Beyond The Prompt - How to use AI in your company

Kevin Kelly’s Mind Bending Predictions of Our Fascinating AI Future

Episode Summary

In this insightful conversation, Kevin Kelly, co-founder and senior maverick at WIRED, explores the transformative impact of AI on society and creativity. Kelly discusses the philosophical and practical implications of AI, highlighting its potential to accelerate all other technologies and likening its impact to the Industrial Revolution's shift from muscle power to artificial power. He shares his experiences with AI-generated art and music, emphasizing the importance of doing something 'useless' daily to spark creativity and innovation. Kelly also delves into the concept of 'AIs' as a zoo of diverse, specialized minds and the need for new frameworks to address intellectual property in the AI era.

Episode Notes

00:00 Introduction to Kevin Kelly

00:32 The Future of AI in Interviews

01:42 Philosophical Questions on AI and Authenticity

05:00 Daily Art Practice and AI Collaboration

06:41 Exploring AI in Music Creation

13:47 The Profound Impact of AI

20:42 AI and Organizational Changes

29:50 The Challenge of Digital Memory

32:08 AI and Intellectual Property

36:11 The Future of AI Models

43:06 AI's Role in the Creator Economy

47:21 Kevin Kelly's Creative Process

53:15 Closing Thoughts and Reflections

📜 Read the transcript for this episode: Transcript of Kevin Kelly’s Mind Bending Predictions of Our Fascinating AI Future |

Episode Transcription

[00:00:00] Kevin Kelly: Kevin Kelly. Thank you for having me on. I am the co founder of WIRED and senior maverick at WIRED and, um, I'm an author of technologically cultural books and my name to fame is I'm a radical optimist. Optimist.

[00:00:18] Jeremy Utley: Kevin, that photo looks old.

[00:00:20] Kevin Kelly: Which photo?

[00:00:21] Jeremy Utley: The photo that's on your screen right now. I'm not seeing your face. I'm seeing your uh

[00:00:25] Kevin Kelly: oh. Avatar. Oh, I am sorry. We, I am really here, .

[00:00:31] Jeremy Utley: This is not ai.

[00:00:32] Jeremy Utley: Kevin, is that what you're saying?

[00:00:33] Henrik Werdelin: How long do you think it'll take us before we will just get AI Kevin doing these kind of interviews?

[00:00:41] Kevin Kelly: Um, Reid Hoffman just did an interview with his avatar AI self.

[00:00:46] Henrik Werdelin: I saw that.

[00:00:47] Kevin Kelly: Very convincing. So the answer to your question, I would say years. But you're not going to be happy with it. I mean, it'll be, you know, it'll, it'll be there, but who is

[00:00:59] Jeremy Utley: that for?

I wonder, you know, what is, who wants that? Who wants the avatar of themselves showing up at the podcast?

[00:01:07] Kevin Kelly: Someone who's in demand and, you know, you can have two versions. You pay a lot of money for the real me and you can pay, um, you know, a 10th of the price and it'll be one 10th as good. Uh, you know, I have a friend who's the doctor, uh, but you know, one of the concierge doctors and, he has a website for 20 years called, uh, ask Dr.

Green for pediatric advice. And so a version of him dispensing his pediatric advice. Would be valuable to many people.

[00:01:42] Henrik Werdelin: You know what?

[00:01:42] Henrik Werdelin: This might be too philosophical a question. I was doing a new a headshot the other day and as with anything I took. 40 pictures uploaded into a model got me in like all these different places with all these different clothes that I would never wear.

Um, and I showed it to my wife and she goes like, yeah, but that's not you. And obviously visually it looks like me. Right. And so we got into like a whole long debate about, you know, like the soulfulness of stuff, you know, like, and, and how It is interesting how we as human are just so good of decoding Ripple and all these other things and, and how like it's difficult right now for, for at least AI photos to kind of like give that, that.

[00:02:31] Kevin Kelly: I think that, the it that we're describing will forever be, um, evident if you really care. And, and, and the point is, is that oftentimes we won't care. Like in the movie. I don't really care whether it's really the actor or an avatar. I don't care. If I can't tell, I don't care. Um, if it's news, and a purported news event, I do care.

If it's interacting with a friend, sometimes I may care, and other times, if you have the avatar and You know, whatever, I may not care. So, I, I, I think, I think, um, I think it's going to be about what we, you know, what the situation is and what we're using it for. And I can imagine lots of uses where a version of me would be better than no me.

[00:03:35] Henrik Werdelin: It's reminded of somebody, I heard this joke once when somebody asked, how's your wife? And the answer was compared to what, um, and I think that sometimes it's also when we now talk about AI doing work and people say, Oh, you know, like. It's not as good, and you're like, compared to what, right?

[00:03:55] Kevin Kelly: And that's actually one of my principles in technology, when you're, we have a new technology, we always have to compare it to the old one.

If it's causing harm, yes, compare it to what? Compare it to the old one? You know, I mean, So, uh, yes, um, people died on the highways, through cars, compared to what? How many people died on horse carriages? How many deaths were there? You have to actually look. Uh, yes, mercury amalgam in your mouth might cause some Problems, but compared to what compared to cavities, cavities are far worse.

So you have to compare to what is a good principle.

