Creative Destruction Lab Super Session 2018: Keynote with Vinod Khosla

Welcome everybody. So just so we have a sense in terms of guests, how many people, just by show of hands, are here from outside of Toronto?

Okay.

So welcome everyone to the Rotman School and the University of Toronto. It's only once in a while in a person's life that when they make a terrible mistake, that they get a do-over. And so, it just happens that this is one of those times. I've had the opportunity to introduce Vinod Khosla one other time. I don't know if you remember what happened the last time I introduced you. But what happened was that we were hosting a conference for economists on economics of artificial intelligence. And it was with my colleagues, Professor Joshua Ganz and Avi Goldfarb. And as we organized the conference, the night before the conference, one of the senior people from the group that we were organizing with, the National Bureau of Economic Research, One of the senior people from the group said, you know, the very senior people at the NBER are a little bit nervous about this conference that you're organizing. It's a bit unorthodox. And part of that was because of the way we had organized the papers. It was a very senior group. Some of the top economists in the world were coming to this meeting. It was in Canada. Normally these are either in Boston, most of them are in Boston, or Washington, or Jackson Hole, or something. And that you've put as your keynote speakers for this gathering people who aren't economists. And so our dinner speakers were computer scientists. They were Yann LeCun, Russell Kutinoff, and Geoff Hinton. And our lunch speaker was Vinod Khosla. And so they, you know, this person said to me, you have a set of non-economists as the keynotes, that's going a little bit outside the bounds of what we normally do. So when it came time to the next day at lunch and my introduction of Vinod, I felt like, okay, I needed to be quite conservative. And so I was very, you know, I'll stick to the facts, and be very conservative and not, you know, express my enthusiasm for having Vinod. And so I did. I went through, you know, I talked about Sun and Kleiner Perkins and Khosla Ventures and that was it. And I said, you know, and now I'm— it's a pleasure to welcome Vinod Khosla to speak here at the National Bureau of Economic Research Conference on Artificial Intelligence. And just as Vinod was walking up to the stage, I don't know if you remember what happened next, but for those of you who recognize who this is, Larry Summers was one of the guests, one of the participants in the conference. And so Larry was President Clinton's Treasury Secretary, and he had also been Chief Economist of the World Bank. He was President Obama's Director of the National Economic Council, and he was also the former president of Harvard University. And so what he did in the next 30 seconds was such a takedown of my introduction of the node that I was sort of stunned and scarred. And I was going to play the video of it, and I really— I just can't sit through that again. But I think he did give an excellent introduction of the node, so instead of replaying it, I'm just going to read you the exact word-for-word quote of what he said. He was sitting in the corner and he put up his hand. So as Vinod was walking up to begin his presentation, Larry Summers put up his hand and here's what he said. I said, can I just say, as someone who's had a little bit of experience, okay, that was in direct reference to me, that introducing Vinod Khosla as a venture capitalist is an entirely accurate statement in the same way that introducing Babe Ruth as a baseball player or introducing Einstein as a physicist is an entirely accurate statement. Vinod Khosla is a legend for having been in front of almost every major trend that has happened in Silicon Valley over the last 20 years and being in front of only a few, some, but only a small number of trends that did not ever happen in Silicon Valley or anyplace else. But he is probably pretty universally regarded as Silicon Valley's leading thinker. Please join me in welcoming Vinod Khosla.

Thanks, Adi.

All right. Well, listen, it's a real pleasure to have you here at the Creative Design.

It's great to be here.

So we're going to just— I'd like to ask you a few questions and then we'll open up for questions from the audience. But I'd like to begin with a question. So a number of these are motivated by things you've said before. And so the first one concerns international trade and supply chain. So I put up a quote from an interview that you gave with the Wall Street Journal. And I'm gonna ask you a question and then I'm gonna turn to— we're really lucky in the audience. We have— I'm not sure where he's sitting, but we have one of the world's leading trade economists, Professor Dan Trefler, where is Dan? Is he here? Oh, there, okay. So I'm gonna call on Dan after your response to the question and for him to weigh in. But can you explain, you make this statement, "Nobody will want to make anything in China," and you refer to what you call the supply chain inversion. Can you describe what that is and how long you think it will take before that's a true statement?

