This series is on how AI is reshaping entry-level work and what universities are changing in response. Could you give us a quick picture of what you have going on right now, the work that ties to that question? Maybe start by breaking down the first initial thoughts that come to mind when I just gave you the main topic of this series.

My research right now, and for the foreseeable future, focuses on the idea of skeptical intelligence. We first, in the 1960s and 70s, had IQ, intellectual quotient. That was presumably raw human brain horsepower. It was actually a fairly good predictor of how people would be able to solve problems. When it was first popularized, the problems were simple as compared to today, and the problems had actual solutions.

The best predictor of a person's IQ was how much money their parents spent on their primary education. So IQ was helpful in figuring out if a certain person could solve problems, but really you were testing for the amount of education they had had. And one of the questions, therefore, was: is IQ teachable to people later on in their life, or is it something that takes 20 years, starting at a very early age, in order to fully imbue into a human being?

The next step was emotional intelligence. This was in the 1980s and 90s, and even more popularized in the early 2000s. This reflected people's capacity to interact with other people and create relationships so that the pair or the team could be more effective. EQ is the measure, your emotional quotient is the measure. And EI, emotional intelligence, is the big idea. So you measure EI with EQ. That was looking at how well you can reflect on what's in your mind and you can control your own emotions, how well you can reflect on what's in the other person's mind, and how well you can control their emotions in order to create a productive working relationship.

And then we have artificial intelligence, obviously, which takes all of this and puts it into a machine. All of the intellectual capacity, puts it into a machine. We came up with the term skeptical intelligence in order to figure out, there is still an ongoing debate as to whether emotional intelligence is teachable, or is that also something that was developed in your formative years or perhaps even genetic?

Skeptical intelligence we're intentionally defining as an individual's capacity for disciplined doubt in the face of output from artificial intelligence. To put it simply, do you believe wholeheartedly what AI is telling you, and therefore do you act upon it? And that unquestioned, unreflective belief is not healthy, it's not productive. AI too often, not necessarily intentionally, comes up with output that is simply wrong.

And when I say non-intentionally, part of this is that AI companies have created their AI engines in order to prompt you to ask the next question. Any AI engine does not know if it's being useful to you. It has no measure for that. Yes, we have the up thumb and the down thumb. Nobody uses it.

So how do we know if it's actually being useful? How does the AI company know if it's actually being useful? Their best measure is, do you ask a follow-up question? That's the only way they know if the first answer is useful. The problem with that metric is, that could be that you're asking a follow-up question because you truly find the answer useful. It could be because the AI output was sycophantic, it's trying to be too nice to you. It's trying to say, oh, that was the best question ever, and here's another answer, and how can I help you next? We don't know whether that's the AI company trying to make their AI engine sticky, and to prompt you to ask the next question, or whether even the AI algorithm itself has drifted because, without any human intervention, the AI algorithm itself doesn't know whether it's being useful.

The upshot of all of this is the AI engine right now is trying to get you to ask more questions. It's becoming an attention-seeking machine, sort of like social media.

There are also a whole bunch of structural flaws in how AI produces data that makes sense for the engineering and for machine learning, but that are not true predictors. For example, AI is the average of historical data. When you ask AI to predict something, it doesn't know what the word prediction means. All it's doing is going through its database and finding all of the instances where some author said, I predict that, getting the average of those and giving that to you. It's not actually predicting. It's all basically matrix algebra and scalars scaling vectors in order to figure out what is mathematically the most likely series of words and sentences and paragraphs that match with the numeric values of the question you asked. It's just math. It is not a prediction.

The result of these two things is that AI is not necessarily solving your problem. If I say to AI, please tell me the best pair of shoes, let's go back to those two issues. First is, it wants me to keep asking more questions because that's the only way it can keep me on its platform and know that it's useful. And secondly, it's only giving me historical data. It doesn't know what my feet are like. It doesn't know why I want the shoes. Maybe it's dress shoes, maybe it's hiking shoes. Maybe it's for looks, maybe it's for comfort, maybe it's for status. It doesn't know. And even if I tell it all those questions, that's only the first layer of my problem. It can't really get into all of my past shoe preferences or the upcoming hiking that I'm going to be doing, and where I'd have to go through the five whys.

