Ben Goertzel: 0:00

I mean no one was filming large objects like this, but they were just like, get rid of this robot and bring on the people.

Joeri Billast: 0:06

Hello everyone, and welcome to the Web3 CMO Stories podcast. My name is Joeri Billast. I'm your podcast host, and today I'm very excited, very honored to be joining by. Ben, hello Ben, how are you? Very good. Guys, if you don't know Ben Goertzel, a big name. Ben, you're already for decades in uh in AI, eh? You're the founder and CEO of Singularity Net, chairman of the Artificial General Intelligence Society, an OpenCog Foundation, and vice chairman at the nonprofit Humanity Plus. You led a software team behind Sophia the robots.

Ben Goertzel: 0:41

That's right. Last time I was at Web Summit was with Sophia actually, and uh we were on the center stage uh sort of preaching the gospel of uh beneficial human-robot relations. And I had Sophia's little sister Desdemona here. This year we did we did a little bit of music performance and so forth.

Joeri Billast: 1:03

Okay, I missed that. But I I remember that speech, you know, that that talk it's with Sophia. Now, you have been working toward artificial general intelligence, AGI, for decades. When you look at the current AI landscape, what do most people misunderstand of what AGI truly is?

Ben Goertzel: 1:21

Well, I think none of us are gonna understand AGI really until it's there, right? This is a fundamentally new fundamentally new technology. But I I do think that uh LLMs, while extremely useful and valuable, have confused people's understanding of general versus narrow intelligence a bit, and in an interesting way. So before the advent of LLMs, there was a fairly clear distinction between what you call narrow AI, an AI that just does one thing, like just plays chess or just uh, you know, bombs people or solves math equations, right, or drives your car, versus an AGI, which is an intelligence that can figure out how to do a whole host of different things, right? So now we have an interesting situation, because AGI, as I defined it in 2005 when I wrote the book with that name, is supposed to be about systems that can generalize beyond their training and experience, you know, at least as well as humans can. On the other hand, LLMs, they're not that good at generalizing beyond their training data. Yeah. But they've just got so much training data. They are what, in a research paper from 2008 or so, I call the broad intelligence. They have a great amount of breadth to their capability, but they got that not the way people do, which is by taking leaps beyond our experience, and they got it just by being trained on so much stuff, right? And that's it's a kind of system that doesn't exist in the natural world, but it's a very interesting and useful kind of system. And I do think that in spite of not being AGIs, LLMs can be used to automate a vast percentage of the human economy, because most jobs don't need a lot of generalization beyond history. Most jobs are just repeating over and over something that had been done before, right? Now, not everything can be done that way, right? Culture and science, these things only advance because now and then we do take a leap beyond our training data, right? But still, the vast majority of the economy can operate based on sort of closed-ended quasi-humans that are operating based on training data. So, from an application development standpoint, sure, it's cool, just you know, milk the LLMs and related deep neural nets as much as you can, and that's what you have hundreds and hundreds of startups here at Web Summit doing. From a fundamental RD point of view, what we want to be doing is making a breakthrough toward the next level of artificial cognition, which either involves throwing out LLMs and replacing them, or adding other things onto LLMs, supplying the missing ingredients, like reasoning based on grounded observations and like more fundamental creativity that doesn't just recombine surface forms. And I'm so I'm interested in both, right? Like in applying AI to music, like I talked about in my talk this morning. I mean, there we're taking existing neural models, we're seeing like what can we do with them in our band, because it's fun, right? As a researcher on AI fundamentally, I'm thinking more like, okay, let's let's make a framework, which is our hypern framework, let's make a framework where you can take deep neural nets, put them together with logic engines, evolutionary learning systems, get these all to sort of interpenetrate and cross-pollinate, and get actual AGI to emerge from a huge N-RAM knowledge graph, right? And so but we're now at an amazing time in history when there's energy and resource behind both building on the current non-AGI AI systems to do so many things, and sort of rebuilding the foundations to get toward real artificial general intelligence. And even the notion of superintelligence has become mainstream so fast, which is quite wild, right?

Joeri Billast: 5:30

Yeah, now everything is going so fast. But another uh aspect is uh decentralization, uh, you've called for benevolent decentralized AGI. Yeah. So why is it essential for you know building intelligence that self-humanic?

