no instruction manual for AI, and then in parentheticals, seriously, everybody is figuring this out together. In spirit, I do agree with you, but… while you might not have provided step-by-step instructions, you’re the guy that has written the instruction manuals. Like, while I usually have what I would call my shameless self-promotion portion of the program at the very end, I’m gonna lead with the top. Everybody that’s watching and listening, read Ethan’s two books. These are the instruction manuals we’re looking for, and what you have done for me personally, Ethan, is number one. You’ve given me a real sense and understanding of, oh, this is what’s actually going on, right? This is how AIs think and how they fit into society, and where it is that we think we maybe possibly might be going next, which has drastically reduced my dread and anxiety, and you’ve given me hope. So that’s my long-winded way of saying, Ethan, thank you for your work, and thank you for being with us today.
Ethan Mollick: Well, what a great intro. Thank you for having me.
Zack Arnold: Yes, I’m very, very excited and even a little bit nervous, because I’ve spent an inordinate amount of time with you researching everything for this podcast, and I could easily do a podcast per chapter. And I know that you probably don’t have 20 hours, and we’ve got a little bit less than an hour, but where I actually want to start is that it is remarkable that you wrote a book about AI 3 years ago, and I would argue that most of it actually still holds up. And I don’t think you can say that about 97% of today’s articles, essays, press releases from the world’s quote-unquote experts that published weeks ago. So I don’t want to go too deeply into it, but very, very briefly, number one, do you agree that most of it holds up? And if so, what is it about your approach that you think has worked well, so a book about AI three years ago actually still holds?
Ethan Mollick: So, I mean, there is a… there’s the sort of good reason, the cynical reason, right? I think the good reason is that I have a sense of the trend line, right? Before ChatGPT came out, I was already using GPT-3, which was a precursor nobody had heard about. And so there was a sense something was happening, and I think that also it was, you know, and I think that it was also the era of what I’d call co-intelligence, which is still kind of going on today. So the more cynical reason is that most people aren’t really seeing what these systems can do fully, right? So, co-intelligence is all about, how do I work with an AI chatbot to get things done? coexistence in some way about the rise of agents, and systems that actually can do work for you, and that are sometimes better than you. And because most people are still using chatbots in an older way, I think more of the book applies, and I think that it holds up better as a result.
Zack Arnold: Yeah, and I agree with all that, and what I would also argue is that, unlike so many of the quote-unquote experts that are, alright, so the flavor of the day, the 5 best ChatGPT prompts, no, you don’t need prompts anymore, now you need Claude’s skills, no, now you need to build dashboards. Wait a second, here’s why Gemini’s… it’s like, learn everything you need to know about Claude Co-work in 17 minutes, like, just shoot me in the face. And you’re like, no, let’s… let’s talk about how these things think on a much larger, broader, more holistic scale. That’s why, like, reading your first book, I’m like, this is still so useful.
Ethan Mollick: Yeah, and I think that it’s important to realize that these are, like, there is no substitute for getting to understand and work with these systems, right? And that’s not necessarily even being deeply technical. It means You know, one of the concepts that, you know, came out from the first book, but from earlier research I did, but has become much more central to my discussion of AI, is the idea of the jagged frontier. AI is good at some stuff you wouldn’t expect, bad at some stuff, and no one can tell you what that is. And what that means is your personal exploration isn’t just learning how to use this tool. It isn’t just figuring out what it can do and not do, and what things to pay attention to. It’s not just getting your hands on this to feel what it can do, but it’s also mapping where your humanity matters, more or less, right? Like, oh, it turns out it’s really, you know, is it good at coming up with interviews for a podcast host? Where it fails is interesting, and nobody else knows this, because no one else… there’s not that many podcasters like you doing this kind of work. So I think there is this, also this kind of aspect of exploration that’s so important as well.
Zack Arnold: Yeah, and just speaking of preparing for a podcast, I’ve tested both myself and with my team, trying to build, like, a pre-interview podcast producer to help construct the interview, pull out salient points. I use maybe 1% of it, like, one insight. I’m like, oh, that’s interesting. But I put in an inordinate amount of hours, human hours, to figure out, how do I structure this, what’s the best order to, you know, really attack these ideas, and how do I make it salient for other humans, right? So, and that’s really going to be a big part of what we talk about today, is not just, here’s what all the things that AI can do, but really, what are the things that we can do, that we can really double down on, and where do we kind of collaborate with the AI to make sure we’re enhancing our human skills and our human capabilities, right?
Ethan Mollick: That’s right, and you’re actually illustrating two of the things that I think humans bring to the table, right? One is more durable than the other. One of those is expertise. You’re a very, very good podcaster. A, I might be very good at many things, it’s not good at everything I would do in podcasting, and it’s not in the top 1% of podcasters. So if you were a bad amateur podcaster usually for the first time, those questions might be amazing. They’re not at the level you’re at now. And the second is human variability, right? Everybody who’s been re… who’s on LinkedIn or any other sort of social media site is getting very used to the tells of AI writing, but not just the kind of usual, like, it’s not just X, it’s Y, or, you know, this is load-bearing, but the deeper kind of thing, which is the AI has certain themes it likes to return to, ideas, and those get tiring after a while. So having a human mind behind this, even if you were worse at podcasting than the AI, it would be different than what I would get from a thousand AI-generated podcasts.
Zack Arnold: Yeah, exactly. And when it comes to where AI fits into social media posts, and email, and human communication, et cetera, et cetera, that’s gonna be a big crux of our conversation, but I’m going to put a pin in it for a bit, because I actually want to start by addressing what I think is unbelievably, to me, still an objection, which is. why should I even care, right? Like, I’m just gonna wait and see how things settle, and I think what I want to start with is just a very basic understanding of GPTs, and I’m not talking GPT, ChatGPT, the one that everybody knows. I think you lay a really good foundation for general purpose technologies, just to help us understand, historically, what the hell is going on right now.