So let me give you a little bit of context. There's a paper I published about a month or two ago. It's 50 pages long, so I'll warn you. It's fairly detailed. And I started with US GDP. And I said, what parts of non-governmental GDP could you reinvent in a way that was more than 100% and hopefully 1,000% improvement? And I expected to come up with an answer that you could reinvent— this is speaking to all the entrepreneurs here— 20-30% of US non-governmental GDP. I was shocked sitting there last August that I could think of ways of reinventing every major category of the US GDP, and that would apply to most GDP, most, most places in the world. In a way where the resource input, whether it was labor, whether it was design, whether it was materials, could be reinvented with 100% to 1,000% efficiency. None of this normal economist thing of 2% improvements a year. Radical reinvention. It was so radical, I was just standing here, and I looked at this steel columns, I said, could you make them with 80% less weight? And I'm certain you can. Why? Because what technology has done is give us, us more and more tools. So this is the context of societal reinvention I'm very, very excited about. If you could design this where every element, every molecule that isn't structural could be taken out. And you can do that with software and AI, like a field called generative design, which is just starting but will get much better. You can now manufacture it because you do— you can 3D print it. If you 3D print it, you, you can do it on site. So the long answer to a short question: we're generally seeing dematerialization of society. From software's eating the world to material, bits and atoms, forget the bits, atoms that can be replaced and do much more for much fewer atoms, dematerialization of society, combined with the kinds of AI technologies which would completely replace labor. So we routinely being pitched assembly line robots that work at $2 an hour. So if you take, if take these trends, dematerialization, labor replacement with robotics and AI, why would you want 6-month supply cycles? If you're a garment manufacturer, why do you need to order in July everything you might need on Thanksgiving Day or just before Christmas. That's what happens. You can't respond. You've got all your money tied up in inventory. When you have money tied up in inventory, you have huge risk. And risk is the principal thing you want to mitigate. If you could do manufacturing at 24-hour cycles, you'd, in addition to all the other benefits, completely change the equation. Hopefully this frames a number of things we can talk about. Great.

Okay, so Dan, I'm gonna ask you if you wouldn't mind and set us up for the conversation on what Vinod just said with respect to trade.

Oh my gosh, I don't know how to respond to that vast vision except for being called—

You can send it to me in writing. I do have— it's called Reinventing Societal Infrastructure, is the paper. You can probably Google it or I will definitely do that.

I will definitely do that. And at risk of being called uncreative or maybe even worse, an academic, let me ask you a few maybe pulled-back questions. And that is, say for example, 3D printing. It has been around for, you know, 3 decades. It has its stumbling blocks in creating new materials. We saw Michael yesterday saying maybe that stumbling block is going to go away, but it has others, you know, not all companies want this, you know, to have this customization. I mean, Venetians figured out 800 years ago that if they did the opposite of customization, they standardized, they could conquer the world by turning out cheap ships. And we saw in the communication satellites example by Kepler yesterday that it was precisely standardization that helped them move ahead. So in 3D printing, you know, I'm seeing these stumbling blocks to its adoption. So my first question is You know, how fast do you think this is coming down the pipe? And then this is sort of the second question, I have to ask it, because we have all these amazing people in this room and especially you. You know, when we talk about repatriating manufacturing in the supply chain, a lot of us, certainly me, have in mind, "Oh, this is going to be great, we're going to bring back American jobs or Canadian and North American jobs." But we're not really going to be bringing back any blue-collar jobs, and I think that's what most of us in this room must be worried about. So I guess my question to you is, you know, coming down the pipeline is something massively disruptive for our least advantaged. The US government is nowhere in a position to deal with it, so it's going to be private sector solutions. I don't even think I have to state the question explicitly, but can you provide us with some solutions, or can the private sector provide us with solutions we're going to need?

Yeah, so how many economists in the room? Only a few, so I can take a dig at economists then. Economics is about extrapolating the past. What this world, the entrepreneurs are about, is reinventing the future they want. So almost all historical notions extrapolated into curve-fitting models of any complexity don't apply when the world changes, when the fundamentals change. So standardization is only for one reason: reduce costs. And if old technologies were still the dominant technologies, you want to keep reducing costs, and there's economy— economies of scale and other kinds of things economists talk about. When 3 or 4 different technologies come to a nexus— and this is my thesis— you reinvent how things are done so radically that none of the past explanation applies. So if you can— let me take an example in manufacturing, which is the one Many— I'd love talking about manufacturing and building construction and things that nobody thinks is open to innovation. I can replace some of these beams with carbon fiber composites that are 90% less material content if and only if I can lay fiber out directionally accurately and compute in real time the density of the fibers and eliminate 90%, maybe 99% of the labor cost of doing composites. Composites as a technology only becomes feasible if you eliminate labor, and only AI and software and, and new kinds of design techniques eliminate that. But if they do, you could eliminate steel. You can now build a titanium-like beam that's 7 times stronger or have 1/7 the weight of titanium in a carbon fiber composite. Why? Because it's a software function and a robotic AI function if you can real-time change the direction of the fiber you're laying down. Along the axes where you need strength, you, you're really changing the equation where no economic theory applies, period. Because economic science is really a series of correlations, it's not causal. You can take an area like CRISPR in biology and say you apply AI to dramatically improving the efficiency at which you achieve your goal, or CRISPRing a gene, it changes the equation. So one general message— and I'll come back to answering your fundamental jobs question— it is at the nexus of things like AI and CRISPR and robotics that the most interesting things happen. It's not within the fields. To the jobs question— and this is, again, a very old economist thing to say we should create more jobs. What if we needed— eliminated the need for work? That's a very radically different solution. And I believe in at least the 40-year time frame, we will eliminate the need for work. And we will eliminate the need for jobs. I wrote in 2014— you can look this up— say AI will cause great productivity improvement— economists love that— great GDP growth— economists love that— and increasing income disparity. This was 2014 in Forbes. It was a long piece, about 5,000 words. You can look at it. But I do believe the problem of creating jobs will go away as a governmental prerogative. It'll be replaced with the problem of how people find meaning in life. That'll be a more important question than how many jobs we have. Because robots and AI and other things will produce all the goods and services we need. And we can consume them. And with dematerialization, they'll be much cheaper. But the question will be, what will give people meaning? It's a very different question. I suggest economists look at the new future. In fact, after that talk, Larry Summers came up to me and he said, you just screwed up every notion of how I had about how education will help solve this problem. Right after that talk. So we talked a bit more. I don't think education is the solution to jobs. In fact, the highest educated jobs, the oncologists, the radiologists, will be eliminated first. So, sorry, long answer. Let's keep going.

