Podcast
Podcast
- 09 Aug 2023
- Managing the Future of Work
Complex systems: From supply chains to artificial intelligence
Joe Fuller: The Covid-19 pandemic made us all interested observers of supply chains. US industrial policy, with its emphasis on reshoring and sourcing, has kept the topic front and center. The influence of artificial intelligence completes a trifecta of sorts. But the logistical, economic, and geopolitical complexities involved defy armchair expertise.
Welcome to the Managing the Future of Work podcast from Harvard Business School. I’m your host, Harvard Business School professor and nonresident senior fellow at the American Enterprise Institute, Joe Fuller. My guest today is Yossi Sheffi, Director of the MIT Center for Transportation and Logistics. Professor Sheffi‘s wide-ranging research spans supply chains, transportation, organizational management, and the business and regulatory impacts of advances in engineering. His latest book is The Magic Conveyor Belt: Supply Chains, AI, and the Future of Work. In it, he deconstructs supply chains and examines the role of advanced technologies like AI and robotics, including how they are likely to change the work humans perform. We’ll talk about what—contrary to most of the media coverage—Covid-19 actually revealed about global supply chains. We’ll also consider whether the US is on the brink of a domestic manufacturing renaissance and the hard choices that would entail. We’ll also parse the workforce and educational implications of artificial intelligence. Well, Yossi, welcome to the Managing the Future of Work podcast.
Yossi Sheffi: Thank you very much for having me. Happy to be here.
Fuller: Yossi, as I’ve followed your work over the years, one of the things I’ve found most impressive about it is its breadth. You commented on multiple major phenomena in engineering, how that applies to business, how regulators should think about it. Tell us a little bit about your journey and what has given you this great purview across so many important topics.
Sheffi: Well, the number one thing is getting old—having many years to cover a lot of topics. Another thing is just innate curiosity. My research area covers many, many subjects. It’s just that, to put it the way my wife likes to put it, I get bored easily. I like to dive into new areas, to dive into new fields. That’s what keeps me going.
Fuller: One constant—going back to your earliest work, and where I first noticed you when I was a CEO in the consulting industry—is your deep study of complex supply chains, global supply chains. That was obviously a major topic during Covid. We saw both grave shortages, frightening shortages. We saw what had been a very stable system have a lot of turbulence introduced to it. And systems, once you introduce turbulence to them, are hard to get back into equilibrium, especially when they’re complex with tightly coupled properties. So what are your observations about what we learned from Covid about supply chains, and what do you think are the issues going forward, particularly as we see a little bit of revisitation of the Washington consensus of widespread global trade?
Sheffi: Well, first of all, one of the reasons I wrote my latest book is that I try to get people to understand what supply chains are about, because it is a very complex system, and try to explain to people that, once they understand what it takes, they are not going to be unhappy when a product does not make it to the supermarket shelf or to the Amazon warehouse. In fact, they’re going to be amazed and thankful when it does. Because once you understand the journey from raw material through many, many countries, many, many tax regimes, many, many customers, going back and forth, you understand what it takes to get a sophisticated product to market, you’re just amazed that it has actually happened. So I try to get people to, in many ways, appreciate the fact that, despite what people think, supply chains were not broken during the pandemic. In fact, supply chain managers should all get recognition for keeping working in this time and getting things moving. And let me just say that people panic. The main reason to panic was the media was an amplifier of small things to make them big things. That’s a lack of understanding and context. The New York Times admitted that, until the middle of the pandemic, they did not have any reporter covering the beat of supply chain and logistics. The Wall Street Journal had several people.
Fuller: Well, certainly we did have all sorts of traditional lines of demarcation between issues erased, in what was a global emergency. And I think one thing that is relevant in your work is that these are systems. And most human beings, certainly most correspondents for periodicals, are not really systems thinkers—like policymakers, when they’re trying to get some specific outcome, very often miss the point of the system’s effects and how they might, in fact, get perverse outcomes or unintended consequences. When we think about supply chains in the future, what are the most important attributes you think [for] people that are trying to get a better understanding of supply chains and how they affect business outcomes, social outcomes? And are there particular themes you think are underappreciated by decision makers?