[00:04:41] Henrik Werdelin: I'm reminded sometimes when people say chat, you be T mix up, you know, like it's not that brilliant and you go like, well, neither is most people. And so, um, yeah, well, we kind of jumped a little bit into the conversation, but thank you so much for joining us.

[00:04:57] Kevin Kelly: I'm delighted to be here.

[00:05:00] Jeremy Utley: You and I, Kevin, we spoke several months back about your kind of, Uh, you know, daily art practice, I'd love to hear how it's, how it's going and what you're learning and what, what are the edges that you're exploring these days?

[00:05:15] Kevin Kelly: Well, so one year I did every day I made art myself with a procreate on the iPad. And, and other, occasionally other physical things. And then last year I did one piece of art a day with, um, co created with AI. Um, that is finished and now I am doing a, um, a design tip for my photographs every day. However, just this week I started playing with the music generators and oh my gosh, they are amazing.

They are like mid journey in the sense that you're not just getting a little, you're getting a full song with lyrics and bridges and chorus. It's, it's astounding. I had the same kind of obsessive late night. I can't stop. As, um, feeling as I had with Midjourney of making the visuals. And, um, for me, I can make a picture.

I can paint. I can draw. I can photograph. So Midjourney was kind of cool, but it was doing stuff I could do. I cannot. I have no musical ability at all. So this is a real superpower. Um, and I feel maybe a little bit less like I'm making it in a weird way. Um, because I have, I'm so ignorant. Can you talk,

[00:06:41] Jeremy Utley: can you talk about that, that ignorance actually, before you get to the, however, if you'll hold that, because I know, I know that we've talked a link about the fact that your knowledge of make of art, making visual art, making painting, et cetera, informed your ability with mid journey.

How have you found your kind of call it, uh, ignorance of song making, has that been an asset or a liability or how do you, how do you think about that?

[00:07:07] Kevin Kelly: What I think will happen, even in the future. little while that I'm doing it is that I will learn. It's a learning mechanism. Um, I, I, if I do this long enough, I know that I will have more musical awareness, more musical education, more musical knowledge than when I began.

I had, I was recently, there was someone who was doing programming using the co pilots like, you know, um, GitHub's or, um, Microsoft's co pilot to help them do the, um, the coding. And they found that they actually said. That having it kind of auto complete, the, the, the, the coding, actually they were paying attention and they were learning how to do it themselves from watching it being done.

And I think, I feel a little bit that that will be what happens with the music, as they get on to it, and particularly if, um, if my prompts, if the prompts are more musically, um, inclined, you know, like playing with beats or, you know, meters or chords or things like that. If, if you get it, I haven't gotten that far to know whether those are some of the controls that you can use in your prompts, but if, if that's where it's headed, then, then I think this would be one way to learn about how to, to compose music.

[00:08:39] Henrik Werdelin: Can I ask maybe a question on that? Never thought about it this way, but we obviously talk a lot about AI and how we will reduce the need for writing traditional code, but one thing you bring up, which I think is super fascinating is that in most domains, there are specific terminology and there is words for how you describe different actions that you only really know if you're in that space.

I'm sure. Somebody who's been a jazz musician for 30 years will be able to say, he or she wouldn't say, Oh, I want it to be like, doom, doom, doom, doom, doom. They will like, have like some kind of way of articulating that specific phenomenon, right? I wonder your perspective on that becoming the new form of coding.

[00:09:24] Kevin Kelly: Yes. Um, people talk about the fact that basically the English language is now the coding programming language, that, that, that. And then, um, for the music. The answer is really how extensive is the training. In other words, the reason why the English language prompts would work is that there's some degree of people talking about music with words, so they can link the two.

So, is there enough data of people doing do do do do do that's connecting to it? I don't know. And, and you would need to have a large source of data like that to train the set to be able to do that. Um, so, so that is one of the issues about music is how much of a correlation, how much of a linkage is there to English language.

talking about it. And there's lots of music review sites and things like that. So there might, there might be, um, um, enough.

[00:10:35] Jeremy Utley: Tell us, will you tell us a little bit about the last, um, you know, you said just staying up all night, just like, you know, with mid journey, walk us through your kind of early explorations with music creation.

And are you, what, what kinds of prompts are you giving and how are you evolving your collaboration

[00:10:53] Kevin Kelly: with the AI? , I'm very convinced that, um, the center of our culture is in, um, moving images and videos and not in text. So my kids, their friends don't read books. They watch YouTube like I do. And I'm, I'm trying to make more YouTube myself and in trying to make the things new to, I found that for me, the way I thought about it, I wanted to start with the soundtrack, start with the music.

and overlay the images to that. And so what I'm trying to generate are basically like little TikTok segments that I can overlay images over and edit, edit to the soundtrack. So that, so that, that works because right now these, the, the version I have only makes it, it's a two minute limit on the, the cut.

I'm going to actually play one of them. And, uh

[00:11:53] Jeremy Utley: Love it. Please do.

[00:11:54] Song: Lost in the darkness, a soul without sight Whisper in the wind, guide me through the night A simple angel's voice, a hymn of pure light Lament and praise, in harmonies ignite

[00:12:23] Jeremy Utley: that's

[00:12:23] Kevin Kelly: amazing. Anyway. So, um, I'll turn it down.

Um, I don't know if that, can you hear it?

