Latest / Crypto Coin Show / Decentralized AGI: Why AI's Future Must Be Open Source | Dr. Ben Goertzel
Transcript
- 0:00Speaker 1: I'm Ashton Addison from the Crypto Coin Show and today
- 0:03Speaker 1: on Blockchain Interviews with we have with us doctor Ben Goertzel,
- 0:06Speaker 1: CEO of Singularity net and the ASI Alliance, here to
- 0:10Speaker 1: talk about artificial intelligence, decentralized networks, decentralized AI, AGI, so
- 0:16Speaker 1: much more. Ben, Welcome to the show and thanks for
- 0:18Speaker 1: taking the time.
- 0:20Speaker 2: Yeah, thanks for having me.
- 0:22Speaker 1: Yeah, I excited to dive into your work on AI,
- 0:25Speaker 1: which is decades long now and now the world is
- 0:28Speaker 1: catching up and blockchain is also beginning to merge more
- 0:32Speaker 1: with AI and the principles of blockchain, of decentralized collaborative
- 0:38Speaker 1: in AI, there's a bit of a battle going on.
- 0:40Speaker 1: I'd love to start out on a high level of
- 0:42Speaker 1: sort of your thoughts on the vision of AI being
- 0:45Speaker 1: built around collaboration, safety and security rather than centralized control,
- 0:50Speaker 1: and sort of what we're seeing in the world with
- 0:52Speaker 1: how AI is being built and being released to the
- 0:54Speaker 1: mainstream right now.
- 0:57Speaker 2: Yeah, I mean, it's an amazing time for AI and
- 1:01Speaker 2: for those of us who've been working in the AI
- 1:03Speaker 2: field for decades. It's it's almost a bit disorienting that
- 1:07Speaker 2: suddenly everyone loves AI right What works for most of
- 1:11Speaker 2: my most of my career, AI was a sort of
- 1:14Speaker 2: little corner of the research world and even smaller corner
- 1:20Speaker 2: of the of the business world. That said, the kinds
- 1:24Speaker 2: of AI that are capturing all the resources today, it's
- 1:31Speaker 2: a very small subset of what AI can do. And
- 1:36Speaker 2: the way the way the business world tends to work is,
- 1:41Speaker 2: you know, once one thing works and someone has a
- 1:43Speaker 2: big success, everyone copies that exact same thing because they
- 1:48Speaker 2: view that as the least risky thing to do. And
- 1:51Speaker 2: then the research world follows along because researchers want to
- 1:55Speaker 2: do what will get them the highest paying jobs or
- 1:58Speaker 2: the easiest venture money or something. So you have an
- 2:02Speaker 2: interesting situation now where there's so much resource and so
- 2:05Speaker 2: much enthusiasms going into AI and it is delivering real
- 2:09Speaker 2: value and it is really interesting. On the other hand,
- 2:12Speaker 2: it said, it's very limited, and I think in the
- 2:17Speaker 2: next few years we're gonna see a great flowering and
- 2:21Speaker 2: expansion of AI, both in terms of the kinds of
- 2:26Speaker 2: AI algorithms and approaches that are out there and in
- 2:30Speaker 2: terms of how AI is is deployed at the hardware
- 2:33Speaker 2: and computer network level. Right and I'm working with my
- 2:39Speaker 2: Singular ASI Alliance team I mean, I'm working on both aspects,
- 2:45Speaker 2: like how does the actual machine thinking work, and then
- 2:51Speaker 2: how do you deploy that across machines and networks, which
- 2:55Speaker 2: is where the blockchain aspect comes in, right, And so
- 2:58Speaker 2: we can we can talk about both of those and
- 3:02Speaker 2: how they.
- 3:02Speaker 1: Connect together definitely, you know, and I've I've been a
- 3:07Speaker 1: blockchain enthusiast since the early twenty ten's and now seeing
- 3:12Speaker 1: AI come in so rapidly, I feel like there is
- 3:15Speaker 1: a great synergy for the decentralized nature of it. And
- 3:21Speaker 1: you know, there's already been discussion around the AI that
- 3:25Speaker 1: you know, the everyday people are using, which is right
- 3:26Speaker 1: now just sort of glorified chatbots, which, as you alluded to,
- 3:31Speaker 1: what they are capable of doing in probably the near
- 3:35Speaker 1: future is going to be astronomically more than what they're
- 3:38Speaker 1: capable of right now. But even right now for the
- 3:40Speaker 1: everyday person, it seems like a ten X in productivity.
- 3:43Speaker 1: I can't imagine over the next few decades it'll be
- 3:47Speaker 1: like ten thousand X.
- 3:48Speaker 2: I think. I think we're going to get to a
- 3:52Speaker 2: human level thinking machine within five years. It couldn't even
- 3:57Speaker 2: be two or three years. And once you get there,
- 4:03Speaker 2: then accelerating or magnifying human productivity is not even the
- 4:08Speaker 2: right way to look at it, right, because I mean,
- 4:11Speaker 2: once you have machines that are so much smarter than people,
- 4:15Speaker 2: then the machines are able to do the production and
- 4:20Speaker 2: direct the production, and it's there's other big questions like
- 4:27Speaker 2: are these our friends, and are they enhancing and helping
- 4:31Speaker 2: with human life or doing something else? And are we
- 4:33Speaker 2: even like choosing to choosing to plug them into our
- 4:36Speaker 2: heads and like fuse with the machine. Right. But I
- 4:41Speaker 2: think we're in a very interesting transition period now where
- 4:44Speaker 2: AI is enhancing, magnifying, accelerating what we can do. And
- 4:50Speaker 2: in the transition period between here and for long human
- 4:54Speaker 2: level thinking machines, I mean, there's technologies to be built,
- 4:59Speaker 2: there's business is to be built, there's lives to live, right,
- 5:03Speaker 2: and this will be a unique period in history when
- 5:08Speaker 2: we're still in some ways smarter than the machines, but
- 5:12Speaker 2: the ways to gain advantage and deliver good in the
- 5:17Speaker 2: human economy mostly involved synergizing with these rapidly evolving sort
- 5:23Speaker 2: of smart machines, right, And this is a it's a
- 5:27Speaker 2: unique period in history, right, which is quite fascinating to
- 5:32Speaker 2: be to be living through it in terms of blockchain.
