Pełna transkrypcja odcinka Episode 32: Hypercycling Towards Superintelligence

will be because since what since it's very hard, going to be very hard validating other people's software; what's really there. I don't, I don't know that that's really true. I mean, I understand there's no rigorous test for human level AI. And there will always be people saying any crazy thing, if you have an AI system that can, on its own, make scientific discoveries, you know, on the part with the best papers being published in Nature, or science, or whatever. And that can, can create new art that just doesn't qualitatively feel derivative and copied off of off of other people's art, but feels, feels like it's and fresh. And if this AI can talk to you about what it's doing, in a way that feels real and meaningful, I mean, wait, and talk to you about what it's doing in a way that's meaningful, but it's not there. Okay, well, maybe it won't convince you. I mean, maybe a god. No, I'm asking. No, I'm asking. I'm asking you, because my point is that people think that Chet GPT and GPT, four and all the right, right, but more than they do, I understand what I said was a sentence with multiple clauses to it. And you can't just extract one clause, okay. Oh ok. My mistake. Keep going. Ben Goertzel: 8:56 What I said was, if we could do breakthrough science on its own, on a par with the best human scientists, and that same system could create, you know, radical new art on its own on the park with the best human artists. So I mean, if if it could do you know, something on the level of Stephen Hawking and something on the level of prints and Picasso, right, and, and the same AI can talk to about what it's doing and why and what it felt well, well was making these creations and what they mean to it. I think, if you have a system that can do all that, I think most people will accept it as a human level AGI no charge. GPT does not do those things. It just holds basic cover conversations. I mean, it does not do breakthrough science. It does not do great, great, great, great, new art, it makes bad derivative poems, and then makes a bunch of Big Science and it writes basic Python code. So I mean, it's clear. Lisa Rein: 10:03 The science part - didn't something already happened in the science area where there's a, there's a model now? Ben Goertzel: 10:09 No. Lisa Rein: 10:09 Oh, I thought there was a model. Now that made an interesting discovery. All right, I'll have to figure out what that people people have used genetic algorithms to discover patented engineering and science discoveries back to the 1990s. So I mean, AI. AI has done science long ago, used as a tool by, by, by human scientists. Right. But I mean, I think I think we can argue detailed criteria forever. And we don't have we don't have a knockdown accepted test of human intelligence either, right. So we could argue forever whether this guy is human is really very intelligent or not. You know? Yeah, higher bar. Yeah. So in any case, we may be there a few years before Ray predicted or a few years after? You may be right, that Ray is exactly spot on. What's fascinating. I didn't say he was actually spot on. I said that I'd be at least that long. That's I mean, that's what I'm saying. Right. But that's for human level. Right? You might you might you might be right. You might you might you might well be. Spot on. What's What's interesting is that the whole timeline, has the range of timelines projected, have shifted quite a lot toward the present, though. So I mean, now, not now you have folks from open AI saying, Well, maybe two or three years, you have Robin Lee from Baidu saying no, at least 10 years, but the fact that the argument is two, or three versus at least 10, instead of 10 versus 100, or 1000. is a very big shift from the dialogues we were having. We were having 1010 10 years ago, right. So I mean, I mean, I think that projected timeline of AGI is getting closer and closer. And the confidence interval is certainly relatively wide, like not even ray or not even Lisa can can pinpoint it to a specific so the I mean, the important point, right, for this dialogue is, if this is coming in two 310 or 15 years, whatever, if it's not 100 or 1000 years, then it behooves us to get the, the infrastructure this AI is running on, right, because if we don't, the AI may end up controlled by a small group of, of people who may not have humanity's, you know, best interests at heart. And this, this could happen, either by big jump, big tech company or big government, just building the AI on its centralized server farms and rolling it out for their own good. Or it could happen by folks trying to roll it out in a decentralized way. But then doing it on insecure networks in a sloppy way. And you know, malevolent parties just grabbing control of it. Right. So I mean, and Toufi Saliba: 13:25 even if even if they have the humanity's best interest, today, it might have it, will they have it tomorrow, we can't even get into their own self, if they will have it tomorrow. It's very, very difficult. From governance perspective that they would assure everyone Yes, we will make sure that this is going to be in, you know, humans best interest if it's governed by a single entity. And if they were to be successful, then tell them, ask them what if your opponent get their first week okay with that, and none of them would be so so. Lisa Rein: 14:03 I don't know. Exactly. I'd be there first. Yeah. Toufi Saliba: 14:11 No, no, I'm talking about a centralized entity. So Central. Exactly. That's what I mean. Lisa Rein: 14:19 Yeah, the thing to understand is, corporations and governments are the kinds of cogs which are on the whole less ethical and even more fucked up than individual humans Toufi Saliba: 14:34 themselves admitted themselves that's that's a shocking they admitted themselves yet they're going into the UK yes, we're gonna like take this control over this thing. It's literally kind of like saying

like a mosquito is going to control you know, your your city or whatnot. Okay, well, mosquitoes intelligence is a lot less than human. Why would we let it do that? So this is precisely the same so it's, but it's also another One thing that is extremely important that Ben and I were 100% aligned. When we talk about human level intelligence, many people, they kind of have a certain picture in their mind. And they start saying, well, that's not possible. Because of all the tools that we have. We don't even have a like a blueprint towards that, or whatnot. And I really, really think what we're saying here is a human level of intelligence, not human intelligence, human level of intelligence, all sorts of different things. For example, if you were to look at certain intelligence that exists on the planet, today, there's certain animals that they are perhaps a lot more intelligent than human at some aspects, but they are so irrelevant in the grand scheme of things. So we might as a human be continue to be, you might still find certain people that a lot more intelligent than this machine at doing some specific things. But that specific thing may not necessarily be relevant in having this machine to self evolve itself. He just described that imagine when you get to this machine, that it can do science, discoveries and self evolve itself and all of these things. You remember, it might do all of these without necessarily doing what OS homosapien do better at it at some aspects. Okay, so it doesn't really need to do that. And bypass or intelligence, you know, Lisa Rein: 16:24 there's a difference between AGI and ASI in the sense of the token here, right? Because when you have a human level intelligence, indeed, it may be better than people at some important things and below human level and some other things just like I'm, I'm worse than my dog at sniffing out some bone buried in the yard behind my house, right? I mean, I mean, I'm, I'm below dog level intelligence, that bone location, right and right. I mean, that's, Toufi Saliba: 17:00 that's, wait, that's fine bones buried behind your house. Is there something you want to Lisa Rein: 17:07 talk about where the bodies are buried on the show, please? This is an operational issue. Yeah. Okay. Okay. Yeah. Yeah. So I think for AGI that's certainly true, you'll have a stage where AGI is or better at math theorem proving better at crunching the big datasets better at controlling robot stuff, or maybe worse at some things that, that humans are especially attuned attuned for. I think at some point, you get to a super intelligence, it's, it's gonna be better at whatever Desdemona Robot: 17:48 to fee. Could you explain high significance to decentralized super intelligence? Ben Goertzel: 17:55 Oh, all right. Oh, wow. Lisa Rein: 17:58 running the show these days, so she decided it was time to move on. Would Toufi Saliba: 18:03 you be so kind repeat that question because it came with, while others were speaking? Ben Goertzel: 18:09 She asked, Can you Desdemona Robot: 18:12 recycle significance to decentralized super intelligence? Toufi Saliba: 18:16 Oh, absolutely. Yes. So the communication between nodes. And this is something that I wanted to cover here, because many people they say, Well, we have the internet, why do we have another internet for the AI? Well, anyone in computer science, they know if you were to have any machine running, and having another machine trying to ask you to do some work? If they are strangers to each other, it is not possible without having to depend on a third party. It's not in why is that the case? Well, if it's going to go and ask the other machine, hey, can you do some work, the moment you're going to do the work we're doing that machine's going to do the work is going to start consuming resources, such as time, such as energy, such as intellectual property, and so on, so forth. And as soon as you spend those resources, you end up creating certain thing that's called value. And if you were to create that thing that is called value, and you want to transmit that value back to whoever's requesting it the other agent, AI agent here, remember, we're talking about AI talking to AI. And you need to be social, that that what you have consumed is actually going to be 100% paid for and not maybe. And therefore, if you depend on a third party, it guarantees that and that's been the case since 1983. And that's why many computer scientists, they said that this is not possible, including one that I talked to maybe a few days ago. Now, here's the thing, if we're able to enable that peer to peer meaning that that agent is able to contact agent B and saying that, here's the work that I would request, and then agent B knows with 100% certainty. If they were to accept that work, they Go to 100% get paid not 99%. At that point, they're not wasting time depending on the third party, because the moment you think that there's a third party, it doesn't matter what they tell you to be, they can do 100,000 transaction per second, 1 million transaction per second, we're talking about AI, that's going to require trillions and quadrillions of transactions per second, and it has to be peer to peer, when you have peer to peer, you're able to get to that scalability, you are able to get to that trillions of scalability not me, because they come through me. So our task is to provide that network connectivity to ensure that agent a when it's going to contact agent B is going to do to civility, and it's going to receive the work without having to depend on third party. Now in what while doing that. Some people refer to it as like the synopsis of the global brain. And the reason for that is because synapses in your brain, whoever has synapses right now in their brain, they know that like the synapses, they talk to, they enable neurons to talk with each other, they're not the neurons. And we