[00:12:32] Henrik Werdelin: Not anymore.

[00:12:33] Kevin Kelly: No, but you heard it before.

[00:12:36] Henrik Werdelin: Yeah. It was very beautiful. Kind of like Irish inspired. Yeah. So those,

[00:12:41] Kevin Kelly: those are some of the prompts that put in was, you know, Irish lilting, ethereal, angelic voice, uh, a ballad and, um, and, and they wrote the lyrics. It was like, it's amazing.

Which service did you use? This is Sona.

[00:13:02] Henrik Werdelin: Oh, yeah. It's amazing.

[00:13:05] Kevin Kelly: Um, so, uh, yeah, and, and that little two minute thing is enough for me to kind of build a kind of a nice little graceful thing. I had some, some kind of natural things, my granddaughter twirling, you know, it'll, it'll be perfect for the little vignette that I'm wanting to make.

And, and also the copyright things is completely solved.

[00:13:33] Henrik Werdelin: You've, uh, you've been a Observe of technology, you know, for a while, I often get asked the question of how fundamental of technology shift is AI.

[00:13:47] Henrik Werdelin: To compare to, let's say it's just things that we see now generation, the internet, mobile computing, what have you, where are you on, on answering that question?

[00:13:57] Kevin Kelly: I would say it's, it's by far the most profound thing we've ever done. It's, it's equal to maybe fire and language, I think, but I like to make a caveat. I think it's going to take many, many, many decades for this to play out. I think it's going to take 10 years for the existing models that we. We have today to be consequential, to actually enter into our lives.

I think it's gonna take 10 years for self driving cars to, to penetrate. So, so, um, so this is not, so we have time to adjust. We have time to learn to train it and, you know, align it and all, everything else. Um, so even though the, the, the laboratory edge of this moves very fast, it's like, You know, cell phones were invented decades before everybody had them in their smartphones.

It's the same thing. This is going to be decades before. This really has economic, huge economic impact. So somebody,

[00:15:13] Jeremy Utley: somebody who's listening, like, just, just imagine a, not a naysayer, but just someone who's, who's, who, uh, is starting the blank slate. So they're not a critic, but I can imagine someone going, wait, Kevin, he just said fire, like walk me.

If I'm, I'm not critical, I'm just. Right. I'm just oblivious. Why? That, that seems kind of foundational. That's like civilizational, you know, civilizational. That's a hard word to say. It's, that's a, that's a really big thing. Why?

[00:15:47] Kevin Kelly: It is because it's an enabling technology that basically accelerate everything else we do.

So it will accelerate all other technologies, um, besides itself. It'll work to accelerate, you know, material science, chemistry, physics. It's, it's gonna touch everything, almost in a way that fire doesn't. It's closer to language. In the sense that, um, we, we, we, you know, language is infused in everything we do.

So, um, so there's both the AI itself, directly working with it and what it does, and then the way that it enables and accelerates everything else in the world. And

Civilization is this collective of, right now, billions of humans collaborating in some capacity. That's collaborate to work, to build roads, and share. And so this is massive collaboration. And what we're doing is we can add, you know, billions more minds. Maybe it's the same, it's the same level or maybe higher or different and we, well, they'll help us collaborating.

Well, they'll, together working with AI, we can solve problems that we cannot solve by ourselves. We can, we can figure out things. with A. I. s working with us that we cannot figure out by ourselves. There are certain quantum gravity, or dark matter, or whatever, may need other kinds of minds that we make in order to be solved.

And then, of course, having those technologies will just enable all these other things. And so, it's, it's, It's sort of like, I, I, I think of them as artificial aliens, and so it's like making contact with another technological civilization, and that we're going to have use of everything that these aliens discover for us, with us.

[00:18:00] Jeremy Utley: You know what it reminds me of, not, not to be, um, grandiose, but as you describe it, and as I try to listen as someone who is oblivious, I think of the wheel. I think of this enabling technology that it's hard to imagine in the area of life it doesn't touch or that it doesn't impact.

[00:18:19] Kevin Kelly: Sure. Yeah, well, another familiar transition is, is the Industrial Revolution, which was based on artificial power.

Okay, so what do we mean by artificial power? For the, until the 200 years ago or so, everything that you made was made with muscle. Either human muscle or animal muscle. So the pyramids were built with muscle power. They were the biggest thing ever made with muscle power. But if you wanted to make a road, you needed muscle.

You needed human muscle or animal muscle to build a road. It was incredibly limiting. And then, um, if you wanted to, you know, build a house, you could only make it so tall because just lifting things higher with muscle was really hard. Well, the Industrial Revolution made artificial power. Where we had wind power, water power, eventually coal power, oil power, and suddenly energy was cheap.

You could throw a skyscraper up. You could make miles and miles of road. You could, um, create things that would just take so much calories from a human that it was just unfeasible. And that has transformed everything. So that cheap. energy is what propels our factories, propels everything. If we had revert back to just using muscle power for everything, it would just be, you know, we'd be back in the stone age almost.

And so, so, so, so we've, we went from natural power to artificial power and that transformed our, our, our world. And now we're going from natural intelligence to artificial intelligence. And that's the second massive enabling technology that will touch everything in the same way that cheap power, artificial power has touched everything in our lives.

