Best of the Pod: Wired's Kevin Kelly on Why AI Is a 50-year Overnight Success

What I've learned with the future is that it's easy to make predictions and hard to make predictions that are true. Jaron Lanier showed me VR in '87. I was completely blown away and I thought, oh my gosh, this is the future. The VR that we had back then is not that much different than the VR you get now. The difference was that was multi-million dollars and it costs now a hundred dollars. I've been very, very surprised about how slow that has been. You could say VR is still waiting for its LLM moment. AI is 50 year overnight success. Kevin, welcome to the show. It's a pleasure to be here. I'm glad to be seen. I am very excited to have you. In addition to being a personal hero of mine, I am a long time Wired reader. I had like a gigantic shelf of magazines growing up. So it's sort of crazy to get to chat with you. There's a lot of things I want to talk to you about, but I want to start with, we have a mutual love of Annie Dillard. She's my favorite writer, and I know she's your favorite writer as well. Tell me why you like her. For those who aren't familiar with Annie, she burst onto the scene with a meteorite of a book called Pilgrim at Tinker Creek that for some reason, it's a account of her spending some year on musing kind of like a Henry David Thoreau intimate investigation of a creek in Virginia or West Virginia. I don't remember. And so it was just the writing and the style and the ideas and her blazing brilliance in being able to capture this into a few words. It was almost like poetry. There was prose. It somehow just worked on my brain. And I don't know enough about literature to describe her in relation to other writers and why it doesn't work for me as much, but there was something about her spirit as well as her writing, which was very expansive, cosmic, enthusiastic, questioning. She talked about feeling as if at one point in the story that she was a bell that someone else had rung. And I felt exactly the same in response to her writing. It was like she was ringing my bell in a curious way. Yeah, I lifted and struck. There's another part in the early part of the book where she talks about the tree with the lights in it. Yeah. And that moment is so very evocative of like one of her big themes, which is you can just be sort of walking around in ordinary life, in ordinary nature, and somehow like the veil is lifted from your eyes and you can experience these moments of transcendence where everything feels like it's glowing from within. Yeah, so there was this, her kind of cosmic poet, you know, ability to kind of take these moments, but then interstitial with that were these findings, these weird little trivia bits that you'd find in some obscure book that you would be grounded in before the next jump. It was a very distinctive way of kind of wrapping rhapsodic ecstasy with very concrete science trivia or oddities. And that combination just somehow appealed to me. Yeah, the thing I remember is she spends a long time talking about Henley's loops, which are like this little part of the kidney. Um, and it's just like, where does that even come from? And, uh, yeah, I think like she sort of zooms out to the biggest stuff and then zooms into the smallest stuff. Right. But another thing that's really interesting about her writing, and I'm curious how you've thought about this in your own work, is she spends a lot of time talking about the beauty of the world, but she also spends a lot of time talking about the worst of the world. Like there's an entire chapter on insects and how gross insects are. She's also angry about it and not afraid to be angry about it. You know? Um, and I think a lot of writers are usually either one or the other. Either you're like really angry or you're really, everything's beautiful and awesome. And she does both pretty well. That's true. She said, I think either she said in the book or elsewhere that she wrote the book as if there was a, a patient of cancer dying in her room. Oh wow. And she's talking to them. So yeah, that was, uh, um, yeah, it is not just saccharine. It's not just sweet kumbaya. It's, yeah, she's, she can be pretty harsh too. And, and I usually, that is part of the attraction of, you know, the sweet sour flavor. How has she, if at all influenced how you write or edit? Well, In some ways it's why I write at all because I, I came across that book in the weirdest of all places. It was in a library, American library in Kandy, Sri Lanka. Wow. And I walked in because there was air conditioning and there was a book. Why they had the book there, I don't know, but there was a book opened up and from the first immediate paragraph, I once had a tomcat, you know, like I was, I was caught and I didn't put it down. And it was like, I don't know what she's doing, but I want to do that. If I could do that, I would, you know, I'm it. I'm golden. I feel the same way. Have you read any of her other, any of her other stuff? Yeah. Yeah. I didn't, I didn't write, read her fiction book, but the other books of essay, teaching a stone to talk and Holy the firm, the writing life. Holy the firm and all those. Yes. And her account of the eclipse as the total eclipse, it's sort of never to be