Welcome back to the third of three firesides here at Big AI at MIT. I am extremely excited about this next conversation. It's very difficult introducing Esther Dyson, I must say. I've known Esther for more than a decade now, approximately. Any one individual bullet point in the list of things that she's ever done in her life could be a normal human being's entire biography is the best way that I like to think about Esther Dyson. She is literally a world-famous, and I don't throw that phrase around lightly, journalist, tech investor, and in every way sort of several steps ahead of all of us on most of the things that she's ever thought about. She's well known for EDventure Holdings, which is a vehicle through which she invested very successfully in technology, but she's always been a very measured investor but also critic of the path of technology in our society, which I think is extremely important. She's also very well known for Wellville, which is a 10-year national nonprofit that cultivated equitable well-being starting in 2014, ending in 2024. She's trained to be an astronaut. And she has a new book out which I've had the pleasure of reading called Term Limits, a design for living in the age of AI, which has some fascinating themes in it that we're going to explore today. And when I think about Esther Dyson and all the conversations that I've had with her over the years, the first metaphor that comes into my mind is I feel like I am talking to the Oracle from the movie The Matrix. Has anyone ever seen that movie? Where you're talking to her and she's just sort of like 10 or 11 steps ahead of you and you're telling her what you're up to or asking her what she's up to and she's asking you sort of like pointed questions and you just get a sense that she already knows like what's going to happen three moves ahead and you're just hoping that you can figure out what the second one is, you know. So I think you guys are all going to be extremely thrilled to listen to what Esther has been thinking about recently. So please join me in giving a very warm welcome to Esther Dyson.
All right. So, it's great to see you. Thank you very much for being here. I'd love to just start with a very big softball. What are you most excited about right this minute?
Well, as so my book is actually only some very special people who read it as a helpful critic and of course you're not supposed to know who gave what responses. So, it's actually I just delivered it on Monday and it really is like a pregnancy. It took approximately nine long months and now it's in the NICU and it's going to be in the NICU until early next year. I had breakfast with my editor this morning and he said it's not going to be too bad.
So, but just it's not out there. You don't need to sign it or buy it. So, it's a good sign. Just enjoy the conversation.
I like the optimistic editorial breakfast. I've never had one of those. Most of my editors are like, 'Wow, we got a lot of work to do.' Well, we've already done a lot of work.
Yeah, that's good. So, the book has been sort of your main focus recently and I assume that we're able to talk about the concepts in the book?
Yeah. I mean, it's, how shall I say? It's purposely designed not to be last minute news, but more what's still going to be relevant next year. And that's kind of hard with things changing that quickly.
Well, yeah. I mean, I think that the theme is going to be sort of relevant for a decade or decades in terms of AI. And so you basically in this book you're contrasting the finiteness of human life with the infiniteness and the timeless sort of scale of AI.
Yes. So in thinking about that for nine months or even more probably in conceptualizing the book what becomes most valuable about being human right as intelligence becomes abundant, infinite and cheap. Yeah. So I learned a ton writing this book. It wasn't oh I know all this stuff and now I just need to write it down. It was more like hm these are really interesting questions and some of them are still questions but first of all just so you know it's called term limits and everybody thinks oh oh my god another one about Joe Biden and don't but and it was slightly Joe Biden refused to resign well ended almost on time December 42nd of 2014, but it was finite. It was supposed to end. And that was sort of the point. It wasn't like we're gonna do this forever. We want to do something useful. It's a project. We want to pass something along. And third thing, Jimmy Carter died. And oh my god, how sad. No, it's not sad. Celebrate the man's life. And it just seemed to be such a this big elephant in the room. Things should not last forever. Things should have a purpose and a time and they should make a dent in the universe. Hat tip to Steve Jobs and beyond that to disturbing the universe which was TS Elliott. Things should end on time, but that doesn't mean they're over. It just means they get absorbed into something else. The way an LLM absorbs everything in some sense. And it seemed worthy of a book.
Yeah, I mean I think it's such a great concept and this notion that you know we are all stardust that we also came from things that sort of ended and evolved, evolved maybe is a better word. But again you've got this sort of like concept of on the other side of this is AI right which is almost like the opposite of that.
Yeah, it I mean it does two things. One, it compresses time into almost nothing. I mean, you know the old joke or truth about women, you can't have nine women can't have one baby in a month. And interestingly, neither can nine women have the same baby. Just this uniqueness. And again, there's another quote is Oscar Wilde. He knew the price of everything and the value of nothing.
