Don’t Let AI Do Your Thinking

Knowledge at Wharton

Ask about Knowledge at Wharton

Episode Description

ABOUT THE EPISODE AI can make people faster, more productive, and more capable. It can also quietly replace the skills, ideas, and judgment that make their work their own.

Ethan Mollick, Associate Professor of Management and Co-Director, Generative AI Labs at Wharton, explains how people can use AI deliberately without giving up control over their work and thinking.

Mollick discusses the rapidly expanding capabilities of AI, the “jagged frontier” of what it can and cannot do well, why workplaces need better approaches than outright bans, and how work, education, and society can adapt to an AI-driven future.

ABOUT THE PODCAST Every day, business scholars answer pressing questions in their research - but what do their insights mean for you? In this podcast, Wharton faculty dive into what inspired their studies and how their findings resonate with the world today. Learn how research insights translate into knowledge you can use, with host Dan Loney.

----- The Wharton School of the University of Pennsylvania is committed to sharing its intellectual capital through Knowledge at Wharton, the school’s online business analysis journal.

Launched in 1999, Knowledge at Wharton offers free access to:

- Articles, podcasts, and videos highlighting Wharton faculty research and analysis of current business trends - Interviews with book authors from Wharton School Press - In-depth series of curated content like Wharton Executive Education’s Nano Tools for Leaders - A searchable database of more than 10,000 articles covering all aspects of business - A weekly newsletter that delivers Knowledge at Wharton insights directly to your inbox

Visit the Knowledge at Wharton homepage: https://whr.tn/3qUpSBE Subscribe to the Knowledge at Wharton YouTube channel: https://whr.tn/3DEAgAU