Ben Goertzel: 5:43

I think it's essential to avoid certain pathologies, and it doesn't avoid all pathologies and isn't a panacea, but I think if the first AGI is centrally only controlled by like one company or one country, you've got some very obvious risks. Like the AGI's mind is oriented only toward making money for a few people, or it's oriented only toward assuring the hegemony of one country over others. And even if you had the most beneficial, loving, wise people in the world owning and controlling the AGI, unless they're also super powerful, something's gonna happen, right? Like someone's gonna get offered a trillion dollars by some evil government over there, or some bad guys will come and kidnap them and force them to give away the source code or something like it. Just seems like we've all seen the movies, right? Seems like not a stable situation. Having the AI be open, be running on a decentralized infrastructure owned and controlled by millions of parties in every country in the globe, like the internet or Linux is, right? This seems like a much more robust way to move us from where we are now to AGI and then to ASI.

Joeri Billast: 6:56

Yeah. Now, Singularity.net merges AI and blockchain. So it's two worlds that often move at a different speeds. What they're both moving very fast right now, actually. They move very fast, but anything that you learned about putting them together in practice?

Ben Goertzel: 7:14

I mean, on the technical level, there's been a tremendous amount of learning, right? And when we launched SingularityNet initially, we sort of hacked it in the sense that our architecture is the SingularityNet node, part of the decentralized network. You know, it has a container running AI. There's another container running the blockchain protocol stuff. And so the AI runs off-chain, but it uses the decentralized protocol to coordinate with other AIs running off-chain on different machines, right? And that works, and it does let you get a decentralized AI network without any an owner or controller. What we're able to do now, which is the ASI chain, which is now in a DevNet beta release, and to be put in the full-scale testnet early next year, this lets you run complex AI processes on-chain, right? So you don't have this separation where a node has the AI part and then the blockchain part, right? You have a situation where the AI itself can leverage the blockchain for its internal operations. And I mean, this was a lot of technical work. Like we built a new AGI programming language called Meta, M-E-T-T-A, and then we figured out how to use this AGI programming language as a smart contract language, and we built a new layer one blockchain around this, which is the ASI chain, and we just we just announced the launch of the developer network of that today. That's on the tech side. I think on the sort of cultural and public understanding side, it's been in a way even harder because the crypto world, while there are high ideals about decentralization and democratization, it's largely full of people trying to make short-term profits from from price fluctuations. And the mainstream tech world mostly considers blockchain just to be about that. Like they think it's just like a quick money thing. And so convincing people that blockchain as an infrastructure can be the golden path to ethical and beneficial AGI is hard when most people are thinking about blockchain as a tool for scams or at best, like uh making quick money on on a on a trade or something, right? And I mean that's a problem we will overcome just by launching better and better products, right? Like if if we launch tools that everybody uses because they're amazing AI tools, and we happen to launch them on a decentralized network, then they'll be like, yeah, oh cool, you can run AI on a decentralized network, right?

Joeri Billast: 10:01

Soon Yeah, it's also, you know, by doing this podcast, by being on stage, by talking about it, like even people like Gary V here in Lisbon, I met him. Yeah. He spoke about, you know, blockchain can be the solution for AI so that people can believe it things to be true.

Ben Goertzel: 10:16

So the thing is, most AI blockchain projects, and I don't want to trash my competitors too much, but most of them are basically taking open source LLMs from big tech and fine-tuning them a little bit on some other data, and then making it so you can pay for them with your tokens or stake tokens and get use or something. And nothing wrong with this. I mean, this is a it's a beneficial thing to do, and you're you can make money and you can help foster growth of decentralized ecosystems, right? On the other hand, it's different than making AI that leverages decentralized networks within its thinking, which which which is what it's it's a more interesting thing to do when it really brings you beyond big tech's home model. Absolutely.

Joeri Billast: 11:09

Now interesting. Um so you had Sophia the robot, we talked about autonomous agents on chain. You have seen human AI interaction evolve firsthand.