So AI is not actually solving the problem that I want it to solve. It's solving an approximation of what it thinks my problem is based on very limited data. There are an enormous number of AI users who don't know that, who think that AI is giving them the best answer, because it's framed, saying, you're so smart, and that's such a good question, and I know you, and I might even know your past shoe purchases. But it's not framing it as the best guess. It's not giving me probabilities. It's saying, here's absolutely the right answer.

Skeptical intelligence is the capacity for any AI user to question constructively the output of AI. We coined the term skeptical. So there's intellectual quotient, emotional intelligence, artificial intelligence, skeptical intelligence. It's intentionally the fourth pillar on the stool of necessary or related intelligence to this new revolution.

The way you measure it is with the skeptical quotient. We have a full measure of 30 questions that we can ask you in order to figure out what is your particular number. The research we're also doing right now, because the prior topics were unclear about this, is determining if and how to best teach skeptical intelligence. Our hypothesis is that skeptical intelligence is teachable. In other words, it's not genetic, it's not based on how much money you spent in your primary education. We think it's teachable.

So this goes to your question about what are we going to do with AI in the classroom. To me the answer is, we have to teach our students at any level how to use AI early and often, how to understand the tools. But more fundamentally, we need to ensure that they have high levels of skeptical intelligence, so that they are using this version of AI and future versions of AI with care and rigor.

Now let me also pause for a second. There's an overlap between critical thinking and skeptical intelligence. The problem with critical thinking is, first of all, the idea is 2,500 years old, Plato, and it has been so layered with other potential theories. Is it reflective? Is it not reflective? Is it an action or an ability or a technique? Critical thinking by itself as a set of theories is too diluted.

We also think that, because AI mimics a part of human intellect, we need a subset, and even some non-overlapping pieces, to really address, to bring skepticism to AI. In other words, critical thinking is both too big and not quite tuned for how humans are interacting with AI. So skeptical intelligence is not just a rebrand of critical thinking. It is narrower and slightly different.

Skeptical intelligence, from our point of view, has three different components. There's how you question the information you're getting. And I don't just mean an instinct to question. I mean the actual pause in your mental biases, your intellectual biases, in order to try and figure out exactly what's what, where the information came from, and its own biases.

The second piece is verification. Okay, so this source looks correct. I've questioned the information. Now, exactly where did it come from? Or are there other sources that would contradict it?

And then the third is a little bit of reflection on your own problem that you're trying to solve. This is not reflection around emotion. I'm feeling tense, I'm feeling afraid, I'm feeling anxious, I'm feeling hopeful. That's not what we're talking about. This goes back to the shoe problem intellectually. Where is the frame? What's the context of the problem I'm trying to solve? So those are the three sub-components of skeptical intelligence.

There are pieces within, so in addition to defining the skeptical quotient, how do we measure this, and the impact of skeptical intelligence? How have we ensured that people with high skeptical intelligence are actually more productive and innovative with AI? We've done that work too. So we've done those two pieces of research already.

The third piece of research is, what do you do about it? What's the technique that people who have high levels of skeptical intelligence use, or what's the technique that people could use in order to develop high levels of skeptical intelligence? For this, we've drafted the Skeptic's Playbook for AI. We just got a book deal, so we're writing a book on it.

You wrote Innovating with Impact. You gave a very good theoretical framework of what students should be taught around AI. More practically, could you bring us into the classroom at Hult or at Harvard? I'm one of your students. You're seeing students boo AI. You're seeing some universities ban it. Can you take us through a day, practically, in your classroom?

First of all, around the ban of AI. I understand how some professors might want to do that with things like writing classes. I understand how you would want students to develop their own ideas and thought processes de novo. Like blank tablet, without AI. I understand that.

I teach strategy in business, where I don't need people to come up with, according to them, brand new ideas. I had a professor in graduate school who said the best way to think you're innovative is not to read anything. Yes. Not to be innovative, but to think you're innovative.

So in teaching strategy, and in practicing strategy, the outcome matters. The outcome matters the most. The process matters because a good strategic process will lead to a good outcome. But I would never want to put boundaries on what people should consider during the process. So use AI early and often, and don't get sucked into thinking that AI is correct. Or even more importantly, with strategy, that AI will give you a novel answer.

