 And there seems to be a lot of artificial intelligence that is coming into the world. And how does one accumulate that into the general intelligence singularity net? And then how do people access the general intelligence? How do they apply it to their practices? Yeah, this is a complex story, unsurprisingly. So I mean, we can really look at what we're making as a, I guess, a network of networks of networks in a way. And of course, human life is that, right? The brain is a network embedded in the larger network of the body, embedded in the larger network of society, embedded in the larger network of the ecosystem, right? So AI has multiple levels like that also. So I have the OpenCog system, whose knowledge representation is a network, a weighted labeled hypergraph, and there's multiple AI algorithms acting on this common knowledge graph, such as some neural nets for pattern recognition, a logic engine to do with abstract knowledge and reasoning, evolutionary learning which evolves and creates new things, and these all act together on this common knowledge graph, which then has a goal system associated with it, and can try to choose actions that it thinks will help achieve its goals, given the current context, where the understanding of the context and the actions is given by all the knowledge in this graph, right? So that's OpenCog, which in itself is very complicated and we worked on it for a long time, and which I think is quite good at generalization and abstraction. We're still working on optimizing it, but I think the core is there. The singularity net is a platform in which multiple AIs, maybe based on quite different principles, can be networked together so they can all talk to each other, they can all communicate, share data with each other, outsource processing to each other. So these AIs can provide services to outside users, they can be used in the back end of various software products and websites, but the AIs can also outsource services to each other in complex patterns. And so that's another level of network, right? So I think you need some AIs in there that have a fundamental power of generalization and abstraction, like OpenCon does, but this can then work together with simpler types of AIs and they can coordinate together into a network whose intelligence in a way is greater than the sum of the parts. So you can look at that sort of like how different organs in the body are networked together and they each have their own intelligence, but the whole obviously has more coordinated behavior than the sum of the parts. But different cities on the economical group. Yeah, as well. Then we're also working on something called the DIA, the Decentralized AI Alliance, and that's an alliance of different decentralized AI projects. So singularity net is ours, which networks together OpenCon with a bunch of other AIs, but then there are other projects like Ocean Protocol, which is a blockchain-based decentralized data network. There's Shivam, which is a decentralized network for genomics-based AI. There's DeepBrainChain, which uses blockchain for fast training of deep neural networks across many machines. So we want to network together these different decentralized AI networks into a sort of network of network. So that's why we have a network of networks of networks. And all these levels, it's complicated, but that's what we see in nature. And I think in different ways, that's what we see on the internet already. So I think that's the right way to think about it. That doesn't mean you don't need core generalization and abstract reasoning algorithms in there. It's just that these, and in some ways you could say these are the crux of it for AGI, but all the other parts are important too. I mean, just as in your cortex, there's certain micro-circuits that are really good at abstract learning and reasoning. But if you just put these micro-circuits in a vat, they're just going to abstractly reason about their own abstract reasoning in sort of vacuous loops, right? It's by putting them in the network of the brain, which is the network of the body of the society of the ecosystem that you're getting. You're really giving these abstract learning circuits the ability to prove their worth. Yes, yes. OK, I liked how you were describing it. It finally started coming together more for me and I like that and hopefully for others as well. OK, now when you're feeding OpenCog, which has the ability to abstractly reason, when you're feeding it a to-do of some sort, how does it parse that and make sense of what it needs to do within its framework to bring about the result? Well, OpenCog can ingest a lot of different types of knowledge. So I mean, there's a language parser where it parses out English language now into nodes and links that try to represent the semantics of that English language. It doesn't always work, but sometimes it does. And you can also take computer programs like little Lisp programs or something and import those and then they turn into nodes and links in its knowledge base. We also have specialized data importers for types of biology knowledge, for example, or accounting knowledge and so forth. And when you import the data, it goes into the whole and it's organized