So when you're interviewing for a job and everybody's good because of AI, how do I stand out and get a job that's still that they still need to hire for? I'm fascinated by that.
Like if Claude is really good at running your company, Claude's also good at running every other company and there's no variation between them. And generically high quality with no variation means there's no boats or competitive edge. I think humans who bring competitive edge to this one way or another just by providing variation if nothing else. Yes. Is a useful way to think about problems. Right. You start to care a lot more about the taste of that person than you do about the entire organization built to deliver the product.
We're all a little afraid of AI. Even the people who love it are a little bit afraid. Some of us are afraid it'll take our jobs or make us dumber or change everything so fast we'll never be able to catch up. It doesn't help that most AI experts fall into one of two camps. They're the doomers or they're the zealots. That's why I wanted to talk to Ethan Malik. He's an AI expert who, refreshingly is neither. Ethan's a Wharton professor studying how AI, entrepreneurship, and innovation impact our work. He focuses on how employees actually use these tools rather than how they're theoretically supposed to improve our work. His New York Times bestseller co-intelligence and his popular Substack, One Useful Thing, have become the go-to resources for all those people trying to make sense of AI without losing their minds. Ethan believes we have more agency over AI than we think. It's our experience, our taste, and our genuine points of view that are things that AI can never replace. And that's important because the choices we make about how we integrate AI personally and at work as a society will shape what it becomes. If you like this episode, please remember to subscribe. This is a bit of optimism.
You are very popular right now. What were you I know you teach entrepreneurship and other things before AI. What was the what were you teaching that was the the the hot thing?
In addition to AI stuff, my other thing I really care about is games and education at scale. Two related things. I've done a lot of work on using video games for teaching. So I was I I was as the brand says not famous but known in that space of thinking about games and you know teaching at scale. How do we teach transformationally? So that has been something I've worked on for a long time and the AI work was sort of in parallel to that.
And was it your own personal curiosity about AI that that sucked you in or was it that you you were forced to use AI in the work that you were doing because it had to because it made it better.
Went to business school and my PhD program at MIT and I worked with the MIT Media Lab with their AI group. So um Marvin Minsky was one of the fathers of the field and I was the non-technical guy in the group. So I was the person who had to explain to other people what AI was, how it worked. when we'd go talk, have technical conversations, I'd come along. I I've been AI adjacent or involved for 20 years, but always in the sort of non-technical like how do we use this, how do we explain it role and everybody else was technical because it wasn't really working out. Like AI had limited use cases. And so when GPT3 came along and started to be make a splash little and then chat GPD after that, I was sort of well positioned in this world of someone who'd been thinking about explaining this for a while in in a non-technical world.
The reason I was really interested in talking to you because I I to be honest with you, I try not to have AI guests on.
And the reason I wanted to talk to you, to be honest with you, is yours is just more practical. It's not it's not the savior. It's not it's gonna kill us all, but it's kind of like, well, it's here, you know, kind of like the internet showed up. What's the best way to use this thing? And it's a little more down the middle.
Yes. It's nice to be in a place where being somewhat pragmatic makes you unusual. Um, that's not usually the place where we get publicity for being the non-bombastic version.
But also those the other problem with those two opinions is they tend to eat the world, right? If you think that we're getting a machine god who's going to save us all, then you know all of it matters is discussing that, right? Just like any other sort of religious belief. And if you believe that we're all going to die, how can you have a conversation about anything other than, you know, this is going to doom us all? And the fact is it's a general purpose technology. It's going to affect everything we do one way or another. It's worth spending some time on that. And some of those effects will be good. some would be bad.
I do remember the sort of the rise of the internet. I'm old enough to remember it. And the conversations were somewhat similar, maybe less dramatic on either side.
There was a lot of positive energy at that time.