Dan, I don't know if you wanted to have any response to that.

You're good?

Okay.

All right.

So then speaking of radiologists, let's move to the next topic of medicine. So this year in the fall, we're launching CDL Health. Actually, we piloted last year, headquartered at CDL West in Vancouver. And the question here— so there are two themes, there are many themes in your— in what you've written on health. This is from your Medium post from a couple of months ago. But I want to draw on two of them. So the first one is an increase in personalization, and the second is a drop in cost. So in your personalization comments, I'll just take one quote here. You have, "We have to get away from the idea that we use symptoms to diagnose disease." Data science should diagnose disease and monitor progress or recommend a dosage of medication and therapy to best treatment. We should get away from the notion of one prescription for all 7 billion people on the planet based mostly on their weight. And then there is a great quote you have associated with that, which is, if I wanted to be a really good doctor in 15 years, I would not go to the medical school, I'd go to the math department. Then you make a comment about a drop in the cost of provisions, such a dramatic drop in cost that it changes the way we think about the delivery of medicine. So you say there's probably a million doctors in the US, give or take, but with AI systems we could create 10 or 20 or a few hundred million doctors' worth of expertise and use human doctors only for what they love to do, which is interfacing with patients. So I'm gonna ask you to comment on how you think this transition will occur and then call on, is Dr. Dave Greenberg, There he is. I'm going to call on Dr. Dave, I'm going to call on you in a moment after Vinod just weighs in on this in terms of the general trends. And I think we're towards— your key points were towards personalization and dropping cost.

My key points go way beyond that.

Okay.

Right? Like the practice of economics, which is extrapolating the past, the practice of medicine has been learning from our experience and keep improving medicine. It is, I call it, the practice of medicine, not the science of medicine. When you have dry skin or a rash or a brain plaque, there is biochemistry happening. There's roughly a few thousand, 3,000 or so metabolic pathways in the human body. They are interacting as networks. When I talk to a doctor about a network of 3,000 metabolic pathways, often with common signaling communications molecules, and I like talking network theory, serotonin does 3 things in the body, at least, doctors can't do that, uh, have that conversation. They say, well, you have a rash, it's correlated with X or Y. Um, that to practice will go away to the science. Today it's not hard if you design modular the billing systems and the charging systems, and so innovation will happen outside of healthcare. You should be able to do a finger prick at home and get 50,000 biomarkers, to pick a number, 50,000 for $50, maybe for $10 or $20. And the biggest piece of that cost is the US Postal Service. No, I'm serious. But we don't set up systems for that. You know, my cell phone is relatively expensive for the compute power. This would cost $1 million not very long ago in compute power. Why? Because we focused on producing billions of these, not hundreds of these. Same is true in Madison. So if you have that much data, two things. First, no human can look at that data. And two, if they did, there's no simple correlation. So it's an AI problem, machine learning problem, at least it's a statistical problem. So reinventing medicine has to happen in the age of data. And humans can't be part of it because they'll make errors. I'm almost certain in 10, 20 years I'll get much better cardiac care through my cell phone in a village in India than I will get at Stanford. Why? Because the expert at Stanford will still be here 20 years from now, and he will want to opine. We've seen this on Wall Street. 20 years ago they said, "No humans can trade," till Renaissance Capital started and started the quant movement. Today, no human can compete, though there's people at Goldman Sachs who still want to compete and still pretend they can compete. That's what's going to happen in medicine. Highly quantified. What is knowledge will be repl— so there's many, many pieces of it. There's biomarkers and tests. We do 30 in a standard blood panel. Why? Because humans can only look at 30. If we did 30,000, it wouldn't be useful. It'd be confusing.

Imaging.

We do the wrong kind of imaging for what humans can look at. Medicine. You know, there's no cardiologist or oncologist who's kept up with the last 5,000 articles on oncology that were published last year, right, that may be relevant to you as a patient. Only machines can do these jobs. By the way, it was 2 years ago, or 2016, I wrote a paper called "20% Doctor Included." I was being polite on the 20%. But that was for the human element of care. You don't want a machine telling you, you have 30 or 60 days to live. But other than that element, most things will be done, most knowledge will be imparted from AI systems. In fact, I was very serious. The dean of the medical school at Harvard thought I was joking. When 4 years ago I said to him, you really should change your admissions criteria. You know, they select for this IQ, off-the-charts IQ. It goes usually with very poor EQ. I said, change your admissions criteria to much more match the USC film school. It is the right way to do admissions in hospitals if the human element of care is which is what humans will do. But he thought I was joking. I wasn't. Anyway.