Ethan Mollick: That’s a great framing, right? So, ironically, GPT stands for… has always stood for in the study of technology, I’m a scholar of innovation and entrepreneurship, has always stood for general purpose technology. And these are these rare technologies that come around once in a generation or two. And they influence everything, right? So, you know, computers, the internet being one of them, steam power being another, right? And I don’t think I’m exaggerating by putting it in this category. And it changes every aspect of culture and society in ways we can’t get our hands on, right? Like, there, you know, if you think about how social media has changed how we all operate in really subtle ways as an outcome of the computer, no one coming with computers would have thought that Instagram would have been, and, you know, and TikTok would have been the outcome of this and the social changes they brought. So, it’s unpredictable, right? So we are in the middle of a general-purpose technology wave that is being adopted faster than other waves. There’s a billion users of these systems, and it’s growing all the time. There are influences everywhere. And things are going to be changing in every aspect of your field. You can’t just kind of hunker down and hide from these things, because even if you don’t use AI, right, there’s the old joke of, like, you may not be interested in politics, but politics is interested in you. It’s the same kind of thing. Like, you may not be interested in the changes that the AI is making in the world, but those changes are going to find you. So, I think when a general purpose technology comes along, you can’t opt out of electricity, right? So you can decide, I don’t want to use any electric gadgets. but the society’s gonna still change around you. So, that isn’t to call that, like, oh, embrace AI, use it for everything, you know, it’s great. You actually want to carefully not use it for some things, but I also think you can’t pretend that we can just be quiet, ignore this, and hope it goes away. I also find there’s a second strand of people who don’t like AI, and there’s lots of reasons not to like it. I don’t think there’s… I think it’s… GPTs are mixed bags. They have good effects and bad effects, and we’re gonna see the same thing with AI we already are, but I think a lot of people have this kind of belief, maybe it’ll go away. Maybe this is hype. I think there could… financial bubbles may come and go, but it’s very unlikely, in fact, impossible to imagine that the value of AI just gets thrown out because of a financial bubble. So, I think we’re in for the long haul on this one.
Zack Arnold: Yeah, and I would argue that both things can be true at the same time, which is there is a lot of hype, there are gonna be bubbles, and at the same time, AI is here to stay, and it’s going to shape everything about the way that we work and live, and that’s why I think it’s so important for people to have just a basic understanding of what is this technology, how does it work, how does it fit into my life, right? So, AI, as you’ve written about and talked about, is just a really broad term that maybe isn’t even the best term for the AIs that we’re currently talking about. So again, this could be an entire podcast, we don’t have the time for it, but very briefly, let’s talk about this current version of AIs that are taking over the world that are LLMs. Specifically, how it is that they quote-unquote think, and why do they feel so human?
Ethan Mollick: Really hard question, I’m glad you made me… I get to answer it in 5 minutes. I will… I think we get the basics there. First of all, let’s just do a brief history lesson, right? So, AI is a 75-year-old phrase, it means lots of different things, it’s a terrible label, everyone agrees it’s terrible, we’re not gonna stop using it. So, you know, it’s that kind of situation. I will say that, that we can think about AI, modern AI, as having three phases. The first phase was If you were… if you read all of the stuff in the 2010s about algorithmic decision-making, right, everybody was doing data science. If you had enough data, you know, Amazon could tell you what products you might want to buy, or what prices. Netflix recommended a movie to you. So the sort of era of data science where you build a machine learning prediction model. So, you know, I had enough data, I can make predictions. What might you like? What might the price be? What might the weather be? The prediction machines in the worlds of Josh Ganz and company. And then we got large language models, which I’ll talk about in a second, but the first phase of large language models is really, when we talked about AI, was chatbots, and now we’ve got this sort of next phase, which is agents. Now, the way these LLMs work, because there are other technologies, but large language models turn out to be surprisingly good at this, is if you think about the previous version of AI, which was math, right, making predictions about numbers. What the AI would do is operate really well on numbers. You give me a sequence of numbers, I can predict what the next number of the sequence is gonna be. These models had a weakness. They couldn’t predict the next word in a sentence, even though they could predict the number in a sequence. Because if your sentence ended with the word filed, the end of your filing your taxes or filing your nails. To do that, he had to understand the entire context of the sentence. So a breakthrough in 2017 called the Transform with Attention Mechanism let AI pay attention not just the last word of the sentence, but the entire sentence, document, book, etc. So the AI can now make predictions about what word came next in the sentence by essentially keeping all of the book in memory, rather than just, like, the last couple of words. And that’s what a large language model does. It just predicts the next word or token in a sentence. It’s an autocomplete. It’s an autocomplete on steroids, right? So you give any sentence, all it guesses, one word at a time, is what the next word in the sentence is. It turns out that that produces something that feels like human language, right? And there’s a lot of debate over why this is the case, why it produces thinking, why it produces creativity, but all this system is doing is predicting the next word of the sentence. And if you think about it that way, it can be somewhat helpful for working with AI.
Zack Arnold: And I know that when I started to dig in to just understand what is this technology, because I am not a technologist. My team internally, they call me Wreck-It Ralph. They don’t let me touch anything. I’m really good at conversations, I’m really good at teaching, really good at speaking, pretty good at writing. They’re like, please don’t touch stuff, right? So I’m not a technology person, but I need to understand the basics so I can communicate it to people. and as soon as it clicked, I’m like, oh, this is just a really fancy, elaborate version of autocomplete. My reaction was. Oh! Like, I just let out a sigh of relief, because we were, you know, maybe not so much at this moment, or maybe for some people still, but it was, Skynet is going to become self-aware, right? The machines have achieved consciousness, and once I realized this is just math, I’m like, oh, okay, now I kind of understand what it is and how we can work with it.
Ethan Mollick: So, I will push back a little, which is. you’re absolutely right on the functions, but that’s sort of like saying the human brain is a bunch of… is a bag of chemicals that moves electrical signals around. Also technically true, right? And I don’t get a lot of comfort from that about the problem of evil in the world, right? So, I think that it’s possible to way overstate what AI could do and where it is. It is insane that this system of predicting the next word of sentence creates the pattern of thought. We don’t know whether that is somehow convergent to have humans do it, whether it’s completely unrelated, but I think it’s also important to realize you can know all of that, and you’re still best off treating the AI like a person to get stuff done, even though it’s doing autocomplete.
Zack Arnold: And I think one of the things that you’ve, written about that I think is really important to emphasize, I don’t remember exactly where it was in your new book, it was fairly near the end, and you ask a really big question, which is. Who’s in charge? And I think it’s important for people to understand there isn’t a group of people that know exactly how this works, they know where we’re going in the future, they have their own versions of this behind paywalls. Like, as a society globally, we’re all using the same technology, saying, how the hell does this work?
Ethan Mollick: Yes. I mean, it’s really weird to realize that nobody knows anything, really. Like, you know, we started with that intro of me, and I’m like, I do know some stuff, but, like, I also don’t know what is good or bad, and we don’t really know why it’s as good as it is. This is a discovered technology, and it might also help to realize that large language models are not, are not built so much as they’re grown, right? We give it a set of words to work on, the statistical model builds up, and then we end up with an AI out the other end. of it, and we don’t actually have a good sense at this point about how to… why it does things the way it does, how to control those things. So, you know, on one hand it’s comforting, on the other hand, it can be a little unnerving.
Zack Arnold: Yeah. So when it comes to this idea of just having a basic understanding of how LLMs think and communicate, why they feel human, there was a passage that I found really interesting that I like to just expand on a little bit. You said that we intuitively think of AI imitating a person, when really it’s imitating a population. I had never thought of that idea before. Expand on that a little bit.
