Outthinkers
The Outthinkers podcast is a growth strategy podcast hosted by Kaihan Krippendorff. Each week, Kaihan talks with forward-looking strategists and innovators that are challenging the status quo, leading the future of business, and shaping our world.
Chief strategy officers and executives can learn more and join the Outthinker community at https://outthinkernetwork.com/.
Outthinkers
#133—Ethan Mollick: How AI Will Enhance—Not Replace—Our Work
Ethan Mollick, is the Ralph J. Roberts Distinguished Faculty Scholar and Associate Professor at the Wharton School of the University of Pennsylvania, where he studies the effects of artificial intelligence on work, entrepreneurship, and education. His academic research has been published in leading journals, and his work on AI is widely applied, leading him to be named one of TIME Magazine’s Most Influential People in Artificial Intelligence. Ethan also writes to a wider audience about AI, including in his book, Co-Intelligence, a New York Times bestseller.
In addition to his research and teaching, Ethan is the Co-Director of the Generative AI Labs at Wharton, which build prototypes and conduct research to discover how AI can help humans thrive while mitigating risks. Prior to his time in academia, Ethan co-founded a startup company, and he advises numerous organizations. He received his PhD and MBA from MIT’s Sloan School of Management and his bachelor’s degree from Harvard University.
In this episode, we discuss:
- What people get wrong when people think about how AI will shape the workforce.
- What he calls the 'Centaur versus Cyborg' approach—the misconception that work must be divided between humans and AI, rather than completed in unison.
- What preliminary studies, in the past few years between organizations like BCG, Harvard, and MIT, have to tell us about how AI powers human productivity and work itself.
- What history teaches us about adopting new technologies like AI to maintain a competitive edge.
_______________________________________________________________________________
Episode Timeline:
00:00—Highlight from today's episode
01:07—Introducing Ethan+ the topic of today’s episode
03:04—If you really know me, you know that...
03:24—Ethan's background and journey into AI
04:53—What is your definition of strategy?
06:20—Common misconceptions about AI's impact on the workforce
07:46—AI's capabilities compared to humans, particularly in creativity
10:03—The best way to use AI for idea generation
10:37—AI's role in the different stages of strategy development
13:12—The BCG study on AI's impact on consulting work Bring Us Together
13:58—The concept of Agentive AI
14:48—Lessons from past general purpose technology adoptions
16:33— The need for organizational redesign in the age of AI
19:18—The impact of AI on profit distribution across industries
20:40—AI's role in identifying market needs and finding solutions
22:38—The importance of experimentation with AI
24:50—Addressing the issue of AI's unexplainable solutions
_______________________________________________________________________________________
Additional Resources:
Link to book: Co-Intelligence
Linkedin: linkedin.com/in/emollick
Substack: @oneusefulthing
X: emollick
Thank you to our guest. Thank you to our executive producer, Karina Reyes, our editor, Zach Ness, and the rest of the team. If you like what you heard, please follow, download, and subscribe. I'm your host, Kaihan Krippendorff. Thank you for listening.
Follow us at outthinkernetworks.com/podcast
Kaihan Krippendorff: Ethan, it is great to have you here. We have tried for quite a while to get you on the podcast. I know you're in high demand, and thank you for being here in person. I'd like to start with the same 2 questions that we ask all our guests. The first 1 is just to get us to know you a little bit more personally, could you complete a sentence for me? If you really know me, you know that.
Ethan Mollick: If you really know me, you know that I'm actually this enthusiastic in real life and genuinely weird about this stuff. I do in reality, just like I do online and on podcast.
Kaihan Krippendorff: Yeah. It's interesting because AI is kind of what you're known for currently, but it's not AI that brought you here. Right? Like, you had this great book on entrepreneurship and what it takes companies to scale and you have courses on that. And that's kind of as I understand it, I got to participate in 1 of your classes a long time ago, and it was an entrepreneurship class. So I don't know if you don't mind just telling us a little bit of the pre AI journey.
Ethan Mollick: I mean, sure. I'm a professor of entrepreneurship, former entrepreneur myself with a cloud roommate. I helped start a company that helped them bend the paywall, which I still feel kinda bad about trying to get over in the late nineties. And I started working at on AI stuff with the MIT media lab. It's, like, the non technical person on the media lab group level in Morgan Minsky on AI stuff.