Mostly positive, but there was lots of conversation like I remember people who, you know, there were the zealots who believed that the internet and everything online was going to replace everything. You know what it is? It's the difference between eating to live and living to eat, you know. And if you leave it to the technologist, they think all technology is living to eat.
I mean, they literally made protein powder shakes. I don't remember a Soyent where you're supposed to remember Soilent. I remember for people for people who thought eating was too annoying. Why are we spending all of our time worrying about eating when we could just get it taken care of, right? It's a different view of efficiency, but it's also a view that kind of blinds you to the fact that technology is the most human activity, right? We're making tools for ourselves. How we use them, how we adopt them, how we regulate them, those are going to have big influences. You know, AI is more self-directed than most technologies, but we still have a lot of agency over what happens next.
I mean, I know from my own experience, most people are misusing or underutilizing the technologies that are available to us. And I'm already getting to the point where I'm turning on, you know, social media or listening to friends or reading an article. And I'm already feeling overwhelmed by all of the advice of people telling me that I should be using it like this. And I be setting up an agent to run my life and setting up an agent to do my marketing and set up an agent to do my finances. I'm already overwhelmed by all the advice of how I should even be prompting the machine to the point where I'm almost backing off and shutting down because one minute Gemini is that you got to only use that one. The next minute you got to only use chatbt. the next. Oh, got to only use Claude. It's all too much already. And for anybody who's not predisposed to like be allin that reaction, I think is pushing some people away, believe it or not.
Well, I absolutely I mean, look, I'm I am a nerd of the old school, so I like getting into the details of stuff and partially like explaining them, which is, I think, part of why you have me here. But also, it is overwhelming. I mean, part of what's actually interesting is AI has gotten easier, right? Not to be too, you know, too evangelistic about it, but like it used to be that stuff mattered. Like you had to like all those little details like prompt engineering mattered. So it mattered how I phrased things. It mattered if I said, you know, you are a physicist. It was better at physics. If I said, think hard about this, that mattered. If I offered bribes, that would matter. Uh we've been testing all that. None of that matters anymore. The models have gotten good enough that if you're good at giving instructions like a human to humans, you probably do okay with this. And similarly, all the models are getting good quite quickly. So whether you use OpenAI or Chachi there like Chachi or Gemini or Anthropic that all three of those are pretty solid. There's like one or two hints I'd give people but otherwise it's really just use it for stuff and don't stress how you're using it because you'll figure it out.
Also when we talk about like AI is going to take away all of our jobs. And by the way, those are the technologists saying that when one of the things that I also love is the technologists are also very fond of saying, you know, things like 80% of the jobs today didn't exist 20 years ago, which mean it's fair to say that 80% of the jobs in 20 years we can't even imagine. And to your point about prompting, it wasn't that long ago where the technologists were telling us prompting is going to be the thing. And people are saying, "I'm going to get a degree in prompting." And the technology got good enough in just a few minutes that that literally went away.
Yeah. And I I think that that is part of what makes this kind of interesting, right? Is we can't imagine the jobs, so we'll keep trying. And also not imagining the jobs does get nerve-wracking when like, you know, AI is here and it's like, what is the job? Well, I'm not 100% sure. I mean, I think there's both legitimate reasons to worry about jobs. These systems are very capable and very competent and they do change. They're impact real work. This is not electricity in the same way of like we need to figure out a way to harness or use this thing. There's a real impact.
I've talked about this a little bit. I actually think that you know nature pours a vacuum markets correct. When we have a bubble the stock market will at some point not of our choosing correct itself and all systems seek equilibrium at some point. I'm fond of saying that in the 70s and 80s robots started to enter factories and the blue collar world said hey we're going to lose our jobs and the white collar world said it's the future baby progress stops for no one reskill. You know, and you know, the pendulum does swing and because your plumber doesn't care about AI, the carpenter doesn't care about AI. The mechanic doesn't care about AI. The people who care about AI are the knowledge workers.
And you know, it's the future, baby. Progress stops for no one. Reskil, baby.