equaled. Anybody who writes about eclipse has to start with Annie's version. Yeah. It's, it's so funny. Cause like the reason I started with Annie is, um, I love her so much and I, and she evokes all the same kinds of feelings in me that I think she evokes for you and, and gives me that same kind of energy to write stuff. If I write stuff after I've read something that she's written, it just, it brings something out in me that I just love. Um, And also no one else I know likes her or even really has heard about her. Like, I feel like a lot of her work is, um, uh, it's maybe taught in school, but it's like kind of the thing that you're assigned, but people don't like as much. Well, they probably also don't read the whole thing. They probably would have understood it if they only get the excerpt. Yeah. Um, you know, probably some one passage about the lights in the trees or whatever. Um, yeah, it is true because, because she didn't really, I mean, she wrote a couple other books, but she didn't really go on to have a huge amount of work and a huge amount of output for whatever reason. I actually sent her a book that I did when I was riding my bicycle across the country. I did a haiku and a sketch every day and I sent her the book of the original and because I thought I was inspired by her. I just thought that she would enjoy it. And she actually sent some nice words in response to the book and sent it back. Um, so that was my kind of my one. It was the one time when I kind of wrote to a hero. That's really nice. Yeah. She did me a reply. I bet a reply from Annie is probably pretty rare. So, uh, that's, that's pretty great. I think I still have it actually. I, I think it's right behind me. I should dig it up and see. She did, there was one little drawing. She said she really liked that little drawing. So that was, that was good. Yeah. Another thing, um, to move on from Annie for a second, one thing that you wrote in your book, The Inevitable that I really loved and thought was interesting is, um, after living online for the past three decades, first as pioneer in a rather wild empty quarter and then later as a builder who constructed parts of this new continent, my confidence in this inevitability is based on the depth of these technological changes. I really love the, the sort of like pioneer to builder transition. I've been watching this show. It's an old HBO show called Deadwood. I don't know if you've seen it. I haven't seen it. Um, but it's, it's basically about Deadwood. Uh, I think now it's South Dakota, but it was like a, it was a mining town in Indian territory before it was annexed into a state. And so it had no law. Um, and the whole show is kind of about that transition from pioneer to like order out of chaos. And I think there's a lot of resonance with sort of what you wrote and, and this, um, yeah, the transition from pioneer to, to builder feels like something that's deeply embedded into tech, even with every single wave, like in the, in this new AI wave, there's this whole movement of pioneers, like kind of figuring out the whole new landscape. And then the builders kind of move in. Tell me about what that's been like for you to participate in and watch over the last couple of decades. Yeah. Um, there are a lot, you know, there are always, um, feelings of loss as the, the kind of the freedom of no laws, the unvarnished, the unconstrained ability to do what you want without having to ask permission. That goes away. And, and actually I was just, just, just had another experience that was like that, which was Burning Man. Yeah. And, um, um, I was reminded because I did a podcast with Burning Man today and they, we were reminiscing about, you know, my, the first in 96 or so, the first, uh about the discovery and or invention of electricity. Because my contention right now is we have no idea what intelligence is. That we're as ignorant of it as um Isaac Newton and others were of electricity when it was first encountered. And Isaac Newton, one of the smartest humans ever, was totally wrong about electricity, right? He had these wrong, weird ideas about it that were just wrong. What do you think about it? I have not heard anything about Isaac Newton and electricity. That's really interesting. He was one of the people who kind of thought that there was this either phlogiston, have you heard of phlogiston? Yeah, phlogiston, yeah, yeah. So it was, it was... The thing that creates fire. Yeah, it was this kind of, well no, it was just this kind of element, and there was another element. And this, and the, the uh discovery of electricity was happening at the same time that basically we were understanding what elements and atoms and compounds were. So Davies and Faraday and those guys, um, were almost discovering stuff about um electricity weekly, and the origins of the Royal Academy came out of the weekly meetings that they would have where they would sell tickets and do demos with electricity and make sparks and stuff. And one of the biggest news items in the shocks was when they, and I forget who it was, maybe Faraday, proved that electricity would happen in a vacuum because there goes the ether. You don't need any ether. It's like, well, then what is it? Okay? And um, and there was, you