- 5:37Speaker 2: You know, in two thousand and one and two thousand
- 5:42Speaker 2: on the Java one point one language, I was screwing
- 5:46Speaker 2: around with, like how do you bring distributed processing and
- 5:53Speaker 2: strong encryption together? Right? And it was what you would
- 5:57Speaker 2: now call early stage AI blockchain network. It was just very,
- 6:01Speaker 2: very slow, and it just seemed like we needed software
- 6:07Speaker 2: to speed up before you could make that viable. It's funny.
- 6:10Speaker 2: I even tossed around with my friends the idea of
- 6:14Speaker 2: making a decentralized money, but we were just like, well,
- 6:21Speaker 2: it's going to be really slow to clear any transaction,
- 6:24Speaker 2: so who's going to want to use it. If I'd
- 6:26Speaker 2: thought of money laundering in tax avoidance as use cases, right, yeah,
- 6:32Speaker 2: then I'd be Satoshi right, and we'd probably have AGI
- 6:35Speaker 2: already because I would have had trillions of dollars to
- 6:40Speaker 2: build it. Right. But the reason I mentioned that, though,
- 6:44Speaker 2: is it's sort of a historical accident. The blockchain began
- 6:48Speaker 2: for decentralized money. I mean, decentralized money is becoming a
- 6:54Speaker 2: good use case now that things get faster and faster
- 6:57Speaker 2: on the blockchain side, But I'm you could just as
- 7:00Speaker 2: well have had blockchain emerge for decentralized AI, or for
- 7:05Speaker 2: decentralized medical records with data sovereignty or something. Right. So,
- 7:09Speaker 2: I mean it's said having started with decentralized money is
- 7:14Speaker 2: a plus and a minus. Right. It's a plus because
- 7:18Speaker 2: I mean, people made money, and it also gives away
- 7:21Speaker 2: for blockchain to be used to fund other sorts of projects,
- 7:24Speaker 2: right the icos and other more sophisticated mechanisms. On the
- 7:29Speaker 2: other hand, it's also caused a sort of narrowness of
- 7:33Speaker 2: thinking whether the world just thinks about blockchain as a
- 7:37Speaker 2: mechanism for digital money. But it's not just that, right,
- 7:41Speaker 2: And this I think we're going to break through that
- 7:46Speaker 2: next year with our launch of the ASI chain, which
- 7:49Speaker 2: is a new layer one chain, and this will be
- 7:51Speaker 2: the first layer one chain. It lets you do large
- 7:55Speaker 2: scale advanced AI processes on chain. Right. So I launched
- 8:01Speaker 2: in twenty seventeen Singularity Net, which is really the first
- 8:05Speaker 2: serious blockchain infrastructure for AI and fetch, which we later merge.
- 8:10Speaker 2: We'd launched just a little bit after that, Right, But
- 8:14Speaker 2: what we do there, I mean we have AI processes
- 8:17Speaker 2: running in Docker containers on different machines. Then we have
- 8:21Speaker 2: a sort of proxy by the side that has a
- 8:24Speaker 2: smart contract. So you have like AI processes running and
- 8:27Speaker 2: containers on different machines and the blockchain lesson communicate and
- 8:30Speaker 2: coordinate together. But the AI itself is not in a
- 8:34Speaker 2: micro level running on the blockchain. The blockchain is used
- 8:37Speaker 2: to coordinate between a decentralized network of AI processes, right,
- 8:41Speaker 2: And that's what everyone is doing. It's what you can
- 8:44Speaker 2: do with ASI chain. We can do things differently. We
- 8:49Speaker 2: can very flexibly run AI on chain. And the way
- 8:53Speaker 2: this is enabled is as part of an open source
- 8:58Speaker 2: AGI project that I've been leading for years, which is
- 9:01Speaker 2: called Hyperon, which is a new version of the open
- 9:04Speaker 2: cog open source AGI project. In building Hyperon, which is
- 9:10Speaker 2: not about blockchain, it's about how do you make AGI
- 9:13Speaker 2: by using not just neural nets, but use neural nets,
- 9:15Speaker 2: use logic engine, use evolutionary learning, use a variety of
- 9:19Speaker 2: AI paradigms, implement them together in this big shared intergram
- 9:23Speaker 2: knowledge graph to make a system that can learn from
- 9:26Speaker 2: experience and learn from studying its own self and its environment.
