- Welcome to MIT's "Supply Chain Frontiers," presented by the MIT Center for Transportation and Logistics. Each episode of "Supply Chain Frontiers" features center researchers and staff or experts from industry for in-depth conversations about supply chain management, logistics, education and beyond. Today's episode features a conversation between CTL Director Yossi Sheffi and Susan Lacefield, executive editor at "Supply Chain Quarterly." Today's conversation was recorded in front of a live audience and covers a wide range of topics touched on in Professor Sheffi's latest book, "The Magic Conveyor Belt: Supply Chains, A.I., and the Future of Work." But first, MIT CTL offers a variety of educational programs for graduate students, seasoned industry professionals and anyone at any level looking to learn more about the supply chain and logistics domains. To find out more about all of CTL's educational offerings, visit ctl.mit.edu/education. And now, without further ado, here's what makes the magic conveyor belt so magical. - Maybe a good place to start is with the title of the book. Can you explain the analogy you make between the supply chain and the magic conveyor belt, and what makes it magical? - So let's start with why I wrote this book. After the pandemic, a lot of people were getting to my wife and asking her, "We understand your husband is in supply chain. What is this?" And imagine if, even after the pandemic, people heard a lot of supply chain, didn't know what it is. So rather than having one-on-one interview with one of the several hundred friends that my wife has, I decide to write a book. So the first part of the book is explaining what supply chains are, why they are complex and, in some sense, why would people should not be pissed off when something is not on the shelf or not available on Amazon, but should be amazed and awe-inspired when it's there. Once they understand what it takes to get something from the mines in China or somewhere to a finished product on a shelf, how many processes it has to go through, how many people are involved, how many different tax regimes, custom regime it has to go through before we get the final product. So this was the rationale. And the magic convey belt is because once you understand what it takes, you think it's magic. - Mm-hm, and it's very true. - That's the title. - That's true. So it was to get away from people asking you why their cat food- - Yeah, absolutely. Absolutely. - So as you mentioned, the first part of the book really talks about the growing complexity of the supply chain over the past few decades. And I was wondering, do you think we're gonna reach a point where companies are gonna push back and say, "Things are getting too complex" and we need to maybe take a step back and look at simplifying? Or is complexity here to stay? - I'm not sure. I think complexity is here to stay. Complexity is here to grow because of unexpected event that's happening. And furthermore, I'm not sure there's a pressure to do it because a lot of the technology that is being available help company deal with the complexity and deal with the unexpected event. So I'm not sure there's a pressure to do it, especially among large, sophisticated companies. So the answer is no. - No. It is here to stay. Here to stay. - You talked about one of the most mind-blowing facts about any product that we touch is the thousands of organizations that have been involved in creating it, and that they have done that without any central control. And I was wondering if decentralization is, do you feel that's crucial to supply chain efficiency and operating in this complex world? - Categorically, yes. - Okay. - The idea that somebody can control, control of supply chain is controlling the economy. We tried it once or twice. Didn't work very well. So we're talking about modern markets. Supply chain is actually a whole set of buyer-seller, buyer-seller, buyer-seller negotiation, transaction, operation. It works because everybody's trying to do the right thing to minimize costs and maximize level of service, by and large. Now there are other things people are worried about, like sustainability and resilience, but everybody is worried about it, so everybody's trying to get the best outcome. I don't see how central planning can work. Even in China, we don't see, it's not central planning. Central control of certain aspects, but not of the transaction. In fact, the Chinese seem to be leery of very large corporations who control more of the larger part of the economy. Has happened to several, you know, tech companies in China. They actually seem to encourage competition between companies. So I think it works, the market works. - But as you introduce decentralization, there's an element of risk that kind of enters the equation. I was wondering how do we balance that risk with all the benefits? - No, it's, au contraire. The risk goes down. - Huh. - Because the risk to a particular company maybe goes up. They are out there on the front line. But the risk to the economy- - Ah. - Goes down. - Okay. - Look, you can always find good restaurant in New York, always. You walk to a random restaurant, the