20VC: Vinod Khosla on How AI Impacts The Future of Healthcare, Education, Income Equality, Geo-Politics, Music and Climate Change

0:00If I'm right, the level of job displacement will be large. AI is a tool for productivity growth. Capitalism does result in more inequality, almost always. AI will only accentuate that. One has to be very considerate of the people being disrupted and accommodate them if you're going to make better progress. This is 20VC with me Harry Stappings and I'm so excited to welcome one of the OGs of Silicon and Valley to the Horteseeds Day, Vinod Kostler. Today we have a discussion on the future, the future of healthcare, education, climate, energy and how AI impacts it all. For those that do not know, Vinod is the founder of Kostler Ventures, one of the leading venture firms of the last decade.

0:42With investments in open AI, Stripe, DoorDash, Commonwealth Fusion systems and many more. And prior to founding Kostler, Vinod was a co -founder of Daisy Systems and founding CEO of Sun Microsystems. But before we dive into the show's day, this episode is brought to you by TIGAS, the go -to research destination for bold investing. TIGAS curates expert in -size analysis and financial data to give you powerful perspective for your investment decisions, with lining fast access to over 60 ,000 transcripts across 20 ,000 companies. You'll discover a wealth of unique insights to fuel your fundamental research, gain perspectives, synthesize information, and even model outcomes all on TIGAS.

1:23As a 20 -minute VC listener, try for free at tegas .com forward slash growth and spare you tools we can't live without like tegas there. May fair helps companies protect and grow their cash automatically. I access up to $50 million of FDIC insurance and earn up to 4 .72 % interest on your deposits from a single account. After the failure of SVB and First Republic, it's clearly the standard FDIC limit of $250 ,000 is simply not enough for VC bad startups. So for founders, that means finding a safe home for cash is now a priority and a major burden. Mayfair can take care of that few in under 10 minutes, through his partner banks and sweet providers, Mayfair automatically spreads customer deposits across a network of over 200 banks, therefore multiplying the level of FDIC insurance coverage and limiting the risk of any single point of failure.

2:13Think for it like a high yield savings account that actually pays a high yield and is actually safe. Go to GetMafet .com forward slash 20VC to get started today. And finally, Secureframe is the leading all -in -one platform for automated security and privacy compliance. Secureframe simplifies and streamlines the process of getting and staying compliant to the most rigorous global privacy and security standards, like SOC 2, KIPER, GDPR and many more. Secureframe's industry leading compliance automation platform paired with their in -house compliance sense experts and former auditors helps you get all it ready in weeks, not months, so you can close more deals faster and secure frame users over 150 integrations.

2:55So you have built in security training, vendor and risk management and more to make compliance uncomplicated. Secure frame makes it fast and easy to achieve and maintain compliance so you can focus on serving your customers. Automate your security and privacy compliance with secure frame and you can schedule a demo today at Secureframe .com. You are now arrived at your destination. Vinod, I am so excited for this. I've been absolutely loving your Twitter content recently, and so first off, thank you so much for joining me again today. Happy to do it. Always like talking to our notes. That is so kind of you, but I want to start a stay on the kind of topic of the day, which is obviously AI artificial intelligence and coastal and you have been one of the biggest proponents and conversationalists of the space in the last 10 years really.

3:42And I just wanted to start with the question of do you believe the recent hype in AI is due to technological breakthroughs in innovation or is it the result of hype cycles keeping up with innovation from years ago? A little bit of both. Let me keep your perspective. In 1982 at Sun we adapted TCPIP and the internet as our native network. We are the only ones. By 1996, it was clear that exponential growth was going to continue and we invested in Juniper, which by the way was a 2500X return, a $7 billion return for us on a three or four million dollar investment. When the internet really took off and I remember meeting Mark Andreessen to invest in Netscape, the browser company on a Saturday morning in Mountain View when it was three people.

4:33It really took off, but the exponential was clear much earlier. The reason I bring this up is I wrote my first blog on AI, Do We Need Doctors and Do We Need Teachers, encapsulating the idea that expertise could be free in 2011 -2012. It was pretty controversial then, but it was a similar moment to when we adapted the Internet and TCP IP in the 80s at Sun. And then when Chatschi BD came out, it became like the browser moment for the internet and it's now obvious to people the Extreme capability that we developed here and will continue to develop and is progressing at a rate where even I can't keep up But anytime there's a rich area for development bubbles fall and I've already a little bit that that will happen here There's obviously real technological progress, real expansion of the opportunity set for real building businesses, but also a lot of hype that follows sort of a rich area to mind, sort of the gold rush phenomena.