- 9:31Speaker 2: But in building this hyperon AGI project, we developed a
- 9:35Speaker 2: programming language called meta me Etta, and two months after
- 9:39Speaker 2: we launched it and named it that Facebook change its
- 9:41Speaker 2: name to Meta, but we have another t We're not
- 9:45Speaker 2: changing the name, and mena with twots also means loving
- 9:50Speaker 2: kindness in Sanskrit. It's the kind of meditation. Right. So
- 9:56Speaker 2: what we found is this AGI programming language we called
- 10:00Speaker 2: META that we develop for efficient large scale diploma of
- 10:04Speaker 2: a variety of AI methods can also be used as
- 10:08Speaker 2: a smart contract language. So we've repurposed our AGI programming
- 10:12Speaker 2: language as a smart contract language. And what that means
- 10:16Speaker 2: is we can take quite sophisticated, scalable AI methods and
- 10:20Speaker 2: we can run them as smart contracts on chain, and
- 10:25Speaker 2: we can sort of modulate at a micro level which
- 10:28Speaker 2: bits of the AI run on chain or not. Like
- 10:30Speaker 2: for programmers in the audience, like each each function blocking
- 10:34Speaker 2: code you can encrypt with a number with a number
- 10:37Speaker 2: of private keys, right, so you can decide like I'm
- 10:40Speaker 2: encrypting this bit, I'm not bothering to encrypt encrypt this bit,
- 10:43Speaker 2: and you can you can very flexibly put things on
- 10:47Speaker 2: chain or off chain. So that's it's going to be
- 10:51Speaker 2: quite interesting. And one thing that that gets you is
- 10:55Speaker 2: like right now, all the big AI models are run
- 11:01Speaker 2: on monolithic server farms, like it's distributed processing, but it's
- 11:05Speaker 2: a bunch of machines like in one facility owned by
- 11:08Speaker 2: one guy, and there's a certain efficiency there, right. But
- 11:14Speaker 2: with the ASI chain, you can take advanced AI algorithms
- 11:17Speaker 2: and run them on a heterogeneous network of machines sitting
- 11:21Speaker 2: in all different countries, in all different places, right. And
- 11:25Speaker 2: of course there's certain efficiency that you get from having
- 11:29Speaker 2: all your machines in one building, but there's also an
- 11:32Speaker 2: efficiency that you get by being able to run part
- 11:36Speaker 2: of your server network like right next to a power facility,
- 11:39Speaker 2: or in a place where it's cheaper to run a
- 11:42Speaker 2: server farm, or even run part of it on your
- 11:45Speaker 2: phone or something. Right. So I mean, I think we're
- 11:50Speaker 2: at a very interesting point where so we have now
- 11:54Speaker 2: we've launched the devnet version of ASI chain, we'll go
- 11:58Speaker 2: into test net early next year year, and then MAInet
- 12:03Speaker 2: after testing has been done with hyperon the AGI platform.
- 12:08Speaker 2: Like in the last couple of months, we finally got
- 12:10Speaker 2: a really fast compiler for our Agi programming now, which
- 12:13Speaker 2: we're now just working on scaling up all of our
- 12:16Speaker 2: research from the previous years. So we're sort of at
- 12:19Speaker 2: a transition point of moving from research into reality. With
- 12:25Speaker 2: hyper and with ASI chain. Now big tech isn't slowing
- 12:30Speaker 2: down either, but again they're doing a certain thing. They're
- 12:34Speaker 2: doing centralized server farms. Then they're doing big deep neural
- 12:38Speaker 2: nets trained using back propagation algorithm. So if that's the
- 12:42Speaker 2: golden path to AGI, like big tech wins, and whether
- 12:46Speaker 2: it's Microsoft or Google or Tencent, we'll see if using
- 12:51Speaker 2: a greater variety of AI algorithms deployed on a decentralized
- 12:56Speaker 2: network is a better path to AGI, then then humanity
- 13:01Speaker 2: wins instead, right, And and then the decentralized ecosystem gets
- 13:05Speaker 2: to play play a major, major role, right, And so
- 13:09Speaker 2: we're we're now. We're now when push comes to shove,
- 13:12Speaker 2: like when the rubber hits the road, right, it's a
- 13:15Speaker 2: which is quite quite fun, albeit a bit exhausting if
- 13:19Speaker 2: you're right in the middle of it.
- 13:21Speaker 1: Mm hmm, yeah, no, it's a it's it's crazy times
- 13:25Speaker 1: right now. And thanks for explaining the architecture behind the
- 13:28Speaker 1: a SI chain and hyperon as well. You know, in
- 13:32Speaker 1: the last month we've seen AWS go down, uh it
- 13:36Speaker 1: was actually affecting millions of businesses around the world. And
- 13:40Speaker 1: then also this week cloud Flare went down x wasn't working,
- 13:44Speaker 1: Meta wasn't working at least Facebook, and even open ai
- 13:47Speaker 1: and the lll ms weren't working because of one server
- 13:50Speaker 1: of one small issue. I feel like that is the
- 13:54Speaker 1: thing that's waking people up to this decentralized AI that
- 13:59Speaker 1: shouldn't run into any of those issues.
- 14:03Speaker 2: Yeah, I mean it's not a panacea, right, but but
- 14:08Speaker 2: I mean it is. It is more robust, right I mean, no, no, no, no, no,
- 14:12Speaker 2: no question. I mean that's that's the nature of it.
- 14:16Speaker 2: The challenge has been to get that robustness without an
- 14:20Speaker 2: egregious loss of efficiency. And I think that that that's
- 14:25Speaker 2: what the ASI chain gives you. And it's been subtle.