Ben Goertzel: 11:22

Is there anything that surprised you how people emotionally respond to these uh You know what surprised me is how much the way you frame things to people affects their perception of the exact same thing, which of course shouldn't be surprising, but as a technologist, you stop thinking that way. So our band Desdemona's Dream, which has a robot as a lead lead singer, the Desdemona Robot from Hansen Robotics, Sophia's Little Sister. We did a few gigs this summer as the warm-up band for Macy Gray, the the pop singer on part of her US tour. And the first show on the tour, we almost got booed off stage. I mean, no, no one was throwing large objects at us, but they were just like, get rid of this robot, bring on the people, right? And I mean, I'd wondered how Macy's audiences would like us, because we're more like uh progressive rock slash jazz fusion with a robot, where she she herself can do every genre of music. Like she's a genius, right? But but her name was made with with a certain sort sort of pop RB soul type thing. But then in the next gig we did, I sort of talked to the audience a bit to contextualize things and tell them, like, well, we know Macy. Macy loves advanced technology, she's always been an innovator, and you know, we're totally not about replacing human musicians with AI. It's just about like, this is a cool thing we can do to explore a different sound, just like a synthesizer and electric guitar, right? And then with that context, the audience perceived the robot and band in a more positive way, and they're dancing and singing along and really got into it. So it's the same show, it's the same robot, it's just a different way of approaching it from your your subjective uh perspective, right? And that and that that made all the difference. And that was again, wouldn't be a shock to anyone with a with a marketing perspective, right? It just uh it was interesting to me because I was thinking about it more as a technologist and a musician, right than from a like how do you adjust the frame of mind of the of the audience point of view.

Joeri Billast: 13:42

So you mentioned as a marketer, I've I mentioned to just before we started recording, I've written a book, The Future CMO. I talk about how trust becomes a new currency in tech-driven ecosystems. And you know, now talking about AI in AI itself, how do you think we can build trust into AI technically, culturally?

Ben Goertzel: 14:04

So technically, blockchain of course gives very powerful tools for this, right? Like any smart contract running on ASI chain, by the nature of the language and chain can be formally verified to do what it's specified to do. So you can actually prove with math that that that that it will do what what it's supposed to do, which is not the case with Solidity, the smart contract of Ethereum blockchain, or or m most of the common things in the blockchain space. We're also integrating a reputation system into the chain so that people rate each other and your rating, you know, affects how much your rating affects other people's ratings and AIs and people can enter together into this reputation system, and that's a different layer of trust. It's a social layer on top of the mathematical layer, and so you can do all these things very nicely when you're sort of launching a whole new economy, like we are with the with the ASI chain. The thing is that's now embedded within the broader global economy, which is quite different. So I think within the network of ASI chain and the hyper-owned system, we can deal with trust between humans, between AIs, and across the human AI boundary in a quite robust way. And you know, we're not trying to take over the whole world economy, we're not quite that megalominiacal. But it would be great if a patchwork of networks like this which have baked-in rational notions of mathematical and social trust, if the economy became that way, right? Because right right now the baseline of the economy is you know, it's a system where trust is given only by centralized institutions like your bank or your government, and these entities are only trustworthy themselves to a limited degree, and they get l their legitimacy by having a whole bunch of guns to point your head, basically.

Joeri Billast: 16:09

Yeah. Uh now some argue that open permissionless AI could be dangerous. It could be. And others see it as the only path to transparency. So why do you stand on openness versus control?

Ben Goertzel: 16:21

I think there's something called the proactionary principle. So there's a precautionary principle, which is if it could be dangerous, be careful. Don't do it. There's a proactionary principle, which is to balance the costs of action versus the costs of inaction. And I look at openness that way. Like, yeah, of course there's a risk. If it's open, you know, Al-Qaeda can take it. Whoever is your boogeyman of the year can take it. On the other hand, there's so much benefit that you get from having the collective brain power and the collective values and ethics of everyone on the planet to help shape your system. So there's there's always a plus and a minus in any complex matter. You have to balance them. I would say you can make the same argument against open source software or open science, right? Like, how can we put fundamental physics research on the open web when someone can use it and figure out how to make the next big bomb, right? But on the whole, putting fundamental science on the open web clearly has been for the benefit of humanity and the worst risks have not actualized.

Joeri Billast: 17:29

Yeah. I think you often mention that AGI will not replace humans but extend consciousness. How do you envision human creativity and machine intelligence co-creating value?