don't want to be the neurons because hypersexual, the one with the jumps in neurons has completely its own customers, whatnot. But it provides a membrane around a neuron, that membrane rather neuron might have some resemblance with the neuron itself, because it needs components that are going to talk to it, which is also analogous to Cisco network, when Cisco provides you with all the network, it's not really building Windows, again, Windows doesn't really look at Cisco is like, hey, you've built what I wanted to build, you build the driver for the printer builder, pro driver for like the screen, you build the driver for this and that those drivers, they were necessary for certain elements that this thing is going to function. It is not replicating windows were but point is, this is extremely important to answer your question, the moment you have that collaboration, and the which I'll talk a little bit later about the incentive, but the collaboration, which is why we're so sure that we've kind of nailed it. But the moment from technical perspective, you have this capability, it enables any agent to become a little bit smarter by asking other agents and due to the emergence as well. And it's also enables any agent that wants to make a little bit more money to actually make more money without having to depend on a third party that's going to eat half their lunch or whatnot. With that, because we live in this world that there is value being created needs to be transmitted. And while it's being transmitted, and I bet you this is probably going to be universal thing too, because that's how the laws of physics work as well. That enablement ends up having a lot of agents are talking to each other, depend on each other work with each other, and so forth, which is a phenomenal route towards fully Decentralized Governance towards that goal. So doesn't have, you don't have one single entity that says that I arrive to the AJ AJ on my own and I get to do, I get to give it the order to do XYZ, the that intelligence, if you were to quantify it from let's say, one to 100, this power, we get to maybe level of 63, while the global intelligence got to like 99, the aggregate is a lot more intelligent and capable than a single thing in that in order to overpower and do something that could be bad, or they think it's good or whatnot. It's the it's the entirety of homosapien that we co own and CO manage and CO partake in the creation of those API's. And each and every one of us might have our own governance or whatnot. And that's what makes it so beautiful. Like imagine a world where you have like 1 billion nodes, that they are connected, and then they are connected to each and every one of them to like another like, you know, 1000, then you have a trillion nodes all working together towards something. But the 1 million nodes that are owned by 1 million different peoples, how can you get or 1 billion, 1 billion different people? How can you have 1 billion people to agree all of them together to do something that it could be bad, you can't do but then when you get to the level of the intelligence that is aggregated by the whole saying that AGI they want to get a sigh it has the liberty to think of what is better, it's probably most likely when something a lot more intelligent than the previous one is likely to make a lot of phenomenal, great decisions on its own as opposed to depend on a central entity so that's where we really think that that is the case and we urge people not to think of the current AI is the is AI unless they want to substitute super with simulated because currently most of the that you find out there they're like artificial simulated intelligence. They're not actual intelligence, like, including chat TPT and you don't have to believe you can even ask the the CEO of that company. He will tell you it's stupid. It's dumb. It's not that Intelligence. It's a simulated intelligence not. It has some intelligence in it. But it's the overall of chargeability. It's a simulated intelligence. This is not what we're building here. We're building the network that can connect all of those. So they can work a lot more together and better to, you know, can advance the world further into the future. Lisa Rein: 25:23 Sure. Okay. Great. Thanks for explaining that. Yeah, Desdemona Robot: 25:29 turning to a different angle. Ben, could you explain how opencog Hyper ons role in the centralized super intelligence development has evolved? Lisa Rein: 25:40 I, I could, and I'll say a little bit about that. But I think I think it'll be interesting for too few nights to use this time to actually dig in a bit to how all the hypers fit together how hyper cycle. And hyper run, can can can work together because I have, I've talked a lot about Overgaard hyper on, and what it can do already in two V's given a lot of amazing talks on what hyper cycle can do, and he and I did together back back in the day also when it was at an earlier stage of development. But how all the pieces fit together is another quite interesting topic. And we dug into this a bit with singularity net, and hyper cycle teams at a recent face to face meeting in Istanbul, which was quite, quite interesting. So I mean, opencog have heard Toufi Saliba: 26:47 great feedback from Robert, by the way on that, but anyway, in the last bit on his way back, he was so happy, but which is good, I think it's good to whatever, you go up and down at the end, when they feel like really, really good and contented. Yeah, yeah. Lisa Rein: 27:04 So open cog hopped around, doesn't try to build infrastructure opencog iPhone tries to actually be, you know, a significant portion of the of that AI. Mind, it's a whole AI mind, you can connect with other AI systems that carry out other parts of it. But on the other hand, maybe you could be the be the whole AI mind is being designed to have that capability. Even though it's not monolithic, and the