Sheffi: Okay. So first of all, from the company’s point of view, the two most important things are cost and level of service. During the pandemic, for example, we saw, because of the situation, costs were going up, and there were shortages, the levels of service, there was no availability there. Companies tried to make sure that the products are always available at the lowest possible cost. Now, because of society’s needs, two other issues came to the forefront: resilience and sustainability. I wrote a book about sustainability, Balancing Green, because sustainability is a supply chain phenomenon. Because a company—it’s meaningless if they cut their emissions, but a company in China who manufactured the stuff has all the emissions. In fact, it is even worse than doing it in the United States, because at least here we have some environmental guardrails. So you have to look at the entire supply chain and judge a company by the operation of the supply chain. And that’s, by the way, not easy at all, because supply chains are so complex. There are many, many tiers of production, of raw material and parts and other parts and more parts, and finally assembly. Most companies don’t have any idea who the deep-tier suppliers in the supply chain are. So it’s very hard for them to understand how to make them, say, more sustainable, how to make sure they don’t go suddenly out of business, how to make sure they have continuity of supply. And the big challenge—as we go possibly into a recession, who knows, but we go into any other disruption—is actually to keep eyes on the ball in terms of long-term, to keep, for example, on sustainability, while people are worried about cost. Europe is already getting into a recession, and you see, even in Europe, suddenly gas is a green fuel. So that’s an indication of people and politicians understanding what’s important. Every politician in a democratic society is short-term thinking. There’s a reason the Chinese are not short-term thinking, because they live for the long term. They’re not going to be thrown out if something goes wrong. So all these issues are adding to what supply chain managers have to contend with.
Fuller: So one thing we’ve seen, as I alluded to earlier, as a result—probably initially because of Covid, and Covid revealing that, for example, a vast, vast majority of the basic chemicals in antibiotics are made in China, that the vast, vast majority of personal protective equipment is made in China, that the vast majority of rare earths are mined in China—so that across multiple sectors, this growing complex global supply chain that had emerged since the mid-1990s had had some perverse effects, and that we started hearing early on in that about reshoring, which is something I must admit I’ve talked to a lot of people about. I haven’t witnessed a lot of it happening, but I’m interested in your commentary on that. But now we also have the beginning of the evolution of some industrial policies, which is an idea that has been bandied about for 40 years, 50 years that I can remember. And the West, and particularly the United States, has always stepped back from it. But now, certainly with the CHIPS Act and the Inflation Reduction Act, with the massive subsidies for green, we seem to be heading in that direction. What are your views of this effort to test decoupling some of the global system, maybe moving to more trading blocs? And what’s the plausibility of actually seeing a lot of change, given this binary set of considerations that you just posed about the security, resilience of the supply chain, but also the cost per unit?
Sheffi: Well, the answer is that it is not going to really happen. The reason is, let’s unpack several of the things that you were talking about. Leaving China, which is what most policymakers are worried about, is something that cannot be done in anything less than decades. Companies invested decades and billions of dollars in not just getting material in China, not just getting low [-cost] labor in China. In fact, China is not any more... In the industrial centers of China, labor costs are getting to be pretty high. But China became a sophisticated manufacturer, an innovative manufacturer of a lot of products. People will argue with this, but I would argue that, in electric vehicles, they’re now close to leading the pack. There are many, many areas that are just sophisticated—military supplies. But also things like textile manufacturing, robotics, and others. So people invested decades in China. Of course, China always helps themselves with all the contracts