Mostly I get asked, 'What should my kids do for a living?' Probably from all of them. We actually have a similar question. I'm sure you do. Everybody asks that, that's right.
And it was even worse because I would basically go to bed at like 4 a.m. I actually sent emails to my entire faculty. 'Education Must Be Completely Reinvented.' A World Completely Transformed. No doubt students are cheating at a rate they've never cheated before. No one's going to show you they're using AI in your company, if they're worried they're going to get fired if they're more productive, right? School and work, turned upside down. How should we live with AI? I think this is a chance for exploration. So I mean, I would go one step further. 'Ambition is rewarded right now.' And even a failure is further than you'd get before and teaches you something.
Hi, I'm Professor Ethan Mollick. I teach and research at the Wharton School of the University of Pennsylvania. Run the Generative AI labs there. I've also written a book on AI called 'Co-intelligence' which is sold in Korea as along with the US.
'AI isn't a competitor — it's a collaboration partner.' How do you define AI? So, it actually one of the biggest problems is the definition of AI is really vague right? So if we talked about AI prior to sort of 2010, we were talking about sort of the science fictional idea of artificial intelligence where only a few computer scientists cared about it. We talked about it from like 2010 to 2022, you'd be talking about big data. And then since then we've been talking about AI, we talk about generative AI like Large Language Models and related technologies that are very different than what came before.
What AI tools do you use most these days? So, when you think about AI right now, you kind of think about three things. You think about the AI model itself, right? So, the big three at the time we're recording this. But it's not just about the models, that's the brains of the AI. It's also about the actual, software applications you use. And then finally, what we call 'harnesses'. 'Harnesses' are the pieces of equipment given to the AI to let it do things. So, 'harnesses' let AI create images if they need to write code, do web searches, all the kinds of access your files. So now when you think about AI tools, you think about all three of those together. So I use all of them, but I tend to use a lot of OpenAI's Codeex and Claude's Claude Code and Co-work, which are versions that use your computer to help you do more work.
Why do you call AI 'Alien Mind' rather than 'Artificial Intelligence'? Because a few things. One is I think we all have this idea of AI from many years of television and movies, right? It was this, it's a machine that if the stereotypical bad science fiction version is it can't, 'I do not understand love' and then the robot explodes if you try and explain love to it, right? Like we expect it to be logical and to not act quite like a human and to be all about sort of cool reasoning to understand emotion. Instead we got Large Language Models which in a lot of ways work a lot like people. They are all about emotional content they are about empathy. They were originally worse at math, than other topics. Now that they've closed that gap and so it's worthy thinking about them as a person, because that's the most easy way to get them to do work. At the same time, you have to remind yourself they're not a person. So, 'Alien' kind of puts that nice distance between yourself and the AI.
The day ChatGPT first launched in 2022 what was so shocking about it? So 'The three sleepless nights' idea from the book is really one where, everybody who uses these systems enough has like a moment of crisis where they're like, 'What does this mean? What am I doing?' And actually, I work with my wife on this a lot. We run the we started the AI lab together. We do a lot of work together. And it was even worse because I would basically go to bed at like 4 a.m. Then she'd wake up at 4 a.m. and start working on AI stuff. In the morning, we'd be like passing each other messages about what we did. And to me, it was, 'Wow, this thing could do stuff.' So, you start projects I couldn't accomplish. Now, I'm doing it. 'What if I test it this way? What happens?' And so, there's a lot of anxiety, especially when you first use AI, it can almost, almost miraculous. Then you start to see the flaws after time, but at first, it's like 'What does this all mean for us', and so that really hit me in November. I actually sent emails to my entire faculty saying we really have to change everything about how we teach the next day and they've come to believe me over the last couple years.
How is AI now different from AI when it first came out? So there's a few things that are different. The biggest difference is the ability. GPT-3.5 which was the original ChatGPT is just dumber on every characteristic. And we could talk about test scores. It was basically it looked like it was sort of a smart high school student when you measured it. And now we've got GPT-5.5 is solving math problems that only two or three people on the planet can solve. It also has become 'agentic'. Now, this is one of those buzzwords that means a lot of things, but an agent is basically an AI tool that can go ahead and do work for you. If you say, 'Put together, you know, a TV script for me or something,' it won't just write an answer. It'll do research for you. It'll look, you know, create images. It'll look up information And it will actually do hours of work for you by just asking. So those are the big gaps.