At the same time, I also started working on my other passion project, which has been about using games for education. So it kinda run-in parallel, which is how do we educate at scale by using game technique simulations, and I've been building a lot of organizations around that. And kind of waiting for AI to take off and when it did always have to be really well positioned for that.
Kaihan Krippendorff: That's awesome. Yeah. People can't see your visual right now, but I see and I've seen this background on other podcasts. You've done a lot of games in your background.
Ethan Mollick: They're all ones that mostly played out on books I've read.
Kaihan Krippendorff: Amazing. Amazing. Amazing. And we're talking physical games. We're talking board games.
Ethan Mollick: Yes. I'm a big video gamer too, but, yes, these are plus.
Kaihan Krippendorff: Got it. Okay. Go. So my son's a big gamer and just started at University of Pennsylvania, and I think there's so much value in kids gaming and what they can learn. Next question.
What's your definition of strategy?
Ethan Mollick: So I you know, it's always a rough question to ask because I am a business school academic. So I have been in rooms where debates over what strategy is. They're taking it for 90 minutes or so. Right? So I am I will not I'm gonna do it off the dome definition rather than try to do something where I'm gonna how are you what a bunch of different thinkers and thought strategy is.
But I think it is it is the art and science of anticipating potential futures and taking action for the best possible outcome across them.
Kaihan Krippendorff: I love that. Potential futures. I love that. Great. Okay.
Alright. We won't go into that. We won't debate it. It's perfect. That's kind of part of the part of the point is that we have these people who spend their lives thinking about strategy coming and talking here, and they all have different definitions of it.
There is Not yet.
Ethan Mollick: I am very much a pragmatist of both in both AI and strategy, which is I think you comes out about entrepreneurship, which is if it works, it's worth working with, then we'll figure out the definition later.
Kaihan Krippendorff: Yeah. Perfect. Great. Alright. So I've got a list of questions here that I created myself, and they're in a random order.
And I'm gonna sort of jump around, and we'll just sort of see where it goes because you've got great work. You've got the book. You've got you know, you've got all kinds of podcasts and other content. So people that wanna just know your content, you know, step by step, they can do that. So I'm just gonna jump around a little bit.
And I'd love to start first by the future of work, the shape of work. What do people get wrong when they think about how AI is gonna shape the workforce?
Ethan Mollick: So, I mean, I think, part of what we get wrong is thinking we have an answer right now, which we absolutely do not have. And so it is yeah. We don't have answers. Right? I think that the reshaping of the workforce is going to happen.
It's already happening. I think probably a good model to follow is the current sort of economic view of the other jobs that bundled tasks, some of different levels of relatedness and important. So the task button's gonna change more in the job's gonna change.
Kaihan Krippendorff: Got it. Yeah. So right now, a job or a role is a bundle of tasks that has 1 title, but where they can be deep decoupled.
Ethan Mollick: Well, I mean yeah. So what are they shuffled? Right? I mean, they already decoupled. Right?
But, like, you would never have had for example, a doctor would never you never give them the role that they have right now. You wouldn't have a doctor where you would where, you know, you'd expect someone to be good at hand skills and diagnosis and running a business and empathy. And, like, that's just too many things in a bundle. Right? So bundle that bundle gets reconfigured.
It changes part of the meeting and being a doctor, but is that always a bad thing? Do most doctors wanna do all of that? I'm impressed. Right? I'm on podcasts like this.
I also talk to companies. I write academic papers. I create pedagogical materials. I do it, like, I run committees. I teach, but, like, that's a bundle.
It's very Mhmm. Gonna be changed by AI too.
Kaihan Krippendorff: Yes. Well, let's dig into that a little bit then. You know you know, there's some things in that bundle that people have thought. Are human just, you know, creativity, for example. The human interaction, the empathy, what what do you think people misunderstand about what AI is capable of as compared to kind of what is uniquely human and, you know, put the domain of humans.
Ethan Mollick: So I think that it is hard to draw golden lines like that, you know, bright shine lines, no 1 crosses. And I think that most attempts to do that will result in you feeling bad. Right? So, you know, creativity, persuasion, apparent empathy, if not real empathy because these you know, we e these are not humans, management, judgment, you know, like, a thing that you would call, you know, call a human. It has quite good at it.