Yeah. That I mean, that's a really set I mean, there is a comeuppance, but there is another thing, right? Like if you look at the last three industrial revolutions, there's either two or three. I mean, that's all we've got, right? And the reason they worked out wasn't because the technology made everything great alone. It was because it was also labor fought against, you know, capital and you had a whole bunch of fights happening. Unionization is how the benefits got spread around, right? The technology doesn't naturally do that. What's really interesting is we're going to have a similar fight here, right? Like AI, we already have good data, right? AI is pretty good at being a doctor. It's getting better at being a lawyer all the time. Like most lawyers I talk to see a trajectory where not too long from now you will be able to get as good advice for not the most complicated issues from the AI if you're not already right and there's ongoing debate about this but the thing is lawyers are not going to be quiet about losing their jobs and it turns out a lot of Congress is lawyers and a lot of people don't donate money are lawyers and I'm willing to bet that you're going to see laws passed in every state that you need to have a lawyer officially a human sign off on something even if they're worse than AI right so part of this like there is a little bit of ah it's your comeuppance white collar workers, but it's also like no no they've got the like they're there are fields that are not going to go easily, right? And and you know doctors, lawyers, it's going to be interesting. Coders really do not have the protection that doctors or lawyers or actors or other kind of guilds or associations do. So we'll see a lot of conflict over these issues in the near future.
Lobbying for our own interests is not a new thing. So for example, for years there have been people who try to to raise money for the government, you know, where we can get more income. the the average lifespan of a dollar bill is about one year. The average lifespan of a coin is something like 30 years. And so the proposal to move dollar bills to dollar coins because it would save the government, you know, untold billions per year in printing and all of the rest of it. And the reason we haven't done it is because of the ink and paper lobby because they like those billions being spent on them to make new dollar bills every year. And so, you know, we're used to doing things that are not to our advantage because of lobbyists. And this is just another thing where those who have access to power and those who who have influence will protect their interests as you said
and not necessarily badly, right? Do we want all lawyers out of a job is an open question. A lot of people will be like yes please. But like you know I I mean already actually there's some early evidence that the number of cases being submitted to judges is expo exploding exponentially where people are pleading their own cases with AI law and like how do you deal with that? It used to be that we had a filter, right? So the secondary thing is you had to convince a lawyer to take your case and if they did a bad job they would be punished and if it you got a 100page legal brief a person wrote that and that was an indicator of real effort but now it isn't. How do we deal with that kind of system? So we're going to have all these ripples in all kinds of areas that require policy changes regulation changes societal changes even if we don't have sort of an apocalyptic AI event the way we were talking about earlier.
So let's let's go from the theoretical to the practicals. And as we sort of said before, which is a lot of people who aren't technologists are underutilizing or or misutilizing this this remarkable technology. Some are still using it as a glorified Google. Google does search. This is more of an answer machine, but still it I I mean we all do that because it is a very simple and fantastic use case. the more nuanced and complex prompts I think I know I'm underutilizing like I ask for an answer even if it's something that has some depth but I don't ask it to write a report or make an interactive you know dashboard of the data that I'm not I know that I'm not doing that and I'm already ahead of some in terms of my utilization of the thing I have two two questions one is you're teaching students you're teaching a lot of grad students at business school is it a false belief that just because people are younger, they're all in on this technology or are they also sort of bumbling and fumbling their way as they as they learn about it?