Well, speaking of 20% of doctors, there's a whole 100% of a doctor. Where is Dr. Dave? Want to comment and question about the fraction of a doctor that will be—

I don't like tradition very much except in family.

So just my undergrad is in sociology and art history. So I've already fulfilled your criteria for the USC Film School. That's first of all. Second of all, the reason I'm here, when we were introducing ourselves at the first meeting, I said, the first time I said, I'm a family doctor, I have no business being here. And the second time I said, I'm here because I'm a family doctor, I recognize I'm gonna be replaced by AI within 5 years, and if you can't beat them, join them. So that was my motivation for wanting to be here. Is it okay to disagree with the smartest person in the world? I'm just asking if— I don't want to sound impertinent.

No one, including the smartest person, one knows everything.

Right.

And no one is always right. Yes.

So with respect, your position that symptom-based medicine is passé, I respectfully disagree with that. I think that for primary care medicine, which is what I do, and I totally get the fact that we don't need doctors anymore or that we will need a different kind of doctor, hence my being here, but the idea that there's a lot of science in primary care medicine is misguided. Very little of what we do has science. That's first of all. Second of all, I get the fact that there's a gazillion metabolic pathways and I get all that, right? One of the problems is, is that in medicine abnormal does not equate to pathological. So we see a lot of stuff that's not normal, but in fact isn't really a problem. So how we're gonna figure out which one of those results are good or bad or how it's gonna affect you, is it telling you you've got 5 minutes to live? You know, that's the old joke, right? I got bad news and worse news. The bad news, you got 24 hours to live. The worst news, I couldn't get a hold of you yesterday. I'm really only here for the stand-up. I'm not that smart.

Stand-up might be a good job for primary care doctors.

As it happens, Voltaire famously said that the doctor's job is to keep the patients laughing until nature takes its course. Nothing's really changed in that regard. But what I would say is, is that in terms of how AI is going to affect medicine, at least in the near term, and I totally get where it's going.

Sure.

10, 20 years from now, 30 years from now. It's a real challenge for me because I've got kids who are crazy smart, committed, dedicated, who want to be doctors. And they come to my office and they work with me. I don't bring up the subject of what a doctor is going to be 30 years from now. Because I really have no idea and I don't want to talk them out of it. But having said that, I know these three brilliant professors of economics, Professors Agarwal, Ganz, and Goldfarb, who just wrote this great book called Prediction Machines. You can— you won't go to a bookstore to get it, obviously. You'll get it from Amazon or wherever people get things. In much the same way, that's where your kids will go for medicine. They'll go online. They won't go to see anybody because That's not what they do. And they talked about AI as being prediction and judgment being the key to the future. The prediction part coming the AI and the judgment being the people that actually have access to that. So whether it's decision support or whatever it is. So we have a current problem right now in the province of Ontario, the election today notwithstanding. There are a million people currently in Ontario that don't have a doctor. Okay, there's 100,000 people moving to Toronto every year. There's supposed to be one doctor for every— between 1,000 and 1,500 people. I'm a bit of an outlier, I've got, you know, probably twice that many, but regardless, the people who are entering medical school this September will be at a minimum of 6 years and at a max— and at least— and possibly as many as 12 years away from actually being doctors. How are we gonna look after those people in the short term? And the answer to my mind is, when you have Alexa or whoever your person of choice is that you call on when you wanna know something, in combination with, who's now been ramped up with AI. I actually did a little, I went back and looked at sort of a quasi-typical week that I had this month. I saw about 140 people that week. Plus or minus.

Now, Dr. Dave, your 7 minutes is almost up.

Sorry, I'll go fast. So, um, I thought I had the whole— oh no, okay. Um, so there were about 35, depending if you're a lumper or a splitter, 35 or 40 different diagnoses. I promise you that there is currently a guideline for every single one of those diseases. I've seen two things— I've seen a couple things today that are going to put me out of business. We've got the triage people who now can take a picture and in real time tell me what it is. The electric— forgive me, I don't remember the name, but the stethoscope that can hear heart murmurs in real time. If you take that technology and give it to somebody with a year's training or 2 years' training, all of a sudden— and you take the 2,500 kids of the 3,000 applicants that didn't get into medical school and offer them an alternative path. We could take care of that shortage tomorrow, and that to me is the interesting part and the, the imminent solution to the problem we have right now.

