Best of: The future of the innovation economy

0:00Hey everyone, it's your host Russ Altman from The Future of Everything. You know, about a year ago, our team was in New York where we taped a live episode of The Future of Everything. It was with computer scientist Fei-Fei Li and economist Susan Athey and Neale Mahoney. We talked about how AI is reshaping innovation, creativity, jobs, and public policy and the economy in general. So we thought we would play it again. You know, this episode won us a Webby Award, and so we're pretty proud of it. We hope you'll enjoy taking a listen. And before we get started, please remember to follow the show on whatever app you're listening to, because you never want to miss the future of anything.

0:45Okay, we're getting started about innovation economy, and it's useful to define what it means. Of course in preparation for the show I did what we would all do. I went to a large language model I went to three chat GPT Claude and perplexity just because I have those URLs Perplexity was my favorite it said The innovation economy refers to an economic system where growth is primarily driven by the generation application and commercialization of new ideas products and technologies rather than the more traditional heavy reliance on physical assets and manual labor And then they go on to say that applications examples might include ride sharing, house sharing, social media, and many others.

1:27Now although the innovation economy is not entirely about artificial intelligence or AI, AI is a very big part of it and a part where our panel is expert. Questions that arise for AI and the innovation economy are what will be the impact on jobs, on the quality of life, on the overall economy, and how do we navigate a future in a way that protects all humans but enables unbelievable beneficial innovations. Today my three guests will help us understand the opportunities and challenges. They are all professors at Stanford University. Fei-Fei Li is a professor of computer science. Neil Mahoney is a professor of economics and Susan Athey is a professor of business.

2:09This is the Innovation Economy Dream Team. Susan, let's get started. You have written that AI is a general purpose technology and so its impacts are likely to be broad and very difficult to predict. But you have looked in your work at other general purpose technologies that have emerged in the past. What did we learn from when these things emerge? So looking back from everything from electricity and industrialization to the personal computer, which a lot of us have lived through, they can have really profound impacts locally, but yet not show up in big changes in GDP growth for quite some time.

2:52And there are many reasons for this, but a big lesson from the past is that when you unlock one bottleneck, you find others. And to fully make use of a technology, you may need to restructure industries, but certainly restructure production in order to fully optimize for it. I take it that these bottlenecks can come up like in all different places and in all different aspects of the economy and the business. That's right. So of course, people, many economists this talk about AI in general as a general purpose technology. One that is less discussed but I think profound is that software development is a general purpose technology.

3:38In general, digitization is something that restructures industries and it helps small enterprises scale, it increases the span of control. It allows firms to grow without just having to directly supervise people. That kind of general purpose technology though has not been fully adopted. It's across the world. Many small businesses are less digitized. And it's generally been something with a lot of scale economies, especially with data and machine learning. And so the challenges have been that there's a lot of fixed costs to get going. If you're running a restaurant off of paper or maybe WhatsApp or WeChat, there's another leap.

4:21you have to be big enough to justify those investments. And so one of the things you might ask today is, all right, what would happen if we remove the bottleneck that software engineering was? If you talk to a bank in Wisconsin or even in Brazil or Argentina and they say, well, are we doing enough with machine learning? And you would find out that, well, they have a lot of trouble hiring people. They have a lot of trouble being able to adopt and And I might recommend five, 10 years ago that really you may not get that much value quickly from making big investments because it takes so long to get going.

5:00So there were many bottlenecks around just getting data organized in the past. One of the things that when you look out in companies, they're using clunky software, if they're even using it at all, because it's so hard to change that you change rarely and you put a lot of effort into it when you do. So the bottlenecks have been around adoption, figuring out what you want given that you're going to be stuck with it, and just the big cost. So now we have to ask if software development is actually very cheap now, what stops us? If we could already build today say a nursing assistant that could be very effective, especially in a country where nurses aren't well trained and there's not a lot of education, why do we think that it's not actually going to be fully adopted tomorrow?