5:41What I'm fundamentally optimistic about is when you look at bubbles. In the 1800s, 1830s, there was a bubble in England when you got the right to build a railroad between two towns. you could issue scripts almost immediately. The bubble did burst, but more railroad was built after the bubble burst than before. My point is the progress of AI applications in utility for society will go on a steady pace based on technical capability and entrepreneurial interest in that, not what Wall Street or investors think about it. So I think investors are important because they fund these developments, but investors cause bubbles.

6:24The reality is the underlying technology in its applications. You mentioned that OpenAI and ChatGPC being the equivalent of the browser in terms of the consumerisation and the access that it provides consumers. You tweeted before that daylight between what we'll be able to do with AI and what we'll actually do with AI is human resistance. What did you mean by this human resistance, Vinod? And how does it show itself? Well, there's two types of things. You know, first a genuine concern that AI will display jobs, and what do we do with that? If I'm right, the level of job displacement over 20 -25 years will be large.

7:01And we dealt with that kind of job displacement before, but over multiple generations, for example, in agriculture, and it was 50 % of all jobs in the US in 1900 and went to 4 % by 1970. If the change is happening rapidly, then job displacement may cause us to have rules and regulations that slow down progress. Slow down the adoption of these technologies. Another example could be physicians resisting AI being able to do physician types of tasks. The AMA is a very good organization, but there is protection built in in certain areas. whether it's in unions or physicians or pick your favorite expertise, media where artists or music producers may object to some of these.

7:54That's what I mean by human resistance because people are being disrupted. You know, honestly, technology is very fun to talk about disruption. It's not a lot of fun to be disrupted. And so one has to be very considerate of the people being disrupted and accommodate them if you're going to make better progress. So I do think we'll have to pay special attention to the people being disrupted and their needs and desires and wants. I want to approach this from a very first principle's interviewer style and ask very basic questions. I hope that's okay. What should we be doing do you think for those that will be disrupted?

8:30As he said, it's fun for us at the forefront of innovation. If you're being disrupted less so, what should we be doing and how should it be handled? So let's start with the math. Let's say the next 50 years of GDP growth without AI and some of the newer technologies would be 2 % a year, per capita income may go from 70 or 75 wherever it is today or 65, I forget the exact number to about 175 ,000. If we change GDP growth to 4%, that per capita income would increase to 475 ,000 or so. So my point is we will have adequate resources, adequate abundance to share much more broadly. I wrote a blog in 2016 in Fortune magazine that said AI will cause, and this was seven years ago, AI will cause great abundance, great GDP growth, great productivity growth, both everything economists measure and increasing income disparity.

9:31It was the first time in 2016 I wrote about needing some sort of universal basic income, which means redistribution of income in a much better way than we have today. The reason I gave you my calculations of the effect of changing DDB growth from 2 % to 4 % over 50 years is we will have enough to share. So my suggestion would be some large percentage of poor capital income growth or GDP growth be reserved in a social security pool for our university basic income or something like that. There's many ways to redistribute some more acceptable than others, but happiness in our society will be a function that equality not just GDP growth.

10:14And I think we'll have to consider that and then I will make that abundance possible. I'm optimistic it can be done. Bill Gates has talked about a robot tax. There are multiple ways to approach this, but I think reducing the rate of increase and putting the excess into a pool for like a social fund would be the right approach. Because it increments are much easier to share than somebody saying, let's increase taxes on everybody. I'm fine with tax increases personally. We can afford to have better tax system or equitable tax system but there are maybe better ways to do this and increase sharing of the benefits of these technologies to other mechanisms.

11:00G -Thing AI does more to harm or to help income inequality. Well AI is a tool for productivity growth. Capitalism does result in more inequality, almost always. Everybody's better off, but the income inequality increases. That tends to be what history has shown us. AI will only accentuate that. And so sharing the benefits of AI in a different way will be important. I think it's not by constraining AI or not using AI to benefit society. It's by taking the product of these technologies and sharing it more equally. The increased abundance should be more equally shared. I'm fascinated, again, I said I loved your tweets, I really have loved them.