So, we actually have a paper on this like so I think what people think I hear this term in the internet we're talking about digital native all the time. People talked about this, right? Like kids these days are good at using the internet and indeed like if you talk to somebody who grew up with Tik Tok, they will know all these intuitive ways that you know as an older person you will not understand like oh that's cringe. There's a bunch of rules and slang and approaches that you want to take. And I think that that model kind of carries through. The younger people get the technology. It does not hold for AI. Like I talked to like a CHRO and he's like, "Oh, the kids these days they're they're AI native." I'm like, "They're not AI native. You're just talking to Claude. They're conduits to Claude." Like if you ask for a report, they'll give you a beautiful report. They have no idea what's in that report. How could they? They have no knowledge of this. They're just giving you what Claude says. And we actually found some evidence on this when we did a study at BCG at Boston Consulting Group. We found junior employees were often much worse at using AI. They seemed like they were using it. They're adopters, but how could they judge whether something was good or bad? I think this is a rare case where the more experienced you are and sometimes the older you are, the better you're going to be at using AI if you decide to use it because you can intuitively grasp how do I give instructions. AI works enough like people that if you give an instruction and you're good at a field, good to be like, no, no, I understand what went wrong there. You're thinking about this, you should be thinking about that. Even though the AI is not a person, it doesn't think you'll know what kind of information to expect to get good answers. So, I think actually experience really does matter. If you think about how our education systems built, we're we're actually schools are in chaos. I mean, they're always in chaos, right? Universities have always been in chaos. This is not new. You know, people are cha, but we actually know the pathway forward, right? Like, it's going to be a little bit messiness, but like we'll do in more in We did this with calculators. We'll do more in-class assignments. Outside of class, we'll use AI tutors, which turn out to be very effective at controlled experiments. We'll make them work better than we do now. In class will be active learning. We'll figure it out. I'm not worried that we can't figure it out. I am worried about the next stage. I I teach people to be generalists at Wharton and I they become a specialist the same way we've taught specialist for 4,000 years which is apprenticeship right I send them off to work for whomever right they go you know they go work at you know Bank of America or whatever and they learn the job and everyone gets a good deal they get a little bit of income but not as much as they would probably deserve but they get a chance to prove themselves and they learn the ropes by doing grunt work over and over again the middle manager assigns them grunt work that they don't want to do anymore and gets to evaluate whether this person is any good or bad for moving up the ladder And it's been a great mechanism and that just broke right because every junior person knows less than chat GPT and they would rather just use chat GPT and they'd be kind of dumb not to use chat GPT or claw to give you answers because it's better than what they could do and every middle manager would rather delegate to the AI than a flawed human who takes forever to give them an answer and it is good and so everyone's just doing AI work to each other and I think that that is exactly the problem that you're talking about here which is the danger is that we lose the talent pipeline. There are solutions to it, but they're going to require fairly radical change in how we think about talent pipelines.
How much of this is kind of like art and and and here's where my brain is going, right? Which is I am an art fanatic. It's the thing that I love more than most things. And I am totally fine with AI making art. It doesn't bother me. I'm totally fine with AI making music. It doesn't bother me. However, when I hang something on my wall, I like knowing that a person conceived of it. I like knowing that a person made it because when I buy a piece of art, I'm not just buying the visual thing on the wall. I'm buying the story that goes along with it. Or for example, I was listening to some music this weekend. Uh I was listening to John Batist's Beethovven blues album, which is, if you haven't heard it, spectacular. Now, could AI make a blues version of a Beethoven sonata? 100% it could. But the joy that I got from listening to that music was not just the music that I was listening to, but I was smiling that a person had the creativity to to come up with this. And that was part of my joy. When we look at the work product, you know, there's two things we're we're neglecting, which is I like thinking, right?
I enjoy debate. I enjoy making my head hurt at difficult things. I enjoy learning. The same way a painter likes painting and a musician likes playing music and and composing. Where is the human desire to want to learn and then will our schools but especially our places of work allow for that to happen or will they all become so obsessed with efficiency that we actually even if we want to learn? You you see where I'm going with this?