5:47and you can think of many reasons actually that it won't be fully adopted tomorrow you have to decide what to do you have to get the nurses on board you would have to train them it's hard to adopt something new and so you wouldn't just pick the first thing up and so yet I don't predict that next year there will be nursing assistants around the world great thank you so on this issue of bottlenecks and jobs Neil you have taken a long view in some of your work on the impact of AI on jobs um how do you think about either predicting or measuring the impact of ai on the labor market in some of the examples that susan just said yeah so it's a great question um you know i sort of joke that the two questions i get the most as a economist in silicon valley is what jobs is ai going to disrupt and what do we do about it i tell them i have no clue on what jobs are going to be disrupted, but I know exactly what to do about it.

6:39Boom. So, boom, what should we do about it? I think we're facing sort of a veil of ignorance moment. So, you know, connecting to the humanities. I think many of you remember Rawls' thought experiment of the veil of ignorance. If we can step back from our current lives and think about before we have our endowments, our skills, what do we want in society? What protections and safeguards do we want in society? And I think that's where we are with AI, that we know that some of us may lose a job or may have human capital, which is less valuable on the labor market. We don't know who it's going to be.

7:21And I think what Rawls tells us is now is the right time to invest in a social safe net. I'll give you one example and then I'll shut up. We live in a country where if you lose your job, you likely lose your health insurance. you know maybe I'm wearing my political hat I think that's crazy now it is going to be hugely crazy when five ten percent of us lose our primary occupation maybe because of AI so so thinking about how we're in a veil of ignorance moment and how we can invest in a social safety net to protect us against I think the disruption we will face is a hugely important endeavor thank you and I want to come back to safety nets a little bit later.

8:07Fei-Fei, you're a technologist who helped shepherd in the AI revolution. You're working currently in the area of visual intelligence and spatial intelligence far beyond what the current things like ChatGPT and all the others do. But these are also areas where living organisms on Earth like humans have been excellent. We've evolved to be excellent at space. We've been evolved to be excellent at seeing things. How do you think the kind of systems you and others are building will interact with the humans who are using them? Or how do you hope that will happen? Yeah, thanks for the question. So, first of all, I do agree right now we're at the dawn of an AI economy.

8:48And I think whether it's the large language models or the next chapters, which are the spatial intelligence models, the embodied AI models, we're going to see more and more innovation in this technology going beyond what we're seeing now. One thing that, also living in Silicon Valley, also getting the same questions that Neo does, that puzzles me is that when people talk about AI, they go straight to the word replacement, replacing humans. And this is just, we got to be a little careful because AI will change jobs. AI will change different tasks within a complex job. For example, being a nurse in a nursing home, there are hundreds of daily tasks a nurse does.

9:39AI will help some, but will not replace the job wholesalely. What is really important is to recognize instead of replacing, AI really augments. And that's what I really believe is that this is a horizontal technology that can superpower humans, superpower our workflow, augment so many capabilities that humans are good at. We might be short of labor or we might not be so good at particular part of that capability but with AI it can help us. I want to just take one example because I'm in the area of visual spatial intelligence we work with creators you know visual creators, storytellers, movie makers, filmmakers, game developers, all of these creators are facing this new era of AI tools.

10:41And it's so incredible for me to meet and work with them because they see AI as a tool to augment them. They see AI as a way to supercharge their creativity and their productivity. So I'm not trying to say that there's not a double-edged sword, we should talk about that. But I do think it's so important we recognize that this technology can do a lot more to supercharge superpower people putting people in the center instead of go straight to the word replacement thank you can i jump in on this yes

11:22one i agree and two i think a useful framework is you know if you think about the questions in innovation policy over the 20th century early part of the 21st century they're about how do do we maximize innovation given budgetary constraints, human capital constraints? But moving forward, I think the most interesting, richest questions will be less about just sort of how do we put our foot on the gas, but how do we shape innovation to be complementary to our skills? And that's a STEM question, but it's also a humanities question, because it requires tapping into what are the activities that give us meaning and purpose.