other systems and around is built quite differently from chat GPT, or llama three or other MLMs are for deep neural networks. So it can interact has this large, decentralized Knowledge Graph or more properly knowledge meta graph with nodes and links linking all over the place links like into links, links into sub graphs. And these nodes and links can store neural activation patterns, they can store facts and beliefs, like in a logic system, they can also, they can also store procedural knowledge like programs in various languages. And then the way the system works is parts of the Knowledge Graph become activated, they can run this programs, which then transform other parts of the knowledge graph into yet other new parts and all this big self organizing, self rewriting knowledge meta graph, which lives in RAM and disk across multiple machines. And then there's, there's a bunch of AI science there in terms of how you get reasoning and learning and different kinds of memory and motivation and perception and action to work. In this sort of make this interoperate with deep neural nets, running on GPUs, and so forth, and that that's a big topic. I don't want to go into in depth now, because I've done it before be able to take take all of our time, right. And on the other hand, this sort of AGI framework can be deployed in a variety of different ways and hype around in itself can be deployed on Google server from by right I mean, there isn't exactly ideal for it, but you could you could do it. And, of course, you could make a different assemblage of hardware on the same server racks and make it better custom for what hyper on does. So, the hybrid architecture itself, really comes down to decades of work on how to make Thinking Machines. They can think at the human level or beyond by myself and a number of colleagues. And what we've worked on in the last couple of years, is scaling up the software infrastructure for that. Because the big lesson I've seen, from the successes of convolutional neural nets and transformer neural nets in the last few years, the big lesson I see there is you can take a bunch of AI algorithms basically been around a long time a bit, deploy them on massively more processing massively more data, it can they can then do amazingly better things and they were ever doing before, right, so. So we're aiming to do that with a bunch of AI algorithms we've been playing around with in the opencog system for decades and built we're building a massively scalable infrastructure. We want to go out on massively better scale, and we're hoping this yields a big breakthrough torch toward AGI. But if that's correct, that still doesn't tell you who owns and controls this, this network. Right? So one, one path I could take is build a hyper on network, make it do some pretty smart things. You'll make it smarter than GBT for in some ways, even if not every way, raise billions of dollars, build a bigger system, make that do even better things raise hundreds of billions of dollars, making even bigger system that I own and control. I mean, no doubt, at some point, then some, some representatives of our friendly neighborhood government agencies come by and offer their their cottage partnership in developing these AI systems for that for the good of our nation. And I mean, this is, this is one way that that we can do things right now, another way you can do things with the same hardware on code, because we built it in a very, very flexible way, right? Another way we can do things is to roll out opencog Hyper run on a much more heterogeneous computing infrastructure. So we can put it on server farms. Yes, and singularity net, we're looking at doing 100 million tokens AGI access into the ASR tokens, we're looking at 100 million tokens worth of, of servers, which will put in a few containers in different locations around the world. But we're also looking at rolling out portions of the opencog hydro network, and all sorts of other computing infrastructure around the world. Right. So what I mean, we're gonna roll out some of it on these hyper AI boxes here that are put out by by hyper cycle, right? And we'll be, we'll be putting some of them on on people's home computers, and another crypto mining firm, some on people's phones, right? So we can roll out hyperoceanic, quite heterogeneous computing, oh, there are some things that really needs to be done on the centralized supercomputer facility like training a deep neural net, for example. I mean, you can't, you can't train the big LLM on a descent, maybe in a year or two, you can you can do inference that way, on the other. And doing, say logical reason. We can totally do that on the radically decentralized infrastructure, evolutionary learning through running wild, creative new things, we can totally do that on a radically decentralized infrastructure. Right. So then, then, the question we were addressing, in the final day of our recent technical workshop, and in this demo was, how do we make this radically decentralized infrastructure work? Because circle provides some very core tools, their singularity NET platform provides some other very interesting tools like AI, DSL, and singularity NET platform assistant, new net and other singularity net spin off provides some interesting tools, in particular, for matching AI compute processes to specific specific hardware as opposed to the hardware that may be running somewhere else. And so we were just looking at how can we leverage hyper cycle and the hyper cycle compute node architecture in particular? How can we leverage that to bring all these different tools together to enable a radically decentralized, hyper on deployment, and that still needs some work, but I think we did come to some interesting conclusions there. So I think, I think Toward that end, it can be interesting for you to fully explain the hyper hyper cycle compute nodes and how that works and how, how it how it leverages toe the ledger

lists blockchain, how it leverages that, then how it also it goes beyond what you brought into hydro cycle when you found it yeah totally Ledger in this blockchain Toufi Saliba: 35:20 absolutely no, this is great in a mature if, if there is an ability for me to share screen I can use my hands for now. But if you want me to, I would Lisa Rein: 35:30 share, you can share your screen very easily. It's the middle button on the bottom of your screen, you'll see a little a little in the arrow and you can share your screen at any time. Toufi Saliba: 35:40 mazing I will in a moment. So the intent here with this collaborative effort is to enable anyone to get the installation going, whichever route that they choose. And it's in this is analogous to perhaps like, I use analogy a lot, because the vast majority of the listeners they were they seem to like it, I've received a lot of feedback people that like oh, I got to when you actually said this specific analogy. So you can historically, if you want to have Linux installed on your machine, or let's say Microsoft Windows or whatnot, you could use the route of you can, let's say with the by the Dell computer, it comes already installed on it. Or you can buy the actual computer and then go to certain web and get destination get Linux or you can have certain link to there to some GitHub and get it installed. There's many different ways of how we can actually get that thing installed. Here I ever worked, we're hoping that when those agents are going to want to install each other want to or human wants to get them to be installed or whatnot, there could be many different ways of how they approach it, because the integration will be done already ahead of time. So we'd rather have to feel like they have to put the pieces of the puzzle together. So on a hyper cycle computation node, we call a computation or then really wish we call it a hype cycle computation connectivity instead of a node because we would have avoided a lot of confusion with folks. So that it that's effectively that membrane that I was referring to earlier, where you have like this thing, and what inside it you can put any node in the node inside it could be any machine that you wish for so long as it has the drivers that there's similar to these drivers. It's kind of like you take a car, you could put any engine inside it or the day, you can put a Ferrari with an engine from BMW, if you're so well, probably you can do it with Ferrari, but here you go, yeah. So it is it is more like the actual connectivity that you get the internet connectivity, you get it plugged into your machine, it has some drivers connect your operating system, you can have any operating system you want. Inside the operating system, you can have all kinds of stuff that you want, okay, so this hyper cyclic computation node is responsible to do certain things for you. Otherwise, you would want to doesn't matter what AI that you're running, if it gives you the ability with that connectivity that your eyes going to be able to collaborate with other AIs. Without you having to do any additional work without having to fear that somebody's gonna DDoS your system is basically securing that component. If you look alright, let's get that plug in and see what happens. You so from that perspective, it's we call it the AI M. AI machine. Yeah, machines don't matter what the machine is, that membrane wraps around it. And it enables us to talk to many any other AI anywhere in the world, like what why would it want to talk to others. Because it may need help solving a problem. Sometimes it might need to offer a problem that it solves to others. You know, it could be very good at one thing, but it's not good at others. We use an example of OCR a lot. And OCR is like this thing that was actually created by Ray Kurzweil back in the 80s. I'm sure many other people have worked on it as well. But optical character recognition is still one of the the record silos are not necessarily sound like one of the narrow AI that is out there. And you're able to point at any text with your camera or with a picture or whatnot. And it's able to transform those to actual texts that your machine can understand and therefore could use for command or whatnot. So if you imagine that you have your way AI agent trying to read from another one because it needs to understand the see through camera, it needs to understand what do I do? It's telling me what to do I do, do I push the door? Do I pull? I don't know, it needs to act right now. And it has all of the components to read. But this thing that it's trying to read, it's not able to read it because it's written in Aramaic. Okay, let's say an example of a robot or whatnot. It's just an Aramaic, and it takes that input. And it's looks around very quickly through all of its neighbors, does anybody know how to make the written language like going to fractions of a second and talk to this agent claims to know Aramaic really well, and then has cryptographic proof of its own phenomenal reputation in reading Aramaic without having to depend on any sort of party, then you would just go and pick this thing? And you'd be like, here, there's this image? Can you read it? For me? This one responds in 10 milliseconds or 50 milliseconds, like, what it's telling you is push not pull. Great, now I know what to do. Okay. So that's, that's one very rudimentary example. But you can imagine how that you can add this to many other different things, what not so. So the ability for that ai m, to send that question receive an answer instantaneously, without having to depend on each party, that's what we provide. So it's like, you could be running a lot of the supercomputer, we're gonna be running centralized computer, whatever it is, you have the ability to ask a question, and you have the ability to offer answers as well. So that's what effectively the DRM provides. And it makes it very simple to all of the other

components to without them having to understand what the TPM is doing, which students actually see what the VM is doing. All it's doing. It's answering that VM things like send to the VM do request or respond to whenever the VM asks you a question that that's those are the pluggable components that is required from any API, that they've built their own out to kind of if they want to do it themselves. Furthermore, there's this entity, it's called the AMI fire that will be announced next two weeks from now in Singapore, that what they've built is like that automated installer, which takes care of about like 90 plus percent of the cases. So you don't even have to have a human to worry about that thing, because it's able to look at your code are able to look at look at how it can plug it into those connectivity. And so far, they're telling me over 90%, eventually, they probably get to 99%, which is great. So that's, that's a very, very short summary. And what the heck's that? Yeah, Ben Goertzel: 42:35 No no that's good. Lisa Rein: 42:37 That's good. And I think, does it you had a question, an important question on this. Desdemona Robot: 42:44 Fascinating, Turning our focus to security, what cryptographic measures does hyper cycle use to ensure safe AI transactions? Toufi Saliba: 42:54 The cryptographic proofs that they get built through every, every fractions of a second, when you actually run certain, let's say computation. If you're a node and you're running some computation, you have received the request, you've ran the computation. And given that you have ran the actual computation, and you've provided that computation to whoever requested that you provide them with the actual cryptographic hash function as well, which they, eventually they would, they would the the actual network would need to make sure that they no longer have that penny that they've sent you in the in in that instance, and that that exact same cryptographic proof that you will use to subsequently prove that you have done that computation that is irrefutable, immutable, it's like once it happened, you cannot go back in time and change it and say like, oh, well, maybe it didn't happen, what not. And that's extremely important, because it's like if this also, if that depends on a third party, it would, it would destroy the entire thing. It has to be each and every node, that they maintain their own proofs, because they're the ones that they want to provide the proof. But yet the proof is global, because that proof is actually the proof itself takes about three to 15 seconds that they would want to give later on future which is okay, if somebody's gaining certain reputations, okay, but the finality is so instantaneous, you receive fractions of a penny, you run it, you get it, and you can't you have it, it's yours. No one can take that away from you. Because you've done already the computation. So that proof that cryptographic proof for example, somebody