There's nothing that we do. We're not using cheap artificial power. I mean, you have electricity every, every, everywhere.

[00:20:42] Henrik Werdelin: So when you are thinking about it and getting the sense that it'll take longer than what some in the industry predicts. Is that because of AI not moving fast enough or because that humans won't be able to interface, absorb it quick enough?

[00:21:04] Kevin Kelly: It's a little of both. Um, it's, you know, it's like, it takes, so one of the things we know about, like going back to the metaphor of the artificial power, um, in order to bring electricity into companies, You couldn't just sort of replace a worker with electricity. The entire organization of companies was reordered by electricity.

You had to change the structure of the company, the physical nature of it, you know, um, the organizational structure, all changed. When you had artificial power. And the same thing with AI. You can't just bring AI into a society, a company, and think it's going to replace people. We have to change the entire organizational structure.

We can't just take cars and replace them with self driving cars. It's going to require An actual restructuring of the highway system, the lights and ramps and everything else would have to be adjusted to it. So that's what takes the time, is the absorption of it. Because, um, you have to, it's a different metaphor, there's different organizational dynamics going on.

And so, we have to absorb the AI into our lives. And it's so, the more powerful it is, the more time it's going to take.

[00:22:37] Jeremy Utley: Do you have any, do you have any kind of inklings on those organizational changes? When you think about structural changes that will need to occur, do you have any kind of glimmers of a vision of what some of those changes might look like?

[00:22:51] Kevin Kelly: Well, one of the things that happened with bringing in, um, like the first generation of, of, um, communication things was that there was a flattening of organizations. Like if you, if you're a, If, if you're running in an army and, um, all your, all your, all your soldiers are illiterate and uneducated, uh, and there is no, uh, there are no phones or, or wires.

The best thing to do is to have a very hierarchical thing where the, where the commands come from the top and they go down to the soldiers, and that's it. But if you're in a world where, um, there is communication between everybody, including soldier to soldier, you can have a much flatter, organization because the information, um, is not just flowing one way.

You can flow up and down and across and that becomes much more valuable. So you flatten hierarchy becomes less important. And that's exactly what we see in companies when they, when they moved from very traditional companies to companies that had telephones and, uh, you know, communication and, and, and, and smartphones now.

And so I think, um, I think that, I think that flattening is one thing that will continue, um, where the, um, everybody in an organization will have the same information as anybody at the top. And so, um, you can distribute the decision making process. And maybe with an additional of AI, you can have more executive function decisions being made.

Maybe there's some way to replicate that. the, um, the leader's decision making into something so that the other people have direct access to that without having to have reports, so to speak. So there, there could be changes in that span of control. Um, structure that would be evident in an AI based company that's not in

[00:25:06] Jeremy Utley: CEO the other day who's done a lot to, um, to train a model, an internal model on his memos, annual reports, company documentation, et cetera.

And if an employee comes in with a question, his first question is always, have you asked my GPT? And he said, most of the time, if someone asks his GPT there, they have what they need. And so he said, whereas before I was, I was dealing with so many one off questions where the information's only in my brain, it's actually not in my brain, but nobody knows where to look for it.

Well, if I train an LLM on that, knowledge base, now folks can query that. I mean, we talked to Guy Kawasaki the other day. He said that Guy Kawasaki GPT is better at being Guy than he is. He's like, he goes, it gives better answers to my own, to questions about my own stuff than I do. Right.

[00:26:01] Kevin Kelly: Yeah. I took all the things I'd written, all the articles, all the books, everything I've written.

I put it into, um, a, um, Google LNN notebook. And, um, there was like a half a million words. So it's a lot. It's a big context window. And so, um, uh, we're using that to, for me, I'm using it to kind of remember what I said somewhere. Like, you know, just, just as like, What do I think about that? I don't know. I'll ask my, it wasn't GPT.

I'll ask

[00:26:33] Jeremy Utley: myself. I'll ask myself what I think. No, I mean, we were actually joking with Guy. Who's it for? Because you might say it's for the world to interface with Guy, but at least for me, granted, I've got a lot less kind of material to train on. I have trouble remembering what I know. And so I just need it for myself to go, what's a story I know about this principle or about this example or what research supports this point.

I, I've actually encountered that. My trouble is I don't know where to retrieve it in my own head or documents, right? Which is to say, it's not about the public being able to interface what I know. It's about me being able to interface what I know.

[00:27:07] Kevin Kelly: Yeah. And, um, Stephen Johnson is the guy who's doing it.

This project at Google, he works for Google now doing this, um, notebook project. And he, um, also has taken all his, um, besides taking. You know, all his corpus and putting it in and asking himself these questions. Um, he also uses it for all his notes and, and quotes and little common book things that he's, so he takes all his notebooks and put it in, which is incredibly powerful thing to do.

Because, you know, you're taking notes, but how often do you go back to looking about here? They're all there. And it's very powerful. Accessible, what's the word I want? Findable, discoverable way, and in a very elevated way because you're not just pointing to it, you're synthesizing an answer to your queries.

[00:28:02] Jeremy Utley: I've long been fascinated by his practice, his obsession with, I think he calls it tools for thought, right? I mean, he's long been obsessed with this. And it's really interesting to see Notebook LLM.