that you have to come here, you want to sell in China, you have to make in China, and by the way, give us your recipe for the product you make. And most companies succumbed to this and did it. But now, in many ways it’s too late. The Chinese are already there. They’re anchored in our supply chain systems. But they’re anchored, not only in terms of they have the rare earth minerals. For example, aluminum. China is one of the largest manufacturers in the world of aluminum. So you can say, we can get aluminum from Australia, too. There are other places you can get it; we can actually mine it in the United States if we really want to. But most of the smelters in the world, the best ones, are in China. They built the whole supply chain for aluminum. When people talk about reshoring, first of all, they think about taking the final process, the assembly, out of China. But the whole supply chain is still in China. And even if we’ll be somewhat successful in this, it means that we will regionalize, which means we will restrict our markets, too, because other people will do it as well. And everybody will start closing their markets, which when everybody’s losing scale, cost is going to go up, inflation is going to rear its ugly head again. All of these things are not well thought out. The Infrastructure Act, which is another way of helping the United States, but there’s so many things that are missing. For example, TSMC [Taiwan Semiconductor Manufacturing Co.] are building a chip plant in the United States, announced that making chips in the United States, even if we have all the material, is going to cost 50 percent more than in Taiwan. Why? Because they have to train engineers. They say, “We don’t have enough trained engineers to do sophisticated [manufacturing].” And those guys make the most sophisticated chips in the world. And they know that when they train these engineers, Intel and Nvidia and everybody else will steal them. So they have continuously trained engineers. They understand this. And then with raw materials, we want to balance sustainability with, say, security. Because the issue of raw material, they also go into military jets. It’s an issue of security. So I want security, I want clean air and clean water, but I’m not willing to completely sacrifice one for the other. So we have to think about maybe opening some mines in the United States, because we know they’re going to be better anyway than the mines in China. So we should do it. We cannot let Europe freeze when they don’t have Russian gas. If we don’t allow fracking gas—which is now considered green energy—we put our allies into trouble, who may turn to other sources. So it’s exactly the point that you mentioned before: system thinking. You cannot think about the environment or security or resilience or cost or service independently. They’re all connected to each other, and there’s trade-offs between them.
Fuller: Well, it’s a very interesting dilemma on two folds. One is, of course, that certainly the way most government bureaucracies [have] organized responsibilities for different parts of the system. Something else you raised, which I think is very interesting, is this notion of the relative merits of a supply chain versus its absolute merits. So, we can say we don’t want to mine rare earths in Wyoming, because mining rare earths is environmentally problematic. But if you’re going to allow the Chinese to mine them in a much less responsible way, are you not just fooling yourself? And I think ...
Sheffi: And creating dependence on China in the same …
Fuller: And creating .... Mm-hmm.
Sheffi: It’s bad on both accounts. But you’re right. The way government is organized, the way many parts of societies are organized, many ... This is what, by the way, attracted me to supply chains in business, because it cuts across. When you’re in supply chains, you need to understand marketing and sales and finance and procurement, you need to understand how the company operates, the whole thing. And not only this, you need to understand your customers, you need to understand suppliers, and their suppliers, and their suppliers. It’s a very complex and big system working over multiple countries and regimes and cultures. I always tell my friends, “Supply chain is life.” It’s complex and frustrating and all of the above.
Fuller: As someone who thinks systemically and broadly about these issues, how are you thinking about the application of AI? Where do you think it’s going to be most relevant, particularly in solving the type of systemic problems that we’ve been talking about?