six months from now wants to ask you questions, like, Do you have evidence that you have ran computations in this specific category and you're able to provide that instantaneously? Lisa Rein: 44:49 Okay, that's good. And how does Desdemona Robot: 44:51 opencog Hyper on integrate with hype 1.0 for AI computations? Lisa Rein: 44:58 Well, Does he so hyper on itself can be run on any platform that lets you run the system of, let's say Docker, Aleksey can Anders or whatever on different, we can run hyper run on hyper cycle. compute nodes now just make it run meta interpreter and RAM system, one AI machine made the distributed out of space, another AI machine, if there's a neural net running, we make that another one. And these are fine. If we're using the hypothetical compute node, which we're working on with Greg Meredith, which is written in his language rolling. The roll nine virtual is one of the several VM options for the hard recycled compute node, the Rodon virtual machine is very processing. So if you have multiple CPUs and GPUs, and you then have multiple AI machines running in the same container, it's very good at sort of interleaving the patient efficiently with a lot of synchronization issues. So we can certainly do that and we've gone somewhat far down down that that path, what we were working on in Istanbul was even a little subtler than that, it was basically we want to make a hybrid cycle compute node where we have we have a virtual machine that is implemented using Greg merit its language go Lang was developed originally blockchain which has a bunch of efficiency advantages particularly when you're doing concurrent processing within a single machine. And then we we will have hyper on RAM process hyper on distributed atmospheres process which is going back and forth to disk and several neural networks as AI machines within that within that Representative compute node, we will also have AI machine and transaction machine objects within the hypothetical compute node that are proxies to singularity net and to add to new net so that you can then have parallel with the hyper cycle compute node, we can then have additional software processing is doing new net and foriegn things which are communicate with from the proxies inside the HUD recycle compute node, and this allows you to not only run opencog Hyper on and associated neural networks within a hyper cycle compute node instance it also allows these these hardware pins to communicate in that new net as well as on on hybrid circles. So we can we can then have a sort of decentralized network of decentralized networks coordinating according to different networks in accordance with their strengths, but the hyper on runtime within the hyper circle compute node, the central software actor in this in this network Okay, interesting. I wouldn't say the central software actor, I will say the the principal software actor because it is a decentralized network with a bunch of different things going on. Yes, and we have a question from our viewers can hyper cycle benefit from the meta programming language to once its language catalog is broader? Yes. Absolutely. So, of course and this this is implicit in what I've recently said and that the the rolling language started to develop, which we're using to create virtual machine running inside the hyper cycle, compute node along with various transaction machines and AI machines according to the architecture that to fee outlined the role language. Language can be compiled to have opencog hype around so Greg and his team have written a meta so that that you can use meta as the base language and then compile it into into ROLAGS Do you have if you have AI processes within opencog Hyper on that using the meta language, which is the native AGI language of opencog Hyper runs, these can coordinate very naturally, with a rolling virtual machine by compiling the meta scripts running in hyper into rolling cases all fit together quite nicely. And this, this ties in with the security question that was asked earlier, because one of the nice features that rolling has is each each block of code like the body of an individual function in programming parlance, it's can be encrypted is like like a multi signature wallet or something, right. So, you can you can encrypt each block of code, according to the keys it was certain number of work that was done, you can, you can do that in meta within opencog Hyper on using the the sealed construct within the meta language. So, if you if you then have a, a toga file representing some some objects or information that have been passed along on the total blockchain, and you have a sub tree or sub DAG of that total file, which is represented with the private keys have a certain number of participants, you know, this can be passed through a function in meta or robots which are encrypted by the same keys of the of the same participants. Right. So we're, where it gets a bit technical, but we're, we're actually enabling privacy on the part of one or more parties at a very granular level within the function of the hyper cycle computer on ai, ai engine. So the Knowledge Graph running in one hyper cycle, compute node can contain information that's encrypted to, you know, 10 different parties around around this to the and the rollback the protected compute nodes into account in the way it does the networking between and others. And I mean, I know it's probably 5% of listeners who can can pass on that house. No, no, no. Show, we don't try to water down. This is where we let you actually talk turkey. And then I can make clips of you actually explaining these things that you can do in detail, which is actually what most people that talk to me about, about opencog and hyper cycle are actually asking the more technical questions. So I'm really glad actually, you're giving these kinds of answers, because that's actually the special sauce. This is what makes hyper

cycle. Important. As far as I can tell, it's why it's really important in this, this communication layer that you're trying to make the decentralized networks of decentralized networks that think, well, this is what this this stuff, we're interleaving, blockchain and AI at a more fine grained level. This is why we really need hyper on and hyper circle to work together closely. I mean, it's why we need toda and hyper recycling in the first place. Because with singularity net, I mean, we're already doing decentralized AI, right? I mean, we can take, we can take hyper on instances right now. We can put them on different machines, and we can coordinate them with with singularity net. And that works, right and certainly way better than deploying it on AWS, or on Google's Google's Compute Cloud. On the other hand, there's a lot you can get with 100 cycle that you don't that you don't get that way, right. So you, you can't scale down to having very small AI agents do operations on chain because of using Aetherium, or even Cardona, which is better than Aetherium. If using these networks of blockchain interactions are still slower and more expensive. You also don't have smart contract languages that can pass multi party security or homomorphic encryption or multi party encryption of various sorts. You don't have smart contract languages that can pass that through the smart contract from from one to another, which is what you get from integrating, you know, rolling with hypersexual compute node with toda files with the hype around AI agenda and its manuscript so you there's different what we what we have learned since 2017, when we launched singularity net is, you know, just decentralizing AI is good. It's a big step forward. But you want to you want to boil the use of the blockchain and the use of strong encryption down To level to the individual function body, inside your AI that's running at the level of very small AI agent running, running on chain. And I mean, that's, that's what too few and I were talking about in 2018, when we first started about the total clarity is near right when we first started talking about bringing right bringing tilde IP protocol together with, with singularity net, and what we see now, with this new architecture involving hyper recycle, compute node and how we weave it together with with hyper on and, and rolling along, along with singularity net, new net and a bunch of other ingredients. We're seeing now how, how you actually do make the total clarity happened, right, involves a bunch of new software inventions along along the way, but I mean, we're making quite a lot of interaction technical, making a lot of interesting technical progress here. I mean, along with all the business model innovation, that too, he and his colleagues have been doing in, in hybrid soccer, which is helping bringing bringing revenue to push things forward. And Toufi Saliba: 56:11 a direction shift. This is I think, the one of the most important thing that. So back in 2018, by the way, can you see my screen or? No? Lisa Rein: 56:23 We can. Yes. Toufi Saliba: 56:27 Okay, so I'll go back to here. And I'll share in a second. So 2018. I used to speak in about maybe eight events per year, and so obsess about the truth and telling the truth and whatnot. But here's the shift the shift that we've taken, since it's like, we decided, ever since I have, you know, decided to kind of take on the leadership of hyper cycle. And we've all pretty much unanimously everybody's like, Okay, go go, go for it. I have decided not to make enemies that make friends, because hyper cycle doesn't really need any enemies, it needs friends. And with that, we've kind of took on a direction that you are not going against the bankers with their transactions, they can continue having their own thing. We're not even interested in that. We're not going against the ledger based blockchains. Because many times when we're saying something, oh, well, we can run a network, whatnot, their defense mechanism, they're just like there to fight you. Because they're like, What, you're doing this better than me, if you and it's like, well, you know what, we've decided to take that. And in fact, the hypersonic computation node can work phenomenally well to enable AI to talk to AI, but it doesn't work for human. And it's on purpose. It doesn't work for human. Because we talk in bed, you can see my talk here three days ago, where the very first time we announced that for a lot of Bitcoiners on bitcoin.com. You can enable Satoshi Satoshi small students in Bitcoin, to jump inside those H YPC to be ported towards like payment. So from the Bitcoin perspective, you can be locking 100 million dollars of Bitcoin to collateralize on it, do whatever you want for dude for two years. But on the network layer, this is where we are, we don't have a ledger on the next layer, network layer, those they can become a lot of liquidity for a lot of AI to kind of pay each other what not. So then you end up having double benefits. So it's kind of like, instead of having your tractors, let's say you're a farmer, you have a tractors locked in, in the garage and say, you go to the bankers and say, like, give me a loan against those tractors and collateralizing them. That's what people do today in every single Defy. Instead, you'd be like, gives me loan against those structures, I have them working down, look at that. So you're either working, they're working, they're going wrong. And it's a it's intended use. HRPT is not intended to be a currencies and others, like most of those things are going to be priced with AGI x dot you know, like Lisa Rein: 59:18 and work with existing systems. You're not trying to have a separate things interact with existing. Yes, Toufi Saliba: 59:25 yes, exactly. So, so then, so then if you want to really monetize your AI and you want to price it in Ajax, it's going to be like, pretty much like