11:45One of them said, the choices, do you let AI do the jobs or not? Some will and some will be laggards. When you think back on that, who do you think will embrace it and who do you think will be the laggards? Well, I worry that the Western political system and capitalism is by permission of democracy. So the Western political system will pay more importance to the short term, which is things like job displacement. While in China, you might get Tiananmen square -type tactics to adapt it and win economically as a country as opposed to each person being fairly treated. That's my worry. I totally understand that as a concern.

12:29I think the hardest thing is political incentives, which is every politician is incentivized to remain in power and actually encouraging job displacement will not help them in the next election. It might help the company economically in 20 years, but the incentive mechanism of politics doesn't align to the GDP and per capita growth over the long period. Absolutely true and if the changes slower it'll be easier to accept than if it's super I'll give you an example. I'm pretty convinced instead of displacing physicians. We could quadriple the number of interactions between a physician and patients.

13:08In fact, it's already two in Australia. There's many more interactions with your primary care physician every year for the typical patient than we do in the U .S. So you could increase the level of service, increase the level of attention with AI taking excess of that. This projected to be under current, course and speed, shortage of primary care physicians in this country. I think we can have an excess, even when we increase the number of interactions by 5X, because AI can take 80 % of that increased load. And that's a good example of how one can work cooperatively to improve the level of service in healthcare and primary care while not disrupting physicians.

13:51Do you believe that AI will be a competitor or a complement to the doctor? And do you believe doctors see it in the same way? I wrote about it probably eight years ago. I wrote this paper 20 % doctor included it's a 100 -page document. And I wrote about the human element of care being provided by doctors in the detailed expertise being provided by an AI. You can't expect your own colleges to be caring, talking about your cancer, and have read the 5000 most recent articles on your cancer. They can't keep that knowledge in their head. So that's the kind of partnership that'll be possible. You know, you don't want a computer telling you you have cancer, you want a caring human being.

14:37I jokingly said to the dean of the Howard Medical School about 5 or 6 years ago, they are to change their selection criteria for more empathy, more mirror neurons and less IQ only, which is what it takes to get into Harvard Medical School. I said, take the model of the UCLA Film School. Look at caring humans as criteria number one for admission to the MD program. What other ways do you think AI will most significantly change healthcare and medicine in the next 10 years? Well, for example, we can custom design proteins for particular purposes. I'm pretty excited that for a large number of genetic defects, and there's probably 6 ,000 or so diseases that are single mutation genetic diseases, we will be able to design with AI's health, custom solutions.

15:28Because we've essentially solved the protein folding problem, we can fully characterize the design and behavior of DNA, RNA, proteins, that's all really, really exciting. So that'd be a huge impact on the pharmaceutical side, not the expertise side of healthcare. Almost every area, whether in healthcare or in media, will be exciting. Ghee, you mentioned before, a brilliant thing, which is I've never seen innovation come from large institutions. And it's really applicable across medicine, across education, across music. But do you think in this next wave of innovation due to AI that holds true?

16:08Or does AI actually allow incumbents or the dominant parties to just become stronger? I believe innovation will be spread broadly in different areas. Now large language models may or may not be constrained to the large platform companies. Right now it appears that'll be the case, there'll be two or three major providers of large language models. But for example, we have an effort in symbolic logic to see if in three or four years it could change the equation of how you train these systems and the amount of data required and all those kinds of things. We have an effort to develop in curriculum for AI instead of curriculum for humans.

16:48Almost everything AI is training on is designed for humans not for machines. So there's a number of large innovations and it's hard to tell today which ones will play out or be important in three or four years. So when we originally committed to opening eye for five years ago, it was very unclear where the world would lead and what the capabilities would be and when they would develop. It was clear to me that there was a reasonably high probability of pretty significant advancement at some point in time, hard to say whether it would be 2022 or 2025, but that didn't bother us. The shape of the curve and the exponential really interested us.

17:31I think there's other approaches to AI, you know, Josh Tenbaum, for example, working a lot on probabilistic programming, others working on formal techniques like symbolic logic. Do they add to replace complement LLMs? I think all those are possible and within LLMs large innovations are possible. So I would say the future is not predictable, but it is discoverable in what I'm really optimistic about is the best minds in the world are now working on this problem. That's more a precursor of progress than any single thing like, hey, we suddenly have a new transformer like paper. Our attention is all you need, the famous paper, it's more about the talent working on it that makes me excited.