12:07And so at Stanford campus, where we have that intersection of the humanities and STEM I think is a great place to be working on that hugely important question. It's the human centered AI question. Yeah, somebody named their institute very well.

12:24Yeah, and I think we have, You know, Fefe and I worked together, Win and Russ, with the founding of the Stanford Institute for Human-Centered AI and John Levin as well. And one of our theses was exactly developing this idea, what is the role of a university in all of this? And of course, commercial interests are often purely profit, which is lowering cost. But a lot of these innovations are fixed cost investments. So if you invent a nursing assistant, it can be applied and adapted in many places. And so if we do some of the innovation in the university for human augmenting technology, then that takes care of some of the fixed costs.

13:11And then entrepreneurs can take it across the finish line and solve the last mile adoption problem. And that can happen not just us building something and throwing it over the transom. But one of the great things about the language based AI is that actually many people can participate in it even if they don't have a big computer science background. And so actually we can see the participation and the value add coming across the world and figuring out how to help it augment humans in those settings. And I just wanted to connect something that came across all of us, which is that government policy also can play a role in this.

13:52And so when people ask me, what are all the people gonna do? Are they just gonna sit on the beach and have drones drop them daiquiris? Well, I hope so. I mean, I have many questions about how we get to the beach and why the drones are bringing us daiquiris. But if we take a slightly nearer term view, there are many activities that scale with the size of populations and where we are sort of under-investing in them today. More child care, more nursing, better doctors, better elder care. You know, all of those things are things that humans could be productively employed at, at large scale. And AI can help humans transition into those jobs.

14:34And governments can procure those things. Governments have a big role in investing in all of those sectors. So like rather than imagining like just a bunch of people not doing anything, if we get ourselves organized, we create the products, we create the government policy, then in principle, we can help people through the transition while making all of us better off. So Susan, you spent two years in government, I happen to know, and I know that you're an advisor and a research associate, Neil. Tell me, with your real-world understanding of the government today, how's that going to go?

15:09You know, it's really hard because at this moment, right, Like we've had the last six years or so, we've had some curve balls thrown at us. And it's not easy to figure out the very best thing to do in the face of big changes and big disruptions. It's really hard. And so we need to work together on solving those problems. And so, you know, it is scary to think about our government getting less functional in the face of places where actually government leadership might be for both research, for universities could be more essential than ever. I just want to ask something, you know, I firmly believe 100 years from now, when historians rewrite the chapters of 21st century, especially the dawn of AI.

16:03a collective success of humanity or this country would be that this era of AI launched a revolution in education. Is that now that AI, even language models has proven that AI can do the standardized standardized tests by and large to whatever passing or even excellent grades. That we should rethink about spending more than 12 years of human capital, that educating young humans to evaluate them to the level of what AI can do today. Human education should be completely rethought because AI showed us that It's not about memorization of knowledge and evaluation of these memorizations. So if there is anything government can do, to me, is the investment of K-12 education as well as higher education, because this is the moment that we can really revolutionize the most important thing on Earth, which is human capital.

17:17Neil? So I want to connect these threads and run with it. There's a great fact from the economist David Otter, who's documented that over 70 years, something like 70 % of the occupations we have in the economy sort of emerge. That is, if you look back 70 years ago, 70 % of jobs didn't exist. So if AI diffuses in a way which I think is geographically spread out and not too fast that we will adapt we hopefully will come up with new ways to educate people to be complementary to that ai ai will adapt in ways which are complementary to humans but history also teaches us that when things are concentrated and rapid so if you think about the hollowing out of factory towns due to Chinese accession in the WTO and automation.