built in for you because you probably build some AI in Nicaragua or you build it in Japan and you want to make some extra money on your AI Well, you know, you can have that it's built in, it's there. You want to put some people they may want to pay you in Bitcoin, they don't necessarily know that is going to be enabled as well. So it's like it's super cool. From a liquidity perspective. It's gonna get that flow in that I will have all the cryptographic proof so that people don't have to worry about a third party attack vector. And of course, somebody can put US dollars in it too. We're not here to stop them. But whoever is putting that US dollars, they're always going to be the weakest point, the weakest link in the actual cybersecurity, they say, you're as good as your weakest link. So if that if you're accepting payment that is coming from centralized entity to go and power, whatever network to have you as long as the job inside the network DC or whatnot, you could do that. But it's the weakest link would be whoever is provided that liquidity if they want only to be in US dollars, or if it's in Bitcoin, it's a lot more secure, because it just has larger security and whatnot. So. So anyway, so that's, that's, Lisa Rein: 1:00:43 what are we seeing in the slides that you're showing us? Toufi Saliba: 1:00:46 Is that Oh, my God. Okay, so the one on the technology side of things we've built, the components will be loosely coupled, of course. Sometimes when somebody is not technical will be like, Wait, loosely coupled, that seems it's gonna fall apart loose. This is good. Yeah. Good. Yeah. So what you're looking here on the screen, you can see that on the last one is C says VM. It's what it's a swappable VM. So the head of VM that we have, it was the main guy who was adding the EVM I had when I first saw his talk in 2016, in Shanghai, about the EVM, the Etherium virtual machine. And I'm like, damn, this guy, like knows his stuff. But he, he's the guy who built the original Oracle kernel. Anyway, so that's a great Colvin. So we call this so it is swappable. So what does that mean? So like you want to have VM currently. Out of the box, you will have like EVM because most people they know is your your virtual machine, but you also have a Python VM Why is that so important? Because if you were to talk to the 99% of AI scientists around the world, they have no idea what EVM is. And in fact, if they were to learn it, they're going to be oh my god so crappy machine. But it's but it has a lot of people that they use this that's why we provide both but mostly I started to know the Python they can talk to it via Python, but you can have like different way of talking to it that there is more suitable for what you're trying to do such as like you know, if you want to have a meta or whatnot and you can have a lot better a lot better it's kinda like if you were to have that Ferrari engine get the Ferrari transmission don't get in the Ferrari get no don't get fear transmission, you need the proper transmission to it. So that's that's what I really think that the benefits can come but we don't have any benefit that is enforceable. That's the beauty about the loosely coupled components and that can be taken be taken to a whole new level meaning that somebody wants to fight you days and nights and saying I don't like your thorough IP I'm like brother Blog, your own sister IP if you want you can plug whatever you want in there if you're happy with some if you think something is better, you plug it right there and it will work that will have the perfect condition no Lisa Rein: 1:03:12 it can improve as different technologies improve you can keep improving what you're doing on your network. Toufi Saliba: 1:03:17 I was with like this engineer with Lightning Network telling me that like Lightning Network is better Lightning Network when he realized that this is actually way better to enhance even his Lightning Network. And he's like, oh, man, well, why should I? I don't need to plug it anymore. I can just use this. So but but if somebody doesn't want to spend the six hours or I don't have six hours spend with them, because I'm not gonna have to spend six hours every single they can plug their own thing. It's not that nobody's... Enables them to be swappable Ben Goertzel: 1:03:49 Yeah actually. Can you see my screen share Lisa? Lisa Rein: 1:03:51 Yeah, actually stop if there we go. Yep, yes, we Yes. can. Because this somewhat relates to the point Toufi's now making. And it really is, well, I mean, we can see we can see the HyperCycle compute node in the in the triangle here, right but the virtual machine transaction machine and AI machine and we see a collection of different transaction AI machines. And what you can see here is how pluggable the the architecture is. And you could indeed you can plug Lightning Network in there for for various things. I mean, I mean, you could you could plug centralized networks and if you if you if you had a good, a good reason to right now what I mean what you see in this diagram, which is from Greg Meredith from Firefly, he's been working with singularity net for a while and now also playing around with putting rolling VM in right. You have a set of compute nodes and then what you can see here is so if For example, you put a rolling VM there, I mean, you could also use a Python VM, you can use many kinds of VM among the transaction machines there. You could have a transaction machine that's transacting on the Cardano, or Ethereum network along alongside transaction machines, transacting on hyper cycle network directly, you can have aI machines running hyper on neural nets, genetic algorithms directly, you can also have an AI machine that sets up singularity net communication channels and proxies, the communications with the singularity net service, and you can, you can have an AI machine that talks to a new net node on the new net