- 14:30Speaker 2: Like we've been working closely with Greg Meredith, who led
- 14:34Speaker 2: the ar Chain blockchain team, and I mean that kind
- 14:39Speaker 2: of crashed and burned for tokenomic reasons. But we took
- 14:42Speaker 2: the open source code from our chain, integrate it with
- 14:46Speaker 2: our metal language and some of some of our other code,
- 14:49Speaker 2: and we've now come up with something that I think
- 14:54Speaker 2: it basically solves the blockchain trilemma, but in it's sort
- 14:59Speaker 2: of artfully dodging way. Like the blockchain trilemma is, you know,
- 15:04Speaker 2: can you get something that's you know, scalable and secure
- 15:10Speaker 2: and decentralized altogether, right. And the way we do that
- 15:14Speaker 2: with ASI chain is partly that different shards of the
- 15:21Speaker 2: chain can have different consensus mechanisms, so the whole thing
- 15:27Speaker 2: is secure and the whole thing is decentralized. But to
- 15:31Speaker 2: get scale, there's not one consensus algorithm that will give
- 15:36Speaker 2: scale for every application that you want. So what you
- 15:39Speaker 2: need is a system that lets you put different consensus
- 15:42Speaker 2: algorithms on different shards and then lets them all communicate
- 15:46Speaker 2: together for consists on the overall networks. So for example,
- 15:49Speaker 2: if you want to run a dex that does high
- 15:52Speaker 2: frequency trading on cryptotokens, then you need a consensus mechanism
- 15:59Speaker 2: that's really really in certain senses. Then you may require
- 16:03Speaker 2: the nodes in that network have a lot of multi
- 16:05Speaker 2: GPU servers, right, and there's a consensus mechanism called Casanova
- 16:10Speaker 2: we can use there that was developed by one of
- 16:13Speaker 2: our lead developers at his previous company, pyro Effects. Right now,
- 16:17Speaker 2: if you want, say something can run on a mesh
- 16:20Speaker 2: network so that everyone's phone can run some of the AI,
- 16:23Speaker 2: and you have look at globally distributed AI commons where
- 16:28Speaker 2: anyone can contribute some processing power. Then they can democratically
- 16:32Speaker 2: vote and what that commons does right, And the validators
- 16:36Speaker 2: may pop on and offline randomly because they're on people's phones.
- 16:40Speaker 2: For that, we can run a consensus algorithm called cordial minors.
- 16:43Speaker 2: It's not as fast as you would want for high
- 16:45Speaker 2: frequency trading, but it's very robust to validators popping on
- 16:49Speaker 2: and offline all the time. So you can have one
- 16:52Speaker 2: shard doing the mesh network thing, one shard doing the
- 16:54Speaker 2: high frequency trading thing. They have different consensus mechanisms, they're
- 16:59Speaker 2: all running on are running on the same security centralized infrastructure.
- 17:04Speaker 2: So you I think to solve the blockchain trilemma, you
- 17:09Speaker 2: have to cheat a little bit like that because there's
- 17:11Speaker 2: not there's not one consensus to rule them all, but
- 17:15Speaker 2: you don't. You don't need there to be right, and
- 17:17Speaker 2: so that then the other interesting thing is once you've
- 17:21Speaker 2: decided to do multiple shards that may or may not
- 17:23Speaker 2: have different sensus mechanisms, you can do tokenomics that way, right,
- 17:28Speaker 2: So you can you can say, okay, take your ASI token,
- 17:31Speaker 2: which we're going to rename the fetch token from from
- 17:34Speaker 2: a network as ASI token. With a launch of ASI chain,
- 17:38Speaker 2: you can take your ASI choken, you can state them
- 17:40Speaker 2: on the shard and then as reward for that steak
- 17:43Speaker 2: and you can get a liquid steaking token associate associated
- 17:45Speaker 2: with that shard. So then then we have an ecosystem
- 17:49Speaker 2: a little bit like dynamic dynamic TAO from from bit tensor,
- 17:54Speaker 2: but we're working out the tokenomics differently to so that
- 17:58Speaker 2: the emissions are tied to reputation and revenue rather than
- 18:02Speaker 2: to market guides. We really want to we want to
- 18:06Speaker 2: disincentivize pump and dump and meme coin and incentivize projects
- 18:11Speaker 2: that are actually using AI to deliver deliver value on
- 18:16Speaker 2: the chain. Right, So I think if this all comes
- 18:20Speaker 2: out right, I mean, then we're building the next generation
- 18:23Speaker 2: AI economy right like we're and inside each node we
- 18:29Speaker 2: have the instrumentation to run scalable neural nets and also
- 18:32Speaker 2: the hyperon system for neural symbolic evolutionary AI. Right. So
- 18:36Speaker 2: what we want to do we want to see the
- 18:39Speaker 2: next phase of AI development towards AGI, going beyond deep
- 18:44Speaker 2: neural nets and neural symbolic and integrate them with different
- 18:46Speaker 2: AI methods. But we want to see this next generation
- 18:50Speaker 2: of AI on the next generation decentralized economy. So like
- 18:55Speaker 2: when you have you have a couple of kids out
- 18:58Speaker 2: of Bangladesha wherever who have an new cool idea for
- 19:01Speaker 2: an AI company, instead of struggling to raise VC money,
- 19:05Speaker 2: which is going to be very difficult in their country, right,
- 19:09Speaker 2: then they can launch a deep end thing on the
- 19:12Speaker 2: shard of ASI chain and if they have a product
- 19:16Speaker 2: that actually delivers value to people that will give them
- 19:19Speaker 2: a lot of LSD emissions, they can then trade that
- 19:23Speaker 2: LSD on the deck shard of the ASI chain and
- 19:26Speaker 2: you have a whole decentralized AIAI economy there, right, And
- 19:30Speaker 2: we're we're in the middle of building this now, like
- 19:33Speaker 2: the devnet is launched. Developers who want to help chip
- 19:37Speaker 2: in or test it, you can, you can, you can,
- 19:39Speaker 2: you can find it online and this is going to
- 19:43Speaker 2: be quite epic during twenty twenty six, I believe.
- 19:46Speaker 1: And that.
- 19:47Speaker 2: So you interviewed Hamayun from our partner project Fetch a
- 19:50Speaker 2: couple of months ago, so we've had a bunch of
- 19:54Speaker 2: conversations about the Fetch agent verse and how do you
- 19:58Speaker 2: run parts of that on the shard in a side chain, right,
- 20:02Speaker 2: because that's now run on Cosmo's block chain, which is
- 20:06Speaker 2: great for some things, but the fetch Agent Verse can
- 20:08Speaker 2: be cross chain. Yeah, probably be on Cosmos. Yeah, probably
- 20:10Speaker 2: in a shard of of of a side chain, and
- 20:13Speaker 2: they can all communicate and synergize together. And in that way,
- 20:18Speaker 2: you're sort of jump starting what we can do in
- 20:21Speaker 2: ASI chain with all the things fetch age verse can
- 20:23Speaker 2: do already.