Kaihan Krippendorff: Yes. Right. I mean, at least just me personally, I think the experience of the creativity that I can what I would call creativity that I that I see in AI is remarkable. And yet, I would think maybe it's, like, 6 months ago, kind of the narrative was, oh, well, humans will always be the creative portion. Just give us, like, just a little bit about AI being creative and what does that tell us about creativity is?
Ethan Mollick: I mean, the problem when you talk about AI stuff is that we have all of these tests that were good enough when humans were the only intelligence of the room that no longer makes sense. Now there's other forms of intelligence available. So we have very mediocre test of creativity of divergent thinking of conversion thinking of ideation, and the AI beats all those tests. Right? Now, you know, we can argue is that true creativity or not?
I I don't know. Right? Because we didn't have to operationalize it. Just like we didn't have to operationalize theory of mind, which is, you know, can you take someone else's perspective the AI apparently does that. We don't have a definition that separates out the ability for humans to do it from AI to do it.
You know, there's the famous turing test. So whether an AI can pull a human to, like, it's human. That was a mediocre test, but it didn't matter because it was the best test we have. So on the creativity side, you obviously should be using AI for idea generation. We have objective data on that.
You should be a little cautious when you use it for objective idea generation that you also join your own ideas outside of it so that you come with your own set of stuff. But it beats most humans. In ideation, and good prompt thing makes the ideas that generates even more diverse.
Kaihan Krippendorff: Would you suggest people use AI before they introduce their own ideas in parallel or after?
Ethan Mollick: I would generally introduce your own ideas first and then have the AI generate so. Because it's the same reason why brainstorming is a terrible idea. I mean, we've known like, brainstorming is about, I think, 19 55. By 19 58, it was clearly a bad idea. We keep doing it because it feels like fun.
But as soon as you're in a room, people start yelling ideas, your ideas become bound to their ideas. So you should always start by bringing your ideas eye down individually. In the same way, I would recommend starting the same way with the with the AI side.
Kaihan Krippendorff: Great. So let me let me just double click or add the other components. If we say strategy is sensing possible futures, and making choices now to get those possible features. Probably the steps within that are, like, thinking about what part of the system to look at, ideating options, and then selecting options. If we think about the different It
Ethan Mollick: does all the steps. Does it does all of those really well. Like, we have paper. Like, it does good idea selection. It does good ideation.
It does good variance. It does good variation. Like, it's good at all those things.
Kaihan Krippendorff: I didn't I didn't I have to personally experience them all except for the idea selection gives us a little bit about that?
Ethan Mollick: Well, we've got, you know, a bunch of research work that suggests that the AI is pretty good. It's like the ideas. You give an extra criteria. Even better if you tell what kind of ideas you like and you say rate the ideas according to this rubric. And it aligns with our humans to rate those ideas according to a rubric.
Kaihan Krippendorff: Amazing. Amazing. I hadn't seen that. And so then where would you see AI fitting into the processors of strategy inside that corporation?
Ethan Mollick: So, I mean, we also know AI does good strategy work. Right? Like, there's a nice paper showing you asking the AI a few times. You get us good answer. Because the strategy was better.
We know that if you're a high performing small business owner in Kenya and you get an advice from AI, your profits go up 18 percent. Like, I mean, you know, to the like, I mean, a direct advisory role. Yes. Simulating potential futures. Yes.
Give you alternatives. Sure. Role playing against you. Definitely. You know, in in the brain knowledge, just a second check on your own perspectives or ideas.
Yes. Do that too. Like, you know, like, there isn't a piece, like, watching your, you know, your situation happening in your in in your environment to telling you whether or not it's you know, you literally watch with cameras as seen and telling you what it thinks of it, like all of those things.
Kaihan Krippendorff: Amazing. So then maybe I'm thinking of it wrong. Because, like, kind of my mindset has been kind of what are the unbundling of the different action activities and where is AI better than humans, where a human is better than we kind of divide up the work accordingly. It sounds like I'm thinking of that incorrectly. How do you think of it?
