McKinsey Senior Partner Emeritus Sven Smit, longtime chairman of McKinsey Global Institute (MGI), has had a front-row seat for all the number crunching, debates, and dissemination of the firm’s comprehensive, data-driven reports on strategy, technology, geopolitics, and economics.
Now, along comes AI to radically change the way we create and consume ideas.
Artificial intelligence is a powerful force for progress, Smit explains, capable of dramatically increasing productivity and enabling a future of greater abundance. But to keep up in this era of rapidly expanding AI tools and information, he says, “we must continue to excel in quality checks and in critical thinking and judgment. Those cannot be replaced.”
In this interview with McKinsey Editorial Director Roberta Fusaro, Smit considers what “insights” will look like in the future, who the new experts will be, and how business leaders should think about knowledge in an AI-driven world.
Ultimately, Smit sees the future as a dynamic collaboration between humans and AI—where judgment, creativity, and adaptability (along with hard work!) remain distinctly human skills even as machines take on more cognitive and physical tasks.
This interview has been edited for clarity and length.
Roberta Fusaro: Sven, AI is changing the speed and scale at which we create content. What are the pros and cons of that?
Sven Smit: I believe, in general, AI is for the good. However, we need to pay attention to some of the risks that might occur. Is the quality sufficient? Are there strange things happening on platforms that may not be truthful and, as a result, create disorganization and disinformation?
When it comes to technological change, there is always fear. There was fear when farming got automated by tractors and other machinery. There was fear when factory work got automated by larger machines and robotics. To some degree AI seems different because it is supposed to replace thinking. Now, we can have a long debate on whether what AI does is thinking or not, but undeniably there is a bit more underlying fear with AI, as the physical aspects of life have already been significantly automated and we get used to that.
That degree of fear really depends on what you think the world will do in 100 years—whether you believe AI is a real threat or a tool for creating more good things and more free time for people.
What about big ideas?
Roberta Fusaro: McKinsey, and MGI in particular, have always been associated with big ideas. Do big ideas still matter in this era of large language models (LLMs)?
Sven Smit: I think there is a significant difference between big ideas and good big ideas. Large language models are very good at cataloging what is already out there. But if we look at the structure of human discovery, many of the biggest ideas were not mainstream. Which poses an interesting question: How do you get AI tools to be contrarian? That is what I do when I play with them. I always ask for an alternative. Then, I explore the proof of the opposite, which is a helpful tool to arrive at better ideas.
Roberta Fusaro: Where will the next big ideas come from? Are they going to come from agents?
Sven Smit: AI will not come up with the answer to big questions about the current world order. That is going to come from interactions between AI and humans. The LLMs could have all kinds of data points. But at the end, humans are going to say, “This is how the game is played.” Many new and big ideas are needed. At some point, LLMs may have much better ideas than humans that we just cannot see because of the data. But I believe it is going to be an interaction. I have never liked the term “artificial intelligence.” I like “assisted intelligence.” It might be that sometimes human intelligence needs some assistance from agents. And sometimes the agent’s intelligence might need help from humans. The balance will change depending on the context.
Roberta Fusaro: As we think about that balance, can you see particular areas, industries, regions, where human judgment will still matter more than AI or vice versa? How do you see that balance unfolding?
Sven Smit: There are certain super-repetitive tasks that we have found hard to automate, that the current AI would be very good at managing—things like maintenance prediction, which in real time is much better done with AI than with a human brain. When you consider case history in law, for example, some AI will be helpful, yes. But when it comes to a final judgment of what is fair, I am not sure I would leave that to AI in the near future.
There will be a 60/40 balance of human-to-agent in one place, 80/20 in another, and 20/80 in still other places. We are exploring so many new ideas. Imagine that we have fully self-driving cars. What is a parking lot with a self-driving car? What is delivery? What is pickup? And that’s just one theme. What if your humanoid robot cooks for you? Would the robot cook for you every day except for Fridays because that’s the day you want to cook for yourself? How is it going to work? That dialogue is always going to have to be there: What do you do? What do I do? What does the robot do? What does the agent do?