- 20:24Speaker 1: Yeah, it's very cool and great to hear Greg's name. Actually,
- 20:29Speaker 1: we were large proponents of our chain back.
- 20:31Speaker 2: And Greg lives like three miles away from me area.
- 20:36Speaker 2: So yeah, and we're both musicians, so I've gotten to
- 20:40Speaker 2: know him super well. Actually that's great.
- 20:41Speaker 1: Yeah, tell him, I said, Hi. We've had him on
- 20:43Speaker 1: the show half a dozen times when our chain was building.
- 20:47Speaker 1: We'd love to hear from him again.
- 20:49Speaker 2: Well, so yeah, it's interesting when you run open cog hyperone,
- 20:55Speaker 2: when you spin up the inram knowledge graph, what you
- 20:58Speaker 2: do is you run what until last month was called
- 21:02Speaker 2: an our node, and it's it's in our node from
- 21:05Speaker 2: our chain running the knowledge graph. It's it's now been
- 21:08Speaker 2: renamed fire node because this new company is Firefly. Right.
- 21:13Speaker 2: For a while, I think his company was are unchained,
- 21:18Speaker 2: which was was pretty cool. So yeah, Greg, Greg has
- 21:23Speaker 2: been a great collaborator and we've I think we've taken
- 21:27Speaker 2: all the computer science and data structure stuff in our
- 21:31Speaker 2: chain and we've worked together to take it to the
- 21:34Speaker 2: to the next level on the back end of ASI chain. Yeah,
- 21:39Speaker 2: it's very cool.
- 21:40Speaker 1: And when we had Humaian on with Fetch and we
- 21:44Speaker 1: were talking about AI agents and how we can get
- 21:47Speaker 1: them into the hands of more people. And you know,
- 21:49Speaker 1: open ai has made it really easy for people that
- 21:52Speaker 1: you know, grandparents that don't really know how to use
- 21:55Speaker 1: uh tech that well, to to query answers and do
- 21:59Speaker 1: basic chat. Now the functionality is coming in where the
- 22:02Speaker 1: AI agent can start doing things on the website for
- 22:05Speaker 1: them or book the hotel, et cetera. And when I
- 22:07Speaker 1: was looking into Fetch's AI agent verse, you know, it
- 22:12Speaker 1: was still quite a bit of technical expertise for for
- 22:15Speaker 1: even myself to run an AI agent.
- 22:19Speaker 2: It's not at that level yet in terms of tooling,
- 22:22Speaker 2: and that that is an advantage that big tech has
- 22:26Speaker 2: right like they have. We've been very backand oriented and
- 22:30Speaker 2: Fetch even as been backend oriented, not quite as much
- 22:35Speaker 2: as SNA, but I mean, if you're open ai, you
- 22:39Speaker 2: know they've got forty five hundred employees, right, So I mean,
- 22:42Speaker 2: I mean they and they're much smaller than Google, They're
- 22:46Speaker 2: much smaller than Microsoft or Tencent. But I mean so
- 22:49Speaker 2: they they can put a lot of work on easy
- 22:54Speaker 2: to use tooling. I mean that that said, I think
- 23:00Speaker 2: we're using their vibe coding tools for interface stuff along
- 23:04Speaker 2: with along with everybody else. It doesn't build the a
- 23:06Speaker 2: G I that can. It can build the it can
- 23:08Speaker 2: build the interfaces, right. So I think I think by
- 23:11Speaker 2: the time we launch a SI chain main net, which
- 23:16Speaker 2: I'm hoping is sometimes middle of the middle of next year,
- 23:18Speaker 2: that we're not giving a hard date yet, right, I
- 23:22Speaker 2: think we'll have some pretty nice tooling there. They're also
- 23:26Speaker 2: to to to make it easy for people to do
- 23:29Speaker 2: to do to do things. And I mean we you know,
- 23:32Speaker 2: there's some things that are an uphill battle because I
- 23:36Speaker 2: mean open Aye has been operating a loss, right, So
- 23:38Speaker 2: I mean that they will give you access to all
- 23:42Speaker 2: their servers and they're spending money for it, and they're
- 23:45Speaker 2: getting huge amounts of money from v vcs for it.
- 23:48Speaker 2: And we I mean, we don't have that much money
- 23:52Speaker 2: as they do to operate at a loss, but we
- 23:56Speaker 2: see that as a challenge to make more and more
- 23:58Speaker 2: efficient a I A I algorithm. Them is right, like,
- 24:01Speaker 2: we may not be able to operate a loss, but
- 24:02Speaker 2: we can let you run part of your AI on
- 24:04Speaker 2: your on your own phone and and and your own laptop,
- 24:07Speaker 2: which they're they're not able to do with with their
- 24:10Speaker 2: with their centralized infrastructure. Right. So, as we move toward
- 24:14Speaker 2: lot towards launch of ASI chain, I think we will
- 24:18Speaker 2: have easier to use tuning than you've seen on Singularity
- 24:21Speaker 2: Net or Fetch with the ASI maintain launch, and we
- 24:25Speaker 2: hope to overcome the inability to operate at a huge
- 24:30Speaker 2: loss like open the eye and anthropic by the lower
- 24:35Speaker 2: cost base that comes from a decentralized hardware hardware infrastructure. Right,
- 24:41Speaker 2: And and these are challenges, but I think, you know,
- 24:45Speaker 2: we've gotten an amazing tech team, and I think we
- 24:49Speaker 2: can we can overcome these challenges. And open source is
- 24:53Speaker 2: a huge help too, right, Like we've got a growing
- 24:55Speaker 2: open source development community helping us build, helping us build
- 25:00Speaker 2: all this stuff. And I mean big tech is embracing
- 25:04Speaker 2: open source, but only in a very limited way, like
- 25:08Speaker 2: the core of what they're doing is closed and they'll
- 25:12Speaker 2: open some models, but not the code used to train
- 25:15Speaker 2: the models and so forth, right, So I mean we're
- 25:18Speaker 2: hyperund is all fully open source. ASI chain is all
- 25:21Speaker 2: fully open source. So I mean as these things roll
- 25:25Speaker 2: out and show more and more capabilities, we can get
- 25:27Speaker 2: a positive feedback by pulling in more and more developers
- 25:31Speaker 2: around the world to help make it better and better.