into buckets, nodes, areas. Not buckets, it's organized into nodes and links which all connect to each other. So not never the buckets. No, it's not partitioned. Exactly. But then the reasoning engine looks at the nodes and links that came from data, including vision and hearing, as well as databases and language. And the reasoning engine takes those nodes and links that came from the outside world and builds new nodes and links. Okay, and that's how you add to it. Yeah, yeah. And then... Because I mean, what reasoning does is takes empirical knowledge and then generates new knowledge based on the empirical knowledge. And then when you're adding the knowledge and then you're trying to get an answer of sorts from that, how does it compute? There is a tool called the pattern matcher in OpenCog. So if you want to find a collection of nodes and links matching a certain pattern, you submit a query and it searches the whole knowledge base to find things matching that pattern. And then there's the logical backward chainer which if you give a pattern, it tries to find things that approximately or probabilistically match that pattern. And so if you're asking a question somewhere or another, that question is turned into a crisp or probabilistic pattern matching query that goes against that whole knowledge base. Okay. And then now... Oh, yeah, we're getting pretty deep in the weeds here. Yeah, but that's really important. Adding the knowledge, making the query and then getting some sort of results. And this is what I want to do a little bit as the roots, as the deep roots. And then also then I see the OpenCog with the SingularityNet able to work with the different AIs that are... Yeah, so I mean if someone makes a great AI for, say, recognizing patterns in videos or, say, for importing data from accounting spreadsheets or something, you want an easy way for that to feed knowledge into an abstraction engine like OpenCog, right? And so SingularityNet makes an easy way for different AIs written by all sorts of different people to connect together. And I mean if someone else makes a different reasoning engine with different strengths and weaknesses relative to OpenCog, that can be there in the SingularityNet also, then some application could consult two different reasoning engines and decide which answer it likes better. Or one of the reasoning engines, if it gets stuck halfway through doing reasoning, it could ask a question of a different reasoning engine, it could help it get unstuck, right? So there's a lot of potential to having different AIs able to compete with and consult each other within a common network. Interesting. Compete and consult with each other in the AI networks. Interesting. And then the decentralized AI Alliance gets in there as well. So this is more new. This is your newest. That's new, yeah. We announced that a few months ago, but we're still forming it. I mean the idea there, so there's a bunch of projects that are using blockchain related technologies to make decentralized AI networks that are owned by the participants and democratically controlled, and they have different focuses. So I mean SingularityNet is a pretty generic framework for interconnecting AI. Now DeepBrainChain, for example, is a decentralized network that is really fast at training deep neural nets. And Shivam is a decentralized AI network that specifically gathers genomic data from people, stores it securely, and then does some analytics on the genomics data, right? So we don't need to replicate everything they're doing. We just let an agent, a node in the Singularity Network, talk directly to a node in the Shivam network or DeepBrainChain. So if a SingularityNet node needs some neural net strain really fast, it may outsource that to DeepBrainChain. Cool. If a SingularityNet node is analyzing genomics data and wants more genomics data, it might ask the Shivam blockchain, hey, do you have any data filling these criteria that I could use? And then the AGI token, which is within the SingularityNet ecosystem, can be automatically converted on the back end into a Shivam token to pay for that. Cool. And of course, if you have reputation and rating, like if one guy has written a bunch of great AI in SingularityNet, where he has like a 4.5 star rating, then if he starts to do something in Shivam, he should get a high rating by propagation from SingularityNet. So you can have both sharing of data, sharing of currency, and sharing of reputation and ratings among different decentralized blockchain-based AI-related networks. I think this is how you can build an alternate AI ecosystem that can really take on the tech giants. Whoa, that last part, yeah. But there's a lot to be done, right? Yeah. And it's not just building the tech, it's coordinating a lot of people with different philosophies and incentives also. Yeah, that's right. I love this, so when a query is sent in, you're not even then limited to what OpenCog and SingularityNet has. Well, that's right. Then you can access... Access a whole bunch of other networks, yeah, that's right. I mean, that's how AI should be, right? Shouldn't be siloed off, it should be open networks connected to other open networks and learning and growing from each other.