chances are it's a very good one. Why? Because restaurants in New York, if you open a restaurant in New York, the chances are within a year, you'll have to close it. The competition is murderous. There are so many good restaurants. So you can say the chances for individual restaurant to succeed is not very high. But going to New York and having a good restaurant, you know, the environment is great. It works. There's no risk. You don't risk going to New York and not finding a good place to eat. I'm not saying a place to eat, a good place to eat, because it is decentralized. - But there is, when you outsource to a supplier and they're outsourcing to other suppliers, there is that added risk of, you know, a quality defect that you can't control or a sustainability issue popping up. Is that a concern with this decentralization? You know, how do you control for that sort of? - I don't see it as a decentralization issue. - Okay, okay. - I see it as the depth of the supply chain, the lack of visibility. It exists. It get slowly better with new technology. But there are limits here. The limits are that for suppliers to tell their customer who their supplier is, not every supplier is willing to do it. It's a competitive advantage to know who the suppliers are. And there always the fear that the customer will go around them, will go directly to the supplier. So there's a kind of built-in opaqueness to the supply chain, which we're trying to get through to visibility and good relationship and all of this, and some people are more successful than others. But this issue is not a technology issue and it's gonna be very hard to solve completely. And it's not decentralization issue. It's the depth of the supply chain. - So in the second half of the book, you spend a lotta time talking about artificial intelligence and the effects that AI is having on the supply chain. And I was wondering, you know, when ChatGPT hit the scene in November, suddenly, generative AI became a very hot topic. And I was wondering if you could talk about some of the applications for generative AI that you are seeing in the supply chain. - First of all, let me just explain that we have been using, even- - Oh, yeah, - AI for a long time, using that. All the restaurants, all the drive-through restaurants are using chatbot. But it's not only drive-through. Every time you call, nowaday, customer service function, you're talking to a chatbot to interpret the results and try to give you answer. And if sometimes it gets stuck or you get stuck and started screaming, "Agent, agent, agent," or something to this effect, a human comes on. And just like when you go to the drive-through and you start ordering, you know, Champagne and McDonald doesn't have it, a human comes onboard and say, "Well, I'm sorry. We don't yet serve Champagne." An interesting application is in risk management and supply chain, trying to look at suppliers and finding out how risky they are. Turns out that when you look at metrics like then financial metrics, they are backward-looking by about two quarters. You want to see what's going on now. We know, for a long time, that one of the warning signs is having coverage about executives' living, about failing some projects, failing some M and A project in particular, having bank covenants that are a little problematic. So now we have several companies are using large language model, particularly, to look at tens of thousands of suppliers at the same time and analyzing all of them, analyzing all the mention in the media of redundancy, of executive living, whatever, in order to generate an alert and have somebody visit there and finding out what's going on, if we need to start looking for another supplier or what. So this is something that could not have imagined before we had this technology. Looking at, if you look at a company like, I don't know, I'm working with Flex a lot, and they've 18,000 suppliers. It's just, first-tier suppliers, just finding out what's going on with them is an issue. Having a much better alert when something goes wrong is something that we were not able to do before this type of technology was able. We could check, you know, 10 suppliers at a time. Checking tens of thousand was impossible. Now it's being done. - So the AI is going to actually enable even more complexity in the supply chain in the future as we're- - Yes. It just, it can enable more possibilities. More possibilities create complexity. So, of course, when people get into, when there's pressure, economic pressure, whatever pressure, we know that during recessions, company reduce the number of SKUs. They're trying to simplify. They're also trying to reduce cost, you know, improve service, but they're trying to simplify. But, you know, there's the accordion theory of management, that when recession happen, the number of SKU goes down, and then marketing comes up with all the good reason why we need more and more and more SKU to serve more and more territories, more and more, 'cause all kind of option. And then it, so that's the accordion theory of management. And it works, actually. - So it kind of, it's like the pendulum swing that kind of