Sheffi: There’s, of course, understandable fear among professionals. After ChatGPT came out, it starts to obviate a lot of white-collar work. I was talking to people in tech who are worried about the computer programming jobs. Who needs us? Or who needs us as writing books? Or who needs us to write media, or who needs us to do TV programming? First of all, it’s very hard to predict. And when it’s very hard, we go to the past, we go to history. Every forecasting is based on history. And if we look at the four industrial revolutions that precede us today, every one of these had fear that jobs are going to disappear, and every one of these created a lot more jobs than jobs disappeared. So first of all, this should be some comfort. By the way, let me mention that in every one of those, people say, “But this time is different.” So it’s also a common phrase. Another thing, it takes a long time. Job changes ... Some jobs are being lost. We don’t have any more elevator operators or telephone exchange operators or computers. Computers used to be a job, not just a device. But it takes a long time for this to change, because these are company processes, government processes. Several examples. I mean, one example is AT&T in 1892 invented the automatic telephone exchange. 1950, there were still 350,000 telephone operators in the United States. Only by 1980, nine decades later, it started to disappear. So it takes time. Taking it in, embedding anything in a process, takes time. That’s one. Second, there are unions, of course, always worried about jobs, and try to slow down jobs, and with some success. Third, there’s government regulation, and we’re talking now about AI. People are talking about, before you’ll be able to release code into the wild, you’ll have to go through clinical trials, you’ll have to try it out, make sure that it works well and doesn’t do what it’s not supposed to do. And finally, there’s acceptability, social acceptability. Not too many people would be totally comfortable with seeing autonomous trucks run at 100 miles per hour on the freeway behind them with no driver there. We’ll see. It may come, but it will take time. Some jobs will disappear, but mostly jobs will change. In fact, the main challenge for companies, for government, for society in general, is how to integrate people with AI. The example of people in the loop, when you go to an Amazon warehouse, the whole aisle moves to the person, the person does something, it moves on. The person and the machine are in the loop, working together. There are cases when the person is operating, when you have an automotive plant, you have people with iPad-like devices who operate the robots. So there’s many, many cases when people monitor and get involved, just when something goes bad, or people are part of the process. In other areas, and when the robot does ... a variation of this is when a robot does the simple jobs. So, for example, when I talk to computer programmers who are worried about ... So they tell me, the veteran computer scientists are not worried about this, because they have so much knowledge. It’s the junior ones who are a problem. Their jobs can be automated. And I’m saying, let me think. How do you think we’re going to grow veteran computer programmers? They have to start as junior computer programmers. They have to learn the ropes. They have to learn how to do it. They have to gain experience. Otherwise, you are not going to have senior ones. And as long as the lifespan of human is limited, you need a constant flow of people. So that’s another challenge. How do you get people with five years of experience or 10 years of experience in every job? Finally, let me just say that it is impossible to imagine all the new jobs that will be created. Ford created the assembly line, which changes manufacturing fundamentally. Instead of a group of artisans basically building a car, the car moves along the assembly line, and each person was doing just a limited part. What happened is, Ford employment went from about few thousand to 150,000 during the heyday of the Model T production. So people think, we got more jobs. This was the smallest part of the effect. Cars became a lot less expensive, people could afford them, and now they wanted to travel. So we built freeways, we built hotels, motels, restaurants. The whole hospitality industry, with millions of jobs, developed. This was not what Henry Ford was trying to do. But there are unexpected consequences. So it’s very hard to imagine. And the anxiety comes because we know the people who are going to lose their job, or the job is going to change, because they are the people that we work with, they are our friends, our neighbors, the people that we see every day. And we can see the people at the supermarket who may lose their job. What we don’t see is all the new industry, because they don’t exist yet, or the new jobs that don’t exist yet.
Fuller: One thing that’s encouraging about this, certainly in our research, in our Managing the Future of Work Project, is, of course, AI has the possibility of replacing, really, tasks done within jobs. And many of the tasks that’ll be replaced are tasks that can be described as dull, white-collar routine work, where you’re creating the monthly report, you’re creating the monthly forecast, you’re creating a summary of the press clippings about the company, that all of those things are highly addressable by existing-level AI, with the right prompt engineering. And what it’ll free those white-collar workers to do, first of all, to be more productive. And productivity, at least the way we’ve measured it in the US, has been pretty unimpressive in its growth in recent years. But also it will allow those people to spend more time doing more creative work, more interpersonal work, and just having more fulfilling careers. We used to think about automation replacing the so-called dirty, dark, dangerous jobs—you could add the fourth D of “dull.” Now, that might be a little bit frightening, because white-collar workers will be more exposed to the advent of AI than they have been to previous cycles of innovation, which means it’s not just the clerk at the grocery store, but it’s also people down the hall from us, or down two floors from us, or ...
Sheffi: Or us.
Fuller: Or us. Well, that brings me to an interesting question, because I know here at Harvard Business School, we’ve been investigating AI relative to its capacity to analyze the case studies we use in our Socratic teaching method. I’ve been using it to get extensive commentary on research vectors I’m thinking about pursuing in our future-of-work analyses, since it can read every paper ever written on labor economics if it so chooses. How do you see AI affecting universities, affecting the way we teach, the research we do, the way we want to capture and propagate knowledge?