What do you make of claims that AI's rapid progress is a 'bubble'? A Bubble? So, I could see arguing six months ago that AI was a bubble. Because we hadn't yet entered the sort of agentic era of AI use. I don't think many people think this is a bubble right now. Companies, I talk to CEOs of companies all the time who are getting huge amounts of value out of the system. I keep talking to companies that are blowing through their entire AI budget in months that they expect to spend on years. So there's a lot of money that people are getting value out of them as users of these systems. So, it's hard to imagine there being a real bubble. Now that doesn't mean financial markets do what they do, right? It's always possible that something's over inflated or not. So, we can separate a financial bubble, But if we want to talk about 'Is there a use bubble? Is AI going to go away? Is it not useful?' I don't see any way you could argue that at this point.
Just two weeks after ChatGPT launched, you co-authored the academic paper on AI in Education with your wife. What made you write it that fast? So, we have actually collaborated for a long time. She has a doctorate in education. 'A New Way to Learn With AI Chatbots' [Co-authored by Professors Ethan Mollick & Lilach Mollick] Turning AI's flaws, wrong answers, hallucinations, Into a teaching tool. Three assignments to sharpen students' critical thinking and boost learning. We ran before this something called 'Wharton Interactive' and we built games for teaching. So, I've always been interested how do we democratize especially business education for the world. So, we built games, that would teach people how to do entrepreneurship. We built a fake space simulator that actually secretly taught you leadership skills. So I've been spending a lot of time thinking about how do we teach people with AI. And when ChatGPT came out, we saw that it solved a whole bunch of problems that have plagued the education field for 10 years. I mean it's causing problems too to be clear but, also solved could solve a whole bunch of problems. No doubt students are cheating at a rate they've never cheated before, but on the other hand it could be an amazing tutor. And we thought it was important to illuminate for all these people trying to experiment what were some ways to actually use this to go forward.
How are you using AI in your classes these days? I actually have been kind of running this. I teach a entrepreneurship class and that's where I've been teaching about AI and I just did a new version of the class called 'vibe founding'. and students launched startup ideas in three days. So they had to go from they were doing a semester's, worth of work in three days. So, AI helped generate the ideas. It built the business plan. They had to actually launch with working code, which a lot of them were never most of them never coded in their lives. They had working code. they had to have a working PowerPoint presentation. All the financial models all of that had to be built. So, I know you can get a huge amount further today than you could.
So how should students actually work with AI? So the first thing that is really annoying to tell students and nobody likes to hear is that, 'Learning is hard and kind of has to be hard.' then we so we call this 'desirable difficulties.' It is the phrasing in, in education theory. And so part of the problem with using AI is if you're just asking for help, the AI is trained to be a helpful assistant to you. And it will give you the answers to problems. and you'll think you're learning something, but you're not actually learning because you're not doing the mental work yourself. So, it's really important to get the AI to act like a tutor or a teacher. And there's a few ways to do that. First of all, all three of the big AI companies have a 'tutor mode' that makes the AI less likely to give you answers. So using it to tutor you to ask answer questions to quiz you, very powerful for AI use and now we have data that shows that's true. There's a really nice study one out of Taiwan, another one out of Africa that show that when you make AI work like a tutor, you get big impacts in education, but if you just ask it questions, you're not going to learn.
Could you give us one example a student could try tomorrow? Sure. So, a really useful thing might be to say, give the information about yourself. 'I am a student in Seoul and, in high school or college or whatever grade you're in studying this.' 'Here's my interests. I'm having difficulty with this.' here are a couple math problems that I was asked to solve or a couple history problems.' 'Can you help quiz me on these topics and then explain what I'm missing and help get me understand the missing pieces and then quiz me about them again?' 'So, just ask and say don't give me any answers. I just need to be able to learn this on my own.' And that will get you part of the way there.
What's the AI role you use most often? So, there's a little update to this too. We've done some research at the Generative AI lab. It used to be personas did two things for you. They would change how the AI responded so. act like Socrates. But the other was it would actually make the AI better at things. So if you said 'You are a physicist' it would do better at physics. That's no longer true. So telling the AI you are good at physics or you're a physicist or you're a mathematician doesn't make it better at math but it still does change how it talks to you. So the personas I use the most often and the one everyone should actually use the most is 'Be a critic'. Because remember, we talked earlier about how AI can be sycophantic, how it wants to make you happy. But what's actually needed is honest answers and challenge. So 'Be a critic.' Is one of the most useful things you could say. When I'm writing I often have different personas for writing too. 'You are a reader who is confused about AI. What should I clarify here?' So, I use personas for both kind of projecting what my work might do but also to help me think more clearly.