Read the full transcriptHide the full transcript
  1. If you want to have your own writing, right, you're going to need to spend some time making sure that you're writing on your
  2. own, even if AI is more efficient, because it does a very good job at all those things, but it just won't be your
  3. job. So I think that that is the big trade-off here. Welcome to the Ripple Effect, the podcast that
  4. takes you on a journey through the minds of work and faculty. I'm your host, Dan Loney, and in each
  5. episode, we'll be diving deep into the inspiration behind the groundbreaking research that Wharton professors have
  6. conducted and exploring how their findings resonate with the world today. Well, now that we have artificial intelligence in
  7. our lives on a pretty much daily basis, the next step of what seems to be the question is to how to coexist with
  8. the technology that at times might be smarter than us. That's a question that is brought up in
  9. the new book, Coexistence, by Wharton professor Ethan Mollick, which looks at the next phase of
  10. that relationship with AI, and Ethan joins us right now to discuss the book. Hi, Ethan.
  11. How are you, sir? Ethan Mollick, PhD Excellent. Thanks for having me. Dan Loney, PhD Thank you.
  12. And I guess we should start out with the fact that you incorporated the use of AI in writing this book, correct?
  13. Ethan Mollick, PhD Right. I mean, but very carefully, right? So the whole idea is being very deliberate
  14. about how to use it. So I write all my own first drafts, for example, but I get feedback, I use
  15. it for research, so you have to be careful about what you're using it for. I also use it for fun throughout the
  16. book as well. Dan Loney, PhD I think part of the issue that some people have is that AI
  17. is seemingly always changing and adapting and seeing new versions come out with chat GPT.
  18. And there's probably a concern of AI doing everything for us. But that doesn't seem to be the case,
  19. that AI has its place in the overall relationship with humans. Ethan Mollick, PhD Well, I mean, I think
  20. you kind of hit the nail on the head, which is it is not just changing, it's improving.
  21. So when I last wrote my last book, Co-Intelligence, that I had to do about four minutes of work with vision percent success
  22. rate. It's something like seven weeks of work now, later. So I mean, it doesn't do everything, there's
  23. jaggedness, it doesn't, you know, it has issues. But I think it is expanding what it
  24. could do, even as there's still a lot of room for humans. So jaggedness, and being jagged is a word
  25. that you use a lot in this book to develop or should say to make us understand the relationship.
  26. What does jagged mean in your mind? So along with my fellow researchers at Harvard
  27. and MIT, University of Warwick, we wrote a paper where we looked at the impact of
  28. AI on consultants way back in 2023. And we found that AI was good, some stuff you would expect, bad stuff you would
  29. expect, and you don't know in advance what that is. So that's the jagged frontier of AI ability.
  30. It keeps growing and changing, but that does mean that there is a place for humans
  31. where AI fails. Part of this is the relationship with human beings and what it's going to look like
  32. as we move forward, isn't it? Well, I mean, that's everything, right? The question is, where do we get displaced?
  33. Where do we get advantage? And how do these things trade off with each other? You tackle the use of AI around work,
  34. around education and safety. Are those kind of the three most important areas, discussion areas, when we're talking about what
  35. we're going to see in the future? No, this is a general purpose technology. This affects absolutely every aspect of technology, right?
  36. And so we can go into micro levels of how we'll change podcasting or how we'll change being a professor.
  37. I talk about society overall and changes in how we relate to the internet and relate
  38. to each other. So I don't think there's like one area where AI is going to matter.
  39. I think it's going to matter absolutely everywhere. But with those three areas, and I guess
  40. safety is one that is talked about a lot, and it's talked a lot about technology
  41. in general. How does AI and safety kind of correlate as we move forward? Well, there's many kinds of safety, right?
  42. So often we divide this into existential risk and other kinds of risk. So existential risk is the idea that what
  43. if AI murders us all, you know, which there's people worried about and we're probably very
  44. underregulated for. And then there's the wider questions in every field. What do you do with cybersecurity?
  45. What do I do with CDR and tests? What do we do about, you know, getting bad advice from AI?
  46. What do we do about, like, there's a lot of other questions there too. So, you know, the technology's upsides and downsides.
  47. How does the regulatory piece, as you just kind of alluded to, fit into this as we move forward?
  48. Because most people believe it's incredibly important. I mean, we have a general purpose technology
  49. that affects everything with substantial risk and substantial upsides. And what guides it to, you know, lower
  50. risks and increase the upsides is regulation, right? So we don't have a lot of regulation
  51. right now around how AI labs operate. Almost everything is voluntary. And so, you know, but we also have
  52. a piecemeal set of regulations that's probably aimed at stuff that is not as critical to
  53. assuring AI success. So getting policymaking right is gonna be important, but it's very hard to do given how
  54. quickly things are developing. How should people think about, going off of the title of the book, the coexistence with
  55. artificial intelligence as we move forward here? I mean, so that's kind of what the
  56. book's about in detail, right? So you're gonna make choices of when to use AI and when not to use AI.
  57. You're going to have to think about how your job is going to change. I wish I could say things are gonna
  58. stay stable, but, you know, we don't have a long-term picture that's absolutely clear about
  59. what happens to the employment, but we do know transformation is going to happen, right?
  60. The nature of your job, if it hasn't changed, is going to change already. And so we have to negotiate those boundaries
  61. with AI and with ourselves. So there's a bunch of stuff in the book I give about when to use AI
  62. and when not to, about some human traits that you can kind of build up that might be, that can give you some more
  63. resiliency. But I think the most important message in some ways is, look, we are on an
  64. exponential curve of increasing AI ability, and there isn't a lot that's sort of certain.
  65. So part of this is about building up sustainable skills, abilities, stuff like that will hold
  66. up in the long-term. It is the when to use AI based on if it's a work situation, what the
  67. company believes, or if it's a personal situation, what the individual person believes.
  68. How does that dynamic play out? So I think part of the issue here is AI use is ubiquitous.
  69. So obviously there's stated that you could potentially ban some kinds of AI models potentially, right?
  70. But in terms of company rules, everyone just secretly using AI inside of companies, right?
  71. And so what you do is you get shadow AI use, that you're not giving people good access to models.
  72. And people have no interest in showing you how good they are at using AI because
  73. they weren't going to get fired as a result. So I mean, I think that there is a role for regulation in deciding when AI
  74. is not used, but there's probably a much larger role in us as individuals deciding when