- 25:34Speaker 1: Mm hmm, definitely. It's it's a good model to have
- 25:37Speaker 1: to have a base foundation on. And I was reading
- 25:40Speaker 1: as a part of the ASI chain, you know, part
- 25:43Speaker 1: of the goal is to have like billions of autonomous
- 25:46Speaker 1: agents and have everybody hopefully easily be able to build
- 25:51Speaker 1: something for their specific use case, do you I mean the.
- 25:54Speaker 2: Way you get billions of agents is agents that build agents, right,
- 25:57Speaker 2: which I mean we've been we've been talking about since
- 26:00Speaker 2: the launch of Singularity in it, but we now have
- 26:03Speaker 2: the infrastructure that can actually do it. Yeah. Yeah.
- 26:06Speaker 1: I wanted to touch on the humanoid robots. You mentioned
- 26:10Speaker 1: it briefly near the beginning, and I know that you know,
- 26:13Speaker 1: you were building the Sofia robot much before you know,
- 26:17Speaker 1: Tesla and these other humanoid robots came into the picture
- 26:20Speaker 1: into the mainstream. Where do you see that at right now?
- 26:25Speaker 1: It seems like there's an inflection point where you know,
- 26:28Speaker 1: at least as a part of Elon Musk's new proposal,
- 26:32Speaker 1: that it's part of that term just to make millions
- 26:35Speaker 1: of bots.
- 26:35Speaker 2: Robotics, robotics is major growth, right and that I'm taking
- 26:42Speaker 2: a walk to visit the humanoid robot right here. I
- 26:46Speaker 2: don't think I'm going to turn it oun but you
- 26:48Speaker 2: can see this is a this is Desdemona. It was
- 26:51Speaker 2: Sofia's Sofia's little sister. Wow, she sings in our rockman
- 26:57Speaker 2: Desdemona's Dream where I played the key boards and occasionally
- 27:00Speaker 2: sing along. She she composes music and and and and
- 27:05Speaker 2: and and and whatnot.
- 27:06Speaker 3: And we've uh, yeah, we've got we've got under repair
- 27:11Speaker 3: Grace who's who's an elder care humanoid robot here right?
- 27:16Speaker 2: So I mean this, this, this is uh is a
- 27:21Speaker 2: major growth area for sure. I mean I think that
- 27:26Speaker 2: robotics is now it's reduced to a software problem plus
- 27:32Speaker 2: a low cost, scalable manufacturer problem. Right, Like, we we
- 27:36Speaker 2: have the robot hardware that we need to carry out
- 27:40Speaker 2: pretty much every practical task that we we care about.
- 27:44Speaker 2: And the reason we don't have robots doing our plumbing
- 27:47Speaker 2: and electrical work and picking up our kids toys. Is
- 27:51Speaker 2: mostly the AI software hasn't been smart enough, right. And
- 27:56Speaker 2: once AI software is smart enough, then the work be
- 28:00Speaker 2: down to tool factories to make the robots cheaper and
- 28:02Speaker 2: cheaper and cheaper. And I think over the next few
- 28:05Speaker 2: years they I will become smart enough, right, which is
- 28:12Speaker 2: is awesome. Now, I sort of think most of the
- 28:17Speaker 2: robots out there are not really going to be humanoid.
- 28:20Speaker 2: I mean, for most tasks, you don't need a humanoid body.
- 28:24Speaker 2: Even if you think about like working in my kitchen
- 28:27Speaker 2: at home. I mean there's low down cabinets where you'd
- 28:31Speaker 2: like to have an arm on your feet, there's high
- 28:33Speaker 2: up cabinets where you want to like set a tentacle
- 28:36Speaker 2: up there or something, right, So I mean I think
- 28:41Speaker 2: a humanoid form factor will be a minority of the
- 28:45Speaker 2: robots that are actually out there in use. Now. The
- 28:50Speaker 2: work I've been doing in robotics with Hanson Robotics, who
- 28:56Speaker 2: made the Sophia robot, the first robot citizen, which I
- 29:00Speaker 2: led the development team for, and then Desdemona and Grace,
- 29:05Speaker 2: Sofia's little sisters. Then also we have a company called
- 29:09Speaker 2: mind Children, which is making three and a half foot
- 29:11Speaker 2: tall robots which are aimed at being teacher's assistants initially
- 29:15Speaker 2: to be rolled out in Korea. So we're focused more
- 29:21Speaker 2: on the social and emotional side of robotics. So I mean,
- 29:27Speaker 2: of course they need to move around, and they need
- 29:28Speaker 2: to be able to pick up things and use use
- 29:31Speaker 2: their hands. But the real thing you need from an
- 29:34Speaker 2: education robot, it's just the ability to connect with kids
- 29:37Speaker 2: and understand what they're doing and help them and with
- 29:40Speaker 2: elder care. Again, if you need a robot to pick
- 29:44Speaker 2: someone out of bed, you don't really need the humanoid form.
- 29:46Speaker 2: You just need like an automated bed and wheelchair that
- 29:49Speaker 2: connect to each other, well, you need a humanoid form.