know, some of the earliest beliefs, electricity was that it was primarily a biological phenomenon, and there's all those, um, all those reflexes and stuff. And so there were just endless theories about what it is, and they all reminded me of all the theories we have about what intelligence is because we don't really know what it is. And I have been saying, I suspect that intelligence is not an element but a compound, that it is uh made up of a complex of different cognitive elements and that we haven't even identified yet in the same way that salt's not a element. Salt is actually a compound of some elements that they had not yet identified. So you can think of the kind of the current AI is we're making some kind of salt and we don't even know what it's made from. Yeah. I think that's true. Where it makes my mind go, and I'm curious how you would respond to this, is, um, we actually do know what it is, but we don't know it in the same way that we know what electricity is. Um, so we don't have a explicit, exact mathematical theory in the same way that we can talk about, talk about electricity, and we don't have a way to decompose it into parts that reduce down and then recombine into it. But we do know what it is on a, in a different way. Like I'm talking to you and I know that you're intelligent. Um, and that's harder to grasp. That's not, it's not the same kind of graspy, um, I can pin it down to the wall kind of thing, but I think that that just may be a property of, of intelligence that it is sort of like this fuzzy thing. How do you think about that? Well, I'm, I don't know, you may be confusing recognizing something with, with knowing it. So I think we can recognize it, but I don't actually don't think we know what it is. And in fact, I think our brains are incredibly opaque to introspection on deliberately. I think we have these complex things that, that deliberately does not allow the organism to interfere and meddle with it. Could you imagine if we had access to the source code, we'd be, we would be completely uh wrecking ourselves. And what the it is, again, I, I think we humans have a very peculiar complex of things that if we map it out in the possibility space of all possible minds, which is a very high dimensional space, that we're going to be our compounds way at the edge. We're an edge species. We're not at the center of anything, the galaxy or the solar system or evolution. We are an edge. Our kind of intelligence will be revealed to be a very peculiar mixture that's um evolved for us. And then what we're going to be doing with AI is making hundreds of various other kinds and filling out that possibility space with many types of thinking. And so we'll look back and, you know, we, we won't even recognize maybe some of these other things as intelligence right now because doesn't, we don't have a very good definition, you know. It's like, what are the definitions or what are the marks for something that doesn't like a human-like intelligence? The, some people say, well, there isn't anything. There's only universal intelligence and that we're just gonna make more of it. That there's only one thing. And I, that's possible, but I suspect that's wrong. I suspect that there is many compounds and that they will be engineered to do different things and to some degree, we won't understand even how they work, but that's because they're different. And so, um, uh, I, I think we are very much like the early days of electricity where we simply didn't have a clue about what it was, even though we could use it, even though we could recognize it. Where does your intuition that it's made up of compounds come from? Well, in several ways. One is it was Marvin Minsky was the first to suggest it called the um Society of Mind. And then these days, the AIs have a mixture of experts, um, where they are already doing that, where they already are taking different kinds of cognition and making them into compounds. And so I think we'll have various layers of this like tissues where you have cells and, and um or molecules made from elements. And so we're going to make some very high chain, heavy compounds of intelligence at some level made up from lots of little bits of elemental cognition. What we haven't done yet is done the chemistry of identifying what some of the basic cognitive units are or the cognitive elements. And, um, we may be starting to do that. Thinking about the way neural networks work, a way to look at them is they learn many, many thousands or millions of rules for what to do in particular situations that they can partially apply and run many, many of those rules in parallel to find the right set that sort of applies to a particular situation. So in that case, like intelligence is a compound of rules that are about like little micro correlations um that are applied depending on how relevant they are to a particular situation. Um, but it's interesting to think about, um, how much of those rules are contingent, like they're just situational versus they're, they're sort of universal. Um, because we do know, for example, how neural networks function at a low level. Like we know the atomic units. And in fact, like the specific architecture, the specific set of atomic units that you use doesn't really