  75. it's important to use AI and when it's important not to use it. You bring up in the book an interesting
  76. component about the speed at which artificial intelligence is kind of being implemented.
  77. It feels like for, I think for most people, and I know for myself, that it feels very fast as to how everything is
  78. coming on. But there's a case to be made that with all that is still to come, that
  79. it's a relatively slow process that's going on here. Well, the thing about exponentials is they start
  80. off slow and they go faster. I think we're already in a very fast improvement phase of AI over the last year
  81. or so. Things are very stable. GPT-4, which was the model that was released fairly soon after Chatsby came out, was
  82. the best model in the world for over a year. And so things are fairly stable. We've had since then, there are now 170
  83. AI models that are better than GPT-4. And 19 of them are on your phone. So things are moving much more quickly than
  84. you think, and there's no sign of that slowing down right now. So when you talk to the AI labs,
  85. they think we are in a continued exponential ability increase. It doesn't mean AI isn't still jagged, but
  86. we do see that gains continue. But is there also the component of how much control AI will have down the road
  87. and that relationship with humans, how that will play out? I mean, yeah, that's sort of what we
  88. discussed in the book here, or what I discussed in the book. There is no we. Either we're the ghost writer or no AI,
  89. right? This is my ideas. But officers get used to talking about we in academia. But it's the point, right?
  90. Is that we do get to negotiate some of these boundaries about what things do or don't do, but the models keep getting better,
  91. right? So we have to deal with both of those things. The first book was a world where you
  92. were definitely better than AI at most things, but it could sort of suck in for humans where you're weakest.
  93. Now we're in a place where some of your strongest abilities might still be better than
  94. that. I don't see a sign of that stopping, but that could stop in the future, right?
  95. So we have to make choices about how you sustain. I found it interesting. There was a spot in the book where
  96. you had written about a variety of different components, but you apparently had asked AI, the
  97. Claude program that you're using to add in its components to what you had done up
  98. to this point, which was very interesting that, you know, that's part of the incorporation as
  99. well, that the platform is going to bring its own learnings, teachings, information to the table
  100. in what you do on a daily basis. Right. And that's why it's important to think about,
  101. if you don't want, if you want to have your own writing, right, you're going to need to spend some time making sure that
  102. you're writing on your own, even if AI is more efficient. If you want to have your own ideas,
  103. you're going to have to, you know, use your own brain to come up with ideas before you turn to AI, because it does
  104. a very good job at all those things, but it just won't be your job. So I think that that is the big
  105. trade-off here. How much or how important will the human component be as we continue to see AI
  106. develop? Because even as we use it as humans, it's still human beings that are behind the
  107. development of the process as well. Well, it is and it isn't. I mean, the AI systems are not hand
  108. -built in the way that another piece of software would be hand-built. In a lot of ways, they're grown, you
  109. know, or evolved. So you have a bunch of training data, you have an approach the AI takes, and
  110. the AI has sort of grown out of the training data. So there isn't really this hand-crafting element
  111. about their software, which is like, what screen do we show the person? Like, the AI does the stuff on its
  112. own, it keeps improving, and increasingly AI models are improving existing AI models.
  113. So a larger, it's a process called recursive self-improvement. So a lot of the work in the
  114. AI labs is about automating the process of creating AI systems by having AI systems build
  115. AI systems. So I think one of the things that's a dangerous thing to think about, it's like,
  116. oh, yeah, it's all humans all the way down. It wasn't already. And one of the things that people worry
  117. about, about AI getting out of control, is the fact that you can have recursive self
  118. -improvement, AI systems improving themselves. What do you hope that people take from
  119. this book? So I think there are two things I hope people will kind of take away from
  120. them. One of them is that, yes, these systems are quite good and they do a lot. And the second though is that we get
  121. to make choices, choices about how we develop these technologies, choices about how we use them,
  122. choices about when to use them. They will make us more successful and happier and better off, or could undermine learning, work,
  123. our ability to stay in control of our own lives. And we get to make those choices right
  124. now. They're important choices. And at the personal level, they matter because they'll determine what skills you're keeping at work
  125. and how you perform and the jobs you do and opportunities for entrepreneurship. And at the organizational level, it matters because
  126. it determines what kind of companies we have and how people are employed. And at the societal level, it determines what
  127. we give up to AI and what risks we're willing to take on. How much is this still kind of a
  128. process that's going to continue to develop over time? And I say that because as you were
  129. just speaking a moment ago, it felt like that middle area is kind of the balance
  130. point of how much we use AI and don't use AI in the future. I mean, I think that AI will continue
  131. to get better. So building the right habits right now about when to use it and when not to
  132. is really important. Ethan, great to talk to you again. Thanks for your time. Thanks for having me.
  133. Thank you. Ethan Bollick, professor at the Wharton School and author of the book that you can see
  134. right behind his shoulder, Coexistence, The Next Phase of AI. Thank you for listening to The Ripple Effect.
  135. We hope you found this episode informative and engaging. Don't forget to subscribe and leave us a
  136. review so that we can continue to bring you the best insight from the Wharton School.

Brief Summary

Ethan Mollick explains how to use AI without surrendering your writing, judgment, or skills as its capabilities and workplace impact grow.

Key points

  • The jagged frontier of AI. AI can excel at some tasks and fail at others, often in ways people cannot predict in advance. That uneven frontier keeps changing, leaving room for human judgment and strengths.
  • AI is moving faster than it seems. Mollick describes an accelerating wave of AI progress: he says 170 models have surpassed GPT-4, with 19 available on phones. He sees no sign of that pace slowing down.
  • Keep your writing your own. Mollick says he writes his own first drafts and uses AI carefully for feedback and research. If you want your writing and ideas to remain yours, you need to make space to develop them yourself.
  • The risks of AI improving itself. AI systems are increasingly being used to improve other AI systems, a process Mollick calls recursive self-improvement. He cautions against assuming humans remain in control at every step.
  • Choices that shape our future. Mollick argues that people, organizations, and society can still choose how AI is developed and used. Those decisions will affect skills, jobs, learning, and the risks we accept.

Tension

27/ 100Low tension

Related episodes