balances. - Yeah, between recessionary period, expansionary period. - So we've talked a little bit about AI. Is there anything about the application of AI to the supply chain that gives you pause or areas of concern? - Not about the supply chain. The areas that give me concern, the areas, other people call, the area of fake news can be done very convincingly. - Yes. - The area of giving instruction, how to build improvise roadside explosives. But while I'm saying this is a concern, a concern around, and the media, and I'm not that concerned about it because, just give you an idea. Unlike the early days of the internet, when we all, everybody thought this is the greatest thing since sliced bread, right? Because we can communicate with everybody, families can see each other. "All the, you know, distance is dead," to quote Tom Friedman. Nobody thought about identity theft and stealing customer data and terrorists communicating to each other on the net. But now, it's different. With the generative AI, the companies, the media, the politician are all aware of the dangers. So there's a lot of work is going on already. Already, the companies themselves are putting guardrails on this. So if you get ChatGPT or any one of the others and ask how to prepare a Molotov cocktails, it's not gonna answer. So this is not give you an answer. So they already started to put guardrail, and there'll be more of this. - Are you seeing that also with use of analytics in companies where, you know, you might have an algorithm? There's an example in the book about two competing bookstores and they're both using a pricing algorithm on Amazon. And as a result, that drove up the price of the book very high. Are you seeing? Companies already had those human interventions in place to make sure that the algorithms don't go outta control. - Let me give you a more even general answer. - Okay. - One of the most important type of work in the future will be monitoring, vetting the automation, AI-infused or otherwise, but having a human monitoring. That's a tough job. It's a tough job because you have to monitor something, and if you don't do every day, and actually you lose expertise. - Right. - It's hard to keep sharp. And we have cases when, you know, things did go awry. So it's important, how do we train people to do it? - Yes. - For example, today, modern aircraft can basically fly by itself, gate to gate. Now, talking about autonomous vehicles, not too many people will go on a, you know, aluminum cylinder that fly 35,000 feet over the ocean without a pilot, right? So the pilots are in the aircraft, actually don't need to do anything. They can just sit there and nap. But what we do, we let them do the communication, basically. It's the number one job. So they always have to communicate and change frequency. So they keep alert. It's one way to do it. Because flying the airplane, it flies itself. So you really don't, once you put the crew in, it flies, it change route, it goes automatically. But you give some jobs to the human, they are not gonna fall asleep. That's part of the challenge of the future. There are several models how people and machine can work together. Now, one such model is what we talk about, the chatbots. The chatbot has a monitor because you talk to McDonald, whatever, in the drive-through, you actually talk to a chatbot in most places, and they respond. And then when they don't understand something, a human comes on and we talk to you. So there's a monitoring of what's going on, and the minute that the chatbot doesn't understand what's going on or gives the wrong answer or whatever, a human comes. So that's actually a monitoring function that we don't even think about, but happens every day. With most customer service function, you know, it used to be that press one for this, press two for this, press three for this. That's rare now. Now you just talk to the computer and it turns it into text that appear on somebody's screen, and then they report and try to find an answer. That's AI. - Do you have any good examples of companies that are doing good thinking around what should be given to humans to do in the supply chain and what should be outsourced to AI? - To me, that's the question of the future. - Yeah. - The question of how. The integration of humans and AI-infused automation is a question of the future. We talk about one model. The monitoring is one model. You can think about when the human is in the loop. The human is in the loop, for example, think about an Amazon warehouse. When the picker stands in one place and there's a, you know, the aisle comes to the picker, does something, then another aisle come to the picker. So the human is in the flow of the work. So that's another example. And a third example is the human operates. When you go to several automotive plants, for example, you see workers standing with iPad-like devices and basically running the robots. That's another way of working with AI. So that's, as I said, the question of the future, how to organize the work- - Exactly. - And how to, in some sense, how to get the best out of the machine and out of the human because they have complementary