For education to truly transform in the AI era, what does the education world need to do? So, part of this is, we have to recognize that there are losses along with gains. Like people are cheating right now. A lot of educational methods that we used to use that worked really well don't work well anymore. If you assign long essays to people, the AI will help them with essays. Like more of your student your students will always wanted to cheat. They're always busy. They have other things to do. They're very incentivized to get high grades. They will cheat. So we have to think about how we're solving that problem. We have to take advantage of the strengths of AI as an individualized personalized tutor, but also think about how we're going to assess people, how we're going to make classroom experience meaningful. The right way is probably something called a 'flipped classroom'. So you'll use AI tutors outside of class to teach you things and inside of class and a lot of people already do this, like my students are assigned YouTube videos to watch outside of class and then in class they do math, right? That's how we're going to end up doing things in a lot of fields. Outside of class you'll be assigned to use AI to learn something and then in class you'll be doing exercises, experiences, working with other students on things. I think that's how things will end up looking.
In the AI era, what kind of people and companies will survive? I think you want to be a 'T-shaped person', right? So, I think you want to be a generalist with one area of deep expertise. So, deep knowledge in a field, have a good understanding of related fields so that you know different things you directions you might want to head in. then having a sense of 'taste' just about what you want to do, right? Like the AI can give you thousands of images, thousands of paragraphs. How do you select the right one? And then finally, the idea of 'agency' of actually going out and giving things a try. So I find people who experiment, especially experiment in their field, tend to be very good at using AI. In terms of organizations, what I find is that if the CEO or top leadership team believes in AI deeply and is willing to take the risk that comes from experimenting, that is a big change that is a big factor in company success.
What does an individual actually need to start? 'Try and use AI for everything.' I wish I could say that there was a cheap way to do this, but I think you have to pay one of the three big AI companies. You know, in the US it's $20 a month. There's $8 a month versions in some countries. But you have to pay and you have to use the better model. And you should try using them for everything. Like try you're going to interview me ask it to do the research In your field of expertise, it's very cheap for you to start to learn is AI good or bad. You'll see the question you'll be like 'It's not very good asking questions.' But maybe if I ask them a different way and prompt the AI a different way, I'll get different kinds of better questions. So experimenting in your field as much as you can with real tasks is the big thing.
Is individual productivity growth actually translating into company performance? So, it's a really interesting question. We know at the individual level both from research we've done and lots of other people that you get individual productivity gains that are quite large. Whether those translate to company productivity gains is at this point you have to think about how you change your company. So let's take coding for example. One of the biggest changes is in software programming. If you talk to any really good programmer right now, they are writing almost none of their own code. Like, literally they instruct the AI to do things. They spend their time thinking about how to manage the AI, manage the process, But they're not writing code anymore. And so, let's say you're 10 times more productive or 100 times more productive. That you're individually more productive, but has your company changed how it's making software? Or are you still sitting in a meeting where every two weeks you have a planning meeting and so it doesn't matter that you're doing a 100 times more work because the rest of the company's holding you back. So, it's not just about individual AI use right that helps you as an entrepreneur it helps you make progress. But, it is beyond the individual entrepreneurial use. It is that bigger picture. How do we use AI as in organizations, in companies in government that you see productivity gains from.
Companies think of AI first as a way to cut costs. Is there a way to move forward without laying people off? To me, that is the most urgent immediate problem of AI, which is easy to know what to do with productivity, because we've done productivity gain. For the last 30 years. What do you do when you have an IT gain? You fire people, right? 'More likely we're five times, you know, 5% more productive. Fired 5% of our workers.' That is the easy path, right? And it doesn't require a lot of imagination. I think that leads you to disaster in a few ways. First of all, AI use we just talked about is being discovered by people using these systems in your company. No one's going to show you they're using AI in your company, if they're worried they're going to get fired if they're more productive, right? So, you're not going to see any productivity gains. So, you're sabotaging yourself. But the bigger issue is that failure of imagination means eventually you're just turning over your business to be run by Claude, right? Or by ChatGPT, and that you have no advantage versus anyone else. So I think the big thing to experiment with is 'How do we make humans do more?' 'How do we help them and support them in parts of their job that they like least?' That's a really big challenge.
For a company just starting with AI what are the things you tell them to do first? So the first thing you need to do you need three pieces to succeed. You need what I call 'Leadership, Lab and Crowd'. So, 'Leadership' means that you need your leadership team to be thinking about this and they have to try it themselves. they have to decide what incentives there are for people using these things. How are they going to talk about what their goals of using AI are? How are they going to start to re-imagine how their company works? For the 'Crowd', the crowd is everyone using these systems. You need to give people access to a good AI model. So you need to give them access to a model and give them a little bit of training, give them encouragement to use it. And then you need the 'Lab'. You need to set up some part of your company that's doing 24/7 experimentation with AI. It doesn't have to be technical people, by the way. Often it's not technical people. It's the people in your organization who are already coming to you every day saying, 'I figured out a way to solve this with AI!' And you're like, 'Okay, go. Go back to your cubicle.' Those people will form your lab. So you need those three pieces in place.