We just said that AI is the average of historical instances. If everybody in a certain market, say the shoe market in Armenia, if all of the shoe companies in Armenia asked AI for a new strategy, it would give them all the same strategy. That's not the purpose of strategy. It's intentionally to differentiate. So AI can be helpful for some pieces of it, but not for all pieces of it.

Let me tie this back to skeptical intelligence and the Skeptic's Playbook. A person who is employing skeptical intelligence, with the Skeptic's Playbook or not, can use AI to actually test some of the outputs of AI. So this is not purely a human intellectual exercise that is dependent upon what a single human being knows. They can use AI, but they have to use AI in certain ways and circumstances in order to test AI.

Even I'm saying that AI can be helpful in this rigorous process. But the first step and the last step cannot use AI. The first step is, what problem am I trying to solve? You can't ask AI that because the whole point is it's very specific to you, to your context, and you can't possibly tell AI enough to give it all that.

The final step is, you have to make your own decision about all of the steps, all of the information you've gotten from AI, even the cross-checks that you've done with or without AI. The final step in the Skeptic's Playbook is, what do you as a human being think? So obviously AI can't help you with that. But to me, AI can be helpful all along the way.

So, this is because of what I teach. So I understand why people might want to ban AI for essay writing. I get it. And I don't have the expertise in what they're doing to have an opinion about it.

I also can understand why some people boo AI. Let's also be careful about how frequently that happens. I think it happened so infrequently that it was noteworthy. Which is why you know about it. This is why it was up on YouTube and why it's being quoted, because it was so infrequent.

It's also, for the graduates right now of college, they're facing a tough job market. Only some of it is due to AI.

It depends on the sector. The Economist last week did a really good analysis of this. Which jobs are most impacted by AI and which jobs are least impacted by AI. So they broke it down, and they used their own primary data, in addition to reporting on what other people said. I highly recommend you look at that analysis.

But also right now, the job market, the job market right now is poor for entry-level, post-grad school and post-college entry-level positions. That also has to do with the business cycle. It has to do with the fragmentation of international trade. It has to do with the newfound immobility of labor due to visa restrictions. It has to do with US debt. It has to do with European debt and fragmentation. It has to do with Chinese real estate debt. In other words, there are so many things that go into the current job market that attributing all of it to AI would be statistically a mistake.

Enable helps you take a written business case into a live multi-role simulation run by agentic AI. This has been the heart of business study for as long as business study has existed. Tell us about it.

I don't want to overdo it. This is a side project that I'm working on with a couple of my former students. They have full-time jobs. I have a full-time job. This was an interesting exercise that we did first for my classroom and then recognized it could be useful to others. We presented it at the Academy of Management, which is the huge annual conference where all business school professors arrive in one place.

We had talked to lots of people at the Academy of Management and they said, oh, I get it. Here was the problem we found. There are hundreds of thousands of very useful business cases in the world. Harvard Business Publishing is the biggest catalog and compendium of cases. But Ivey and the Case Centre, there are thousands, hundreds of thousands of useful historical examples and illustrations of either how a theory should work or doesn't work, or could work, that are useful exercises for students to go through. So they have a very context-specific instance into which they can apply a theory and learn how it operates.

Fewer and fewer of my students read. They don't want to. They have videos, they have AI. So I'm in a battle for their attention. I struggle to get them to read a case on paper. And in order to have a fruitful discussion during class, what we did was to build a tool that can take any case, business case, legal case, medical case, unpack the problem that the protagonists are facing, apply the theories that are in the teaching note. All cases come with a teaching note for the professor, and then put that into a multi-agent simulation so that the student can actually talk to the specific people in the case.

If the case has a CEO and a customer and a supplier, the student can actually address their questions to one of these three people because they'll all have different answers depending on the context, the case, their role. And then the student can, through that, unpack the mystery, apply a particular theory, see how that theory would have performed in that particular instance. So it's much more interactive. And this is now using the hundreds of thousands of existing cases and meeting students where they are, the channel, the manner, the media in which they now prefer to operate.