skills. You know, machines work all the time, don't get breaks, don't go to the bathroom. They just work and- - Get sick. - They don't get sick. And they're usually very accurate. They do, you know, repeated work over, and overtime. What machines don't have is context, understanding when something does not belong, has to change. When we start think about, you know, something change in the economy, and suddenly people order things differently. So many standard automated ordering system use the point of sale data and order based on this, put it into some forecast. But this forecast is based on, at best, say, on machine learning, which is basically looking at past data. All forecasts are based on past data. When something is changing structurally, suddenly there's a pandemic, suddenly there's something else happening and people change their buying habits, then humans have to intervene again because the machine does not have context. As the machine is concerned, nothing changed. I mean, it gets, you know, point of sale data, I keep going. But something has changed, and people understand the context. Now, there's other things they worry about, empathy and bias and things in general that human can make sure that happen or don't happen. It's harder for machines. - Do you think we're gonna get to that point, where machines are gonna be better at mimicking that empathy piece? 'Cause it feels like the people who are working on AI are trying to get there, you know? See AI used in mental health these days and. - Yes, there are some actually automated psychologists that try to help people. Who knows. - Who knows? - I'm not sure about this because that's exactly a question of context. - Yes. - Two people coming and saying, you know, "I hate my children." - Or, "I hate my supplier." - Or, "I hate my supplier." Well, you hate a supplier, don't go to a- a psychologist. But, you know, you hate your children, you go to a psychologist. But the context may be entirely different. You know, I hate my child because he's a thief and a liar, or I hate my child just because I don't like tall kids. I don't know, who knows? - I mean, it's the context that- - Get a crick in my neck when I talk to you. It's hard to imagine some of these things moving to AI completely. And they talk about supplier. Again, it is hard to imagine, or let me put it strongly, I don't think in the next 10 years, in five, 10 years, we will be able to have an algorithm setting up a contract with a supplier in China or Vietnam, let's say. To set up a contract and relationship for a long while, it requires somebody to fly to Vietnam, to negotiate like hell for two days, and then sit and have dinners or two dinners and talk about their kids and talk about the family and create relationship. I don't see it changing in the near future. I mean, AI will have to be so much better and have to, but not the quantum jump in capability to be able to do it, which, right now, not clear it's possible. - It's interesting, though, because there's been a movement with technology of making decisions more fact-based as opposed to, you know, I like Joe over at such-and-such trucking company, so we're gonna use him. But it seems like that human relationship is, you're saying is still gonna be an integral part of supply chain management in the future. - Yes, it is still integral because, for example, if something goes wrong and there's some disruptions, how do I make sure that this supplier knows my situation, knows me? And if I'm calling and say, "Look, I really need it," and everybody else call and say, "I need it." - I need it, yeah. - But I know this guy and I know that he really needs it. So the knowledge is, I think, still very important, the personal relationship. Now, one has to be . There may be critical suppliers and maybe suppliers that are not so critical. And if they, maybe supplier, if I have some part, some commodity that they have dozens of suppliers, and if that supplier goes down or I have some shortage, there are many others, maybe that I don't need to be close to them. But for most important suppliers, I don't see any other way. - Sometimes it's hard to know which are your critical suppliers. You might need that little screw, and then, suddenly, that screw goes down. - Yeah, it's called, in the automotive business, they call it the golden screw. It's one part that's missing and you cannot make a car. - Right, right. The example in your book about the Ford not being able to ship out because they didn't have the little Ford logo that sticks on the truck. - This was last year. You know, Ford has the blue little oval that they put in the front of the truck. They didn't have them during the shortages. They couldn't make trucks. I mean, the trucks were standing in the yard and they couldn't sell them for a month, actually. - So can AI be helpful identifying who it is that you need to spend your time developing that human relationship with? It might not be who you think it is. You also have to. - You can take AI, I think it's simpler, but as an aside, let me say that AI became the buzzword- - Buzzword. - At the time. We used to think about blockchain or RFID or