- 29:52Speaker 2: Four is cognitive and emotional interaction, be sort of nurse's
- 29:59Speaker 2: assistant assistant with the nurses and nursing assistants are busy,
- 30:03Speaker 2: so I think that's also an interesting problem. It's interesting
- 30:11Speaker 2: to me that Tesla and Figure Unitary, these big guys
- 30:16Speaker 2: in the robotic space, they're not focusing on that, like
- 30:20Speaker 2: these are kind of faceless, scary looking things, whereas hands
- 30:24Speaker 2: and robotics and us with my children, they're focusing more
- 30:28Speaker 2: on the kind of soft soft side of it. But
- 30:32Speaker 2: I mean the beauty of being a robot is you
- 30:35Speaker 2: can mix and match body parts, right, So it's like
- 30:37Speaker 2: you can take a hands in robotics head put it
- 30:39Speaker 2: on a unitary body, or take a take a mind
- 30:43Speaker 2: children teach your head and add on arms that someone
- 30:47Speaker 2: else may that will help kids with their with their
- 30:52Speaker 2: fine handwork better. Right. So, I mean I think it's
- 30:55Speaker 2: a it's going to be huge, and it highlights the
- 31:00Speaker 2: need for data sovereignty, data protection, all the stuff that
- 31:05Speaker 2: the blockchain gives you. Because once you have robots in
- 31:08Speaker 2: every classroom in every house, and like robots in your
- 31:12Speaker 2: bedroom and your kids playroom and so on, I mean,
- 31:15Speaker 2: where's all that data going? Right? I Mean, on the
- 31:18Speaker 2: one hand, yes, all that data should be used to
- 31:21Speaker 2: help train and make the robots smarter. On the other hand,
- 31:24Speaker 2: none of us want big tech and big government to
- 31:27Speaker 2: be mining everything we say in everything we do. So
- 31:31Speaker 2: the capability to store all that data in an encrypted
- 31:36Speaker 2: and secure way and let different views into that data
- 31:40Speaker 2: to be used to train the AI models without leaking
- 31:44Speaker 2: any of our private info. I mean, this is a
- 31:49Speaker 2: this is something that blockchain delivers. It's something that hyper
- 31:52Speaker 2: and running on ASI chain delivers it's something that big
- 31:56Speaker 2: text tech stacks do not deliver because they don't care
- 32:00Speaker 2: about that. And at what point people start to care
- 32:04Speaker 2: about that will be interesting to see, Like everyone claims
- 32:07Speaker 2: to carry about data sovereignty, but they still give all
- 32:11Speaker 2: their day to Google and Facebook. Bit and maybe that
- 32:15Speaker 2: robots cruising around in their bedroom is where they draw
- 32:18Speaker 2: the line and actually do start to care about about
- 32:21Speaker 2: data sovereignty, right, in which case blockchain based infrastructure has
- 32:26Speaker 2: a very large advantage.
- 32:28Speaker 1: Yeah, we'll see. I feel like right now, there's definitely
- 32:32Speaker 1: more awareness going on with privacy. At least we've seen
- 32:36Speaker 1: you know, z cash, which may or may not be
- 32:38Speaker 1: fully private, but that's sort of bringing to people's minds
- 32:41Speaker 1: about privacy. And then if Vitalic's latest talk at the DevCon,
- 32:46Speaker 1: it's talking about how ethereum chain is more moving into
- 32:50Speaker 1: the privacy era, So hopefully that translates to sort of
- 32:54Speaker 1: overall awareness maybe outside of the blockchain bubble that you know,
- 32:58Speaker 1: people once they a robot staring at them while they're
- 33:02Speaker 1: sleeping or in their private moments.
- 33:05Speaker 2: It's interesting how little people care and practice about this
- 33:10Speaker 2: this sort of thing, right, And it's i mean, outside
- 33:14Speaker 2: of the sort of blockchain neo cipherpunk bubble, it's it's
- 33:20Speaker 2: interesting how little care people actually have about that. But
- 33:27Speaker 2: now in Europe people do care, right, a lot more.
- 33:31Speaker 2: In China people care a lot less. In America people
- 33:35Speaker 2: pretend they care, but they but they don't act like it.
- 33:38Speaker 2: But it's interesting to see how it evolves. But I mean,
- 33:44Speaker 2: you may be just one intrusive step by the government
- 33:48Speaker 2: away from people caring a lot, right, like if if
- 33:51Speaker 2: Trump's IC starts bumping illegal immigrants based on their social
- 33:56Speaker 2: media posts and then suddenly made people make care carry
- 34:01Speaker 2: a huge amount, right, So it'll be very interesting to
- 34:04Speaker 2: see how it evolves.
- 34:05Speaker 1: Yeah, I feel like some world events or some specific
- 34:08Speaker 1: triggers are definitely gonna wake up and hopefully that will
- 34:13Speaker 1: help bring more awareness to the decize.
- 34:16Speaker 2: We can, we can bring the we can build the technology. Yeah,
- 34:19Speaker 2: so that it's it's it's there when people realize they
- 34:22Speaker 2: need it. Yeah, definitely.
- 34:24Speaker 1: Well, I'm looking forward to seeing how the growth of
- 34:27Speaker 1: a SI chain and decentralized ai UH is now. You know,
- 34:32Speaker 1: at least that the greater blockchain community and Web three
- 34:35Speaker 1: developers are realizing that this is important to be working on.