18:14What do you think of the barriers that we need to overcome for AI to have the impact we all want it to have on healthcare? Well, in healthcare in other areas, AI is doing something really well. Other things not so well. That's where development needs to be rapid. For example, in healthcare specifically, guardrails is a very, very important issue. Safety is a very important issue. A lot more attention has to be paid to those factors. Explainability is a huge issue. Those things will need to be developed by somebody applying AI broadly to say, primary care on ecology or pqf area, but all of those I'm pretty convinced will happen in next five years and we'll see the capability developed to provide very, very low cost, much more frequent accessible care, an accessible 24 -7, and accessible at a very cheap price to every human being on the planet I think.

19:14Then we've spoken about healthcare. You've written able to apply education. You've written extensively about both many years ago. ahead of time. To what extent do you think about market timing when investing? Because, I think as Mark Andreessen's quote, there's no such thing as a bad idea, only a bad time. Do you think about market timing? You know, you think a little bit about it, but I mostly feel I'm not smart enough to predict it. In retrospect, you always feel dumb about some things that really smart about others. You know, we invested in OpenAI and we invested in Conveld Fusion about the same time.

19:47I'm really happy about both, but both, you know, I'd like longer term bets than they've turned out to be. And progress has been more rapid, both in AI and in fusion. Very different areas. Then I would have expected. So suddenly I feel smarter, maybe when the overconfident about my predictions. Today I sit here and say for example, the amount of compute power AI systems need, the expense of training. Good one have in -log computing and we are working on that. Dramatically change that equation by 100x, both cost and power. And I think that's entirely not only possible, but likely. Vents a little bit harder to predict though we are dipping out chips this year.

20:30I have to as you mentioned fusion there and you mentioned commonwealth. I thought it was fascinating when I thought about kind of the future of energy and when I was doing a lot of work prepping for this. Because you said before, forcing non -economic solutions is the wrong approach to climate, and solutions will need to be at what I call the gindier price. My bet is on fusion and geothermal. I just wanted to break this down because I didn't understand some of it if I'm honest, which is why is forcing non -economic solutions the wrong approach first? Well, energy is such a large scale problem that there's no way it will be traded off, you know, we will spend the money to scale it.

21:10It's great for the first one, two, three percent of the market whether it's the electricity market with solar and wind or the hydrogen market or picky of everything. I believe subsidies should be declining with scale so they can be high when it's one percent of the market because the tax on the overall energy costs would be low but they have to start scaling down and no subsidy is being designed that way. We are as it gets to two percent or four percent or 8 % or 15 % of our segment, whether it's hydrogen to replace natural gas and coal or solar or fusion to replace electricity from coal, it has to scale down.

21:50Because if it's not, it's going to become too expensive. You get clean energy and no healthcare spending? What are you going to straight off? Those are the kinds of practical things at scale that matter. Having said that, only subsidies help get technologies developed when they're subscales, but by the time they're 5 % of the market, they really shouldn't need subsidies at all. Also, practically, if you're a politician in India, you're not going to tell people they can't have clean water, but their kids will be better off because they'll be a less warm climate. It's a very hard political trade -off to make, and consumption's been increasing in some with the developing parts of the world.

22:34You can't defy the laws of economic gravity at scale. You can do it in small, start -up sort of situations, but not at scale. For it to be contributing positively to economic development in countries like India or China, it has to be competitive globally with other sources, or it won't be adapted. And this leads to a very important point in climate. We don't need incremental solutions. We need large breakthrough technologies, will make these new technologies comparative at scale. So we can take small risks and have incremental decline in costs, but that doesn't solve the problem because it won't be broadly adapted.

23:15Or we can take larger risks and if successful, really have solution, fusions that kind of thing. Deep geothermal, a 400 or 500 degree centigrade geothermal, would be that. Finding a new geothermal site incrementally doesn't really help. It's not scalable, it needs to be scalable by 100x more than exists today, at least. And so my interests are technologies that scale and are economic and don't defies the law of economic gravity when scaled. And what we are talking about it, these take a long time, so sometimes the venture community ignores them. but they're profitable. If I look across our portfolio, some of the investments we made more than 10 years ago, there's a whole number of them.