18:14Those impacts can be devastating and we need policy to come in and provide a safety net. So I don't know which timeline we're on, we're probably some combination of the two, but I think thinking about those extremes and sort of shaping the technology and education is sort of useful to triangulate in an uncertain. Thanks. And now Fei-Fei, I want you to put your hat on as somebody who has a startup and you're trying to make a go of it. And then we're talking about safety nets. We're talking about governmental policies. I'm sure you have part of yourself and your cohort that gets worried that there will be premature regulation, premature policies that actually take away your ability to do the things you want to do.

18:55How does that conversation go and how do you think about it? Great question. First of all, I'm still partially involved in Stanford. so I'm not 100 % on leave, just to make it very clear. And it's also one of the most fascinating conversations that has been happening in the state of California, as well as federally about the tension between AI regulation and AI innovation. And I want to start with, I'm a parent. And when your child is about the age of, I don't know, six or seven, one of the most important lessons you need to teach them is to turn on the stove and cook an egg. And that's to, you know, use fire.

19:44And it's a pretty dangerous thing, right? But you all have to bite the bullet and teach your kid to use fire. And then there are many other things we have to teach our kids. The reason I'm using this example is technology is always a double-edged sword. That's since the dawn of human civilization. We, in our DNA, were compelled to innovate so we can live and work better. But we also use that to sometimes inadvertently hurt ourselves or sometimes intentionally hurt each other. No matter what, that tension between the driver innovation and the need for establishing norms and guardrails is always going to be there.

20:33So as an entrepreneur, as an innovator, I think it's very important that we arrive at a healthy balance. And with Stanford HAI, we have been actually advocating a policy framework for AI, which is very simple. is first science, no science fiction. You know, to do good regulatory or government policy, we should use data, use measurement, like what Susan has been doing in government to guide our regulatory framework instead of those far-fetched science fiction extinction doomsay. Second is that be pragmatic, not ideological, right? For example, in AI, we have so many regulatory frameworks. You're involved with the healthcare, FDA, and we should just maximize the interaction and partnership with these pragmatic frameworks instead of going ideological.

21:34And last but not the least, in this audience, I always believed investing our public sector, investing our innovation engine. Like Condi said, there's no plan B. government policy in AI should include the investment of our country's innovative engines, including the universities and public sector. Thank you. And I actually want to go to a slightly different topic, but Susan, you were in the Justice Department for a couple of years, and in all the examples of innovation economy successes that I gave in my intro, social media, ride sharing, home sharing, the winners have tended to be either monopolies or near monopolies.

22:16Is this a feature of the innovation economy? And are we okay with it? Or is it not a necessary feature? I think that we have often concentration because of the scale economies. But I do think we have examples where even a little bit of competition is a lot better than none. and we have seen a lot of concerns when one firm can put a tax on the whole economy. I mean, probably there's some people here in the credit card industry, but that's sort of an easy example because there's a few basis points coming off of every transaction. I've heard it referred to as a toll booth. Yes. And so in the end, when those tolls get too high, that is also bad for innovation.

23:06So I think you can think about we need firms to be able to get returns from their innovation in order to want to do the innovation. But then if one firm stops, slows down all the innovation around it, that can be problematic as well. So I think here I have a few comments about competition in AI. One is that there was a bit of a push to hold back open models, open source models. but those things can, although people were scared of them, they can pull down the prices for everybody and that means that every single business in the country that is using large language models can get them at lower price if there's a free alternative that's pretty good.

23:47And so that can be very impactful. That conversation changed once DeepSeat came out, but still we also need to worry about other kinds of market power. There are lots of potential for bottleneck. When I think especially about smaller countries, the countries that aren't going to themselves be generating the profit from the AI stack, there's a huge risk if those countries are buying AI at sort of a high price and then automating a lot of their labor, they might have real wages fall because if they're paying a lot for the AI services, their goods prices don't fall. So stuff people buy stays expensive while wages fall.