With AI blocking entry-level hiring, how should young people build their careers, and how should companies develop talent? So, those are two really hard questions. first is, it's really bad idea for companies to give in their junior pipeline. They're going to have to think about how to train people differently, but if you don't have junior people, you're not going to have senior people in the future. So, we're going to have to think differently about how we bring junior people in. They may not be as useful compared to AI, but they are important for the future. We're not seeing a huge collapse yet, in sort of junior jobs. There's sort of signs it might happen. We don't really know, right? But I think that, the idea is being flexible as a junior person entering the field as a student, not betting only on one thing, which is good idea anyway. I mean, there's lots of people who wanted a job in finance in 2007 and when the financial crisis came along, those jobs weren't there. So, I think maintaining some flexibility is important, but it's really on companies to think about how do they keep their junior talent pipeline going?
If you were 22 today, what would you actually do? I think this is a golden age for entrepreneurship. I mean, if the honest truth is the best thing you could do with your time is launch a startup company, right? It's an incredible time for that. By the way I don't mean an AI company like it doesn't mean that you have to launch yet another AI company. But I am where I'm seeing a lot of AI transformation is in boring businesses where people say like 'What if I think about insurance?' Or what if I think about you know how do you use AI to better manage a golf course or how do I use AI to better manage a laundry mat? At 22, you don't know a lot of things you have a lot of disadvantages, but the AI does know a lot of things. So suddenly it can support you and help you write code, it can do all kinds of stuff for you.
Some scholars warn about an AI 'doomsday,' while others frame it as humanity's salvation, both narratives are emerging at once. How does your view compare? So, there are serious people worried about this, which should be enough to be to take this somewhat seriously, nobody knows anything. Interestingly Nick Bostrom is now like in favor of fast AI development. So, even though he invented the idea of the paperclip maximizer. I think it is worth taking that seriously. I think that I worry a little bit that conversations about doomsday or 'The AI saves us all and cures all disease' that stuff ends up taking all the oxygen out of the room. Because, it's really hard have an argument about how to best adopt AI and education 'When you're like in five years we'll all be dead or saved.' I think the far more likely scenario that we should worry about those kind of things is that the world looks like this one but projected forward. That the world is not ended or changed and we need to spend a lot of time thinking about the world we want to build as a result.
We may be the last generation able to define what AI becomes. What kind of presence do we want AI to become? And in some ways my really annoying answer is, this is what I want everyone to be debating. We're all going to die who cares, and then or it's all hype it's going to go away, and I don't think it's going to go away. So I think it's lots of little choices that make this decision. Is it okay to use AI to interview me? Would it be all right? You know I've got the a great crew of people here filming me, a lot of work went into this kind of meeting. Is that a thing we would let AI do? Is that a good idea or a bad idea? We get to decide those kind of things, right? It's not the technology is not automatic. It doesn't do things to us. We get to decide what it does.
I heard you have teenage children. How do you hope they'll live in the AI era? So I mean all the usual. Nothing changes about what you want as a parent. You want your kids to be happy and kind and flexible and successful and all the things that you want. Humans don't change that much, right? Now, what do I want them to do to navigate a technological environment that's changing? Flexibility. But, careers are long I teach students very few of them end up going into their career the way they think it's going to be. People change jobs, life changes where they go. The things alter and vary. So, I think 'resilience' is important.
Some people say 'Will my job survive?', then how would you say to him? So I think you should think about a job as not one thing but as many things. A job are bundles of tasks. So, as a journalist, we were just talking about there's a bunch of stuff that you already do, right? You have to fly here for these and fill out your expense reports and interview me and prep for an interview and, go to story meetings and whatever all the other things that you end up doing, right? And so, if AI is good at some of that stuff, it's probably not good at all of them. It transforms your job rather than destroying it for most people. And also, you just pointed out earlier that you can expand your job, too. Suddenly, you can do things you couldn't do before. So, I think job transformation is more likely than job destruction. I think this is a chance for exploration. So I mean, I would go one step further I think it's a great time to expand what we do and to think you know big. 'Ambition is rewarded right now.' And even a failure is further than you'd get before and teaches you something.
Thank you so much for making time for us. These were some of the best questions I was ever asked. So thank you. You're amazing. I appreciate it. - Thank you very much.