became, and, you know, people who are doing blockchain project, they're actually just fixing up their systems. To get funding from management, they call it, that's a blockchain project. Now they call it, that's an AI project for doing some optimization. - So that's the learning to go away. When you go back to your company, make sure your project is AI. - Use AI, okay, what exactly are you using and is it appropriate? Can that tell you how many company out there that tail is wagging the dog? I used to go to boards, and people would ask the CEO, "What's your China strategy?" Or "What's your blockchain strategy?" Now you ask, "What's your AI strategy?" And I always said, "Stop it, what's your problem? Start with the problem. - Maybe the solution is AI, maybe it's just hiring another person. I mean, you don't start with the solution. But it's amazing how many people still do because, I don't know, in part because Wall Street pays premium for having an AI strategy or something of this effect. It's not clear to me. It makes no sense. - Is figuring out the problem an AI issue or a essentially human issue, is that something that's gonna- - AI issue, you know, operational research issue, statistics issue, you know, people issue, process issue, can be anything. So that's why I don't like having an AI strategy or a blockchain strategy or whatever is the current fad. I should say AI is not fad. It's been growing for many, many years, and we got to the point that it could make substantial changes in how people work and the relationship between people and machine. - Right. Just like a year earlier, I think the buzzword was all robotics. So it's kind of, or cobots, and so it's the same sorta thing. - Robotics are also now fused by AI. I mean, so, - Right. It's not the actual hardware of the robot, it's the software. - Of course. - Yeah. So kinda taking a step back, to your point about context and the pilots and training, sometimes you have to do all the low-level jobs to get that context to know what to do next. So- - I do talk about it. - Yes. So what can we do with our supply chain pilots, so to speak, to make sure that they have the background, the knowledge to be able to take over those unusual events? - Again, I take the problem a little further from your question. - Okay. - So I was interviewing a shop, basically a software provider, asking them about ChatGPT taking their job because it can now program. And so the senior computer scientists are not worried about it, but it may take the job of the junior computer scientists. Now we're saying, "Guess what? Senior computer scientists don't come as senior computer scientists. They start as junior computer scientist. We don't hire junior computer scientists, you don't have work for them, you are not gonna have senior computer scientists. And even for monitoring, you need people with experience. In the book, I talk a lot about how to do it and how to upgrade skills, but there has to be recognition that you need to hire people at the lowest level. One of the suggestions that I made is maybe pivot in the United States for more of the German system of people spending half time in a company and half time in a university. And they come up, and it's called the dual education system. That's about 52% of the German high schoolers go into this system, which is government-controlled. The government defines Germany. So the government define 365 professions where this can be done. And the university, you apply to, actually, to the company, and they work with a local college or university. You spend half the time studying the theory, basically, and half the time doing the work. 70% of these people get hired by the company that they do their internship with. But they come with experience, knowing the culture, knowing the company. It's much higher to move them and much easier to move them along. The United States, we suffer another problem, is this, is every mother wants to say that their child goes to college. My child goes to so-and-so college, and your child just goes to trade school." I always say that people should meet my plumber. Yeah. - My plumber drives a Bentley, and we don't have enough plumbers. And they can set the price, and they do set it high. So we send, there are too many people who go to college in the United States and unfortunately, in many cases, come back with, have to call it debt for a long, long period, rather than go to trade schools and community colleges, or combination. Actually, there's a university here that does it. Northeastern. - Northeastern, yes. - The combination of work, and it's not as organized as in Germany, but it's the same idea. You work one semester, you study one semester, and you flip between them. - It's interesting. I feel like Northeastern is becoming a school that more and more people want to go to nowadays. - I know. - So that comes back to your main job of training students. How have what you feel are the necessary skills for a supply chain manager changed recently? - If I go over the history, this program here started a very analytical program, and then we realized that our graduates are very analytically savvy, end up working