I there's 10 million directions from here, right? So I want to put a pin on the art thing because it's actually really important interesting which is you know obviously there might be a rise of more artistal humanmade things. It's more the most direct version of this by the way is when you have AI poetry or long form fiction there's often a lot of things wrong with it but because we're used to if we read something that reads beautifully and it's effortful to read we assume that there's a purpose behind it. So we spend our own effort figuring out the holes. Like for example the eyes are very famous at weird analogies right? So it might say this conversation is like a gap to smile. Now that is not meaningful. But if you spend some time thinking about you're like oh how is it like and you will reach a feeling of meaning right? If that was you know lazo krnikai or someone else writing this set of stuff and I was reading I'd be like oh this person thought hard about that analogy and I should spend the work to do it.
If it's the AI doing it you know in some ways it's beautiful but the meaning comes from me and am I being cheated because I have to create the meaning that has no intention. the intention the intention no longer belongs to the artist. the intention now is shifted to the to the listener
viewer right maybe it always has death of the novel stuff but like so that's one angle that's kind of interesting right and then I think the sort of second one is you know is on thinking about developing some of these kind of you know how do you develop intuition and we actually have a way of doing that we we know how to train people like we could teach people to be experts but the problem is it's effortful um and so you you like there's always been this sort of view that you know I told you early on I make g I made games for education and one of the most depressing things you learn is you can make something incredibly fun, but first of all, it's only 80% fun. It's not as fun as actually doing a thing for fun. And second of all, learning is effortful. And if you're not doing effortful work, then you're in trouble. And now for the few areas that we intrinsically care about, for you, you know, it might be art history or music or maybe for some people it's math and science, maybe for some people it's, you know, it's a sport they care about. Whatever you we are effortful about and intrinsically motivated, you're like, why isn't all learning like this? And the problem is you don't care enough about it. and so but I still want you to learn math even though you don't want to learn math I still want you to learn American history and if you shortcut that through AI giving you the answers you learn nothing we we have enough experiments to show that so making people essentially lift mental weights becomes the problem in a world where you there are shortcuts
you know I find there's a great irony in all of this which is the problem actually doesn't lie with AI which is we've been on this steady drum beat this this path to this point where we are you know discomfort avoidant. The concept of ghosting is is a thing where you just avoid a difficult conversation or you see it now particularly among young people where they're more comfortable with quitting a job than having a a difficult conversation or getting negative feedback and and then the idea that we've become so endultoriented, you know, as as capitalism has become short-term focused and more focused on a shareholder supremacy, shareholder value over the quality of the product or customer satisfaction or employee satisfaction, you start to see we become more resultsoriented and we've left out the work product. This is not a new concept. AI is just the most exaggerated form of being results obsessed at the expense of the effort, the work or the journey to get there.
Well, and and just to take another path from that, I mean part of this is what makes AI at work so challenging, right? Because if you want productivity gains, you just you'll get 100 times more PowerPoint, right? if my job is producing PowerPoint like so it requires you to rethink what the work is
and so what the work product is can't be the same thing I mean even the most basic way coders can write 100 times more code than they could before if they are embedded in an organizational process where it takes two weeks to do a you know a product sprint as they as they often call them right so each there's stand-up meetings every day and the assumption is the coder will write x number of work the product manager will do this the designer will do this the marketing people will do this and suddenly one person is 100 times more productive. What does that even mean for an organization becomes a problem? So part of this is our systems coming back to that theme we've been developing throughout which is human systems are not built uh for an AI world. We have we're effort you know school wasn't built for a place where anyone could write your essays right like you know um like work wasn't built for people be able to produce PowerPoint on demand without thinking about it right fiction wasn't built that I could write as many papers law clerks and courts weren't built for anyone to be able to bring up a case that's not a problem that happens at every industrial revolution it's just all happening at once everywhere and sometimes the AI wins sometimes human systems win sometimes we both lose but that's where I'd watch the adjustment happening