24:02QuantumScape, Werdagy, LandsATech, Impossible Foods, that are 5 to 10X multiples, though they took a longer time to get there. So climate 1 .0 was successful if one was patient and picked the right technologies. Things like fusion, new plant proteins, I'm very excited about new fertilizer production techniques, new steel production techniques, we are taking radically different approaches in cement, steel, those are really, really exciting breakthroughs and there's no question they will be cheaper than its fossil alternative. If the technologies work all already I can say the cement we are producing will be 70 % less carbon and our first plants will be operational this year and cheaper than traditional Portland cement and that's when things start to scale.

24:50Can I ask you on fusion? What is the core of the bullishness and how does fusion bankrupt entire industries and destroy massive companies like you said before? Well, there was one breakthrough that enabled fusion to happen, which was high temperature superconductors. It wasn't possible to build a fusion reactor with a 20 Tesla magnet, which is what common wealth fusion is doing. With that magnet, the size of the reactor scales to 1 over Tesla to the power of 4. So if you quadruple the power of the magnet, which one can do with high temperature supercomductors, the reactor itself becomes 1 to 50th the size.

25:32So 250 times smaller. Because of that, you can now hold the economic, but also experiment much more rapidly, build them much more rapidly. Convall Fusion was able to build a 20 Tesla magnet for Fusion and demonstrated it in 3 years, 3 .5 years. It is taking the large I -TAR project with smaller magnets, 25 -30 years to get to the same place, so 10x faster. I believe demonstrating positive energy by 25, maybe 26, will be also 10x faster. Not only has that happened, there's a dozen other credible projects in Fusion that have been started mostly because Convelled instigated this change towards Fusion and I have this theory you need Instigators and it's spocked up and feel that was slow and not attracting the best talent to attracting the best talent now And there's probably a dozen startups and that would make me optimistic At least one or two or three will be successful in this.

26:36How does fusion, how does geothermal, how does new energy resources and innovations within their extraction, how does that rearrange international relations do you think? Well, if we can reduce our dependence on oil, we can dramatically reduce international tensions in what are called strategic interests. There's so much fighting over strategic resources like energy, the minerals and metals is the other area. We can eliminate that tension. Imagine the following. If you take every corn plant in the country, and it's boiler, the corn boiler, or a natural gas boiler, is replaced every 5 to 10 years.

27:17If you could replace it with a fusion boiler, you wouldn't have to build. 5 ,000 new power plants in the US. You just have to build 5 ,000 new boilers, which is less than the number of Liberty ships we built in five years during the Second World War. That's how you achieve these at scale and completely change industries in our source of power by being clever about it. I had Larry Summers on the show not too long ago, and he said that China is a prison. Japan is an Earthing home in Europe is a museum. Do you agree with that statement and how do you think about that with regards to the discussion we just had on how energy changes international power flows?

27:57I think what Larry was probably talking about was demographics. Europe is an aging demographic. China and Japan are rapidly aging. Tinker if I read right by 2040, which is 15 years, 70 years, years from now, the average age in Japan average will be 54 or something like that. The world needs younger people to have a vibrant society and what is called the dependency ratio, a number of people working to support each person not working whether they're elders or kids. That dependency ratio in Japan, China, most of Europe is heading in the wrong direction. The US has a reasonable demographic. India has a great demographic.

28:40you know the mid -east and Africa have good demographics but this anti -immigrant sentiment will hurt these European countries from really renovating invigorating their societies. That's a complex problem. Why are you so bullish on India moving forward and will India develop into a tech giant do you think in the next five to ten years? What I love about India is the demographic is young with a huge focus on learning and education. Now, India's pretty significant problems, I think, leaving a population like the Muslim population behind in India would be a huge disaster and could completely change the optimistic projection I have for India if we leave a part of the population behind.

29:29And that becomes a source of wild lands riding things we've seen. But generally a well -educated population and a course set of technologies in English as a language, common language, Western values, democracy as a value, all those make India a pretty good candidate. Brazil's similar in many ways. So India's not the only country with that advantage. Do you think the US and the rest of the world will be increasingly reliant on Indian -born tech talent given what you just mentioned there in terms of the amount of young people on that talent. The great thing about technology is you need to develop it once and deploy it a hundred times.