[00:28:14] Kevin Kelly: Yeah, he went to Google originally to work on writer's tools, and then while they were doing, or developing it, the LLMs came on the scene.

So it was not originally an AI per se, or an LLM project, but it became that very, very quickly. I'll say one other thing, and that is, you know, the, um, there's about 30 million images generated per year. Maybe more these days. This is a couple years old when I did the last count. Uh, with image generators like Midjourney and Dolly and There's probably 50 million at this point.

A day. Images a day. And 98 percent of those images are for an audience of one.

They're for the people generating the pure pleasure. It's like me with my, um, making the music. It's, most of it is for the pure pleasure of The co creator and, um, and I think going back to what you're saying with these, um, personal. Gbt is that most of the, the customer is mostly going to be a 1. A customer of 1,

[00:29:32] Henrik Werdelin: I've been thinking a lot about how you do this with emails because. A lot of my world documentation is now. Stored there, right? And same thing, you know, like, from the basic of. I talked to a friend who's a senior recruiter, big recruiting firm.

[00:29:50] Henrik Werdelin: And she was saying, she's getting the same email basically every week from her partner saying, you know, does somebody has somebody that's profiling or like, we literally just talked about that last week.

Right. But even down to the, the unsearchability of your own digital archive as it is right now, like if I was to buy a new. camera, I bet you, I, over the last 10 years, I've talked to two or three people I know well, who are obsessed about cameras, probably don't do it professionally, but they know everything there is to be know about cameras.

And I have no way of really thinking about who they are. I just don't recall them if I don't, if I don't store it in memory. And so it'll be fascinating as these LM started to mine our digital lives to help us do all these different things.

[00:30:41] Kevin Kelly: Does anyone doubt that Google will mind Gmail for you individually, each person saying, here's your entire Gmail, which is now accessible with this ai.

It's like, that's such a no brainer. Um, and, and by the way, I know the superpower that, that Google has, I think it's unappreciated, is YouTube. And, and they'll sort out, I mean, they will, they will, they will, they will train one way or the other. They will use that. To train, um, AIs to semantically parse the frame by frame.

Every single frame will be understood by the, by the AIs. They will, They will memorize and understand every single frame in every single video. So you can ask it

[00:31:40] Jeremy Utley: that, that, that makes me think, you know, I've heard recently about some of the, you know, the Sora controversy with open AI. Have you trained it on YouTube?

Right. And we've even, I've heard about, you know, internal kind of arguments at Google because they don't really want to pursue open AI perhaps because actually they're doing the same thing, but Google actually has, there's kind Walls are supposedly walls inside where they're actually in a way they're they have less access to YouTube.

Maybe then the public does in some sense. How do you earlier earlier?

[00:32:08] Jeremy Utley: You mentioned the challenge of kind of call it copyright with with there's no copyright infringement when it comes to creating sound to accompany my moving pictures. What's your thought on the question of? Intellectual property. I realize that's a complicated question, but do you have any kind of frameworks or heuristics for navigating those questions?

[00:32:31] Kevin Kelly: I would say that, um, copyright is broken in terms of trying to be a framework for, for this stuff. I think the best we could do is develop a new framework, what I would call the reference right, right. Or referencing it because there's no copies involved. And so there might be, you know, a new idea of the right of reference.

And what are the rights and obligations of, of referencing things or having things be, be referenced. And, um, I haven't seen very much movement on that. But, um, the idea would be that in addition to copyright, there's this other thing, rights of reference. And, um,

it, it might be partly payment and partly control. Like, you know, like opting out. You know, those kinds of things. Or, um, The problem with payment is that the,

the, the degree, the, the, the, there's not that much money being made in this per, per image input. And so, um, yeah, it's, it's more of, it's, it's probably more going to be along the lines of, of controlling credit or something. I know Microsoft and Darren Lanier have been working on trying to do that. Reverse engineer the influence of a particular image on another image.

I don't think that's going to work, but I don't know. Um, there might be some things like, like that, that technologically are capable of giving a ranking, a probabilistic number that, um, this image has a high probability of using these other images above a certain level. But the problem is that those images themselves were influenced by other images.

Yeah. Yeah. I was,

[00:34:42] Jeremy Utley: I was going to say, I was going to say, just take it conceptual for a second. I think one of your biggest ideas, probably one of your biggest, most influential ideas is the notion of a thousand true fans, right?

[00:34:52] Kevin Kelly: Sure.

[00:34:52] Jeremy Utley: I can imagine a world where you could imagine where anytime a thousand true, that phrase called a thousand true friends is mentioned.

There's like, if you hover over it, it says Kevin Kelly. Right. But then to me, the interesting thing is. You got it somewhere. You got, you were inspired by some version of that somewhere else. Right. And does the, does the point of inspiration upstream from your thinking of that way of putting it? Does that get credit?

Right. It's almost like, where do you, where do you define origin is a really tricky with concepts and images and styles. It's like, there's so much influence. How do you give proper attribution?

[00:35:26] Kevin Kelly: So, so I, so I don't think. Getting down to the level of individuals being credited is going to really work.

However, there might be something where you can, you know, like, like robot protocol, you can opt out. You have a right of, uh, your right of reference, then you can opt out of it. And I think that might be, um, maybe as far as it goes. Um, but in any case, I think copyright's not going to be the frame, the legal framework that will work for, for this.