Ethan Mollick: Well, I would tell you the difference between cyborgas, center of work. Right? So cyborgasbord as you divide up the work, half person, half horse, and cyborgasbord as you blend it together. And so, like, I think you blend the work together. I mean, I the idea that you'd be building strategy without you have a super intelligent I mean, not only super intelligent, but quite intelligent, like, extra adviser, why are you not using that?
I mean, our BCG study showed it was basically operating at the level of, you know, of high end BCG consultants. I mean
Kaihan Krippendorff: Yep. Yep. Yeah. Tell us a little bit more about that BCG work. You I think you did Yeah.
Ethan Mollick: It did work with a whole bunch of people at Harvard MIT, University of Warwick, and BCG. And we did an experiment back when GP4 came out that gave those x to g p to GP4. We took 8 percent of the global workforce at BCG that we created 18 rails consulting tasks that they actually use for hiring. And some people got access to GP4, some did not. The people got access to GP4 had a I mean, a 40 percent improvement of the quality of the work is continuously, 26 percent more work done, 12.5 percent.
I just was 26 percent faster, 12.5 percent more work done. Like, instantaneous effect across every 1 of 52 variables we measured. And, you know, it basically was doing the consulting work out of the box. Right? It was lifting the lower performers because it was basically even in our performance.
Kaihan Krippendorff: Amazing. Let's talk a little bit more about that. So we talk about Agentive AIs. You know, AI being something that can induce things as opposed to what most of us experience right now, I think, is non Agentive AI.
Ethan Mollick: Right. So we already see agent AI, cloud with computer users, agentic, and so on. And an NGINX system is 1 that considered their own goals and big action on their behalf. Right? So this is already coming.
Like, this is not a fiction. I can already like, I signed an agenda AI to go on to, you know, go do some comparison shopping between Walmart and Amazon. And it actually goes and opens the web pages, does some searches, tries to buy some things, writes that notes to me without me doing any work at all. So the logistics systems are already on their way. They're I mean, they'll be here they're already here, but they'll be prevalent in a few months.
It's pretty impressive.
Kaihan Krippendorff: Amazing. Okay. So I like it like, what can we learn about you know, the kind of evolution or the adoption or the incorporation. I don't wanna use the words. I don't know what the right word is for it.
Of, you know, you say, like, this is a general purpose technology. It's their steam engine, the Internet, were such general purpose technologies. You look at the pattern of the evolution of these technologies and how they change things. What can we learn about other past general purpose technology adoptions, revolutions, if you will, and that that can inform how AI will evolve.
Ethan Mollick: So I think the biggest thing, you know, to the strategy folks listening here is, I think, in some ways, the biggest danger is to keep doing strategy the way you're doing. You know, I'll make the connection in the short, really short second here. But, like, the idea that is what I worry about is that firms have given up large parts of their internal strategy, but how they organize and how they think about work to outside consultancies and to toolmakers, right, as SaaS providers. So, you know, who tells you how to do sales is actually Salesforce. Tells you how to organize your HR department's Workday.
Right? Maybe you bring in McKinsey to help you with some of the stuff. And that will not help you, Brian and large, in the world of AI. Because it's when you look at what happened in the industrial revolution, the steam engine was important. But what made the industrial revolution happen was thousands of skilled artists and figured how to adopt the back and forth motion of the steam engine into running mills and equipment within factories.
And you have to do that. If you're waiting for the SaaS providers to figure out how to make AI work for you, you're lost. Right? So we have to think about organizationally, how to reorganize around the world of AI. How do we make sure our work works the right?
Like, we have to we have to go deeper than where we are right now on this. And it has to be experimentation decision making inside the organization.
Kaihan Krippendorff: Okay. I got it. So you as you said at the beginning, we don't really know what the answer is. So maybe this question maybe the answer is we need to find the answer. We had Adi Goldfarb on the podcast, and he had this nice analogy of when electrification came to be, there was a slight, maybe, little bit of efficiency in elect in replacing a steam engine with an electric engine.
But then the real opportunity came as you said, when we reconfigured the entire factory and to align, what do you see that did any inklings or beginnings of the contours of what an organization looks like when it's reconfigured?
Ethan Mollick: We're getting an idea, but just barely at this point. Right? Like, it it's a fundamental reality of work. Right? AI is an empathetic manager.