Yeah, so a couple of comments. First, not everything's rational. I'll give you the example I always cite: the placebo effect is completely irrational at one level. If you go further into brain function and immune function regulation, by the brain, uh, and mood and depression and all those other things, it becomes science again. Um, the second point I'd make is you're right in one of the problems today. If you have osteoporosis and you have diabetes, and the average Medicare patient in the US has 7 conditions, those 7 specialists don't know about each other. So I would argue that the only place where integrative medicine can happen is at primary care. Person knows you have osteoporosis, and person who knows you have cardiac disease and diabetes. By the way, you add oncology to it, you end up with a whole bunch of non-chronic complications that might happen that no single doctor would ever detect. A friend of mine who's one of the best experts in decision— medical decision support His mother had 3 complications, right? An embolism in the leg after a cancer, cardiac event much later. She saw one person for the embolism, a different person for the cardiac. They did not detect till too late that the cancer was progressing much faster, 10 times faster than anybody had expected. So what was— he moved locations to be with her expecting 5 to 10 years, and it was 5 to 7 months. So primary care is integrative. Having said that, with your 170 visits, if you practice for 1,000 weeks, it's 170,000 patients you'll see in your lifetime. A good AI system can take 170 million or a billion patient cases and learn from that. So you will never be as good as that AI system. And by the way, there's two levels of primary care. Today's primary care— and my son has a startup, full disclosure, doing a primary care doctor AI. That's his startup. And I'll come back to the shortage problem in just a minute. Anybody with AI background who wants to work on that, email me. Also feel free to email me on anything else. My email is the best way, vk@khoslaventures.com. But the initial part of medicine will be primary care medicine. Captured in an AI. There have been poor attempts at it, but data is getting so interesting. You can reverse engineer medical records into coming up with decision support rules. Decision support rules exist in various systems that have been done. Lots and lots of ways to build a primary care doctor the way primary care is practiced today. 10 years later, we'll have enough data to reinvent primary care, which won't be symptom-related. The fundamental problem with symptom-related prob— problems is you go to a car and see, uh, the, the front headlights missing and say, oh, the front headlights missing, you must be sick. That's how medicine's practiced, opposed to saying, what are the causal things that happened a year, 5 years, 10 years before. And most of the really important medical cases— diabetes, the expensive medicine, diabetes, cardiac disease, Alzheimer's disease— things happen 30 years before, when you could do something about it. And that's when you want to start acting, not when you have the visible car crash. Oh, you got Alzheimer's now, you can't remember anything. That's way too late. And we know things can slow progression. So that's why molecular medicine is what's going to be important, and that's why data science is going to be important, and that's why humans can't play a role. So the next 10 years will be a certain type of primary care, which is a replication of current primary care with AI. After that, it'll be all different. And the notion of symptoms will go away. Except in acute care cases. If you broke your leg, you broke your leg, and that's a symptom. So trauma is a whole different field. But anyway, as to the shortage question which you raised, that's probably where we might agree. That the first generation of primary care systems will actually be what I call upskilling systems. They'll, they'll let a healthcare worker in Uganda— so I have a friend called Raj Punjabi who trains hundreds and hundreds of healthcare workers to go out into villages, but he arms them with decision support for 30 conditions only, because that's the most a high school graduate can handle, and that's all he gets access to. And he'd love to go from 30 to 300 conditions. A nurse practitioner can do most of a physician's work if you equip her with AI assist. So bionic assist and upskilling of physicians will matter. A primary care doctor will be able to do most of what a cardiologist can, and at the same time do most of what an endocrinologist can. So the skills shortage will be addressed with bionic assist for upskilling people into doing more than they currently can do. And there, most doctors feel pretty good about that.

Except for—

But it's on path to a much larger goal.

The doctors feel good except the ones who are at the top of the food chain that are the super specialists. They get out by now.

And it is a shame primary care doctors make so much less money than the oncologists at the top of the food chain. It should be the reverse. The primary care job is much, much harder. By the way, nurses do an even more amazing job.

All right, sit back down, Dr. Dave. OK, let's move on to cities. Creative Destruction Lab is launching CDL Cities this year. You've had a very expansive— again, I refer people to Vinod Khosla's Medium post where he covers a number of these topics including this one. So transformation to cities and you had many reasons and ways in which cities will be transformed. I'm just going to begin by focusing on one of them, which is a substantive drop in the cost of commuting. So from getting from point A to B and thinking about what— how cities are transformed when the cost of that falls to almost zero. And so I put some quotes up on the slides for those that are just interested to see how he arrives at that. And then I'm also going to call on— we also have a world expert in this field, here is— where is Professor Strange? Will Strange? Right in the back. Okay, so I'll call on Will in a moment, but if you can just give some of the key ideas here in terms of how you think that a drop in the cost of commuting changes the layout of cities. And let's also move into your investment philosophy. So given that, what are the types of investments you're making now as a result of your vision of the future?

So let's look at what's sort of wrong with transportation today. We're producing tens of millions of cars, and there's a lot of people who still don't have cars. A car's doing the following. On average, carrying a 200-pound person— 150 pounds, whatever— and 2,000 pounds of steel. Right? It only needs 2,000 pounds of steel because humans crash cars. If you could avoid crashes or make crashes mostly inconsequential, you can get to a rickshaw level where you're carrying a 200-pound person and 200 pounds of steel.

Of steel.