We're going to get to that part. Yeah. Okay. So, but there's not just time. There's kind of this duplicity and humans have sentience and purpose and these things that evolved. And to some extent, if you're infinite, you can't have a single purpose.
That's interesting. And at the same time every you know this whole multiple universes thing every time you create a duplicate that duplicate encounters different circumstances or interfaces or answers different humans questions or and then becomes no longer the same entity because the data they've collected every every prompt is not just a question but it's also new data for whoever you're interacting with.
I feel like this could go in a bunch of ways in terms of the effect of this sort of timeless scale infinite AI on one hand and the juxtaposition of the finiteness of human life on the other hand could go in one of a couple of ways. We could sort of reflect on this and say, 'Wow, we can really understand the inherent value that we have as sort of this fleeting moment in time.' I'll give you a little personal tidbit that relates to this, which is that on my car keychain, I have a skull and it's what we call a memento mori. This notion that hey, remind yourself that you will end every day. And I was actually at a meeting and somebody was like, 'What is that? Some sort of like satanic like situation you've got there on your keychain?' And I'm like, 'No, it just reminds me every day that I am finite and that therefore every day is almost infinitely valuable.' That's one way that reflecting on AI's timeless infinite scale could impact us. Another way is to sort of be overwhelmed in a sense by it as people many people are like there's too much and then there's business models that take advantage of that.
Yes. Yes. So we're going to get to that as well. But are you hopeful or are you sort of worried when it comes to this either? Okay. But I'm also very practical and if I'm worried and pessimistic, I'm definitely not going to help. I'm optimistic and write a book to try and persuade people to think more deeply about this and find meaning in their own lives. And this is not a 10 timely tips book. This is 10 infinite pieces of advice for a finite life.
Yeah. Okay. Well, I think that's good because there are enough of those I think in the world of publishing. So in terms of like another element of AI it's this notion of measurability is extremely elevated now right because quantification, measurement, you know, being able to count is an essential input into AI in a lot of ways even if we're talking about language. I mean people mistake what I just said to think I'm just talking about numbers or math.
Yeah. Sure. Exactly. So you can then quantify language and now we need to be quantifying emotion and then we will be producing emotion in AI as personas or personalities which are distributions over emotions which are then distributions over words and so on. And that's a very like quantified way of thinking about things. And then you mentioned Oscar Wilde. And so the question or the danger or the perspective I'd love to hear from you is about sort of where do you see AI pushing us towards mistaking what is measurable for what is valuable.
All the time. All the time. I mean it is and it's the fundamental conflict there. Well, perhaps the fundamental there's so many fundamental things but one is nothing is so good that we need an infinite amount. And having an infinite amount of it is going to devalue it almost immediately. Again, that's quantifying quote that just you can't because you're human and an AI can and but I think they have some other deficiencies. But there's the corollary is nothing is so bad that you should shun it, refuse to consider it, banish it from your mind. And death is certainly one of those.
Yeah. Death is a good ending.
It can also be a bad ending. But thinking about it not as a tragedy but as a again a conclusion to what you hope was a good life. But even happiness happiness comes from overcoming obstacles. It doesn't come from sitting around spending a lot of money doing pleasant things. And yeah, so all everything friction is good, friction is bad, friction, the friction this morning of not being able to find my way to my meeting with my publisher. He was walking anyway, we found each other because I had a phone. The friction of interacting with people. And one thing we're doing too much of long before I wrote this book, I used to tell people, you know, I don't pay people to play tennis for me. And there's some sense of, well, I can hire somebody to manage my affairs, to donate to charities on my behalf. I got a press release come email about an app that would save children the awkwardness of communicating with others online by creating avatars to represent them so they could answer their friends questions.
That's very depressing. I have to say that this has been a theme at this conference in a sense which is this notion of cognitive offloading which is hey we can give AI the thing to do but that doing that prevents us from experiencing both the challenge and therefore the joy of doing it. And as well the benefits of learning by doing and learning by struggling and doing all the messy things that help us evolve during the short finite span of our individual lives. I think we've seen a number of those types of arguments emerge here.