24:29So that's going to be a really bad situation for countries and economies. And so if this is something that like every small and big business is going to buy, like the price of that thing is very important. The quality is important. The price is important, too. So at the moment, it seems like we're doing reasonably well at that. But keeping our eye on the ball to make sure that this technology is actually going to allow everyone to innovate on top of it will be crucial. Thanks. And Neil, on that note, you've been looking at the innovation economy and economy more general, and you've expressed wonder about whether America and Americans are in something of an economic funk.

25:10And I can't help but think of that after Susan's comments. What do you mean by this economic funk, and do you see a way out? So if you look at data, there was a piece in the Wall Street Journal, it was now nine days ago, that documented belief in the American dream had gone from 70 % of the population to 25 % of the population over a generation. That optimism about our economy has, it cratered after COVID and it hasn't recovered. What's going on? I think we're still figuring it out. Probably some combination of three things. One is there are, I think, real risks in the economy tariffs, people concerned about uptick in unemployment.

26:03It has to be true that social media is skewing our vision of what is a good and meaningful life. I quip that on Instagram you see more selfies from the Ritz than the Motel 6, and that gives people a unbalanced view of what is normal and what is successful but I think there's a lot we don't know but look these these conversations I think political leaders innovators I think are all like really important in helping this country get its mojo back thank you well I think we're gonna have to leave it there I know we want more but we only have 26 minutes this discussion has been fantastic but before we finish up as promised I want our move to our new segment called the future in a minute I will ask you some quick questions and I'm praying that you will give me some quick answers so we're gonna we're gonna start with Fei Fei and the first question is I'm gonna do both questions for each of you what is one thing that gives you the most hope about the future well it's unequivalently humanity there's nothing artificial show about artificial intelligence and I want to paraphrase Dr.

27:17King that the arc of history is long but it bends towards benevolence and I believe the hope of AI is in humanity.

27:30If you were starting over again and you needed to get your degree or training in some other discipline what would it be? Okay I gotta use 30 seconds. Six weeks ago a Stanford sophomore interviewed me and as usual as a professor I'm like what is your major and the the student said no none of these majors are good at Stanford I'm gonna have my own major so I'll create my own major which you are allowed to do and I asked him what are you gonna create he said I'm just gonna use AI to maximize money making for me I was like dad why didn't I have that idea So, okay, my real answer would be if I were to start again, I would do a combination of physics, computer science, and art.

28:22Thank you. Neil. My own major though. Neil, what is one thing that gives you the most hope about the future? I was going to say students, incredible students, but this morning I was chasing around my kids, 11 and a 7 year old at the American Museum of Natural History and seeing dozens, hundreds of little kids excited about STEM, about history, about dinosaurs. We face headwinds as innovators, but if we reflect on the fact that inside we're 7 year old boys and girls that like dinosaurs, we're going to be alright. If you were starting over again and you needed to get your degree or training in a different discipline, what would it be?

29:05Robotics. I care about the overlap between STEM and science and the real world. That's why I study economics, but I think robotics is another great field for that intersection. Neil, you can come to my lab. I would love to. Susan, what is one thing that gives you the most hope about the future? So I do think that AI is more accessible and has more potential to be helpful across countries and across the income distribution than the previous rounds of technology like machine learning. I think that it can allow small businesses, people who don't code, who don't build Excel spreadsheets and don't buy enterprise software to run a business through a chat application and natural language and to grow and scale and get more efficient.

29:52And that that can help people rise up out of poverty and it can help poor countries grow. If you were starting over again and you need to get your degree or training in a different discipline, what would it be? So I've had to do this twice already. I trained myself in machine learning and AI technical work and then I also had to learn law. But if I, one more thing if I wanted to add it to the mix, I think going forward anybody's going to be able to build a great product. If you have a good idea, you're going to be able to make it reality in the digital space. So product management. Thanks to Susan Athey, Neil Mahoney, and Fei-Fei Li.