I'm going to go backwards a little bit here let's go back to practical which based on classroom and in the in the business world because I know you study that as well. What are better simple ways that we could be using the technology available to us? Like I said, most people are miss or under utilizing the tool and I know I'm overwhelmed by the people giving me advice as how I should be doing things and what I could be doing. But, you know, from a from a just a a basic standpoint, how can somebody level up just one to 10%? You can get more than one to 10%. I take no money from AI lab so you don't sound like a I don't sound like a shill but you have to end up paying 20 bucks a month to one of the big three companies is what I recommend. Either Google's Gemini, Open AI, Chat GPT or Anthropics Claude and you have to actively pick the best model available at that point which is what you'll have access to what are called thinking models and those will change over time but you have to actively select that. It defaults to a lower one. You will get huge impact improvements just from picking the most recent model and using it. The second thing I would say is AI has gotten quite good. So there's kind of three phases of AI. There's prior to chat GPT where mostly we talk about AI we've been talking about how you use you know data analysis basically there's all this talk about algorithmic fairness and price mining and all the like you know customized pricing all of that came from prior to chat GPT. and chatb kicked off generative AI and what I will grandioely call because the title of my book my previous book um co-intelligence where you'd work back and forth with a chatbot to get an answer right I typed something a chat it would give me an answer now we're in a new phase which which is called agentic AI and it's really just three or four months old practice
what does that mean what does that mean
an agent is an AI system that can independently go do work if you ask it to
so agentic is just the the adject of of agent
agent right so agentic is an AI agent and there's marketing terms around it but it's an AI that can do work. The most important thing to realize is how good the work is and how long the work is. There's this paper by and test by open AI, so you always take with a grain of salt, but there's been independent enough assessment. I feel good about it called GDP val. And what they did was they took people representing 5% of the US economy, so journalists and product managers and lawyers and private investigators and with an average of 14 years of experience, they had them each create a really hard problem that they face in their field. They hired another set of people with 14 years of experience to do it. took them on average seven or eight hours to do the work. And then they had the AI do the same thing. Took about 15 minutes for the AI. Then they had a third set of experts come in and spend an hour evaluating the outputs from each of these, not knowing whose was whose and voting on which they like better. And when this came out a year ago, the best AIS in the world were getting about 48% 48% of the time they were tying or beating humans. The latest models as of when we're recording this are at about 84%. So 84% of the time the work that they do seven hours of human work are equivalent to or better than a human. What that means going back to the practical piece is you would probably save three times effort and three times cost if every complex job you would give it to AI and even if it took you an hour to put it together and evaluate even if you had to give up 30% of the time you would still save time and effort. So one of the things I think you're doing is not using AI and giving it hard enough tasks to do.
Okay. And and so that's the other thing which is the value of the AI that where efficiency is not how quickly it can solve the problem but how quickly you can evaluate whether it got the problem right which again brings back to expertise. An expert can look at this right away and be like not just it's wrong but like often it's wrong because of a specific problem that you should have either specified better or the AI is stupid about something. And sometimes you can instantly get oh it's never gonna get this because it's too subtle and I can't communicate the point. I'm just gonna do this myself. But sometimes you're like, "Oh yeah, yeah, this is a rookie mistake and I should remind it that when it writes articles to not just factually explain everything, but explain it with a story or whatever your thing is and then it's better." Right? So evaluation, feedback, these are things experts are good at and the AI responds really well to that.
Here's the other problem, which is I remember when I wrote my first book, right? Everybody told me, everybody in the publishing world said the most difficult thing for any author is quote unquote to find their voice, right? To have a voice. Now, it's a very hard concept to understand you know what voice is. Essentially, it's when you read my words, they are of me. They my they're my personality. They're they're my point of view. It's not just nicely written, but it is it is of me, right? And it's very hard to do for an author. And I have found that AI can write beautifully, but it has no voice. And if you ask it to have a voice, it's going to always have voices that are available to it in the in the world. In other words, published people, but not you. And so most writing will start to just sound the same. I mean, I'm already seeing it. I'm getting AI generated emails in my inbox, and they're all basically the same email.