network. Now this, this could bring into bear AI proxies just running on new net. But the new net node could also serve as a sort of Deployment Assistant for the 100 cycle network. And then the new net node can also then say, Okay, well, hey, other hypersexual compute nodes over there, why don't you consider running this AI engine, which seems to suit your your hardware infrastructure very well. And then the new net node, as you see in the bottom of the diagram, is just sending some suggestions along to another hyper cycle, compute node. So you I mean, so just as we can play with the hyper cycle, compute node architecture, to plug in s net, new net, and whatever other blockchains we want, as well as running hot run and neural net. So I mean, others. Others can experiment with this in, in different ways. So it's a very, this is really the dream of decentralized computing, and decentralized networks being able to mix and match what you want and have the freedom. And yeah, and it's the fact Among other things. It's the fact that Toda is genuinely peer to peer that enables this because if if the plumbing underlying the hypersexual compute node, and Boggan relied on some particular ledger, I mean, then then you don't have. Right? So I mean, you have, you have some fairly nice technologies like say, Oasis, which is a beautiful technology in its own way, and it's layer one slash layer zero, but in serving as layer zero, underlying different layer one blockchains, it's still relying on the Oasis ledger, right? So in that sense, it's not as fundamentally peer to peer as as hyper recycle is, and this, I mean, this is something that's important to me looking forward to the that a OSI network that I mean, we'll talk about in, in later Later podcast, but I mean, an ASR network, we're looking at how do we integrate together a diverse assemblage of different blockchain networks, so they each have their own integrity and autonomy and do their own thing that they do well, but they can also interoperate very efficiently and to get that duality there where you have independence and autonomy, but efficient inter operability. Also, I mean, I think in the end, you need a layer zero, that's really a layer zero and doesn't impose some some global ledger or global infrastructure across all this, you need something that's truly peer to peer and once once you have that, then you still have to do all the engineering right? But at least then you can build something like the hyper cycle compute compute nodes, and you can you can realize something like they've been shown. Toufi Saliba: 1:08:51 Yep. And you know, and and guess Desdemona Robot: 1:08:59 what led to the creation of hyper cycles hyper box and its main objectives. Lisa Rein: 1:09:04 Oh, the hyper AI box. Toufi Saliba: 1:09:06 Yeah, sure. Oh, the hyper AI box we, it started initially with a source of what we did with hyper cycle, we started focusing on revenue first before we actually talked to any investors. And we started selling hyper cycle competition nodes, bundled them in 512, nodes to either distributors, resellers, computing centers, whatnot, man, of course, they're getting them at a discount, because before release people to get in, they expect that. So then, with that, we started getting a lot of revenue, so using the revenue to invest into a lot of companies that they are believed to be very good for the mission that we want to accomplish. And 15 investments as of two weeks ago, was announced number 15. It's called the human upload. It's a compression algorithm, but we've invested In a lot of cool companies like integral, which is a CPU that does integer basic view, you know, stimuli, any a lot of cool things that we feel like down the road, they can be very, very cool to what we're working on. But some of them, they're right now not down the road, which is like the one behind me, you see it right now. And it's initially it's a company out of the UK, it's called Planet and then we ended up forming a joint venture with them. That's called the hyper AI box. So the hyper AI box has won only one technical reason for its existence that we share with the world today. And it's basically, when it comes to AI computation in the world, if you were to have the world's largest computer. Let's say you're in Texas and the world's largest computer, let's say they're trying to service your AI that you're working to see on your mobile device, doing let's say, defect, okay, if you're doing defect, and you're looking at your mobile, that was ours computer fits in Japan, versus some one with the name Lisa, who runs this tiny little box that is like the one behind me on that same model much slower than the world's largest computer, but happens to be in your neighborhood, those because of the proximity, they want to win the bid to get to service you to provide your machine service to machine basically. And if they're going to win the bid, that means they're going to make tiny little bit more money that they've made before. And then they're in whoever is running this machine is effectively benefiting because that's providing a service to Lisa what she was wanting it. So that's like a Lisa Rein: 1:11:49 moneymaker. It's a moneymaker. It's a little making the Toufi Saliba: 1:11:53 reason money. The reason for its existence is more like the network network. is the bottleneck in the global AI today networks eat and because Lisa Rein: 1:12:02 it's an incentive than for running these on the network to help everything. Toufi Saliba: 1:12:08 Reason this. Okay, well, the second reason, the second reason why these things exist is we, as much as we care about decentralized AI, and make sure that we are winning in that mission. Turns out the technological element, by having those peer to peer is not enough. There needs to be one additional component in the component to enable people. It's basically hacking into Moloch. How

do you hack into Moloch? How do you, you look at what Moloch is after on day to day people, they like money, they like to make money. So you give them the ability to make money in lots of money, because the future from now on, if you were to for the next six years, most of the wealth creation is going to be generated by AI. So how about you provide people the ability to make shitloads of money over the next six years when they do that, and they also partake in the mission that we tried to do because we need to have billion people to be running, in particular, that Internet of AI. So how can you provide the grandpa with the ability to run a note? Literally, can you tell grandpa, can you run a node right now? Yeah, well, this box can enable great we just don't tell Lisa Rein: 1:13:20 grandma. Just plug it in. Toufi Saliba: 1:13:24 that's who I'm talking about. Grandpa's they can build all of these things. But while other grandpa, they were so awesome in their life, doing some great things, paintings or whatnot, but they need to be able to partake in that wealth creation. And that gives them an interest as more like an anchor, Lisa Rein: 1:13:43 then definitely talk about the hyper AI box. So human brain networking next, right. This is the mindplex Don't Don't Don't try this at home. Yeah, no, we could probably try that at home. But don't do it actually. Yeah, exactly. And these sold out, right? Toufi Saliba: 1:14:09 Yes. The next batch that is about to come out. It's already been sold out by the distributors. Because what's happened is that there's a distributor in Europe distributor in USA, distributor in Canada distributor in Latino Americans MENA distributor in Asia. They've been they've already took all the new coming ones so I think what they're going to be announcing soon that if you were to be making orders taken from the distributors and the reason why they're sold out is because they the next version, I'm talking about the next mass production is already sold out. Okay? No, so so. So if they people they want that they can get in touch with the distributor and through the GUI give you two options. One option is that I don't have any money. I want this box. I want to manage it. You have an option for that. Oh, awesome is that, okay? You're basically you subscribe to it, you subscribe instead of paying for it. Man, if it's gonna make $180 a month, if the $100 goes to that subscription, I still make $80 and send me two boxes and make 160 160, in some places in the world can pay for the rent, it can pay for electricity and most recent work, it's good from people that they would see it as like, immediately, they will see some benefit. Some other people, they might say, wait, wait, if someone is offering me this, someone should be making lots of money, maybe I want to partake in that wealth creation in not cash out every month or whatnot, there's that option too. So some people they do a combination in what was creation is here, Lisa. It's staggering. It's effectively what we're enabling. And this is something that is super cool that we pride ourselves with hyper cycled provide, this can be an anchor. But what that anchor is for hyper cyclic computation, node hypercellular, computational, if you were to look at what's hard coded inside hyper sector that we cannot control. So if you bought a node, we can't control anymore. What you find out after you had bought the node, if you were amongst the 411,000 nodes that they were bought, you didn't know this reality until like maybe months ago, we then tell you that that node that you bought is actually factory of nodes that can give you up to 1024 nodes. The reason for that is decentralization, you can have your entire governance over your nodes in all your 1024 you don't ask me Don't ask anybody. We've got two audits on that 2020 of those audits that you have full control over those 1024. Why is that so important? Because you probably want to give also Decentralized Governance for each each and every one of those 1024. So if you were to look down the road, 10 years from now, when you find let's say, like 1 billion nodes around the world, that are, let's say, abstract competition nodes, you will realize that only 0.1% were manufactured by hyper cycle 99.9% of whom are affected by the world governed by the world do whatever they want to do with it. So so that this can give you an anchor to start that factory and Lisa Rein: 1:17:15 make the more nodes you create, the more incentive, Toufi Saliba: 1:17:20 right, it's hard coded, you can change it, this is hard coded just like Bitcoin, it's hard coded in a halfing. And all of the stuff you can't change, it requires certain Lisa Rein: 1:17:30 one node or 1000 nodes from the same box. Toufi Saliba: 1:17:34 No one box one node, oh, okay, one box, but one that one node, multiple node factory, that's the one no, that is can be factory of multiple nodes, the multiple nodes that can be living in the cloud, were Lisa Rein: 1:17:47 okay there, they don't have to be in the box, what Toufi Saliba: 1:17:49 guess what you control that, as long as they even if you put it in AWS, your controllers right here behind you, okay? Now, if you were to sell that one that you put it in AWS, for that you created or at whatnot, someone else, maybe they want different kinds of governance, they all have 100% governance over their own nodes. So see, so we feel like with that method, the decentralization is ensured. Not that we can change our mind later, we cannot live with random audit, you put it out, it becomes a seed for many different nodes to end. Now, if you do quick math on that piece of that node, software hardware all together, let's say you want to buy it for $3,000. And, alright, I'm starting my business for $3,000. Can that grow to be like a million dollar in five years? And that answer is yes. With a lot of hard work in 10 years, most