Yeah. And I think that one of the this isn't it's interesting to put the finite individual human life into the context of the generational sort of marching forward of human life as a species in the sense that those struggles enable human beings to experience joy and accomplishment and fulfillment. And as well attain new knowledge and new skills. But then that when that human life ends as a finite life, presumably that is at least one of if not the main mechanism for our evolution as a species from generation to generation to generation.
Yeah. And some gets internalized so that there's so much we have in common as human beings and David Waltz who worked for thinking machines that probably went to MIT said it so nicely words are not in themselves carriers of meaning but pointers to shared understanding and there's so much shared understanding in each of us that an LLM just the bandwidth is so low.
Yeah. And I think that somebody mentioned yesterday I think it was Nicole Imora. She was talking about how one of the most fundamental aspects of being human is in relating to each other. And that to replace that with AI is a mistake. But also there may be ways that AI can help us do that better if we do it right.
Yes. If the AI asks questions, I mean, it's just like working with a shrink. The sort of the danger of psychiatry is transference where you interact with your psychiatrist the way you have the same reactions to your psychiatrist as you had to your father or, you know, your ex-boyfriend or whatever. And as a psychiatrist has to be an external advisor rather than I mean just as you get addicted to food or all these other things you can get addicted to a sycophantic bot but it's ultraprocessed. And we've had the whole ultraprocessed social media thing. But somehow in the bottom of your brain, you know that they're fake and it messes up your intellectual metabolism as opposed to your Yeah. Whatever insulin does for sugar, you've got something in your brain that does that for.
It's interesting to think about AI as like a processed version of social. And like human beings, real human beings being like organic organic produce. Yeah, farm to table. So you've been thinking in terms of like measurement as well. You've been thinking about a little bit about the GDP.
Yes. So I want to do an ad slash prompt. I don't know. So this notion of trying to figure out the GDP in some sense it's a single number and we always want it to go up, right? It's sort of the same way people measure their weight. You know, it's a single number and almost everybody wants it to go down. But anybody with common sense is going to want to know, well, how much of that is bone mass? How much is visceral fat? And the GDP is full of good things, housing, food, but it's also full of health care that is basically repair work on things that should never happen. A lot of it diabetes that you got from eating ultraprocessed food that is also in the GDP and some of it is cheaper than it should be because it is ultraprocessed and so it benefits from farm subsidies for corn interestingly. So I would love to find some economists who would create what I call in the book the GDVP the gross domestic value produced and you subtract from it like half the food supply a third of health care spending gambling opioids have some use but most of them anyway if anybody's interested in doing that.
Well that's super interesting I mean you could imagine that sure like a poor a nutritiously poor meal is probably a zero value or a negative value because it's calories. But it should be like discounted heavily compared to like the organic meal, right? And then opioids depending on the use can have very positive effects can have negative probably negative effects or be discounted heavily.
Many negative effects. Yeah. Yeah. That's interesting. Yeah. And it would just get people to think a little differently.
Yeah. Have you seen anyone attempting to do this? I know you have a degree in economics, right? Is that right?
No. But I found this stuff completely confusing. And the data are hard to get. The gun lobby doesn't even talk about its revenues. Talks about numbers shipped or something. And the better numbers come from some other direction interest group that lists how many people died by gun violence in one way or another, right? I talked to a banker from a well-known bank about this and he completely didn't get the point. He said, 'Well, you know, the problem with the GDP is it's all about consumer spending and the real profits are in the business side.'
Yeah. Yeah. I mean, it's interesting because GDP is a well-developed methodology. There are known deficiencies even in the mental model of the GDP as it is. For example, the inability to account for goods that have no price or given for free like social media. Yeah. For example, that is a well-known thing that economists are working on. This is very different what you're talking about. But what's interesting here is the advertising revenues are also not counted because they're not quote consumer spending, but they are in the price of the stuff that you buy that's advertised.
Well, it's interesting because I mean, not to go too far down a rabbit hole, but it is a rabbit hole. Yeah, that I'm sure that the advertising would be counted in the GDP, but the advertising is one part of a much larger product, much of which is given away for free. But that doesn't even come close to talking about what you're talking about, which is just a different conceptualization of what we're trying to do. The interesting thing to me is that GDP as currently formulated is used as a metaphor for welfare.
I know. And it's crazy. And that is known. I mean, it's definitely a proxy for welfare. But it is to call it a crude proxy is probably pretty generous, I would say.
Yeah. Anyway, prompt for whoever wants.