AI, if it makes cars crash-proof, would completely change the equation, completely change the planet's mining industry for the amount of steel we need, especially if you also do that in buildings, which I think is possible. I already talked about that.

If you—

then you have this notion of you have 2,000 pounds of steel., and you move it around for 1,000 miles a month when you could do 10,000 miles a month if it was shared transportation, which is where Uber comes in. Now you look at Uber. If you buy an Uber car as an Uber driver, you're paying $300 a month in leasing the car, it's costing about— if you're going to make a reasonable wage, even $10, $15 an hour, it's $300 a shift for the Uber driver. Because if he does 20 shifts a month, that's $6,000. Take out some expenses, he gets $50,000 left over. Pure economics. But if you even use the car all the 3 shifts a day or 2 shifts a day, you, you got 40 or 60 shifts at $300 an hour. The cost of the car is inconsequential to the cost of driver. If AI replaces the driver, the cost of an Uber drops by 80% or 90%, especially when you make the trade-off for low capex, uh, high capex and low opex instead of what cars are today, low capex, high opex. This is why electric cars cost more. Uh, all this boils down to, and I've done the math in my paper, you could rethink cities by saying 80% of the roads can't be used by regular cars. You could dramatically improve the capacity of a road. Each mile of road would carry so many cars per hour if they were all self-driving. It costs a fraction of the cost. And I've done the math. City transit wouldn't be scheduled from point A to point B at certain times of the day. It'd be anywhere to anywhere, on demand, off your cell phone, at under $1 a ride. That's what city transport services should be. Uber would go away. Volkswagen would go away. That whole chain would change dramatically if you could achieve this. I do think there's a practical path on how this will happen. Somebody will go and say, I'll do a— a number of people, including Google, are doing taxi service. But I'm suggesting we go one step further. Cities do this and say, this is the municipal transit service. Right? Then it becomes dirt cheap, but more importantly, much faster. The city could say, as long as it's 200 pounds, it can go in the bike lane without even disrupting traffic today. When bike lanes are full, they can add one more lane to it. So, or say, in San Francisco they've already made it where you can't pick up, Uber can't pick up on Market Street. You could take 3 streets in San Francisco, there's probably 1,000 miles of road in San Francisco, you could take 100 miles of that and say these arteries, Market Street, Embarcadero, Van Ness, only driverless cars. Others can cross these streets, but can't drive on them. You'd so change the city, make it so effective to make it public transit, that it completely reinvent everything from Goodyear rubber tires to steel industries and mining, the whole supply chain, because every molecule every atom would go 10 times further and it costs 1/10 as much and most municipal cities would make money, not lose money by having much, much better public transit so you wouldn't need cars. You might still need Maseratis to have fun on weekends if you can afford one or hack your Honda Civic for weekends, but it's not What kind of transportation you need it for, you need it as a hobby.

So, okay, Will?

Thank you very much for the nice comments. I actually think you've understated the potential benefits of AI and so forth on cities.

And I always have this problem of understating because people think I'm too radical. So I try and, that's why I say 20% doctor included, I really mean 2% doctor included.

I mean, the thing that I saw that you did not talk about is one of the crucial inputs in taking a car to work is your brain. You know, that you have to pay attention. Well, most Toronto drivers pay attention, maybe not all of them. And if you have a smart car system, then you don't have to be paying attention to the road, and the opportunity cost of travel, which to an order of magnitude is about half of the travel costs that people incur, might not be incurred anymore.

By the way, I principally take Uber everywhere so I can do email in the backseat.

Now, the follow-up question that I would ask you is the following. You've described a process that allows big cities to get big at lower cost. And, you know, Toronto, we all pay huge amounts for our housing because a lot of people want to live here and because of scarcity. Scare housing that's close to the places where people work. Now the process that you give says a city can be bigger. It is unlikely that all cities will be bigger. What do you see about which cities are going to win in this process and which cities will be vacant? Or do you see other processes causing cities themselves to be less important to production and to consumption?