for Harvard MBAs who are half as smart and get paid twice as much. And said, "This is not working." Furthermore, companies came to us and say, "Your graduates don't go up the ladder in the company because "they need the soft skills." They need to be able to communicate, they need to be able to sell, they need to be able to explain a position, they need to be able to work in a team. So the programs change, started doing a lot more of this. I think that as AI and automation is getting more and more into the workplace, is the soft skills that will become even more important. How do you work in team? How do you make sure that your people can work with AI? You know, the promise of AI is that it will do the job that nobody wants to do, and people will do the more interesting and fulfilling job. How do you make sure that this actually happens? So all of this is part of the challenge of the future. We don't have all the answer yet. We don't even have some of the answer yet, but we're thinking about it. So people will need to understand, we're not training computer scientists, but people need to understand the capabilities and where it can go wrong. So people need to be sophisticated users. It's like my colleague Chris Caplice always talks about driving. There are mechanics who actually can fix the car and know what's inside. And then drivers, you don't have to know what's going on. You can just operate it. We like to train drivers, people who understand what the system can and cannot do, but they don't need to be builders of AI or generative AI system. But they need to know the promise, the limitation and how to best use them. - Yes. - People always ask me, in classes, if we allow people to use ChatGPT. That's a big debate in universities. Some universities absolutely disallow it. It's ridiculous. You know, when I was your age and actually younger, they used to teach me how to take square root by hand. None of you studied it because there are calculators. None of you are studying how to do a financial statement by hand because there's now Excel and spreadsheets. So the question is, why do you need to do to replicate what ChatGPT can do by hand? What you need to do is when it goes awry, you need to test it. You need to make sure that the results are not what they call hallucinations. So because ChatGPT can hallucinate and invent stuff, invent sometimes reference that don't exist. So you need to be sure of this because if you can submit to me a paper written by ChatGPT, as long as you realize that if something is wrong, open AI is not getting an F, you are getting an F. Just so we understand each other. So in short, the responsibility is still on the user, but not using a tool that's available, for me, it's a losing proposition. It's very hard to work. Another example, you know, of how automation is de-skilling jobs, but having other benefits. So if you go to London and you go to a black cab, you know, to drive a black cab in London, you have to study for three or four years and pass an exam, which is considered the toughest exam in the world because you need to know every point of interest in London and how to go from everywhere to everywhere. And you sit in an exam that you have to show that you can drive from everywhere to everywhere in the shortest route. And you have to understand congestion. And people who are doing this and spending four years of their life doing it. And then came Google Maps and Uber. Everybody can do it. Now, there are still, the number of black taxis in London went from 25,000 to about 8,000, but the number of Ubers available is now about 60,000 in London. Lots more of them are available. So win some, lose some. - Another thing you talk about in the book is how technology has had an impact on enabling supply chain strategy. Like, we wouldn't have been able to do all the outsourcing and offshoring if we didn't have advanced communication technology. Do you see some radical changes on how companies will be structuring their supply chain or organizing it because of the AI or other emerging tech, like robotics? - It's already happening, in the sense that the number one use of robotics is in warehouse automation. I mean, warehouses are putting robots like there's no tomorrow. Autonomous vehicles. Autonomous vehicles are robots. So there's a lot of work on autonomous tracking. - Yeah, - Let me just say, however, that I talk to a lot of people, a lot of interviewing, people are worried about their jobs. It's the number one, you know, fear, jobs. And, again, people should chill, at least for the short term, because it doesn't happen fast. Give you one example. In 1892, AT&T invented the automatic telephone exchange. Until then, there were, you know, women putting plugs. "Where's Mrs. Smith today?" "She went to the supermarket." They'll connect you later. Very personal service. - That's what my grandma did. - Yeah, okay. By 1950, there were still 350,000 operators like this in the United States. Only by the 1980s, it started to go really close to zero. Nine decades from the invention until it really, all the jobs went or most of the jobs went away. So it takes time, and it takes time because