So that's Enable. It's a tool that we have found useful and we're offering it right now basically for free. It's for the compute cost. This is not a high-growth for-profit venture. Mostly it is helping all of academia improve what they're doing in the classroom. For me, it's that it's running on agentic AI. And it's multi-agentic AI. So there's an agent that unpacks the case, there's another agent that grabs the context from large language models to add more flavor to it. Each of the roles has an agent. There's another agent that administers the game. There are other agents that grade, that evaluate and grade the students' performance. And there are other agents that will even evaluate and provide suggestions for how they could have done it better.

So it's multi-agent. What you should be hearing here as an AI company is, the compute here is large. That's our cost and that's our pricing. How do we cover costs with a little bit extra so we can keep doing R&D? Enable may have a huge market potential, but there's lots of competition in this space and really, maybe I'm just too old, but I don't have another full startup left in me. I'm having much more fun doing research and teaching students. At the age of 56, I don't want to be working 80 hours a week to try and generate market share and revenues and raise investment. That's not interesting to me anymore.

You said you love teaching and research, yet Enable seems like it replaces a lot of the teaching. So with a tool like Enable, what's the role of future professors? What's the role of a future classroom?

Let me go back to your premise. That there are some people, and I've even done it. I did a couple of TED talks. My second TED talk started with this premise that my role is one of the first to be ejected and diminished by AI.

There are a couple of things that AI cannot do. The first is to tell you the right question to ask. And in order to know the right question to ask, you have to have a broad set of theories and ideas.

As a quick side note here, I learned French as a student, but my vocabulary in French is incredibly limited. When I try to think in French, my thoughts are limited. I literally can't have a complex thought in French because I don't have the words.

The same is true if I all of a sudden approach AI. Yes, it can potentially answer any question I ask it. But if I don't know enough context, theory, possibility, if I don't have creativity, and creativity isn't just raw blossom from my mind, creativity is a disciplined process. If I don't have any of that, I am asking simple questions, and I'll never get the nuance and the innovation from AI that I seek.

So one of the first roles in education is to expose you to an enormous amount of different theories, different ideas, so that you can seemingly randomly ask AI more and better questions. I say seemingly there because it's not random at all. When I have students talk about strategy, I want them to bring in operations. I want them to bring in, if they happen to be big soccer players, I want them to bring in some of their soccer expertise, because there are strategies in soccer that could actually help ask a good question for business. If they've played music, I want them to bring in music theory and music creativity and music improvisation to ask AI because that gives them, again, richer questions to ask with unexpected answers.

So part of our job in education is to give huge amounts of information, not just what AI can deliver. I think there are some people who right now think that AI could replace an education. That may be true for a specific job with a specific current set of skills. It would be a huge mistake for any student to bypass this huge body of knowledge and the techniques and processes and mindsets that come with all that knowledge and just use AI. They'd be incredibly well-tuned, very efficiently. They'd save a ton of money and time and they'd be perfect for that job. But if that job evolves, nope, they don't have the broader context in order to evolve with it. Or if that job goes away, they have no capacity for the next job.

So I am a huge believer in a broad-based education. That's the first thing, why AI will not replace the university.

Second thing about why AI will not replace professors is that professors ask hard questions. If you come into my classroom, I teach Socratically with randomized cold calls. So I will literally out of nowhere say, David, what just happened? How could skeptical intelligence interact with, how do you see your kids learning skeptical intelligence? But I won't use the word skeptical intelligence because I'm leading the witness. I'm encouraging them to have their own biases that are based on my preference. It's called social desirability. David wants to please me as the professor.

So instead I'll say, how did your kid learn how to tie their shoes? What happened? That's not a genetic skill. It's not a, nobody evolved to tie their shoes through evolution. How did they learn how to tie their shoes? And can we take how they learned how to tie their shoes and put that onto how you are using AI? I would do that Socratically in order to get David to recognize that what I'm talking about isn't just around strategy. He experiences all of this stuff all the time in different ways. To make connections for him.

The third role that I think is under-heralded for a professor is to improve each student's sense of dignity, and part of our self-worth. Sense of self-worth, which is theoretically different but in the vernacular the same as self-confidence. If I give you a tool, and it even could be an AI tool, and you don't have faith in yourself to use it well, you won't use it well. It means first of all you won't even try it. And then when you try it, you're overthinking or under-thinking all the time. It also may mean that you believe the results and therefore you don't touch skeptical intelligence. Like you don't question anything.