Sheffi: Okay. So, let me start by being a little provocative. One of the debates in many universities is, on a lower level, do we allow students to use ChatGPT to write articles, project reports, what have you. Because people say, “They’ll never learn how to write. They don’t know how to put two sentences together.” My answer to this, and I started thinking about this, when I was in school, they taught me how to take square roots by hand. Nobody’s teaching anybody today how to take a square root, how to do the calculation by hand. It’s really not necessary. Would we get to the point that it’s not necessary to write? That writing will be, who needs it? As long as you can read, as long as you develop the critical thinking that allows you to judge what you read, and to see if it’s ChatGPT hallucinating or it’s being able to critically read. That will be the most important thing, maybe, taught in school. I’m thinking always about driving cars. There’s a mechanic and a driver. The driver doesn’t have to know how the machine works, just has to use it. In terms of using it in education, we can certainly use it for exactly the type of thing that you mentioned: preparing class notes, preparing even a class syllabus. We have a huge online program—that just had, in November, we had the millionth registrant on supply chain. So it is very successful. And we use a lot of AI to analyze, for example—and this was developed before ChatGPT, it’s just standard machine learning, you might call it, before large language models—in order to find out when people are going to drop the course, to see the warning signs before they drop the course, so we can intervene. To see the warning signs when they start having difficulty. Honestly, that’s much easier online, because online you have a record of every click that they do, and you can have unbelievable data. Now, what you don’t have is looking at the face of somebody, which you can do in class, and say, “That that person is in trouble.” But you can do it, actually, pretty successfully. We were very successful in ... well, success almost 100 percent in identifying the people who may drop out, not 100 percent in convincing them not to drop out. This was about 50 percent or so. But identifying is usually the most important part. Intervening, OK, we can intervene digitally, we intervene with them, talk to them, try to see what’s going on, and try to help.
Fuller: It strikes me that there may be some exciting opportunities—particularly as the very impressive rate of learning in the generative AI systems—that you could get patterns where the AI is able to deduce more rapidly than the instructor what it is about a concept an individual student isn’t grasping. Or what are inferences or arguments that a researcher may be unaware of, that are in an adjacent but relevant topic. And the prospect of making both learning and the generation of knowledge more efficient and more systematic, to go back to a concept we touched on, might be something we can look forward to.
Sheffi: And we do it. We do it already, but we do it in the online, because of the richness of data. So, for example, we give problems that student have to solve. Automatically, if they solve it, and there’s certain errors, we know that they should get another problem of that kind, of a different kind. In many cases, we just introduce, try to walk the students through a problem set, and get them to the answer, rather than give them the answer. And this is now being done through large language models, or we’re in the process of doing it with large language models.
Fuller: Yossi, I’ve been so grateful for the breadth and vision of your work. Where are you heading next? With this restless mind of yours that doesn’t like being bored, what is the next glittering distraction you think you might focus on?
Sheffi: Well, I’m trying to ask ChatGPT, where should I go? I think that I’m still going to keep tabs on the development of how people get human and AI-fused algorithms to work together. The feel is that there’s a fundamental change that may be going on, and I’m trying to keep on top of this, to be part of it, to see how it’s working, to document it, to do all the things, how it affects supply chain, in particular, but in general, as we say, supply chain is life, so how it affects everything.
Fuller: Well, the emergence of more, what some people have call “bionic” jobs, the blend of biological talent and technological talent, is certainly going to be a principal theme in our research. Always a pleasure to hear from you, Yossi, and hear about your thinking about current events, the economy, technology, and very anxious to see what emerges from your restless mind going forward. Professor Yossi Sheffi, of MIT and the head of the Center for Transportation and Logistics, thank you so much for joining us.
Sheffi: Thank you for having me. Appreciated it.
Fuller: We hope you enjoy the Managing the Future of Work podcast. If you haven’t already, please subscribe and rate the show wherever you get your podcasts. You can find out more about the Managing the Future of Work Project at our website hbs.edu/managingthefutureofwork. While you’re there, sign up for our newsletter.