there are many hurdles. You see already hurdles. You see what are the writers and actors worried about? They're worried about using AI. And they are, you know, stopping the industry, putting the industry down. And the industry will have to come to some kind of agreement. My guess would be part of the agreement will be somehow slowing down or putting guardrails on the use of AI. - Kinda like dock workers with- - I was about to say. - Sorry. - Dock workers also fight automation. In Long Beach, it's nothing like Rotterdam or Singapore or Dubai because of the afraid for the job, afraid for the immediate job, and not taking into account that you can increase the throughput and get even more jobs. - Right. - In general, that's the most difficult thing in this area, in this, people are worried about jobs. And I understand it. It's anxiety because you know that people are gonna lose their job. You see it in the supermarket when you get to, when you can check out yourself. People are gonna lose their jobs. So these are people that you know. What you don't know is all the new industry and the new jobs will come. So one quick example of this, that is old example. So Ford came up with a assembly line system. Changed manufacturing, of course. But it used to be the specialty team used to build one car at time, and Ford employed several thousand workers. During the height of the Model T, using the assembly line, Ford had about 150,000 workers. But this is not the big impact. The big impact was that automobiles became less expensive. Highway developed, hotels, motels, restaurant, the whole hospitality industry created millions of jobs. This was not what Henry Ford had in mind. I mean, but it was a side effect of what happened. That is why it is so hard to imagine all the new jobs that will come. Many of the jobs that exist today did not exist, you know, few decades ago. We talked few decades ago about people who will optimize ads on Google or people. There's so many jobs that are totally new because of new industries that came up. So it's hard to predict what would be . the one thing about supply chain coming back, because that's what you ask about, is it still involve physical movement? Product have to move. So there are some things that will be still grounded for a long time until we start having 3-D printing at scale. This can change supply chain, but it will be a long time because 3-D printing is still very slow. Technology cannot replace mass production, not even close to replace mass production. So I don't see fundamental changes. The changes that may come will come because of geopolitical consideration, resilience consideration, sustainability consideration. But to get this, done we'll need to have some more system thinking, which is in very short supply among the political class, the media class. People are talking about. Give you an example. Rare earth minerals are used in every sophisticated product now. China controlled 80, 90% of the world supply. Aluminum, China controlled most of the world supply, and stone, most of the smelters are in China. You know which country has more rare earth mineral in the ground than China? The United States. But we don't want to mine it because it's environmentally problematic. Even though one should say if it were done in the United States, it would be probably done in a lot more responsible way than it's done in China, but, still. So we have to decide. We have to stop saying, and green is, and we just go green. - Right. - We go security, we go standard of living. We have to think more holistically. And this is system thinking that, as I say, in short supply, because there are pressure groups, whether the green parties in Europe or environmental lobbies in the United States. There are the security hawks that want everything to be from here. But, again, from supply chain point of view, moving the assembly or the last stage of manufacturing to United States is meaningless, or to Europe, is meaningless because there's a whole supply chain that was built after investment of billions of dollars and decades that is still in China. Very hard to get out of this. It will take billions of dollars and decades to get out of there. So we need to stop talking about totally separating the Chinese and the Western economies, and starting to work better together. It's just not realistic. - So no two, what are they? Two-pronged or? - Two-pronged supply chain. It's a nice thought, it's just not realistic, I think, because people don't realize how much is there already that is very hard to move. And, by the way, even if you move some how much of the resources are coming are mined not in the West? So you still need that. And as long as you depend on something, you're not really independent. - Thank you, Yossi. This has been a great conversation. I've enjoyed it. - Thank you. - Thanks for listening to this episode of MIT's "Supply Chain Frontiers," presented by the MIT Center for Transportation and Logistics. To check out other episodes, visit ctl.mit.edu/podcasts. And for more on the center's research, outreach and education initiatives, make sure to visit us at ctl.mit.edu. Until next time.