matter for the high level behavior. It does to some extent, but you can get basically the same behavior no matter the simple components you use. Right. So there's something like always hard to understand in between the simple components and the like outward behavior that we observe. I mean, you know, Danny Hillis made a computer with Tinker Toys. So you can make computers out of all kinds of elements, logic gates and stuff like that. So I don't want to confuse sort of like a neural nets. I don't think that is the basic unit. I think there is a type of reasoning or learning or something that happens with the neural nets that we haven't quite identified yet that I would say that's that that would be the element is, is what is that process of pattern matching or if that is what it is, or deduction, or what's the logic or what's the workflow for doing deduction. And it could be, you know, it could be agnostic to the actual platform. It seems like it must be, um, in, in some way. But yeah, that's interesting. Um, one of the things I love about um the way that you've set up your career and, and I think probably also the um The way that you think about just creative work in general is um it's about being honest and authentic to who you are instead of what you think you should be doing. And I found that over and over again, like, especially like at the beginning of a career, it's like really easy to be like, I need to do things in this particular way. And then at some point you're like, I don't know, this kind of, at least for me, you're like, this isn't working as well as it should and this kind of sucks. And like I guess I have to like do the thing that's like more honest to who I am and more shaped to who I am and, and it, it removes all of this like daily friction. Um, so like, honestly, like hearing you talk about that in various forms over the years has been like quite helpful for me. How does that look now after many years of thinking that way? Is that still something that's on your mind that's difficult that you have to like unstick yourself from, or are you kind of, do you get used to living that way after a while? Um, Yeah, I mean, my own life was never very planned or deliberate in that sense. I had, I had more directions than destinies or destinations. A phrase that I like or advice that I like to give these days is about don't aim to be the best, aim to be the only, be the only. But that was not something that I was doing consciously when I was younger. That was something that I only realized I was doing much later. And it's part of the book about wisdom I wish I'd known earlier, is that I really did wish that someone had told me that earlier. Um, and so I think I did so a bunch of things The spots that I'm most ignorant of, and, yeah, for me, it's a really great way to start. Is there anything that has worked particularly well as part of your workflow to get it started? Well, I use some of them for research, too, and that's another way, while you're researching, you're summarizing stuff and synthesizing, so there are these elements that begin to take place. The LLMs are really good at organizing things, and that's another, see, I'm not that organized normally. I'm a gardener rather than an architect in terms of things. And so I can get a little bit more further along in thinking about it in an architectural way. So for me, it's a, yeah, it's a great way to start on things. What are you using day-to-day AI-wise? Like which models? I'm using mostly OpenAI right now. Do you use any particular one? Like are you using 40001, 03? I have the 01 Pro. Do you like it? Yeah, I mean, I've done some of, you know, for very deep research, and it's astounding. I'm just reminded again and again that this is a skill. Using them, getting the most of them is a skill that will take your 10,000 hours, and it's definitely not just pushing the buttons. It's not just clicking. And so it's very apparent to me that, you know, I need to spend a lot more time whispering and understanding how it is and how to use it. And so my needs right this moment are not so that I'm using that level of research every day. Do you have specific examples of like when you've used it for research that it has like blown you away? I did something, and this is a little bit of something I've been meaning to write about, which is I had this fantasy. I had this observation that I realized that Leonardo da Vinci, Martin Luther, and Christopher Columbus were all alive at the same time. And I said, this is my conversation with the AI. I said, imagine it's a snowy evening, and all three are stuck in the same hotel together, and they have a conversation. You know, give me the conversation based on, you know, based on your writings and what are their interests and their personalities. And so they did a conversation, and I said, that was amazing. But they got along so well that they decided to collaborate on a project. What would that project be? And this was the AI's idea. The AI's idea was that they would want to start a new city in the new world that was based around science and religious freedom. And I said, okay, you know, start writing Wikipedia articles about this city and start, we'll start filling it in. And then we had characters and