And so far the courts. They've also been rejecting the copyright arguments because there's no copy being saved.

[00:36:10] Kevin Kelly: So, so that's the weird thing about these ILLMs is that, um, I, I'm, I talk about the latent space, but you know, they, they take, they're huge training sets, right? They're, you know, it's like billions of parameters, incredible number of tokens, this massive, massive amount of data and, and massive amounts of compute, a hundred thousand.

You know, deep GPUs, and then they compress it, they compress all this knowledge to a very tiny little, uh, you know, 100 gigabyte file. Okay, so in the 100 gigabyte file is all human knowledge in it is every single face, but it's not, there's not a single face and it's just all the information about every face.

So, there's no copies of Alice in Wonderland, there's just all the information about it. And, it's, it's kind of a weird magical thing that there is this compression, and I think that that latent space is itself going to be a medium for, for something. Just, just that compressed version of all that we know.

Um, and, and I just saw something last night that was really interesting. There's a guy at OpenAI who does all these trainings, and he says that, In the end, the results coming out of all these different models. Don't matter on what the models are. They only matter on what the data source was. So they say that the models converge on their output where they diverge is where they begin with the data.

On that note, like you, they kind of, in other words, say that another way of saying is, is they converge onto the data, the distinction converges onto the data's data sources. And so. So we're going to see lots of different models here on curated data

sets. So there'll be, I'm sure there'll be people who make conservative AIs, conservative large language models, that are trained on only a very curated set of information to begin with, and that will produce a certain kind of Of intelligence in mind. And so I think, so I think of this as educating you, you're sending your AI to school.

It's like, what kind of school is it? Do you feel already

[00:39:05] Henrik Werdelin: that we, uh, I read somewhere that you were a student of kind of Asian culture, but surely very aware of the difference between American culture and rest of the world. Do you feel that those, the models, because a lot of it is trained in English language is already having that kind of influence?

[00:39:27] Kevin Kelly: Yeah. I mean, the Chinese have been very deliberate to try, you know, like there's only, I think I saw some figures like, you know, 0. 1 percent of the material was Chinese that, you know, in the open AI version. And so the Chinese are kind of like, this is unacceptable. We, you know, so they're, they're not just a hundred percent Chinese, but they're a majority of, of, of stuff in Chinese.

And that will produce a different. a different mind that will produce, um, uh, you know, a culturally different kind of, uh, of answers. And, uh, and, and I think, um, that it'd be interesting to see what happens, um, when these, and part of it will be, You know, the market sorting out and people, and also I think what, what, what, what they're used for is, is, I mean, I always insist that we talk about AIs plural.

There are hundreds right now, there'll be thousands of different species of them doing different things. We're going to engineer them for different things, different purposes. And some of the AIs don't need to answer questions. You want a A. I. for a strawberry picking robot. It doesn't need to know anything about mathematics or philosophy.

It, it, it, in fact that's a distraction. It wants to be the world's best strawberry picker. And maybe a general farm picker. But it doesn't need to know very much philosophy. And so, and, I think the Chinese are going to be very good. And making those kinds of AIs. They don't have these issues of um, political um,

[00:41:07] Jeremy Utley: Have you all, have you all seen stuff like this?

Japanese image models coming up and things like that where they're a little bit more tailored perhaps to some cultural inputs and what's your, what are your thoughts on things like this?

[00:41:21] Kevin Kelly: This is inevitable, as far as I'm concerned. , I think the picture you want to have in your mind is a zoo of hundreds, if not thousands, of different species of AIs, some which will have huge memories and very short attention spans, and others will have long attention spans and small memories, and some will have really fast and deep memories.

And some will, have little glimmers of consciousness, and some will be as smart as a grasshopper and ubiquitous, and So, so, so they're just, the, the possibility space of minds is very, very large, and we're not at the center of that space. We're off on the edge, our, our combination. So they're, what, what we're going to be discovering is, is different.

elemental cognition primitives, elements that are recombined into compounds, into different mind complexes. And we have a peculiar complex of, elements in our minds that, , would be hard to replicate in silicon. So, uh, we're gonna just make all these other varieties of, of thinking and cognition in order to do different stuff.

And sometimes it'll be very abstract, like solving quantum gravity. And, um, other times it'll be very, very basic of like super perception for picking strawberries.

[00:42:54] Henrik Werdelin: In terms of, uh, I had one more question on a thousand true fans. , In terms of AI's influence on getting people to a thousand true friends, Where do you think, and maybe this is kind of like a an odd question, but is it in the zero to a hundred or is it in a hundred to the 500 or 500,000?

Where, where do you think AI first will allow people to help?

[00:43:19] Kevin Kelly: Actually I just had a conversation with Jack Pontier Paton about this. I think the place where AI. We'll have this impact in this kind of creator economy is in the matchmaking and collaboration field.

So, um, according to my theory of a thousand true fans is, is that, um, even if you're weird obsession or idea or business appeals to only. One in a million people. In a world of several billion people, there are going to be one thousand of those, one thousand people will be ready and eager and desiring your weird thing.

The issue is like making that match. It's like them finding you, you finding them. And, and, and here's, and here's the challenge. The challenge is, is that those thousand people may not even be aware of what it is that they are looking for until they see what you're, what you're doing. You're doing. So you can't just like, have them have a better search thing.