I like how you prompt the audience like you would prompt an LLM. Give me a new measure of GDP. Chat GPT. Yeah. So, what would a society built on the idea of term limited humans do differently than the one that we're building now?
Well, you never know. But in my version, it would focus a lot more on basically raising children and the whole, you know, value of people, education, child care, health. And so it gets difficult. Outlawing stuff is often not the best approach. So 10 or 15 years from now assuming that the valuation pays off and people have to do a lot less work because we've reduced a lot of stupid friction. But how do we keep the good friction? And it seems to me we need something not so much universal basic income but universal basic service. So when I was talking earlier about intermediation and outsourcing everything I'm on the board of charity navigator which I joined because they help focus on impact. But at the same time I go back to the story of this Narayana Murthy who co-founded Infosys. Long story but I happened to have dinner with him once in Bangalore where he lives and he said oh you know come to my house it's quite modest of course our children live in another house nearby but and it was modest his wife cooked the meal and brought it out and we ate and we talked about various stuff and one thing he said was well you know when we lived in the village we were also the most privileged people there but we didn't hand food out the back door. If people came to us who were hungry, we invited them to have dinner with us. And that did as much good probably for Narayana Murthy and the things he was telling me as it did for the hungry people. So something where you have to spend if I were writing the rules, two years of some kind of service in a community completely different from where you grew up. And for that you get free college or you know whatever is relevant at the time. But just and of course there are difficulties. Some people are really horrible and you know you send them to a village and they're going to wreak havoc. But a system where there are guardrails and you know that's the whole point. We need guardrails. It's not untraveled. Go show up and do what you like. But something that perhaps replaces religion and religion itself has been corrupted. You know, the notion of tithing is kind of the same thing as yeah, I'll outsource my charity. I'll outsource my giving to the church and then I'll get to go to heaven. That's not the deal. Or I want my name on the building or but that's the sort of thing I would like to see more of some notion that it is your there's so much ancient wisdom so this notion of learning, earning, and serving the serving part is really important.
It's so interesting because that exact triplet was mentioned by Rana el Kaliouby yesterday. Oh good. First you learn then you earn and service is like and she literally said almost the exact same thing yesterday. So you guys are telepathic I think in some way which is great. You and I have talked a lot about social media over the decade and decade plus and a lot's been going on in social media like what's your take on the state of social media these days?
Yeah. So, how many of you noticed the Facebook losses in court in the last couple of weeks? Yeah. Great. Hallelujah. And this isn't Oh, it's wonderful. We're going to have a whole lot of lawsuits and the problem's going to go away. I think there's a better way to go back one step further and that's the notion of so government regulation is always two or three years late. It's too complicated and it doesn't deal I mean ask anybody in the mafia there are lots of ways to commit crimes without saying things. If you Google mafia code words, you'll get some interesting words and concepts. It's and the way the issue in this lawsuit wasn't simply the news was all about, well, is social media addictive or not? And I think it is, but the real news and the cases hung more on the culpability, the willful simply, you know, shunning the bad news that, oh my god, there are children being harmed. There are these things going on. And these companies, it's not that they said we want to get these kids sick. It was simply, you know, we need to make our quarterly budget and it's only 3% of some number and you had parents and real people in the jury and they were horrified at what they heard and saw and okay, how do we take that forward? Again, it's not so much government regulation or a regulation that says two things. One, every company like that needs liability insurance and they need to disclose it to investors. If that's the SEC requirement, private investors are going to ask the same questions and if not, you don't want them as your investors. Now, you need to get the insurance industry involved and that's where quantification is valuable. What is the risk that this company is going to have to go to court? Because not because I looked at the rule book, but because I talked to their chief marketing officer who was oblivious to the notion that any kid could ever get harmed or you know, it's always things are always a trade-off and you can't be perfect, but you want to find people who are alarmed when they hear about the danger and trying to figure out something what to do about it. And an insurance company's motivated by if we do this well enough, you need a fairly long-term limit for the damages, not a year or two. So, it's very much a reinsurance business, but the insurance company ultimately needs to price the risk. And if someone's insurance bill is very high, don't invest in them. Because it's a very bad sign. And a good insurance company is known by the clients it turns down, not just by the fancy names who are their clients. And somehow just building that into the woodwork of how our society and how our financial system works. But it's a perfect job for AI. Just go through all the company's records and don't look for the code words, look for the code attitudes. Look for the sentiment analysis about the welfare of the kids versus the revenues versus.