people and histories. And then I was going on with starting to tell, actually to tell research stories based on these characters. And then I introduced other things like other contemporaries like Queen Victoria, who was alive at the same time. She was the main rival trying to take down the city. There was, they reached China 10 years before the Portuguese, and then they would bring all the books from China. And so I was just making this bigger and bigger and bigger thing, world building. And I started, I got 10 different novels from it, and then I had them, I had synthesized and ironed out all the contradictions between the novels and make a big epic saga. And then I had it write the, making book covers and writing the marketing materials. And the point of all this is that I'm not going to show it to anybody because the, I don't need to, the joy of creating it was better than reading it. It was, it was the audience of one. And so what I'm hypothesizing is that a lot of the generative stuff, the 50 million images that are generated each day with AI, 99.9999% have the audience of one. They're generated for the pleasure of the co-creator. And this idea of people will be making feature length movies for themselves. And the pleasure will be in the generating of the movie, the co-generating of the movie that you're directing. You'll be directing the movie for yourself. And anyway, so that was a project that I was using Pro for because the degree of historical realism and fantasy was mind bending. That's really cool. I love that. It seems like there's an interesting line from a thousand true fans to an audience of one. Yeah. How have you thought about that? Well, of course, there's no economic model for the audience of one. There is for OpenAI. Right, exactly. I think this is, you know, this is the abundance mindset where you have the time to do this. So I think this is not so much a business or someone's career. I think this is kind of a different form of entertainment or self-expression. It's just like Sunday painting or keeping a journal or, you know, someone doing ice skating. It's a form of self-expression and relaxation and enjoyment, entertainment. So it's closer to entertainment than it is to actually, you know, a career. It reminds me a little bit of, are you familiar with active imagination? No. It's a part of like Jungian psychotherapy. The idea being you can do active imagination where you take a dream that you've had recently and you sort of like re-enter the dream world while you're waking and you explore some of the archetypes and themes either by writing them out or just like exploring them to and with yourself. And in doing that, it kind of reflects back to you things that might be like a little bit more latent in your psyche. So the fact that you're playing around with these characters and these novels are going in different directions or whatever and the decisions that you're making might say something to you about what you're processing or like what you're currently thinking about that, in addition to just being really fun, it might tell you something about yourself. Yeah, so it may be something like that. In fact, you know, I mean, I think there will be very, very good AI therapists and maybe to the extent that we may not even make that distinction, maybe some particular AI companions, AI buddies, AI partners will perform some of that and people will use them in that capacity, even though they're not maybe nominated as that. And again, this is something else I've been saying, is I think people are gonna be shocked by the degree of emotion, emotional bonding that we will have as we put emotions into the AIs. And some people will be very, very close to these on an always-on basis and very dependent on them to do their best. Yeah, one thing that, I don't know if you saw, but just came out today, like a couple hours ago. No, I haven't seen it. I've been podcast all the time. What happened? You're in for a treat when you get off the podcast grind, but OpenAI released a new memory system for ChatGPT. So previously, ChatGPT, you could ask it to save memories, but it would like do it when it was really explicitly told to. And, you know, when it like really thought, okay, this is definitely something I need to store. But what the new memory system does is it just, it is able to access all the past history of all your chats that are relevant automatically. So you can ask things like, what do you know about me that from all of our chats that I might not know about it myself? And it will just go through all these historical chats and tell you a lot of really interesting things. I was playing around with it all day before we got on the show. I think you'll really, I think you'll really like it. And I think to that sort of emotional attachment point, I want to use something that feels like it knows me. Sure. I get, I get more attached to it. I personally think that's really good, but there's trade-offs, obviously. And I'll be curious to see how that changes how we relate to each other too over the next couple of years. Yeah, yeah. That's cool. I've been, I've noticed that you're also doing a lot of tweeting with, it looks like AI generated images. Are you using, are you using native image gen