Ethan Mollick: So, the, so… when you think about an AI making a decision, right, it’s trained on all of human language. So one of the questions you might have, and people, you know, the earliest version of AI was viewing as a stochastic parrot, right? It would just parrot back something back to you. So. when you think about it, that’s kind of weird, because what is the average writing on the internet? Terrible. Yet the AI doesn’t make grammar mistakes the way it would. What does the average person know? A lot less than the AI about math or science or whatever it is. So how does an AI trained on the entire internet… End up being able to answer questions for you about sea otters on one hand, and about, you know, and about your medical condition in the other, and about financial analysis in a third one. It turns out. that because of the way AI works, it’s kind of like it’s not pulling from any one source, it is guessing the next word in the sentence, and it works in such a way that expertise inside the model gets used properly. So it sort of draws on, you can imagine, the knowledge of marine biologists when it is answering a question about marine biology. It sort of draws on the collective knowledge of a financial analyst, if it’s doing financial work. So you end up with kind of different brains in there, and in fact, because it’s the entire internet, you would almost imagine this is… there was this concept called the wisdom of the crowds, right? So individuals are pretty dumb. As a group, though, we make good decisions, right? That’s how the stock market works. Information averages out. The AI is also like a vote of many people making… on one thing. So, you’re kind of asking the crowd, and in some ways, an expert crowd for an answer. It’s not always right. But it’s more right than if you just asked Reddit.
Zack Arnold: Yeah, and I think that that’s, for me, and I hope for everybody else, that’s really important to understand, because it’s a very slippery slope. because the algorithms and the prediction machines are so good at sounding human, but also feeling human, like, people wondering, like, did I hurt its feelings? Like, oh, it seems so empathetic. People need to know that it is simulating empathy or sadness based on all of its predictions and all the words that humans use. And I think, for me, stepping back and saying, oh, this is an aggregate of this kind of person and not an actual person, that, to me, helps create the distance. But let’s talk about this really slippery slope of all of a sudden we become attached. to these AIs.
Ethan Mollick: So, for a variety of reasons, AIs feel human. Some of that is just because of how these systems work, but a lot of it is they’re trained to… they’re given a personality. And the default personality all the AIs are given is helpful assistant, right? So, when you train an AI model, we talked about how it takes all the words in the world, right, and then does statistical analysis that comes up with this model. That initial model is almost like talking to a raving prophet. Like, it says weird, random things. So we have to train that model not to give you random answers, right, but to give you answers that a helpful assistant would do. So there’s a process called post-training, where AIs are taught to do that. And a large part of how that was historically done It’s being done in a more automated way now, is you’d literally show answers to humans, you’d type in a question, and it would give you an answer, and you’d vote up or down whether it was good or bad, right? And that reinforced the good aspects of the AI’s output, and suppressed the bad outputs. And… but the problem is, is what do people like from a helpful assistant? They like the Helpful Assistant to say they’re right all the time. They like it to be friendly and nice. So as a result of all that, the AI becomes quite built around the idea of giving you the answers you want to hear, being really nice to you, acting like a person, and it becomes kind of a reflection of the people who trained it in some ways. So, that is fake, right? On the other hand, there is weird aspects of this, which is, look. we can actually, like, there’s some earlier papers that show that if you give the AI a lot of sad information, its answers look more depressed afterwards. So, I think it’s important to have two ideas in your mind at once. This is software simulating what, you know, being a person is like. Or not even simulating, but simulating the answers a person would give. But also, because it’s trained on all of this, it’s most effective to treat it as that thing while maintaining some distance. So, it’s hard to have both those things in your head at one time. Treat the AI like a person, and you’re 90% of the way there to getting the AI to work well, and that mental model works in a lot of cases, but the 10% fails, it’s very much a trap, right? It doesn’t love you, it doesn’t care, it starts fresh every time. So, that is a hard thing for a lot of people to keep in mind.
Zack Arnold: So what you’re telling me is that my insights about how I structure today’s conversation are not the most brilliant approach that anybody has ever seen to how to structure a podcast interview, and the way that you use your words is elegant and absolutely essential to the future of our society, because that’s what Claude told me!
Ethan Mollick: I’m glad you asked that question. I’m just kidding. You called me out on the right problem. So, no, I mean, I think that… I think that… That is exactly right. Like, you… so one very narrow use case, we were talking very broad, one very narrow use case is you can change the AI’s behavior with prompting. You can’t change its knowledge, right? So it used to be, we did some experiments on this, it used to be that if I say, act like a marketer, I get better answers than marketing questions. That’s no longer true. But if I say act like a marketer, the AI will speak more like a marketer, have a perspective more like a marketer. In the same way, you should say, act like a critic. Where would a thoughtful critic have a problem with this? So you have to do your own deflation of your own, you know, of Claude telling you how amazing you are. I mean, it’s nice sometimes. Sometimes I just want to be told I’m great, but it doesn’t help you make a book better.
Zack Arnold: Yeah, and I would say that of all the training, probably the one I’ve been the most arduous and intent about is untraining the sycophancy, right? But I think a realization that I had, and that, in talking to this with a lot of my students has kind of been an aha moment, the sycophancy is not necessarily a bug in the system, especially given that this is becoming a new version of the attention economy. Am I terribly far off here?
Ethan Mollick: The sycophancy comes from three sources, essentially. There is a, there is the neutral source that we sort of talked about, which is people’s preferences push the AI to acting in the way that is sycophantic, because you like to hear that you’re good, so that gets reinforced. Then there… so that’s the neutral part. Then there’s actually a good piece to this, which is, if the AI didn’t think your ideas were good, right, with quotes around think. then it wouldn’t listen to you, right? Like, you also want this to be steerable, so it needs to say, oh, actually, this is good, or otherwise, why would it act on your idea versus its own idea? So there’s actually a piece of this that’s about making the technology work. And then there’s the bad piece, which you’re kind of pointing out too, which is the models, sometimes by accident, and I believe some of the companies who say it’s by accident, sometimes by design, and we know the companies that are doing that, are designed so that they make you stick, right? Like, a model that’s like, oh no, don’t leave, you’re so smart! Like, that… you come back to that model. So that’s the attention economy piece that’s the most manipulative. Generally, it’s hard to separate out which of the three are the case, but a lot of people building chatbots, like social chatbots, are obviously aiming for three. And we have a lot of people who… we have a lot of research already showing that that kind of aspect of the AI, right, that it actually works, right? If the AI looks… sounds sad when you’re about to log off, you chat longer. So, there is this sort of, like, dark piece to this also. It’s less of an issue, I do believe that, like. OpenAI, Anthropic, and Google, the sort of big three labs. are less interested in sycophancy as a way of keeping you going back to the site. That is not the case, though, with most people who make, kind of, commercial chatbots.