So let me make a couple of comments. This is a very, very complex area. And I'll recommend some reading before I finish. First, I should say, notion of space. This building, this room is not occupied most of the time. Right? It's a shame. Our utilization of buildings is really— none of you have given up your home today. Sitting empty. Maybe your pet, your dog's using it, but other than that— so I think with robotic kitchens, you may need eliminate the need for restaurants. With Airbnb, you're eliminating the need for hotels. It's reusing space. With WeWork, it's a very similar kind of model. So there's the business model changes. But even more exciting, we have a startup in Boston that turns a 400-square-foot studio apartment into a 1,000-square-foot studio apartment because you don't need 3 feet in front of your closet other than for 60 seconds in a day. So the robotic furniture is like those Transformer toys we used as kids. It actually changes. From a closet to a desk to a bed to— it's making space more efficient. So there's fundamental economies. When you go to the other side— and I could give you a dozen more examples like that of how cities can be made much more efficient. Look, when I was talking about driving 10,000 miles a month instead of 1,000, I really mean reusing streets with much more capacity and throughput per mile. Right? That's what I was talking about. And I haven't even talked about delivery robots and all that other cool stuff that I cover in my paper. But you ask a really, really important question of what size is optimal for cities. And there's two very different ideas. One is from a sustainability perspective, density helps. So more density will improve. The other idea I love, yeah, is much more important for humanity. Ideas build on each other. If you have enough variables together in a soup, think of it as a city. Almost every idea becomes autocatalyzed or catalyzed by some other thing in the city. This was all work by an economist, Brian Arthur, at the Santa Fe Institute. And one of the best books to define the, the physics of cities is a new book, probably the most important book I've read in the last 10 years, called Scale by Geoffrey West. Jeffrey West is a physicist at the Santa Fe Institute studying the physics of cities. I highly recommend to say, do large-size cities explode in terms of generation of new ideas, new value creation, benefits for humanity? So without going into all the details, let me recommend those of you interested in that topic, Scale by Jeffrey West is Highly recommended. He goes from why trees only go to a certain height, the number of blood vessels you have in your body of each size. There's physics to all that. And it also applies to cities, the physics of cities. It actually ties back in many of my ideas around medicine, which we talked about earlier. Came from complex systems theory at the Santa Fe Institute when I took a sabbatical. I just became a postdoc there. It was very cool. But these ideas are all related in biology, whether in a tree or in the human body, or the communication network that the internet is, or a city. Is all very, very similar math. And network theory is behind this, not simple medical pathways like this protein causes this problem. But let me leave it there.

Okay.

Sorry, I'm giving long, long answers.

No, this is— I think people appreciate the detail.

Yeah, that was wonderful. Thank you very much.

Okay, Barney, you had a question?

Oh yeah, I just had a quick question. I'm really enjoying— it's great to see Vinod. I'm really enjoying the talk. Technically on this one, so I get that car sharing will enable us to have fewer cars, okay, more people sharing the cars. That doesn't actually change the amount of people on the road. So unless you're actually having carpooling, more people per car, or you're having smaller cars where it's sort of one car is the entire footprint of a person like a bicycle, that doesn't actually change the density and hence the congestion. So can you just talk about that?

Well, a couple of things. First, I only peripherally referred to robotic delivery. The number of people going to pick up food or pick up groceries or pick up other things will decline. Uh, we can come back to this notion of community spaces because I think it's a very important notion of community spaces will be much more important in cities than they are today. We see them in unusual places like yoga studios. But community will be an important function of space. But we'll have a lot of delivery. Telemedicine will eliminate the need to go to a physician. You'll have grocery delivery, food delivery, all that. But throughput in streets will go up dramatically. Per mile of road, you'll transport many more people with much less steel.

Um, sorry, the steel, the steel doesn't matter. If the steel— if you use carbon fiber and these cars had almost no steel, but you're still having one passenger per sort of 20 square foot or 10 square foot of a car or something, doesn't change that.

Except if they can go and synchronize among them, your throughput per mile per hour would go up, and that increases capacity.

That seems like it's more of a linear effect, like that'll go up 10%, 20%, but that's not solving the problem of congestion.

I think it can be more than linear. I do think it can be more than linear, but it's not exponential and forever. Whether we need that or not, what are the social dynamics and how local is local, is a harder question. Some of the people, Sandy Pentland at MIT Media Labs has done a lot of work on social physics. I'd recommend his book on the topic. Look, one thing I would say, when enough variables change, the world isn't predictable. In what ways? Robotic delivery changes when you need to go out, how far you need to go will change dramatically. One of the things Geoffrey West says in his book is if you do the parameters and models of cities, one, they are unusually constant, whether it's a city of 100,000 or 10 million. The parameters don't change. Commute times don't change, though distances change dramatically. So if you could change commute times, then— like flying cars, you know, point-to-point helicopter service. That's exponentially more space, because you add 2 more dimensions. So— Yeah.

That's actually—

What I want to say is it's very hard to predict anything. You can speculate about various scenarios, and they're mostly historically, when you look back after the fact retrospectively, they tend to be wrong.

I mean, I just want to say I agree with all of that. I think my concern with this analysis is that it actually, because it doesn't talk about the passenger density issue except, So if behavior changes and people are traveling less because they need to travel, then obviously we have wins. But simply replacing less steel, higher utilization of the same vehicle, that actually misses the— it undervalues mass transport. It's like, hey, get rid of these sort of large passenger-density systems without actually giving a high-density system in replacement. And that can lead to actually the wrong sort of policy decision-making in cities. The other point I would make— and so I think flying cars is probably is probably a solution to congestion. And we're farther away from that in terms of technology. So it's interesting. The other concern I would have is almost the opposite problem. I think that with self-driving cars coming, we're going to have longer commutes. Because you'll go out to the lower-cost real estate. We'll have longer commutes. No problem camping in my car for a 1-hour drive. Everyone will be happy to commute to save property costs. And we have a complete disaster heading for us in terms of cities. So we're gonna have to get to rethink that and rethink it soon because it's gonna get way worse before it gets better.