So part of the research that we have done is, how important is a person's sense of self-worth in whether they will ever develop skeptical intelligence? And the answer is, it is significant. And this is true across any skill you ever learn. Could be coding, could be parenting. That your own sense of your self-worth is vital in your capacity to use a skill and to improve the skill and to create other skills.

So one of the huge roles of professors as this sort of person in authority, but with this bigger context in mind, is to, and this is what I do every day in my classroom, I engineer circumstances where I can affirm students. Whether we're talking about strategy or platforms or pricing, these are all just the topic of the moment. Each of my classes is to help my students build their own sense of dignity. This is especially important for the people in my class who come from cultures with hierarchic cultures in which they are not at the top of the hierarchy. So for example, young women in patriarchal Asian cultures frequently don't have a high sense of their own self-worth.

As a quick aside here, I once had a young student from Vietnam and I asked her a question and she just froze. And whatever, we kept going. And she said that she had already, this is an MBA program, she had already gotten a master's. But in her 12 years of schooling up to my classroom, no professor had ever asked her a question. We were in a classroom in San Francisco in an international master's program. That's nuts. What a huge disservice her prior professors gave to her. She had spectacular skill. She was incredibly well prepared. Once I did get her to talk and to express herself and she gained self-confidence, it was clear to everybody in the room, about 80 people in the room, that this was probably the smartest, most well-prepared person in the room. They hadn't seen it because she had never uttered a word.

My role for her was to give her the self-confidence so she could use the skills and then improve the skills. If I did nothing else for her in her education, that changed her life. And it wasn't through me teaching her a theory. And it wasn't even my affirmation. It was providing an opportunity for her to step into her own capacities and to develop self-worth. AI cannot do that. It has no power. Even AI can be sycophantic. AI can say, that's the best question you've ever asked me. But because you know that it doesn't mean it and it doesn't know you, that doesn't improve your self-confidence.

I want to push back on something. You have this theory that the testing and the calculation side of things can be replaced by the robot, whilst strategy cannot, and that is where the human really shines. You have a Harvard Business Review paper where you say that strategy beats sheer volume of market tests and that the Lean method can kill good ideas too early. It seems you're implying that the human is giving the strategy side of things rather than that the machine can. Right or wrong?

So the Lean Startup method is disciplined doubt. The Lean Startup method is the scientific method just applied to entrepreneurship. The Skeptic's Playbook is also the scientific method applied to AI. These work hand in hand. You can use Lean in order to test and refine an idea. And now you can use Lean plus AI to test and refine an idea. You can do that, however, unskeptically. And therefore unproductively and unprofitably.

A human plus AI is better than just a human and better than just AI for producing novel, productive strategy.

Too simplistic. AI is great for giving you a hundred different possible strategies. AI is okay at narrowing some of that hundred into things that are more context-specific. The human being's skill is discernment. Which of the hundred strategies that AI first gave me meet my problem? Because AI doesn't know the whole problem. Even if you say, okay, I'm going to let AI bring this down to 50, where did it come from? Did it come from all of my, it's the average of all of my competitors. That's not helpful.

So the human, it's human discernment that is evaluating what AI has done so far and figuring out how to ask AI future questions, and also recognizing that AI structurally is going to give answers that aren't necessarily useful, like historical averages.

Crystal ball. If you had to bet on what a business or entrepreneurship degree actually certifies in 2035, what would that be? What would it be certifying?

A couple things that the degree certifies. The first is that the human being did the work and learned the skills.

If I'm gonna hire somebody, I don't just want them to have written a good essay and be able to solve the problems that I already think I need them to solve. That's IQ, right? That's where we started this conversation. A company that's hiring somebody is basically saying, here's what I think I'm going to have this person do. If they're finite, discrete tasks, I don't need to hire the person. I got AI for that.

Instead, the degree certifies that this person did the work. In other words, I'm hiring a human being with creativity, capacity for relationships, capacity to evolve in different ways, capacity to make unexpected connections. AI's connections are totally expected. It's math. Human beings can make unexpected connections. How do you connect what just happened to the World Cup to what I think Coca-Cola should do in South America for its beverage strategy?