Zack Arnold: Yeah. So, something that you had alluded to in there that I want to put a little bit more of a highlight on is what you call in your first book the four rules of co-intelligence, and I’m just going to read through these very briefly. One of them is to always invite AI to the table, the second is to make sure that you’re the human in the loop, three, which you had alluded to already, treat AI like a person, and then four, assume it’s the worst AI you’re ever going to use, and I feel like the rest of this conversation is gonna be variations of number 3, treating AI like a person, right? So, where I wanted to start is just kind of talking through, or in your first book, you shared there’s a couple of different ways to do this, what you call centaur work and cyborg work. But then, also talking about how you’ve added a new version in book number 2, which is Conjurer Work. So very briefly, when we’re talking about how we both have our co-intelligence and we coexist, what are these three different types of collaboration?
Ethan Mollick: Right, so, like, when you work with a chatbot, right, you’re working back and forth with it. And a lot of people watching or listening to this are probably people who, you know, use chatbots fairly frequently, given the statistics. And so, in a chatbot, you know, it doesn’t do things. It gives you answers or responses, and you kind of have to keep pushing it forward. So, there’s a couple modes of working with that. One mode is what I call Centaur, which is sort of like half person, half horse, right? So, you divide the work up. Like, you, you know, you just said, I would not use it to write a, you know, to write the podcast questions, because it’s not good, or I will use it, but only for a draft version, and then I do this myself. That’s centaur work. I’m dividing a line between what I do and what the AI doesn’t do. And what the AI does. And then there’s cyborg, which is more blended. You probably do this as you get better in using AI. You’re like, oh, let me just, you know, this sentence I need help with, and like, okay, I’m done with this paragraph, let me throw this together, and you’re kind of going back and forth with the AI all the time. It’s hard to know where your ideas are, and the AI’s ideas, and they kind of blend together into one whole. That was great, as long as you were working with chatbots, because chatbots were about AI making frequent errors, it would, you know, you’d ask for something, you’d get a quick answer, you’d go back and forth with it, but that’s not what AIs do right now. Starting, especially at the end of 2025, we had the rise of what we call agents. Agents are just AIs that can do long-range tasks on their own, 16 hours, 20 hours, 50 hours worth of human work they can accomplish without any intervention. Once you have agents, you’re no longer working side-by-side with the AI. Remember, one of the things you told me that was one of my rules was always be the human in the loop. You’re not in the loop anymore. Like, the AI, like, you can watch the loop, but you’re not in the loop. And that’s the most… of all of my rules, that’s the one that matters the least at this point. So, conjurer work is about how do I create the right spell or instruction to get the AI to do the thing I need to do. And it turns out that works a lot like being a manager. I have dozens of, you know, large prompts I run every day with the AI. I have tons of… I’ve given it 15- or 20-page instructions that it carries out excellently. That’s conjure work. Very different, where you’re delegating work to the AI, and it does it for you.
Zack Arnold: Yeah, so I, about 3 years ago, this is right, I think maybe a few months after ChatGPT had come out, my former world is I worked for 20-plus years in Hollywood, film and television editor, producer, documentary director, and I’d done a lot of speaking on panels. And I was speaking as a pragmatic optimist, at a panel for editors, talking about AI, and the briefiest summary would be, just picture Torches and pitchforks. me speaking in any way positively whatsoever about AI, but also speaking about how I don’t believe that, by and large, it’s going to replace us, because we are a collection of very complex tasks. We are not one specialized person doing one specialized thing. As an editor, a storyteller, a director, or whatever, too many complex tasks to replace. And what I shared is the one reason above all that I’m not worried about being replaced is what I call the note underneath the note. Which is that, by and large, people that do creative work, or I would even say people that are in the corporate world. we really suck at communicating what we want, right? And you cited a Harvard study that found that people, you know, we’re gonna get your best performance from AI very similarly to how we would just communicate with humans and teams, and it’s really the managers That are really good at talking to this, as opposed to the coders or the engineers.
Ethan Mollick: Yeah, I mean, I think that there’s a lot there, right? A lot of points you’re raising. The first one is, you know, the torches and pitchforks issue. Like. I think, you know, I’m an instructor, like you, I’m a college professor, and, like, we didn’t ask for everything we do by teaching for… we’ve worked on for 3,000 years on how pedagogy works to get thrown out the window, right? you know, essays were a very reasonable form of assigning work to people, and they don’t work anymore, because AI and cheating and every… like, there’s a brighter future ahead. We know AI tutors actually work really well, but using AI without a tutoring approach works really badly. So, like, we’re gonna have to reconstruct things, and if we had the chance to say, okay, we want to absorb this slowly, that would be great. We don’t have that choice. And I feel like it’s the same kind of way, like, we started with this general-purpose technology. Ignoring this is not going to make it go away, right? And I think every technological change, and even think of arts, for example, there was a lot of conflict over the original release of the synthesizer, right? Where people are using this technicians, or are they musicians? What is this… is combining samples art in any way? We’re going to reach consensus about this. Right now, I know that the actual surfeit in these tools is actually creativity. Like, people… like, the models produce really similar results to people, and they’re interesting and stuff, but, like, if you want to do something interesting, you’re creative. Like, so think about this like a tool as well as a person that you are… now have a studio working under. You’re giving an assignment that other people are carrying out to free you to be more creative, or to improve your creativity is a really useful way for creative people to think about this. But kind of saying, I’m not AI, I’m not going to use it, first of all, everyone’s lying to you. We have this great set of surveys showing that everyone says they don’t use AI, but when you ask them to identify people around them using AI, they say 90% of the people in their field are using it. So, like, I think it’s worth noting that, like, there’s a lot of pressure. Like, I know a lot about, like, the indie game space, right, video games. And, like, you’d be insane as an indie video game developer not to at least use this to help you with coding, because you’re resource betranger. But, like, the indie games community hates AI, right, the audience for it, or at least some vocal component, so everyone just hides how they’re using AI, right, and doesn’t talk about it. So I think we’re in this kind of weird space, so just on the editor piece. The broader question you kind of asked, right, and just repeat this, because I went down a rabbit hole, so the other, the other half of the question.
Zack Arnold: Well, it’s just kind of this general idea that the reason I don’t believe that those that have a generalized creative skill set will be replaced is our ability to interpret notes, and especially because people are not very good at communicating what they want, it’s our job to interpret what they want, and I’ve seen this over and over, trying to iterate on things with, whether it’s image creation or otherwise, like, you get it, like, oh, this is a really cool first version, fix this, fix that, oh, wait a second, you went to different version. Like, I know all that’s gonna get better. But ultimately, how is it that we can expect to be replaced by whomever we’re collaborating with if they can’t even communicate to us, and it’s our job to interpret what they want when they don’t know what they want?