'cause, 'cause they, they aren't searching for it indirectly. They're searching it, kind of like wanting it, but not knowing what it is that they want. And so AI could be instrumental in making that match happen in the sense that they could it, that the ais could, could, could say, given all the other things that you're interested in, and the fact that I know that this is happening over here.

You might be interested in, in this, even though you've never thought about it.

And then, I think also, it can be instrumental in kind of doing those kinds of things to help people make. New stuff that hasn't been made before.

[00:45:32] Jeremy Utley: I wonder if there's some part of what you're getting at. I think Kevin is something like a discovery mechanism marketplace. You know, Patreon is a good kind of analog, but that's, I think a sliver probably of all the market spaces that need to be created.

And I think, uh, you know, someone said, Henrik, I can't remember who said it to us, but maybe it was Ethan Mollick. He said, we're going to need to be marketing to AI's. I think, and I thought that was a really interesting question of how do you start to market actually not to people but to AIs? Um, I don't know, Kevin, if that sparks anything for you.

[00:46:06] Kevin Kelly: Well, I, I think it's much more likely that our AI agents are marketing to each other. I mean, that's, that's very clearly that we have, uh, most marketing will be AI to AI, agent to agent, and behind the scenes. Um, If, if, if, if we're, if you're marketing to AI, it's just for sure that you are having an AI do it.

So, uh, it's, I think it's going to be symmetrical in that sense. It's like, um, for an image generator to generate an image, it has to be able to recognize the image and see the image to perceive it. So, its ability to perceive things is equivalent to its ability to generate. So, there's a symmetry there that I would expect.

If you're marketing to mostly AIs, it's going to be mostly AIs who are marketing it.

[00:47:01] Jeremy Utley: Can we shift maybe just, you know, we've kind of gone philosophical. I want to get back to the hyper practical as we start to wrap

[00:47:09] Jeremy Utley: . When you think about your own practice and inspiring your experiments and dabbling, can you talk about how do you go about seeking inspiration and where do you go to get sparked and provoked to try new things?

What's your own practice for discovering new things to try?

[00:47:29] Kevin Kelly: I would say there's a couple different sources, but I would say one of the, um, one of the newest sources has been YouTube. I am deep, deep into YouTube, and I actually find the YouTube recommendation of, uh, very good of recommending things that I knew nothing about that I might be interested in. And it's, you know, it, it's It's knowledge of me is very minimal.

It's just basically from what the other YouTubes that I watch which is just one dimension out of my multi dimensional life. So it, you know, it's in theory, it could be a million times better, but even as it is, it's, it's pretty good. And, um, uh, you know, then just in general, you know, I have a two story library of stuff that I look at to, um, be inspired and, and, and to, you know, start down different directions.

Um, and I, and I, and I think, uh, the, the, the best way to kind of get somewhere that someone hasn't been before is, is, is not by thinking about them, but by, by, by, by doing stuff. So, so I like to think by doing. I like to do my way forward rather than just think my way forward. Um, you know, it's like you, you do something and that the doing of it generates a new idea and then you do that and then that.

that doing generates something else. And so, um, as a maker and a doer, that is often a very productive way to, to, to arrive somewhere where it's the, it's the act of making something that then sparks the next idea in the next direction.

[00:49:38] Jeremy Utley: And do you block time on the calendar for that? I mean, I know you're, you know what I mean?

Like, how do you preserve space? Because I'm sure you get invited to way more things than you can do and way more, right? So how do you protect it? Is it just implicit now at this point?

[00:49:53] Kevin Kelly: Yes, it's mostly because I'm not a very good, um, scheduler of things. And, um, I'm a little bit more up to opportunistic in terms of the creative part.

Um, I'm very good about saying no to things. Um, and so, somehow that combination works for me. So I'm not, I'm not a Tim Ferriss who's, who's planning out the, the number of hours for this and that during the day. Um, I have something that, I have some things I need to do and then I, I try and, you know, spend time doing something useless is the way I describe it.

I want to do something useless every day. So

[00:50:46] Jeremy Utley: that's such good. Can it, I don't know if that's one of your pieces of it. I don't think it's an advice for living, but what a great goal. You need to add that in the third edition, my friend, because there's one end of the spectrum, which is try to do something useful every day.

I think a lot of people go, absolutely. You know, very good advice. Right. But to say, try to do something useless every day.

[00:51:05] Kevin Kelly: Right, right, right. Yeah, that's, and, and so, um, yeah, I have, I had this slight, not an argument with David Allen that, you know, get things done, and a guy, and, and, and my thing is that, um, rather than structure my life so that I spend as little time on doing things as possible, I want to I structure my life so I do things where I want to spend as much time doing them as possible.

I want to do the things where I just keep wanting to do it. I don't want ever to stop. That's, that's what I want in my life. I'm not trying to, you know, make everything efficient. There are certain things I want to make efficient, but I want to do that only so that I have time to really spend most of my day doing things where I want to spend more time doing them.

[00:52:00] Jeremy Utley: It's beautiful. It's beautiful. For every get stuff done, there needs to be a leave things undone, perhaps, or something like that. No,

[00:52:08] Kevin Kelly: it's not leave things undone. Or find stuff to

[00:52:09] Jeremy Utley: do, right?