Um, it's possible. I'm more optimistic. I shouldn't be given the situation with housing and transportation in San Francisco. It may be the spectacular example of economic development causing other kinds of problems, including a real lack of diversity in the population pool.

Okay, we're almost out of time. I'm gonna take two more. I'm gonna take both questions at the same time, and then you can react to them however you like. So first Ron, and then Haig.

So there are cities and there are cities. In other words, cities you like being in and cities you don't. And one of the joys of a city Toronto, for example, is walking around and discovering things, where stuff isn't all organized, and where you, where just, you're not going anywhere specifically, or you, by walking, you discover new things every day. My problem with your analysis, with respect, is that that kind of social side doesn't think about efficiency. I mean, just efficiency or arguments on efficiency miss a big part of what makes a city really nice to live in. So I think that might actually affect any predictions that you might make that are based purely on efficiency.

OK. And a comment from Hagen, and you can take it to the next question.

It's not a comment.

I would just be curious to know how you see the impact of quantum computing on the world in the future?

I'm sorry, the impact of quantum computing on the world of the future? Two very different questions. I'll take a minute on each. I only obliquely referred to this notion of community, and I think the number of community spaces will increase. Again, it's speculation. I can't say I'm right. People like interacting with other people, and I think this notion of discovery is important. I would also add that a 50-year-old brain is very different than a 20-year-old brain. And depending on when you went through the 10 to 30 age group, which era, your programming is dramatically different. I know 12-year-old kids who in the same room talk to their friends over text messaging. You don't do that. I can say this because I'm much older. So how this evolves is hard to say. I do think community will be much, much more important, and we will have much more room for community spaces and community interaction. Jeffrey West talks a little bit about this old notion of city design, of putting in parks, and why that is the wrong way to do things. And why evolution of cities is important, more important than design of cities. But I won't spend more time there, but Jeffrey West covers some of those topics. And there's a couple other very, very good books on the city planner's view of the world, which is historic and extrapolation of past and why it's wrong for today's evolutionary environment. Rapidly evolving environment. On quantum computing, I think to me quantum computing is a really important tool, but we have to realize it's only a tool. It's what we use it for. If we get much better drugs, if anytime somebody has a patented drug and you can do 10 equivalents, the cost of drugs won't go to $10,000 biologic dose.— right, Rituximab is like $10,000 a dose. It'll be $10 a dose because there's 100 alternatives. So quantum computing will be a powerful tool for us to do various things with if we do the right things well. We could use it to design drugs. We could use it to build better nuclear bombs. It's up to us whether to use a tool well or poorly. So let me stop there.

OK, let me— I'm just going to conclude with one last question, which is you've made a number of investments now in companies based in Canada, and particularly in Toronto. And so if you can just wrap up with your changing views, say, how you view Canada today relative to how you viewed this as an investor a decade ago.

Yeah. So here's a couple of comments. People talk about Silicon Valley, but to me Silicon Valley isn't a place, it's a state of mind. Right? It's a culture. When somebody drop out, I just met with somebody who dropped out of a PhD program because they had this vision for creating something new. That's a cultural thing. Growing up in New Delhi, responsible parents would never let their kids drop out of school. So it's a cultural thing. You allow, you have permission to do radical things, operate at the fringes outside the system. Now most innovation happens at the edges of the system. In fact, I can't think of one large innovation to come out of any large institution ever, or at least ever in my life, which is last 30, 40 years of productive work. That's never happened in any area that I could think of. The most important one I could come up with, Bank of America doing credit cards, if you consider that innovative. They went from credit— offering credit to putting credit on plastic. That was the largest institutional thing I could come up with in the early '70s. So this permission culture combined with a knowledge culture, and I love the number of people who serious academic study here in Toronto especially. That with investors who are willing to fund speculative ideas as opposed to run spreadsheets. That nexus of things is very, very important for society to move forward, and that exists here in Toronto. You see a lot of AI PhDs, a lot of material science PhDs. You see investors willing to fund those ideas and people willing to take risks. I think for society as a whole, the biggest risk society can take is to not take risks. And risk-taking comes from diversity, a cultural mix, a skills mix. It's amazing how many ways diversity is important other than saying we value diversity in itself. Because it catalyzes the soup of activities that are possible. This is where I go back to Brian Arthur's work in his sort of complex systems theory. You're seeing that in Toronto because you have this education, you have a big influx, as somebody said, 100,000 residents a year. That's adding to the talent pool, to the diversity, to the skill set here. A number of people I've run into came here for education and then stayed. Education combined with— which is knowledge combined with culture and financing and risk-taking. Toronto seems like a great place for it.

Well, let me just say on behalf of the group that's here, this is a reasonably diverse group. This room is entrepreneurs, investors, there's a lot of scientists in the room from coast to coast across Canada, and also many visiting from other countries. And so I know you flew in last night, landed just after midnight, and you're flying out tomorrow. You came here to join us today. We really appreciate it. Thank you very much.

Thank you.

And—

thank you.

I know I can't meet too many people today, but my email is the best way to get hold of me, vk@khoslaventures.com. Thanks.