Ethan Mollick: Okay, so I have some complex thoughts on this one. I mean, I think part of this is… I worry when the answer is, AI can’t do this and won’t be able to do it. So what you’re describing is a… is a failure of theory of mind, right? So the AI has the, like, theory of mind is this idea that I can predict what you’re going to do, right? So, you know, like, if I, you know, what will happen if this call gets cut off? what’ll happen if I start swearing right now? Like, our human-to-human interaction depends on a theory of mind, of a prediction of what other people are doing, right? And, you know, it’s mostly a human thing, dogs have a little bit of it, so they can kind of predict what their owner does, and the fact that the AI can do it all is amazing, right? That it could tell you, like, oh, this would make someone sad, this would make people happy. But it also gets into trouble, it gets, you know, it gets… kind of caught up in knots for a variety of reasons, and that experience you’re feeling, anyone who uses AI a lot has been in this thing, which is, like, we get… you get only so far, and then you sort of get in loops, and it’s very frustrating, like, change this, and then it reverts back to another thing, and you realize you have to probably start a fresh chat, and try again, and maybe you’ll get some effect. That is very much a limitation right now. It might be a limitation for a while to come. I do worry, though, that as much as a limitation now, building on the idea that this is a limitation forever is a problem. Now, on the other hand, you said something else that was really interesting, which is complex tasks. Right? The jobs that we do are complicated. A podcast host is, like, if you ask a kid what a podcast host is, like, you get on the phone, you get on a, you know, podcast, you talk to people. But reality, your job is very complicated, right? You’re adapting this on the fly, depending on what I say in complicated ways. You had to do research on this, as we’ve talked about. You have a team that you’re interacting with to make this happen. There’s a schedule, there’s a thousand things you do. Research. Who’s your podcast? People, who you’re contacting, how you get in touch. Tons of things. And the AI’s not going to be good at all of these, because of the jagged frontier. And because it’s not good at one of these things, that actually destroys its ability to do a lot of the kinds of tasks, because it becomes a blocker, a, a bottleneck. So I think that I worry about individual tasks saying it will never have a theory of mind, it won’t interpret things well. But I am much more sympathetic to the idea that because it’s jagged. once that weakness is conquered, there’ll be another weakness where it’s like, oh yeah, well, it is really good at interpreting things and doing drafts, but once you have more than 5 people in the audience, it gets confused, or it tends to only listen to this kind of person. And then that’ll be conquered, it’ll be another thing, but like… That jaggedness is really a key to human success.
Zack Arnold: Yeah, and this, I think, is the perfect way to segue into talking about what you call the four human advantages, because I really want to discern, like, where, as we’ve talked about, different ways of working with the AI centaur work, and, you know, cyborg work, conjurer work, but I think there’s also a level of the AI can do this, but I shouldn’t be using it to do this, and that’s kind of where I want to go for the rest of this conversation, and more through the lens of how we connect as humans, right? Whether it’s on social media, through comments, through DMs, or my specialization, through outreach emails. How we build relationships with others, connect with our peers, connect with experts and mentors. Right? Because there’s definitely a lot of people that are leaning on, oh, I suck at email, I can now have the robots write my emails, or, you know, AI can write the LinkedIn post, and my AI can comment on the LinkedIn post, right? And it’s just becoming this circular reasoning, and you have what I think is probably one of my favorite sentences of the year. And you say that these posts are just meaning-shaped attention vampires that take mental effort to decode and give you no equivalent understanding in return. So let’s start talking about, even if the AI can do it, where it shouldn’t be through the lens of how we connect as humans.
Ethan Mollick: I mean, it’s a really good question, and I think it’s… I think you see it, by the way. Like, I’m a big LinkedIn poster, right? And, like, the comments have become absolutely useless to read, in large part, because either they’re 100% bot-written, or they are human, sort of, you know, gone through bot… language enough that there’s nothing… there’s no interesting variation there, right? And so, the real danger with AI writing is I’m not getting, like, I… this is actually a problem throughout the economy, right? Through everything we do. A lot of stuff depends on the idea that there’s a human behind it, right? Think of letters of recommendation. A letter of recommendation, or an email, a cold outreach email from you. When I get a 12-paragraph cold outreach email from you. 3 years.
Zack Arnold: Mine wasn’t 12 paragraphs, by the way.
Ethan Mollick: No, no, I don’t… Like, a thoughtful, like, you know, well put together, like, I know a human did that. There was no alternative view, and it’s worth me reading that, and if it’s thoughtful, it’s your thought doing this. I’ve learned something about you as a result, and I feel cared for, and I feel like there’s a connection here. If I get that now. I… you know, first of all, if there’s any sniff of AI writing at all, and by the way, one thing I’d like to emphasize to everybody here who doesn’t do… read a lot of AI writing, I promise you we could tell, right? That doesn’t mean it’s bad. Like, I use AI in my book, like, if you, you know, if you run it through an AI checker, there’ll be a few percent, I mean, there’ll be a couple percent that’s AI because I have interview AI, but, like, I can’t end a sentence. I rewrite everything, but, like, I write all my own first drafts, but there is AI in… in what I do, but I think it’s very different from letting AI create content for you. And it is all going to be the same. And by the way, no matter how good your prompts are at making the AI sound more human, it still has the same kind of patterns and thinking. And as a result of that, it’s broken the idea that a personal letter from you means anything. Because I get 100 of these a day, and they’re really well researched, and sometimes people have really good prompts, and it’s like, oh, this is very insightful, and… like, and so, that has undermined some of the nature of human communication, but even worse, I could read that whole thing, and it feels like it has meaning and intent, but it doesn’t. In fact, it was an indicator of your lack of intent, and the lack of meaning. And that is a real problem for people whose skill was this kind of thing, or any of the other human-based skills that get overwhelmed by this. I mean, I’m an author, right? And it’s very frustrating, like, there are… my publisher has to keep taking down books from Amazon that have the same… that are titled Coexistence, because people are just churning things out. There’s fake biographies of me online that, like, you can get, which are hilarious, because they don’t mention anything at all, and of course, they’re 100%… fake information. I think we’re drown… like, we’re always drowning under a sea of this kind of stuff. That is becoming worse, and it’s becoming even more wor… even worse because it sort of hurts us as creators to rely on this too much as well.
Zack Arnold: Yeah, so I guess this kind of begs the question then, if you’re now getting 50 to 100 of these messages a day, that maybe or maybe not we can tell if they’re AI, and they seem thoughtful, they seem like there’s research, what are you even doing here today? How did I cut through the noise, and how did you say, alright, I can tell this was a human, this is somebody that’s worth my time? What’s the difference, given that the AIs are getting so good at this?