AI is a good adviser. Like, it has to change where you have been in management now. I mean, we invented the order form and, you know, the more about an order chart of 18 55 for the, you know, New York and Newrow railroad, and we haven't changed it since effectively. We, you know, we went divisional versus functional you know, and we invented the matrix, you know, which was our Mankind's greatest mistake. We went through a period where we, you know, developed in the 19 tens of time clocks and assembly lines in the early 2 thousands to agile, but the we have to challenge the assumptions where what happens when humans are the only people who can come to the table.
Kaihan Krippendorff: Yes. Gotcha. Alright. Yeah. So, yeah, we have to kinda start with a blank sheet and redesign with this future reality, current reality.
Do you have any sense though where profit is gonna pull in industries? Like, who's gonna is it the tech just gonna make the make I
Ethan Mollick: I mean, I think I think that this is this is inherently a human technology. If you think about, like, like, if you think about, like, software, you're in trouble because it doesn't work a software. Software shouldn't argue with you. Software shouldn't make the right like, and coders are often really bad at using AI, and they have very kind of unmanaged of uses. It's a it works like working with people.
So where the benefit comes from, what happens to do? 10000 PhD analysts? And if you don't have an answer to what to do with 10000 PhD analysts, be thinking really hard about that because o 1 is a pretty good PhD analyst in lots of fields. What can you do with that? That's a fundamentally different question.
Then how do I improve performance 12 percent by acquiring people and replace them with AI tools? Right? And I think the difference is that companies who view this for a period of abundance as opposed to viewing it from an efficiency gain are much better shape. Give it to IT and it becomes an efficiency instead of tools. Right?
And, like, the like, it has to be the level. And, by the way, the number 1 piece of advice that gives all your strategy all the strategy of listening is if you haven't used frontier model for at least 10 hours, you're making a terrible mistake. Like, you have to use this yourself. You can't delegate use.
Kaihan Krippendorff: Got it. Yeah. I wanna get into some of the specific tools that we should be aware of that we're not aware of. But let's just zoom out a little bit, you know, from the organization to the industry. Do you have a sense of, like, where profit starts pulling across the industry?
Like, you know, is it the technology firms that are developing the tools and technology that are gonna profit? Is it the you the companies that adopt it intelligently that are gonna create the advantage? Like, where does competitive advantage shift?
Ethan Mollick: Competitor shifts to who figure who people who figure out how to use this in fundamentally new ways. Right? You I mean, what do you do with intelligence? Like, it has to be like, there's organizational redesign, but my biggest fear is talking to companies who are, like, Well, we've allowed a temporary Copilot. And, like, no one's doing experimentation.
They're just kind of, like, hoping someone else develops a solution. If you're the first company who could figure out how to steam power something, let's go with what's the steam power equivalent of a thousand mines a thousand extra mines for your training firm? Do you do with a thousand extra months? Like, it's possible right now. Like, you know, what do I do when every ATM machine can offer financial advice?
What do I do when I can produce individualized marketing at scale, you know, in any way that I want translate information in between thing? What I do is have an agentic system that every system can go on the Internet and talk individually to people or do work. Like, there's an imagination problem here that I think has to be conquered first.
Kaihan Krippendorff: Yeah. Yeah. I see. Sort of, like, the fixed cost over the long term become variable costs, but that long term is now getting really short. And it's gonna be painful to reconfigure.
It's somewhat of a selfish question here. There is 1 area that I've been really interested in recently, and I'm looking at is if we think about companies organizations existing to take needs, like, sensing customer needs and finding solutions to those needs, there's, like, a matching problem there. And I'm wondering it seems to me that, you know, AI should play a role, you know, the fact that pharmaceutical companies can test thousands of potential molecules and kinda shorten the match time from the application to the to the solution. Do you see AI playing like, how do you see AI playing a role? So a sick of a company as being having a function a central role of identifying needs in the marketplace and then creating or finding a solution to those needs.
Most such marketplaces are highly inefficient. How can AI help?
Ethan Mollick: I mean, have you asked that? I guess is the question. Right? I mean, There's agent next steps that we could find needs. You could simulate customers with it.
They're I mean, look, I've been throwing HPS cases into o 1, and it cracks them. Right? Like, new ones from the last couple months. Like and, by the way, it shouldn't be able to do accounting, but it pulls off some degree of accounting. Even though that system have been optimized for, and they will be soon.