for that? I think they're mostly mid-journey. Mid-journey. Yeah. Which for me, I just have a habit of that. It's, it's comfortable. I know Dolly, I mean, not Dolly, Chat now has some, which I've tried and it's pretty good, but I just, I actually like the public aspect of the mid-journey Discord where you actually, it was a huge quick learning curve because you're seeing what the prompts other people were doing and how you could get there. And I liked that public aspect of it. Yeah. I'm curious if we zoom back out. You've seen a lot of different technology waves. I think one of the interesting things about this particular one is how, how scared people are. And that's probably been true to some extent of previous ones, but I think in a lot of ways, like with mobile, it was like, people didn't even care. It just like, wasn't even on their minds for a while, at least in my lifetime. And I'm curious what you've learned about, like seeing all these different waves come and go, like what you've learned about the future and how to think about the future. That's a pretty big question. I think the first thing I think about all the places that I was wrong. Where were you wrong? Oh, so many, so many times. I was very wrong about VR until Jaren Lanier showed me VR in 87, something like that, in his lab. And I was completely blown away. And I thought, oh my gosh, this is, this is the future. This is amazing. And to be clear, the, the VR that we had back then, which was 30, 40 years ago, it's Okay, so AI didn't work for 50 years, and then suddenly in like 2010 or so, it started to like, there was some like glimmers of hope, and then ChatGPT a couple years ago is like, really wow, this is a thing. In VR, like, if we wanted to measure where we are in that cycle, do you have a thought for that? Yeah, I mean, you could say VR is still waiting for its LLM moment, right, where there's some technical breakthrough in the lenses or focusing or something or projection that allows for that. So, it's interesting too that the LLMs were kind of not working on reasoning directly. They were doing language translation and that's, they noticed that there was some reasoning happening in language translation, which was completely unexpected. And maybe the key technology for VR, AR will come from somewhere else, but it hasn't happened yet and doesn't look like it's happening right now. So I, yeah, we, you know, AI is, yeah, it's a 50-year overnight success. What about, going back to the other two examples, so eBay and blockchain of like, hey, like, I don't, this isn't really that interesting. Right, right. So eBay, I think with the auction stuff, and I think the only reason why eBay worked is that people moved beyond the auction. I use eBay all the time and I've never used auction. I just have no patience for it. And so I think the idea that it was this idea that it was auctions and that I didn't connect to and I was never really interested and I could be wrong, but for me, eBay has only now succeeded because most people aren't using the auctions, but I don't know if that's really true or not. Do you use auctions on eBay? I only, I got really into eBay when I was like 10 and I was like, I wanna sell everything in my dad's garage, you know? Did you? I didn't end up doing it because it just, selling things online as a 10-year-old was, it was a complicated situation, but I was really into it for that reason, but I've never really bought anything on eBay. I've never, I've never bid. So eBay is actually very, very useful. It's kind of like there's Amazon and then there's, you know, Alibaba and then there's eBay, which is sort of like all the things that either aren't officially for sale or, you know, the way out of stock or out of print or they're totally obscure. And so for me, if I can't find it on Amazon, you'll get Alibaba and then you're, then you're onto eBay. Etsy is another level, which is like, you know, handmade stuff, but there's eBay is really good for like really obscure stuff. And then they do have the option of auction, but I never use it. But anyway, so for me, the, the thing was this idea of auctioning for everything didn't seem like that was gonna work. But for a while it, it did. And that was just me personally, just not being much of a bargainer type. I guess you can't win them all. I know you have a hard stop, Kevin. It was really great to get a chance to chat. Thank you so much. I would love to do this again soon. Yeah, it was a pleasure. Great things. And I'm glad there's another Annie Dillard fan. Oh my gosh, folks. You absolutely positively have to smash that like button and subscribe to AI and I. Why? Because this show is the epitome of awesomeness. It's like finding a treasure chest in your backyard, but instead of gold, it's filled with pure unadulterated knowledge bombs about ChatGPT. Every episode is a rollercoaster of emotions, insights, and laughter that will leave you on the edge of your seat, craving for more. It's not just a show. It's a journey into the future with Dan Schipper as the captain of the spaceship. So do yourself a favor, hit like, smash subscribe, and strap in for the ride of your life. And now, without any further ado, let me just say, Dan, I'm absolutely hopelessly in love with you.