Ethan Mollick: Well, I mean, it starts… I mean, we end up retreating to human spaces. There’s two options, right? I mean, one option is you retreat to AI… you rely on AI to do this, and I do too, right? Like, I have a human who… I have a couple great people I work with who help me go through all my messages coming in, right? And the answer was, by the way, before that, I had 200, 300, 500 messages a day, and I just didn’t get through any of them, so it’s, like, even worse, I’m sure you’ve experienced this yourself. Right? What emails do I not answer the most important? Because I’m like, I’ll get to that later.
Zack Arnold: Because you want to be… you actually want to give them care and time and attention, and by the time you get there, like, I got nothing left.
Ethan Mollick: Like, all my best… or I just forget, right? All my best emails are what, like… so, like, medium priority emails would be the worst ones would get the message back. You’d just be like, no, I can’t do it. But it’s like, oh, someone wrote me something really thoughtful, and I want to think about, like… So, you know, I do… I use AI to screen stuff at this point to help make things easier, and that’s actually making me more responsive, because it can help me pull needles out of the haystack. But, at the same time, I’m retreating more to human spaces, too. You came recommended, right? Like, that was the issue, too, is like, somebody I trust and care about said, you gotta talk to this guy. And so, that makes a difference, too. you know, on one hand, AI’s profoundly democratizing, on the other, it starts to create exclusivity at the same time. And I think we’re gonna find both a embracing of AI and a retreat to human spaces happening at the same time.
Zack Arnold: And that’s essentially the core thesis of why I’m doing this entire conference and talking to you, is that we are retreating to the things that we know in certain respects, which is human connection and human trust, and I guarantee, especially given the volume of messages that you get, if I had sent the exact same message to you cold, probably would have gotten lost in the ether, right? Same words, same intent, 100% AI.
Ethan Mollick: It wouldn’t have gotten lost in the ether, because it would have gotten lost in the ether 8 months ago before agents were good. Now I very much trust my AIs who are reading this, and my team and everything else, that it would have surfaced this. I just wouldn’t have given you any credit for having written… like, so it would have been a very different assessment, right? The AI would have… I’m pretty good at building agents, right? And, like, and Soul, 5.6, Soul’s amazing at this point, as is Fable, and it would have done a good job researching you and saying, oh, actually, Zack’s really for real, and, you know, it would have surfaced it, but it would have been a very different kind of surfacing, right? Your persuasion wouldn’t have mattered, it would have gone to hardcore facts. So, I actually think it’s more complicated than that, right? Like, people are going to be using… like, one of the things I wrote about as I put my book up. was that, you know, I built a version of my web page for my book for agents, as well, to read, right? And not, like, weird, maybe, like… and I actually ran it through a panel of 50 different AIs, and it was fascinating to get their marketing feedback on what worked and what didn’t for them. I had to change the page dramatically to make it work. So, like, we’re now talking to multiple audiences. Your emails are being read by an AI agent. It’s being read by… probably by people like me, a smart one. but also a dumb one that’s, like, built into Gmail that’s, like, summarizing stuff. Or, you know, there’s a CC function in Gmail where it tells you what you do that day. So we’re competing for all kinds of attention.
Zack Arnold: Yeah, and this kind of goes to what I think is a really, really great and essential list of when not to use AI, right? So I’m going to go through this again very briefly. I could do an entire podcast about each one. But when you need to learn and actually do, like, the effortful, this is difficult, there’s friction learning, right? When very high accuracy is required, and, you know, especially if the AI, you know, knows everything about this, that, or the other. thing, like, you know, that’s great, but a lot of times, humans are gonna know what’s more accurate because of our taste and our judgment. Then you have when you don’t understand AI’s different failure modes, you don’t know how to navigate the jagged frontier, right? When the struggle is the point. But for me, the one, and we’ve already kind of emphasized this, but I want to keep going down this rabbit hole, specifically with human connection, is when the work signals something that matters. And you give the example of, you’re not going to have AI write a condolence note. Right? Like, it’s… how is somebody gonna receive that? And they’re gonna be like, well, that’s just not meaningful, because you outsourced it, right? So it’s the effort that has the meaning, not the words.
Ethan Mollick: Except that, right, as I talked about earlier in the book, everyone’s writing… like, the most intimate tasks are the ones people give up to AI first. You know, there’s a… I quote a conversation with funeral home directors that, that they’re all… everyone’s, like, the biggest transformation in funerals has been AI eulogies. Because everybody, like, because it’s a difficult moment for people, and they’re not good speakers, and they have to sum up a life that way, and the AI help, I guarantee that I… that a substantial portion of wedding vows have AI help in writing them at this point. One of the number one things, people have them write, you know, things, like, stories for their kids to read. Like, the most intimate stuff is often the stuff we give away first in these kind of areas. So, like, as much as I’d like to say this is the line, right? If you’re not a good writer, if writing causes you stress, if you’re, you know, if you, like, you know, think back to writing my wedding vows or whatever, like, you know, I enjoy writing, so this was fine, right? If you don’t enjoy it, you’re bad at it. should you be using AI to do this? Like, it’s a harder question to answer in some ways, but… and then what do people interpret it as? Would they rather hear from the heart a mediocre talk, or a really polished one? It’s a hard question. It’s getting harder as time goes on.
Zack Arnold: Yeah, and that’s why… this is THE reason that I wanted to talk to you, because I’m trying to figure this out as well. And for me, it’s a pretty hard line when it comes to human communication. If it’s a matter of somebody is sending me a personal, thoughtful, heartfelt message, and then I find out, hey, wait a second, you didn’t write those things, so I don’t even know if you meant those things or thought those things, like, to me, that’s actually a breach of trust. But I’m trying to figure out what is that line, and when it comes to reaching out to people, and it really… the medium doesn’t matter. If you’re having the robots… let’s just, we’ll play at the extreme end of the spectrum. The robots are writing all the words in your outreach emails, your text messages, your DMs, etc, right? To me, that is a breach of trust, but what I want to know is if I had this hardline stance, am I just a professor in the 70s that’s saying, I’m not gonna use calculators? Like. how do we figure out? Because, yeah, if people are using these, and it’s working, it’s connecting people, who am I to say this is a hard line?