Like, I just I mean, I think that it you know, the answer is, like, yes. Talk to the system. Right? It stimulates customers really well. It does you know, like, beating all your market data.
Where am I finding a mix of a need in this kinda case. So, you know, III had an experience when I asked 2 an agent, I asked 1, give me a strategy to make money with me doing no effort and you doing all the effort with your current tools that's as safe as possible. And I created a Python code to automatically buy ETFs and short term bond funds. And then I said take the most dangerous method possible, and it created an elaborate code on the blockchain with smart contracts that would let me do flash lending to support crypto sales. I have no I have no idea.
Like, you go under what these things can do.
Kaihan Krippendorff: Amazing. Amazing. Alright. I've got so many more questions, but we're reaching the top of our time with you. I think you'll so I think we kinda my conclusion here is, look, we don't know.
But if you don't dive in now, you're gonna be left behind. So you need to start getting familiar as the And you're conscious of what I'm trying
Ethan Mollick: to do. That that feels mild. I think you have to be doing r and d. I don't think you have to dive in now. I don't think like that's a thing.
Right? There aren't a lot of products about it. Your job is to figure out how useful this is. They'll only be looking to tell, like, AI makes up of all the time. It basically, you know, hallucinates, and those little stages are plausible.
I, you know, I just did a which is an experiment where we had, you know I had 2 PhDs suggest that this novel, like, prove this new this existing theorem about neural networks in a new way. And they've been debating for 2 days over whether or not the novel proof is a completely new, you know, new approach or not because you have to be a PhD to understand this, and I have no idea. So part of what I what the what I would say is that you have to be experimenting. In your area of expertise, experimentation is cheap. And outside your area of expertise, it is it is not.
So you have to be actually experimenting. What are your benchmarks for AI success? Have you tried using it? When new model comes out, have you tried to figure out what that does? If the answer to those questions are no, then you're I think you're falling behind.
Kaihan Krippendorff: Yes. I gotcha. So what can you just give us some starting points? Like, you know, I saw you give a lecture at Wharton and you pulled up lots of different tools that are available now that, you know, probably are just you know, matter of fact for you, but for other listeners, they might not be familiar with who would be, like, 2 or 3 ones that we should start getting familiar with.
Ethan Mollick: I really think this is less about, like, a tool because the truth is the foundation models are really good, and you should probably just stick with the foundation models. And so you should probably be using either Cloud or Gemini or maybe I'm sorry. Cloud OpenAI or Gemini. Right? Those are those are the options you should be using.
Kaihan Krippendorff: Okay. And just start pushing the boundaries. Just start seeing what they can do and be surprised at what they But
Ethan Mollick: But methodically, like, what, you know, what jobs are like, it's not I think there is this methodical angle of turning this into research.
Kaihan Krippendorff: Okay? Just 1 1 last question that sort of that sort of come up is, you know, a lot of times the solution can't be explained. Like, we can't understand the solution that AI is giving us. How do you see that evolving? Do people just start trusting?
Ethan Mollick: I mean, I think that you have to this is, again, where it's an organizational problem. We're used to really control systems for flawed people. Right? They're called organizations. Right?
Like so the issue is, like, how flawed is the AI in your space? How do you know? Have you tested it? How what's the human benchmark? When do you trust it?
When do you sure not? How do you check the work? Like, this is not this is literally it's just for the last 200 years solving this problem. Like, we have answers..
Kaihan Krippendorff: Yep. Yeah. Yeah. Sure. And I think that humans are more flawed than we like to think, and so the benchmark maybe isn't as high as we want to believe.
Cool. Awesome. Well, Ethan, thank you very much. I know we just scratched the surface and appreciate you spending the time to give us a little bit of inkling into where things might be going and what we can do to be prepared to be part
Ethan Mollick: of Amazing. This is a lot of fun. Thank
Kaihan Krippendorff: you. Thank you. Thank you to our guest. Thank you to our executive producer, Kareena Reyes, our editor, Zac Ness, and the rest of the team. If you like what you heard, please follow, download in scribe.
I'm your host, Kaihan Krippendorff. Thank you for listening. We'll catch you soon with another episode of Out Thinkers.