[00:52:12] Kevin Kelly: It's, um, yeah, what's the alternative? I mean, you know, he agrees with it. It's just a little shift. It's saying, um, Yeah, uh, um, spend more time, spend more time doing things.

I don't know. Get things

[00:52:32] Jeremy Utley: done versus find things to do. There's

[00:52:36] Kevin Kelly: a balance there. Let's do the fun things. Uh, or only do. Only do the fun. I don't know. It's, it's, it's, uh, you can't just only do the fun things. You obviously have to do things that you, um, would rather not do. And those you do want to minimize.

[00:52:55] Jeremy Utley: Thank you.

[00:52:55] Kevin Kelly: Thank you guys for a great questions and I appreciate the opportunity to share, um, some half baked ideas.

. Bye bye. Bye.

[00:53:04] Jeremy Utley: . Henrik, how do we, how do we wrap this fantastic conversation?

[00:53:08] Henrik Werdelin: . I'll throw you a few things that stood out to me and then love to hear you play back to, um, I mean, his whole analogy on the muscle and the artificial muscle and thinking about it in that way and think about artificial as kind of like something similar, I thought was such a powerful kind of.

Way of thinking about everything that is happening and is about to happen in the world because of AI, um, you know, connected to then obviously the old charge and, and the old change and how we can't just think about it in a very one dimensional way. We have to think about how will this change kind of like the environment around everything that's getting introduced, which.

I thought was very interesting. And then I think the third one was this idea of thinking of AI, not as the AI, but as a zoo, like of an, like an alien technology or an alien with all the permutations and the personalities and the specificities of, you know, what a species might look like. I thought that was very, very fascinating.

[00:54:16] Jeremy Utley: Yeah, he's, I've, I've often had a chance to talk to him and he's, he's often emphasized the idea of AIs. We should not be using the phrase AI because the truth is there's going to be thousands. And that came through very clearly. And I really appreciated his description of, you know, some have a lot of memory.

But little, you know, context. Some have tons of context, a little memory. Some like grasshoppers, they proliferate. I like the zoo metaphor. I think it really helps us understand the characteristics. And then even that to put ourselves in this kind of the possibility of mind space, we occupy some space there, but there's probably many kinds of mind that could live in that space.

I thought that was great. I also just love his insatiable curiosity. I mean, I think it's made him. And his career and his reputation forever. But the fact that, you know, last year when I spoke to him, he was obsessed with image generation, now he's obsessed with song generation. Now he's moving more towards YouTube and understanding kind of videos.

And I think there's really something special about that, that attitude. He mentioned at the end, do my way to learn versus think my way to learn. And, um, and of course, I mean, you knew it was like catnip for me when he said, do something useless. It's just so, it's such perfect kind of counter conventional wisdom that I think more, more people need today than ever before, because all of the conversation around AI is about efficiency gains, productivity, blah, blah, blah, right?

Which is, it can be wrongly reduced to be more useful. I think what Kevin was saying is the reason to be efficient is so you can be useless. It's so you can do useless things and we do useless things to discover and spark our imaginations and tinker. And it ends up actually being what informs and, and inspires useful things.

[00:56:12] Henrik Werdelin: Yeah, I think in a slightly different way of saying the same, , but kind of think about it a slightly different way is. As you know, I'm deeply inspired by Kenneth Stanley, uh, the professor and author of, uh, White Greatness Can't Be Planned, and he will, I think what he would, he would not even define this as useless things, he would think about its interestingness, and that by Pursuing what you considered interesting.

So interesting this with great intensity, you'll kind of like map out your path. Like you would put down the stepping stone in front of you to something that's incredible. And so I think he, you know, he would even object to the world of calling it useless, but it might not have a specific objective at the moment.

[00:56:57] Jeremy Utley: Well, I think what a word like useless does is it gives permission to break your own rules. Interesting is kind of like it can almost, I mean, I, I agree. I think they're roughly synonymous. Interestingness, I have to have some amount of liberty that feels like an indulgence. Right. To be, to do, what did he say?

I wrote it down. Try to do something useless every day. That feels like a counter cultural, rebellious, it's, it's a, it's a war cry against the establishment. Interesting is like a luxury. Be, try something useless is like, It has some kind of revolutionary vibe

but I think if you tell an American, for example, be irresponsible, it's like it's, it sounds like try to do something useless, right? There's like a, there's a rallying cry there that I feel is a helpful counterbalance to the tyranny of productivity that's being leveled upon us by the technology. Complex.

Rant over.

[00:58:00] Henrik Werdelin: Maybe podcast over to, um, did you have anything else that you wanted to bring out?

[00:58:06] Jeremy Utley: No. Thanks folks for joining us. A lot of fun to have you on this adventure. If you enjoyed this episode, please hit like, please hit subscribe. Please tell a friend, please tag Kevin Kelly on Twitter and let him know how much you appreciated his zoom metaphor, doing his way forward, AIs, plural, and anything else that struck your fancy until next time.

[00:58:25] Henrik Werdelin: That's awesome, man. This is a great episode.

[00:58:28] Jeremy Utley: Yeah, he's cool. He's really, he's

[00:58:30] Henrik Werdelin: incredible. Thank you so much for, uh, for sorting that out.