It's also certifying that the person has a broad set of skills. There's a big difference between a certification for entrepreneurship and innovation and a certification for Python coding. Python is a very discrete, known set of skills. Innovation, we have no idea, and we've been doing research on this for 50 years. We have no idea what the recipe is for innovation. We have lots of techniques that we think can help people get there. We have no idea, and I don't think we ever will, because the whole point of innovation is it's something unknown. It's got to be differentiated. Surprise. If there's a recipe, you're not being innovative. That's sort of oxymoronic.

It's like the market, any market only has already embedded all the information that's known. But then why do markets move? Because there's unexpected stuff all the time in order to develop unexpected outcomes. This is, so far, a human capacity.

The other thing that I'll say about 2035: in the last, oh, recorded, human beings have been around as we know them for about 100,000 years. Like the Homo sapiens have been there for a million, but sort of with some communications for a hundred thousand years. We keep coming up with new ideas that change our existence and yet we don't die off. We came up with fire, changed everything. We came up with, let's jump more recently, we came up with writing, changed everything, came up with electricity, changed everything. We didn't die off.

I don't think AI will ever replace humanity because humanity will evolve and it's evolving quickly. But I think AI in 10 years will have created not just new jobs, but new human cognition. Some people are going to be thinking differently and we have no idea how AI will be thinking and no idea how human beings will be thinking. You just proved that to me five minutes ago when you said you abuse Claude. You're now thinking differently than you did a year ago. I don't know if that's better. I don't know what you're thinking, what you're coming up with. I don't know if it's better. I don't know if it's replaceable. I don't know if Claude will take care of you altogether. I'm pretty sure that you will figure out, maybe with the help of Claude, how to have a new role.

Now, to go back to the initial question about the job market. There will be huge dislocation first. I have no idea where that dislocation will end us. Did you ever see the report from Citrine Research that moved AI markets? This is sort of a minor market analyst group. So normally they evaluate stocks. And they focus on AI stocks. And they wrote a report about six months ago that was unusual in that it was a future scenario. They were predicting in 10 years what's going to happen. And they wrote it as though the 10 years have already happened. With AI, it's pretty cool. And they were focused on the markets.

One of the things they said is, this is the first technology in history that can replace the upper middle class. Electricity replaced labor. So the upper middle class, these are middle managers, these are knowledge workers. And it wasn't trying to say how they're going to be replaced or what they're going to do. Instead it was purely macroeconomic. But the exercise was interesting.

Basically it said the upper middle class is what drives economies. The middle class provides economic resilience. The upper middle class is discretionary spending. So this is what drives SAP, Oracle, Google Search. Right now Google is making a ton of money. Who are the people who are really paying for Google? It's the upper middle class. Middle class, they're still labor or they're low-skill knowledge workers. They don't use Google that often. It's the upper middle class that's really using Google. They're the people who are really buying the things that sponsored search companies want them to buy. It's discretionary income.

What happens when that collapses because of AI, that income collapses? So it's an interesting exercise. It's not my area because it's macroeconomics, but I recommend you read that report because it's predicting something that pretty quickly spirals out of what we possibly could predict. If upper middle class all of a sudden has no money to spend, the economy collapses. Who's going to pay the huge expected valuations for Anthropic and OpenAI? Did we just crater the production of AI because nobody can afford it anymore because the economy cratered? Both of these things can't be true. That AI is worth something, but there's nobody around to pay for it. That can't happen.

So you're sort of asking me that similar question in 2035. What will this look like? There are a whole bunch of predictions I could make, some of which are mutually exclusive. I have no idea which path we're going to go down. I have no clue.

That's the full thing. When you paste into Word, the markdown formatting will carry the bold on the title, byline, and questions; the italic on the journalist line and the lightly-edited note will also carry. Blank lines between each Q and the answer that follows separate them visually so they don't sit on top of each other. The three "---" lines between Q&A blocks are visual separators; you can keep them or delete them once it's in Word, whichever you prefer.

* This interview has been lightly edited for clarity and readability. The interviewee reviewed and approved the transcript before publication.