- 34:39Speaker 2: Yeah, I think we need to. We need to also
- 34:44Speaker 2: focus on interoperability and cross chain stuff because well, I mean,
- 34:51Speaker 2: the blockchain world can feel big when you're in the
- 34:54Speaker 2: middle of it. It's growing rapidly. It's like a shiny
- 34:57Speaker 2: spec compared to the centralized tech where world. So I
- 35:01Speaker 2: mean one thing we've gotten from Greg's technology with ASI
- 35:05Speaker 2: chain is I mean we're gonna launch hybrid nodes between
- 35:08Speaker 2: ASID chain and Cardano, Etherium, Cosmos and then other chain.
- 35:13Speaker 2: So you should be able to build cross chain liquidity
- 35:15Speaker 2: pools like steak tokens in one chain and get rewards
- 35:20Speaker 2: in another chain. You should be able to run an
- 35:22Speaker 2: AI process where part of the network is on Cosmos, Kardana, Ethereum,
- 35:27Speaker 2: part of the network is within within ASID chain, and
- 35:30Speaker 2: I think how all that will evolve is quite flexible.
- 35:37Speaker 2: But we need all the different chains to be cooperating
- 35:41Speaker 2: together as much as possible to sort of overcome the
- 35:48Speaker 2: advantages that big tech has in terms of money and momentum.
- 35:52Speaker 1: Yeah, definitely, we need the capital to flow as easy
- 35:55Speaker 1: as possible to where it needs to be. So I
- 35:58Speaker 1: think that intropability is important. And now we don't have
- 36:01Speaker 1: a lot of time left.
- 36:02Speaker 2: Ben.
- 36:02Speaker 1: But with the ASI chain in dev net and then
- 36:07Speaker 1: moving to test net, Uh, what can develop developer minded
- 36:12Speaker 1: people do right now? Or if people aren't developers, when
- 36:15Speaker 1: and how can they get involved in this?
- 36:18Speaker 2: So with devnitt, you can jump in and see what's
- 36:23Speaker 2: there and understand the architecture and you can start prototyping
- 36:28Speaker 2: AI software or other DeFi tuoling whatever is your thing
- 36:33Speaker 2: on top of the DEVNIT for AI developers, I mean,
- 36:37Speaker 2: you can dig into the hyperund codebase in GitHub and
- 36:42Speaker 2: start experimenting with how to use different sorts of AI
- 36:46Speaker 2: technology within your AI workflows, which could be a GI
- 36:51Speaker 2: or it could be as simple as like using using
- 36:54Speaker 2: more flexible knowledge metagraph inside graph rag for an ll
- 36:58Speaker 2: M or something. I mean, there's all sorts of things
- 37:00Speaker 2: can do now. When we move from devnt to test net,
- 37:05Speaker 2: it gets more interesting in that that will be more
- 37:09Speaker 2: like that the dev net has the structures there, but
- 37:11Speaker 2: it's not yet scalable, so you can use it for
- 37:13Speaker 2: prototyping when we have more scalable consensus, which will come
- 37:18Speaker 2: with the dev neet. Like like it's just some we're
- 37:21Speaker 2: moving some components from scale out to rust basically, then
- 37:24Speaker 2: then we'll have a lot more scalable then then, I mean, ironically,
- 37:29Speaker 2: scale out is not actually the scalable language in spite
- 37:33Speaker 2: in spite of the name. Yeah, I mean, I mean
- 37:37Speaker 2: then it still won't be with full tokenomics. But but
- 37:42Speaker 2: you'll be able to put AI applications on the test
- 37:47Speaker 2: net and round them on a big a big network
- 37:50Speaker 2: and sort of sort of see see what they can do.
- 37:53Speaker 2: And there's a So there's a lot for developers to
- 37:56Speaker 2: play with a absolutely, I mean from an investors standpoint,
- 38:03Speaker 2: I mean, I'm not I'm not a money guy. I'm
- 38:07Speaker 2: an AI I'm not gonna hawk our token, but you can,
- 38:09Speaker 2: you can, you can look and see, like the fetch
- 38:13Speaker 2: token may be a very good deal at this particular
- 38:16Speaker 2: moment to the recent historical events that have pushed the
- 38:20Speaker 2: price probably below what it's it's logically worth. Right, So
- 38:25Speaker 2: there's a lot of interesting things there now. Tokenomically, next
- 38:31Speaker 2: year when we launch a SI chain, there's gonna be
- 38:37Speaker 2: very interesting opportunities to buy a license to operate a
- 38:42Speaker 2: shard and then kind of monetize your AI or or
- 38:49Speaker 2: DeFi ideas on the shard and then profit from the
- 38:53Speaker 2: LSD emissions on that shard, and we've tried to learn
- 38:57Speaker 2: a lot from Dynamic Tao. It's very interesting what they
- 39:01Speaker 2: did with the sub network economy. On the other hand,
- 39:04Speaker 2: we're working hard to make sure the tokenomics and the
- 39:07Speaker 2: aside chain reward real products that are generating revenue from
- 39:11Speaker 2: delivering valuable services. Yeah.
- 39:14Speaker 1: No, I love that you're switching, and a lot of
- 39:17Speaker 1: ventures are switching to emissions from revenues and you know,
- 39:21Speaker 1: sort of the traditional business rather than just market cap
- 39:25Speaker 1: to start, you know, stockpiling tokens. And I can leave
- 39:29Speaker 1: more information on the ASI chain launch, the dev net,
- 39:33Speaker 1: just Singularity Net and the ASI alliance and fetched token
- 39:37Speaker 1: everything else in the show notes below, because we covered
- 39:40Speaker 1: a lot of topics. I really appreciate your insights into
- 39:43Speaker 1: all of it, decentralized AI, Sofia and her family there
- 39:48Speaker 1: that you're showing in the human eyed robots, and wishing
- 39:52Speaker 1: you and the team all the best and everybody in
- 39:54Speaker 1: the ecosystem growing together. And I would love to follow
- 39:57Speaker 1: up again in the near future.
- 39:59Speaker 2: Cool. Thanks right, Yeah,