Ben Goertzel: 17:42

I think the process of humans and machines co-creating is beautiful and fascinating. And that's what I talked about in my talk this morning, where I had Des Demo the robot, and we were giving some demonstrations of joint human-machine creativity in music creation. If you subtract out competitiveness of whose creative product is better, and subtract out what can I make money with my creative works, then just playing with AI to co-create new stuff is amazing. And then new tools are great, and you know you can use them one time and not use them the other time. Like in music, sometimes you use a robot in AI generation, sometimes you just sit there and play the piano, right? And it's it's it's it's all good. When you bring the economic bit into it, it gets more complicated because I do believe that AI will be able to create accessible pop music, probably more reliably and cheaply than humans can, right? And then there's the question, is AI taking away musicians' livelihood? Well, first you have to say the music economy is already broken, and almost no musicians can make money anyway, so it's not like AIs are messing up something that functions, right? Like if AI is obsolete radiologists or advertising copywriters, in a way they're disrupting a somewhat functional subset of the economy. With music, AI is and all creative professions, basically. Music is just AI is disrupting an already utterly broken section of the economy anyway. But yeah, it will disrupt it even worse in some ways. I look at look at music production. I mean, we're a couple years away at most from an AI, we'll just take the stems your band produced and just mix it down into a basic mix for you. And it may not do as well as the highest end producers, but it can do as well as a typical producer. Now, on the one hand, yes, this will eliminate jobs for all but the highest end most artistic famous music producers. On the other hand, it will give every band the ability to mix down all of their recordings, which now they just don't have because they can't afford to pay a producer to do it, right? So, I mean, there's a cost and a benefit. Like, is it worth every band being able to mix down all their songs into a listenable thing? Is that benefit worth eliminating the jobs of all but the highest end music producers? I mean, to me as a musician, it totally is, because we've got all these tapes we haven't mixed down, right? But on the other hand, if I was making a job as a producer, I'd be like, well, fuck, fuck, fuck you, right? So and you have that same that same issue over and over. So in the end, our society is gonna have to deal with the wholesale elimination of jobs. And once that's dealt with, then humans and AIs creating stuff together will just be perceived as a beautiful thing it is. But in the interim, it's sort of mixed up with broader economic issues, which makes it complicated, right? And that and this occurs all over the economy. Like take this take this interview as an example, right? I mean, I could in a few years each of us can have our digital twins conduct this interaction. Like your digital twin will know the style of interview that you give. My digital twin can know everything I'm saying, because none of it is that out of distribution from things I've said before. Yeah. No. But that doesn't mean we won't get together and talk to each other in five years, right? Because we're humans, we want to get together and talk to each other, and other humans will like knowing that they see actual humans talking to each other, right? So, I mean, but the thing is, humans will be doing stuff only because they want to, or because other humans want to see them do it and and and and and and they want to, right? It won't it won't be because AIs couldn't do the thing, and that's that's a it's a different setting. But that I mean that's analogies occur in our life right now. Like I could pay someone to take my kids to the playground. Yeah. Most sometimes I do, right? But often I want to do it myself because it's rewarding, even though someone else could do it in a more economically efficient way than me.

Joeri Billast: 22:06

Yeah. Cool it. Yeah, it's uh it's the way that that you look at it and you know how it can benefit AI and all our lifestyle change is so exciting. Unfortunately, Ben I could talk for our CTU, but we will be kicked out about this podcast at a certain moment. So if you know my listeners, yeah, they want to know more about AI, about Singularity, everything that you are doing, where should I send them?

Ben Goertzel: 22:31

Yeah, you can go to singularity.net or superintelligence.io and you can find out about Hyperon and the ASI chain and all the other stuff that my various teams are now building to try to bring about a beneficial decentralized AGI and ASI future.

Joeri Billast: 22:51

Well, great. I was also at your press conference. So, guys, if you're now listening to this episode, you know there are show notes, there is a blog article. A lot of things that uh Ben said during the press conference will also be mentioned in a blog article. Ben, it was really a pleasure to have you on the show. Thanks a lot for having me. Guys, what an amazing episode. An historical episode, I would say. I started actually my podcast four years ago. Here at WebSummit. Uh-huh. Yeah. Guys.

Ben Goertzel: 23:21

I wasn't in that one. I was in 2019.

Joeri Billast: 23:23

I know you were there. I was also here at 2019. Then I was hearing this podcast who was actually interviewing uh Melanie Perkins from Gamble at the time.

Ben Goertzel: 23:32

So 2017, that was when we were doing the initial fundraise for Singularity. That was a very high-energy web summit.

Joeri Billast: 23:39

Yeah, man, I remember all those website, but this is my 10th web summit in the middle. So yeah, this podcast episode, I'm sure you have people around you that will benefit from this episode we should share the middle. Friends, family, all the members, all the marketers, people interested in my blockchain. If you're not yet following the show, this is a really good moment to hit the subscribe button. If you haven't given me a review yet, five stars, please be a real audience. And of course, I would love to see you back next time. Take care.

Ben Goertzel: 24:12

Five summers, you haven't