Combatting AI slop
Roberta Fusaro: There’s a lot of AI noise out there, what’s being called AI slop. How can researchers and executives ensure they are getting the right data? How can they avoid being distracted by too much or incorrect data?
Sven Smit: In 80 percent of the major research pieces we have done so far, the most interesting data we use is not available to AI. It either sits in proprietary databases, or it is a newly created set of data. Or it embeds a lot of knowledge that is in the human brain. That expertise needs to be extracted and applied to the data, and I am sure that will happen more and more. But I am not entirely convinced that all the quality data is already accessible to AI. The main sources AI draws from are published data, but how does an LLM judge, say, which scientific paper is the right one? Based on citations? How does a human discover which paper is the right one or not? I think that is the essence of discovery.
Roberta Fusaro: How have your own research practices changed with the advent of AI?
Sven Smit: I like to play a game called “trap the AI on the contrarian view.” For the A Century of Plenty book, for instance, I asked, “What sources of energy will we have in 2100?” The answer was 100 percent solar. Okay, so then I asked, “Did you get that answer based on levelized cost equivalency? Or did you get it based on levelized system cost, including the grid, no subsidies, and so forth?” And the answers I got back were completely different. The AI assumes parts of the question.
One of the most important things in strategy is to ask the right question to get to the right answer. I can basically, on the same topic, get two completely different answers with a very small modification of the question. I play with that a lot to discover the real source of data the model is using to make the judgment it comes up with.
As a result, the AI trains me. It comes up with data that I have never seen, which I can then judge. Is that part of the total repository? Or is it something new that I need to study a bit more? However, if you do not do A/B testing, it is not going to work.
Roberta Fusaro: More generally, how are you consuming content differently?
Sven Smit: In the old days, I always tried to read multiple books per week. The depth of a book cannot be replicated by short bites. But the equivalent of a book now would be engaging with long-form podcasts where a person who might have written a book talks about their entire lifetime of study in a particular field.
If you are just browsing randomly, you are overconsuming hundreds of topics. You need to focus on the topics that matter for your work and that you find most meaningful. For example, one of those topics for me was the future of work. I had a set of agents continually debating the future of work and providing updates: What is in the news on the future of work? What is the technology frontier in energy? What is the technology frontier in lifespans? You can define your questions and ask AI to work on those questions—but you can continue reading books and talking to people.
Roberta Fusaro: What do you tell your kids about AI? How do they think about AI?
Sven Smit: One of them hates AI and says, “It’s not good, and it prevents my learning.” In that case, there might be a bit of fear about being replaced. I tell my kids they need to be good at using AI tools and understand where the frontier of these technologies is. I tell them they need to use multiple tools rather than just one.
I also talk with them about bias. I do not want them to know only what I think or only what an LLM thinks. The art in the knowledge domain is to develop critical thinking. I want them to come back to me and say, “Dad, I think you are full of sh-t. You saw that, but I saw this.”
To the larger question of what will happen with our kids, sometimes when I present to CEOs or leadership teams, I raise the bias that suggests everything technical will be automated and that you should just focus on human judgment and the touchy-feely stuff. I am all for sensing and emotional intelligence. Humans should be very good at that, and it is part of who we are. But I do not think you can survive in this world without being technically competent.
AI and the CEO agenda
Roberta Fusaro: Speaking of CEOs—what kinds of capabilities or skills do you think business leaders need to focus on in this world of AI?
Sven Smit: In this environment, I think it’s important to be the fastest learner. Right now, there is no Toyota, which pioneered the use of lean end to end. It was the best in the world, and then everybody else started to do the same thing. There were two banks that were known for being the most agile in banking. But I don’t think you can point to a company right now that is the most AI-driven or automated. Some insurers have done claims processing well. Some banks have done audit well, or compliance. Some companies have automated the recognition of a rotten apple in a sorting center. There are many places where you see the AI movement happening. But there is not one iconic company that can say, “I am the AI winner,” other than the people who are providing the AI tech itself.
Roberta Fusaro: How do we get to that point? How can CEOs manage the core business but also integrate AI so they can claim to be the first mover in their industries?