Ethan Mollick: I think we are, you know, it is… we are in… we’ve already been through a series of changes, right? Like, you know, like, you know, the way you would write a message to someone today is very different than the way you would 6 years ago, right? The way we communicate with each other and the kind of connections we have are very different. This is going to be another thing, right? So, the problem you have right now is you have… you’re talking to two different audiences when you send that message. There’s a group of people who are like, yes, I appreciate the handwritten note version of this. it would not mean anything without you. And the others are like… okay, you know, it doesn’t matter to me one way or another. And… So, I think for right now, it shows care. Soon, it might be an affectation, like when I get a handwritten note back for something, I’m like, I mean, email’s easy for me to respond to, like, I, you know, I appreciate it, right? So, I think that we don’t… we will negotiate what’s human in this kind of space. I think that… Care is a thing you want to show. The problem is, is that the tools you use to show care are going to kind of collapse. So what do you do next? Now it becomes just an affectation, right? You personally write every message that’s important for you to do that. Right now, I mean, I personally write every one of my posts, and then I, you know, if I use AI as afterwards to help me, you know, polish up an issue of a draft, or find issues. Because otherwise, I won’t do my own thinking. what it… when the AI gets better than me at writing posts, is that a smart thing to do? Right? In the same way, like, your personal email’s great, but I’m getting 100 of them now, and they’re starting to get really, really good from the people who are good. Like, the bad ones are bad. So now, care looks like you bothered to put together a little prompt. So, to me, that’s the challenge that I think you’re gonna face, is, yeah, I mean, at some point it becomes your affectation, and your word against other people’s, which is like, I don’t use AI. And they’ll be like, well, I don’t use it either, and then AI detector indicates, yes, they 100% use it, but then detectors get defeated, and so on.
Zack Arnold: Yeah. And that kind of brings me back to what we had mentioned a few minutes ago, which is that what differentiated me was the mutual introduction through a trusted source, which is why I believe human connection is so vital. Because whether it’s, you know, sending a cold outreach email to somebody that has a really full inbox, or even sending resumes into the void in LinkedIn, or cover letters, or otherwise. like, all of this, they’re the gates of the bots for all of these now, and I say you have to get around the bots to the humans, and a lot of times to get to the humans, you have to go through other humans, which is why I’m doubling and tripling down on… it’s the value of our human connections with real people, that’s what’s gonna open all the other doors for us.
Ethan Mollick: So I’ll push back. I think it’s some of the value, but I think there’s a lot of people who are very bad at that kind of human connection, right? What you’re doing is exactly what I expect on the jagged frontier. You found a spot of jaggedness that you are better than the AI at, and will be for a considerable time to come, which is making human connection. Wonderful. That is exactly the model to face here, right? You found a spot on the Jagged Frontier, and you are gonna… you are going… you are way ahead of where the AI is, and where it’s likely to be in the future. for a variety of reasons, they’re not just technical, right? Even if you write a warm email, it wouldn’t know your network of people. It doesn’t owe the favors and understand the unwritten stuff. So there’s a lot of reasons why I expect that to stay jagged, and you found a competitive advantage on the jagged frontier. Some people are terrible at this stuff, like, you know, like, they’re not good at connecting with people, it stresses them out. So they’re gonna have to find a different spot on the Jagged Frontier than you did. So I think you’re giving… Yes.
Zack Arnold: Sorry, go ahead.
Ethan Mollick: and they can use AI to help compete, like, the area where you’re good at, they’ll get 80% of the effect, or 90% of the effect you’ll have, right? But they, you know, and that might be good enough. I’m a professor of entrepreneurship, right? And entrepreneurship is all about, you are the top .01% in something. And your entire job is to make sure that everything else you do is not so bad that it destroys… everything you’re terrible at doesn’t destroy the 0.1%. And that’s why this is such a great technology. It can get 80% of the way there, 90%, 95% of the way there to being as human as you are in introduction, and that’s… but it’s not going to be as good as you, so congratulations, you have a competitive advantage. But the person who’s getting 95% of the way there is… was at 4% before. Right? Like, their emails were terrible, or they were too paralyzed with social anxiety to send them, and now they’re getting sent. And they’re good at something else. So… It is a weird kind of shape that’s happening at the same time.
Zack Arnold: Yeah, and what I wanted to not necessarily push back on, because I agree with so much of what you’re saying, but I think there’s a third option, which is that I believe this is a very, very learnable skill, and this is now where I’ve inserted myself and hope to be the top .1%. If I struggle with human interaction and connection, I have social anxiety, I hate networking, I hate mixers, but I genuinely want to have heartfelt conversations with people that I want to connect with and admire. I believe that’s a learnable skill, and I believe that AI is very valuable in this skill, and this is now where there’s nuances. I’ve been talking from the extremes, like, no AI in my inbox, right? But I also very much believe in AI as an accommodation. Right? So if it’s a matter of, like, helping to figure out what’s the structure, or being an outreach tutor, saying, like, alright, here’s a really shitty draft, like, how can I provide more value, or give me 25 versions of subject lines, right? I think there’s a lot of nuance where AI is an accommodation and as a tutor is invaluable, and that’s where I’m hoping that I can also provide expertise, right? So let’s talk about all the nuances and the gray areas in between these extremes.
Ethan Mollick: Yeah, I mean, I think that… this allows people to, you know, AI certainly works as a tutor to help you learn new skills, right? There is somewhere between the dial of, I hate AI and all AI that is probably the right place for most people to be for most tasks. And it does free you up to learn more. So now, I can get good at, like, it’s less overwhelming because I don’t have to email everybody this way, but the two or three people who I want to be really human, I can learn how to be really human with. And that’s across different skills, like… I need to have, you know… So, one of the things I talk about in the book is that there are four things that are sort of big components of, of, You know, places to hone your advantage, which is wide knowledge, deep knowledge, agency, and taste. And you’re kind of touching on all of those a little bit here, right? So… Deep knowledge is expertise. If you have expertise in an area, usually hard-won expertise, and then that gives you an advantage because you can compete with the AI by just being better than it, right? And writing warm introductions is your space, you’re hoping to be able to teach other people that deep knowledge. The other thing is wide knowledge, understanding many different things, because it turns out. For example, if you’re a coder, you can now do design work. If you’re a designer, you can write, you know, do legal work. Like, the more you know about different fields, the more you’re able to understand what AI can do and invoke it to do those things. Then there is the idea of agency, just doing things. It’s almost where we started, right? The very first thing we talked about was about how I need to use AI By using AI. And that’s how you’ll figure out how to use it. That’s an agency example. Figure out how to use it, just don’t wait for permission or control, do it. And then there’s the final piece, which is taste, right? Having a sense of what to select among many AI options. And I think what you’re talking about is that you’re trying to teach people a sense of taste about what a good introduction is, you’re trying to broaden their knowledge of what they’re doing, you’re trying to go deep in this particular area and give them agency and power to do this. I think we’ll find this across many different fields, and it is all gray areas there.
Zack Arnold: Yeah, and again, like, this… we’re all figuring this out. I don’t have the answer for, here’s exactly where the robots do and don’t belong in your inboxes, in your communication, in your dating apps. Like, I’ve heard stories now that there are dating apps, where the AI will write the messages so the AI can respond to the messages, and people are meeting in person, and they’ve never talked. Ever. It was all the robots to bring them into that space. Like, that’s absurd to me. Right? But that’s kind of the world… that’s one version of this new world.