Sven Smit: As with all change programs, if you do not make it a priority, it will not happen. If you do not ask everyone, “What did you learn yesterday? What did you do? Where did it go wrong?” and make this a central part of your conversations, it will not happen. And it’s not just about focusing on the “happy AI”—those incremental tasks that companies have started to off-load, like summarizing documents. That is not the same thing as the complete automation of an insurance claim process, which I would call “transformational AI.” For this, the standard tool does not exist, and it requires hard work.
Roberta Fusaro: Everything is competing for attention on the executive’s agenda—AI, the economy, geopolitics, energy. Where do you think leaders should be spending their time now?
Sven Smit: To some degree we are overfocusing on AI. It is not only AI that is changing the world. It is automation in general. Productivity is another point of concern: Am I producing 5 percent productivity growth annually? In a company using AI, one should at least say, “I am going to aim for 5 percent.” Geopolitics is also significant. Your customers are also important.
Always look on the bright side
Roberta Fusaro: As you mentioned earlier, fear and pessimism naturally arise with change. How do you think about that when it comes to AI?
Sven Smit: If I look at my own life, there was a lot of bad stuff in the context of good stuff. In the 1960s and 1970s, we went to the Moon, and there was the excitement of getting the first car, the first freezer, and so on. There was never a discussion, not even when my mother was rebuilding walls in Germany with stones from the bombings, that the future was going to be bad. The feeling was, the future is going to be great, better, and we are going to work hard to get there.
Now we live in a society where we have somehow informed the world that things are going to be worse, or at least not much better—especially in the West—despite the enormous amount of catch-up growth happening in Africa, India, Indonesia, and so on. The poorest in these countries are getting better. Yes, the inequality gap and other issues exist, but the sum total of the issues cannot be larger than the progress. For some reason, we have talked ourselves into a ditch, with a common narrative of limits and complexity, which is totally unnecessary in my view.
Roberta Fusaro: How do you combat that pessimistic narrative?
Sven Smit: The first step is to talk about optimism. I understand being pessimistic if you fundamentally believe things cannot get better; then you are going to batten down the hatches and try to divide and conquer. What I am doing is having lunches with young people, as well as older people, and just saying it out loud: “AI is complex, but it is going to be great” and another “century of plenty” is possible. Increasingly, it is being received as a positive counterpoint in complex times. If that happens, I am very happy because then we are at least debating the possibility of progress. Which progress we want—blue, orange, purple—we will see. How we exactly do it and who participates in what way, we will see. But if we stop discussing the possibility of progress, we are in a really odd place.
AI at McKinsey
Roberta Fusaro: How do you think AI will affect the way McKinsey works with clients?
Sven Smit: Over my 35 years at the firm, the tools we use have changed, but nothing has changed the real, core work that we do, and AI probably will not change it either. We help companies see the waves in the world and jump over them and adapt to the new environment. We excel at helping larger organizations, and smaller organizations, jump to the new challenges that are there. We know how to change companies, and we are experts in the methodology of that change. Will AI now do some of the work? Yes. But if you look at a McKinsey team, on average, it consists of a few partners and two to four people in support. Looking at the scale of our engagements, how much smaller do you want to make that team for a billion-dollar change?
When I joined the firm, “data analytics” involved using a ticker to count revenues per till on a cruise ship to figure out whether that ship was competitive compared to another. OK—so now I can do that with AI, and I only need to check what it writes. But transforming a large organization is not something AI alone can do. It still requires a chief executive, a leadership team, hundreds of managers, and thousands of people who need to adapt.
Roberta Fusaro: Given the explosion of knowledge from AI, how do we differentiate McKinsey’s ideas from everything else out there?
Sven Smit: Part of what we do is observe the outside world, which AI will help with. But you still need to judge which observations are real, which are relevant for a company, and the idiosyncratic observation of the company itself, how it works, how its team functions, how the executives interact with each other, what works and what does not, and what the capabilities are. I have helped dozens of CEOs become CEOs. There is no database in an LLM where the LLM can read how to do that. It can read five books, but it cannot read my experience or that of my colleagues. When we discuss these matters, we talk about a lot of things that are not in books.