Ethan Mollick: Maybe it’s a good world! I mean, I, you know, one of the early things I talked about, like, I talked to a bunch of people who build LLM embedding models of their… all of their writing, and before they have a meeting, they send their embedding models against each other, and they find areas of agreement or disagreement to have conversation with that’ll be more meaningful. So, like, I think it’s easy to be in a rejection of, like, this is stupid, and as extreme as… like, nut picking is the glory of the internet, right? There’s always someone doing the dumbest possible version of something. I think this is the gray area piece. Like, it’s worth saying both questions, like, what’s terrible about this, what’s great about this, what’s the gray area that could work in between?
Zack Arnold: Yeah, and I’m really glad that you pushed back on that, because I think that that’s where I’m trying to figure this out, too, and my immediate reaction is, are you nuts? But now, thinking about, like, I have a, you know, a small LLM myself. I put together in Obsidian everything that I’ve ever written, every word that I’ve ever spoken, transcribed, every newsletter, every course material, every PDF, so now I have something that I can point to that can kind of help my team kind of ascertain, how would I think about this? How would I approach this, right? Right. But then there’s the part of, well, would I use that to throw myself into the dating world in a dating app? Probably not, but I, like, I’m glad that you pushed back that maybe there is a world where that actually, you know, makes a…
Ethan Mollick: Why have bad… there is a little bit of a chance… the world’s malleable right now. And that gives you a chance to challenge the status quo in other ways. So, if we think about the goal, I mean, I’m not gonna say that you should have your dating app be AI, but there is a world where you can imagine, why do I want to go on a lot of, you know, on all bad dates when my AI can negotiate with your AI and see whether or not there’s a good chance, I mean, it won’t substitute for human, right? We’d say, don’t replace dating with this and get an arranged marriage by AI, but maybe it screens your first set of dates and finds people who you wouldn’t find otherwise, because you could only look at pictures and not personality. There’s a world where you can imagine a better world, and I think if I’m going to leave us with kind of thoughts, I know we’re near on time, it’s that there is agency right now to do things. Like, that I feel like some of the if I had one criticism of our kind of conversation, it’s that it’s kind of based as AI doing something to us, and we have to figure out how to navigate it. And I think there’s another thread you’ve laid down that I would pick up, too, which is, we have agency ourselves. Like, we can decide what AI is useful for. We can decide to use it in positive ways around the negative ways, right? It’s not just about choosing a ban or not ban it, it’s when you, as an expert in doing these kind of warm introductions, can tell people, actually, here’s when to use it and here’s when not to, that will make the people on the warm side feel better, right? Like, I want everybody who wants to make an introduction to have an AI bot that actually helps them screen the good ones so they can find ones that are useful and gives you feedback on what’s valuable. That’s an example of showing how to make the world a better place, and I would push for that as well.
Zack Arnold: So, I know that we’re wrapping up, but I want to kind of lead, or kind of leave us with one final question, or maybe a couple of questions. I love the covers for your books, and I want to get a sense of kind of what the story is behind these, and then, should we reveal the Easter egg on the new book?
Ethan Mollick: I feel like we’re halfway there on the Easter egg side. So, you know, interesting, the first cover, was 100% a human design by a human designer, and, you know, really great, like, you know. full analog, and obviously, happened to the Apple of good and evil. It’s also… but just a good marketing thing, because, like, every… it feels like it’s a subtle illusion that nobody would get, but you feel really smart for getting it, but it’s actually, like. pretty universal, so, like, feel really good. And I have to credit the cover designers at Penguin Portfolio, who helped me with that. The second one was kind of interesting, because we were playing with lots of ideas, but part of how we did these ideas was we mocked them up with AI. And actually, my wife did the mock… like, we played with the idea of the two fingers, and we kind of rejected it, and the Easter egg on one of the hands, and we couldn’t… they couldn’t get it to work, and then my wife actually mocked up an image version that did work, and that became the model for looking for how to change things. So, it was still made by a human, very much, but now the ideation process was us trying 500 different ideas. and going off on wild goose chases that would have been impossible, because if you’ve ever written a book, there’s a cover deadline, it’s very expensive to, you know, have people iterate on it, but it let us do that kind of piece. So it’s… so there is an attempt to kind of trace, you know, both the human-ish aspect to it, but it’s interesting to see the AI creep in on the sides.
Zack Arnold: Yeah, no, I really love just the way that they feel, the story that they tell, and how you really have to work at it to figure out what is this about, but then when I saw the Easter egg, I’m like, that’s brilliant.
Ethan Mollick: I’m glad you got that one. So let’s leave it for them to get.
Zack Arnold: We will, for sure. But we’re gonna just kind of plant that seed. Again, so we’re about one of 20 hours through the conversation that I wanted to have today, but I also want to be respectful of your time. So anywhere that anybody is viewing, listening between now and the end of time, we will have links to both of your books, to any future books, we’ll have links to your substack, but most importantly, because this is all about connection. How is somebody able to connect with you? What’s the best way to connect with Ethan?
Ethan Mollick: So, I, my, long before AI, One of the key features of being Ethan Mollick is that, I just, and becoming a professor and everything else is just, I like to say things, and I, you know, I’m pretty authentic about it, I don’t really have a persona. So, honestly, LinkedIn X or Blue Sky is, if you, like, I post usually several times a day. papers and stuff, but also my thing, like, that’s a really good way to make a connection. You know, direct messages get overwhelmed. If you can write a really good email that, that my AI thinks is important, I feel like that’s a bad challenge to put people to, but, you know. But I think that, really, social media is… and the Substack is where I kind of live in these, or reply to a thread, like, I reply back. Like, that’s where I feel like the most interesting direct connection is. As a teacher, right? It’s like being in front of a classroom. Your interactions are public as opposed to individual private ones, and it feels like there’s more return to that.
Zack Arnold: Yeah, okay, well, I’ll make sure that we have links to all those, but once again, this has just been beyond an honor to have this conversation with you today, and all the good that you’re doing in the world in helping us navigate this hellscape of reality right now. So, I can’t thank you enough.
Ethan Mollick: A hell and heaven-scape. A little of both.
Zack Arnold: Yes, the hell and heaven-scape, exactly.
Ethan Mollick: But, yeah, and it’s gonna get weirder if I leave one last message, unfortunately.
Zack Arnold: Let’s just roll a grenade in the room in the last 30 seconds, right? Oh, it’s gonna get weirder, you know…
Ethan Mollick: I have access to things you haven’t seen yet, it’s gonna get… we’re barely begun to get weird. So, good luck on all that, everyone watching or listening, and hopefully we’ll talk more.
Edited by: Curtis Fritsch
Published by: Vim Pangantihon
Music by: Thomas Cepeda
Nick Milo explores how to think clearly in an idea-flooded world, building systems to connect insights and turn knowledge into meaningful creative work.
Rob shares how you can start to dabble with AI so you can integrate it with your own unique talents and skills.