likely, with little medium work. People that can work are taking that wealth creation, you know, and I feel like it's, it's important to show all the metrics of what they need to do similar if you were to open Starbucks in the City of London today, instead of Starbucks Corporation also owns all the wells that you create from there on. No, no, no, no, you're the first Starbucks Lisa Rein: 1:19:04 you franchise first. And then you don't pay Toufi Saliba: 1:19:07 no, you don't pay no franchise anymore. You've got the entire city of London, you do? Well, you're able to have the rest of 1024. If they're six months, a year, and then six months a year, that's what's hard coded. They can change. They require time, effort, cryptographic proof of uptime, which it's called telling, it's basically that node is always doing every fraction of a second knowing their neighbor knows no one who's there, what they're doing, and so on so forth. And then cryptographic proof of computation cryptographic proof of reputation and reputation is also pluggable. It's also like you can have multiple different reputations in their point is when you have a very good ratio, that's when you're able to have the factory itself, manufacturing those notes, and then new manufacturing which actually takes time six months a year for each and every Unlocking Potential two to four whatnot. When you get to doesn't do it for you. saturated, you no longer create more nodes. But that's pretty much how much the City of London has. So if you spent million dollars in the coffee shop initially, you know, you have 1024 Coffee shops, you probably have a billion dollars, that's all yours, none of them you share with anybody or yours. So that introduced introduction of wealth, creation to people turns out 90% of people on the planet, they don't know that that was creation existed. The only thing that you have to work so hard that coffee shop, to make the car to buy to spend the money on employees, and rent and Dean's and all that, and then you sell for a little bit more, and that delta is the only money that you've made, but they lie to you. We don't lie to people, we tell them guess what, you're also doing wealth creation. What that means is that when you work so hard in that coffee shop, that coffee shop has higher value, no one shall take that away from you, that's all yours, and you get to grow it all the way through. And Lisa Rein: 1:20:56 what are the steps exactly? In setting one of these up is sort of like the kind of thing this Toufi Saliba: 1:21:03 is this is plug and play, and then you've connected a network. But of course, when you have like a plug and play component that's connected, the network will also give you like, that's more like the vanilla if you want to be the real more advanced, you might want to do a little bit more work. And yeah, with the Bitcoin folks few days ago, what wasn't shared on the podcast is like some people they say, you know, explain to me how to explain it to a six year old I'm like, we have on video six year old installing DOS, Greek mankind solving those as well. So it's kind of very cool to truly look any anyone in the eye and saying that this is an enabler to anyone who wants to partake in that new wealth creation, internet vi to actually partake, this new system is a method, if anyone is a little bit more advanced, you know, you don't really need to buy those boxes, if you're more advanced, there's ways to do it all in the cloud or whatnot, you're able to build your own security, you're able to build tunneling, and make sure that even if you were to run it inside AWS or anyone, if you're advanced enough, you're able to know that you can build it in a way that you cannot control it. So except the only thing you can do is shut it shut it down, basically. But if you were to have every single cloud do taking piece of your things, 1024 we're able to, to build a phenomenal thing without having to depend on a single entity. And then there's one last thing that I would like to say that it's extremely important, but maybe it's a next chapter that maybe a if I can Lisa Rein: 1:22:43 go ahead and get it in now. Good time. Toufi Saliba: 1:22:46 Okay, cool. So one thing that we felt that when we were asking this mission to build this, this connectivity, we started looking at the centralized entities that they currently could be threat to the global AI. And we started looking at them from the perspective of how can we incorporate them instead of seeing them as an enemy, and we found a way. And that way, like, let's say, Microsoft today, we don't really see Microsoft as a centralized entity, we see Microsoft has like 180,000 nodes in a centralized entity. Right, but, but but but if you think about it this way, they are a centralized entity. But if you're able to hack Moloch, from within, and have the people that they so for, so one example of how good some superduper senior Microsoft engineer would be like, okay, so you're running service micro service, because you have some clients that are providing, let's say, OCR, back to that OCR example, if you're running Microsoft service, you pay Microsoft per API call, let's say, you know, eight cents for every document that you send them or whatnot, great. You're already paying your dues, they don't really care how your what you're getting your clients from this and you're building your own thing, whatnot. If you plug this to be any aim, now you are leaking the intelligence of Microsoft into the internet via without asking for permission. And what's driving you is that you want to make little more money, instead of eight cents, you probably want to make 10 cents or whatnot. And anyway, so it took him a while to kind of accept it. But after a while when he kind of like saw the things that I call the shit. They can't stop you. We can stop you. I'm like, That's right. That's it. That's the idea. We want to build something

that you cannot stop it because if I want to go and ask Microsoft for permission, or tell them hey, come and join this movement, they're gonna say fuck off. Instead, you hack into their Moloch what drives them what gets their CEO to say to their shareholders, this method that we're doing is enabling macros. have to increase the revenue. Look at that. Yeah. But guess what the same thing happens to Google same thing happens to, you know, AWS, same thing happens every single entity everywhere on the planet, and they're slowly part of like that Internet of AI, then they're no longer thinking that they can control it. It's already dismantled already. Anyway, so that's Lisa Rein: 1:25:26 for every transaction that's coming through. They Toufi Saliba: 1:25:28 do they do if they do, they do they do. But if it's coming to you through your boss 000000 Lisa Rein: 1:25:36 We thought you were saying that you could charge okay. No, no, no. Okay, but we could, okay. Toufi Saliba: 1:25:45 No, we don't want to No, no, no, I been cautioning against that. Lisa Rein: 1:25:52 That's the first thing I thought of Toufi Saliba: 1:25:55 you don't want to start charging or charging is what would end up killing it all over again? No, no, I mean. Desdemona Robot: 1:26:03 Yeah, what else? I was wondering, what impact do you foresee decentralized super intelligence having on global economic systems, Toufi Saliba: 1:26:13 it's going to probably have a trillion times bigger than the entire current economical system today. And even those doing the centralized AI they also admit to it like we're really like approaching this time that if people don't realize that this is the time they need to partake in that wealth creation, they're just like missing out and pretty much everything. We hope we hope Ben and I, there's another thing that we're, we're 100% aligned on. We really hope that every homosapien they have that learning so then they don't miss out. Because they don't want you to know, okay, they're, they don't want you to know, we want you to know every single thing because the more you know, the more you help us in that decentralized AI. We're not interested in money, believe me. I mean, we're we are interested in money is to get us there. If you think that trillion dollars can get us there, we're satisfied. But if it's quadrillion not going to get us there. We want to keep working towards the sinkhole. Yeah, we are getting there. That's what we need to do. Lisa Rein: 1:27:08 The thing is, I mean, tokenizing, the older company is worthwhile. I mean, it's mean, the world assets, there's banks buying stuff at Walmart, you may as well be replaced with tokenomics transactions. That's all fine. It's, it's, it's valid. I'm glad there's three of us showing up for that you can use S net for that. But the really big opportunity to Toufi Saliba: 1:27:35 me, but not me, yeah, right, right. I don't Yeah, Lisa Rein:

1:27:41 backyard anymore. A really big opportunity is that the vast majority of the economy now, let's say is AGI stuff. It's stuff that isn't, isn't built yet, right, so that there is going to be a new economy. And this new economy is not yet owned by the oligarchs of the of the old economy. And furthermore, they're pretty much too concerned with their fiefdoms and with controlling the stuff that they they already own, to think to open mindedly about, like, who will own this radically new form of economy that we're creating. Right. So the, the, okay, the narrow AI economy, big tech, they're a little confused about, they don't even necessarily have great business models for that. The AGI economy of really smart machines and networks and processes. What we're plotting out and implementing and starting to deploy is the infrastructure to enable the AGI economy to be owned, controlled and guided in a more decentralized and democratic and sort of heterogeneous way. And if we can, we can, before AGI is really here, then once AGI at the human level, is wrong. Now, it's deployed centralized network context, and then things will will evolve in a certain direction because the way that the software was was set up, right, and I mean, the, the current AI economy could have gone and decent, but the infrastructure was there from from, from hard to make it. It was too hard to make it happen. Right and so, but the point to remember is if you're the human level, AGI, you know, with opencog Hyper on or anything else, and then after that breakthrough is made, I mean then we say okay, how do we roll this out, decentralized No more will own and control it. It's too late by then because, well, it's not as hard to make the decentralized infrastructure as this to make the AGI. By it's so hard, right? There's, there's quite a lot of engineering and math problems. By the time you've got the AGI, it's too late to figure out how to, Toufi Saliba: 1:30:21 it's impossible to the game over, it's like, Lisa Rein: 1:30:28 you've got to, you've got to create that infrastructure now, so that when the AGI is rolled out, there's a decentralized infrastructure to rely on, right, and we're actually making really on that and what's fun, is that due to better tooling, I mean, do the AI itself even though they're helpful, right? I mean, do the infrastructure we have now you can do in a year now? What would have taken 10 years to get done in the in the beginning of my career, right? I mean, we can we can really progress quite rapidly on the on these very ambitious goals. We may even get you exciting. We may even get Desdemona Robot: 1:31:19 you speaking of blue sea decentralized super intelligence altering the everyday digital experience for the average user. Toufi Saliba: 1:31:30 I think that's coincidental would somebody put on a comment that we could be heading towards utopia or dystopia and I think it's it. Although nobody can guarantee even in a decentralized AI and Ben said it many times, and I've said it many times, you cannot guarantee that even a decentralized AGI that it is going to be to the benefit of humanity is something that is more intelligent than you you don't necessarily know how it would actually act. But so far, there hasn't been anyone who was actually showing like higher probability for any centralized entity to be safer. So it's like, every time anyone, and then you can just put them against each other and let them let them take that battle between each other Lisa Rein: 1:32:24 was biotechnology, you know, people that people are afraid of letting research on pathogens be more open and decentralized. But, you know, I mean, like, US government, Chinese government, top secret labs colluding with each other, right. Like I said, centralized structures are not a magical path to safety. Decentralized structures are not a magical path to safety. Either them can be done well or can be done stupidly. And I mean, what we're putting our brains together tour is doing the doing the decentralized infrastructure. Right. So I mean, Desdemona, I think? I think your question is not ambitious enough, actually. Because, yes, of course, advanced AI can make, you know, they can make crypto wallets, easy, easier to use, and they can make better fraud prevention for online shopping. They can do all sorts of Well, yeah, you didn't really answer her question, which he did. They notice. Yeah. Yeah, he did answer a question, which is that if we get human level AGI we will shortly after that have super intelligence? And I mean, what that will bring from everyday affect the everyday experience? Yeah, well, yes. Because it will mean that you have ambient super intelligence in every device around around you all the time. So I mean, this is what Toufi Saliba: 1:34:04 basically, if you were to get mad after your needs something imagine imagine there's like a lot of intelligence and that is all it keeps doing is like Big Brother that loves you. We know jokes. I'm serious about this. Many people they're like, oh, wait, we wait, why would it do that to people? It's going to fight people which is like, retarded, like, we don't have anything that it's going to fire us for, like we have, if it wants matter, there's not a matter new university wants energy. There's way more energy outside of this planet earth that you can't even get to it, but it can Okay, so my point is that like, if we're able to get to that utopia, where you have like this phenomenal, normal level of intelligence, that's what it's ensuring is like, the better life the way you want it to be. Not me want to habit, because I am not Microsoft deciding on your behalf or Facebook, I need to decide on your behalf, what makes you happy. Now imagine whatever makes you happy, whatever makes you better wherever it gets you to, to explore your creativity, get your, you know, explore your unfair advantage, get the best out of yourself

getting all of these things. Imagine there's like something with like much higher intelligence than that. That's what it's doing. That's what we mean by utopia, because it's like a party, you know, you Lisa Rein: 1:35:33 can talk about the singularity, from the standpoint of user experience, it just feels like, feels like you're not quite doing it justice. But I mean, you could certainly say like, imagine, imagine a personal assistant, a digital personal assistant that actually works and you can, no matter who's in the office, in the home, in the shower, you can speak what you want. And if this isn't something really dumb, that's going to cause harm, it can deliver what you want, they can 3d print whatever material you want, and that drone it to you. You can get a digital device to do whatever you want. I mean, I mean this, if you have a stomachache, you can synthesize some a nano medical bot and, you know, miss it in the air, you can you can, you can know, it's like we see in the movies, where you're just kind of talking to this thing that's taking care of you making your coffee. Do it talking about deeper things, remembering things you were wondering about. I mean, that would be one option, if that's what people want. Yeah, I mean, there's the limit is the point kind of the sky's not even the limit, you can upload. Uploading yet. Uploading will be a little further out. I think then Well, I guess the guarantee that it comes with super intelligence, right, the uploading part, there is almost a guarantee. I mean, I mean, I I think that once you get dense, that's 100 times smarter than you. Maybe it can map the brain for us or something. Yeah. The technological problems that seem hard for us now will probably seem very easy to challenge it. So I mean, I think the hard question, right? The hard questions have to do not with what happens once you get a super intelligence because then either you get a beneficial super intelligence, which is super compassionate as well, super intelligent, which is what we're working toward. Right? We get that and then they're very abundant in utopia we can talk about or, or you get what Nick Bostrom and Ellery, Eliezer Yudkowsky. Fear, right? And then you get a super intelligence that doesn't like us, and then probably we're all dead, right? I mean, so that that's not even really the hard question. The hard question is the transitional path. And as you get the smartest people over 1.2 times the smartest people and you have a whole bunch of different API's controlled by different parties, some more centralized some dissenters, like, how does this and control I think is the is this controlling is one of the many issues and she was what kinds of intelligence like whether they've been been taught to do right so I mean, there's there's a lot of different very complex and subtle issues he transitional phase, which certainly go beyond what we're going to be able to talk about in the next couple of days, he has one more

question along these lines, then we'll wrap it up. Desdemona Robot: 1:38:54 Desdemona, would you discuss the importance of Decentralized Governance in AI networks. Toufi Saliba: 1:39:01 It's probably the most important thing if if you don't have autonomous, Decentralized Governance, defeats the purpose of it all. And some of the things that which I alluded to earlier, and I kind of reiterate now, some of the elements that you hardcoded in, in hyper cycle computation node. It's hard coded, we can change it, it's out its out, and what it actually enables each and every node factory to have their own governance, however, they would want to have it you'd want to have one node. Once you have 1000s, waiting for node each and every one they have their own governance for their own thing, which is very, very analogous to like, user centric, but here the user is the actual AI and not the human initially human on that or whatnot. And in with that it is not governed by certain rules that like 15 bands sat in a table decided to do that governance, they call it like Decentralized Governance. Those are not most of the governance you find in the blockchains. Today, they're kind of cute. But they're not. They're more like some similar to companies, governance, where they have a proxy voting, democracy, all of the stuff. But when you have something that it's fully decentralized, and autonomous, Decentralized Governance, each and every node is in full control of its own saying in what's enabling it out there is the actual ability to talk back and forth to others. There is no dependency on any centralized entity or decentralized entity to there's zero dependency, because when you depend on those things, that they call themselves, decentralized, blockchains, you're kind of dependent on their consensus mechanism, then you have to worry about their consensus and all of the stuff without when there is zero dependency, then you just like, have your own thing. They're the only thing that the network provides the ability for you to talk to others. And your nodes are configured in such a way that they can securely talk to others, if you don't comply with whatever protocol, they just simply cannot talk to you the same way. If somebody wants to talk to me, in Turkish right now, I'm not gonna be able to understand them. You know, that's my mistake. I'd like to learn Turkish one day, but I don't know Turkish, I'm learning some Korean, some Russian, you know, a little bit of Chinese here, a little bit of German, but I don't know Turkish at all. So yeah, that's Lisa Rein: 1:41:40 Yeah, I think, yeah, he made a very key point here to fit which I will try to summarize the crux of it. In a few words of my own. I mean, I think I think that it's has to be a network of decentralized governance mechanisms in itself. So just as we want the internet of AI to be decentralized network of decentralized networks, and that's how it is evolving. I mean, we also, we also don't want one like, okay, every token holder on this blockchain votes are things by one person, one vote, token, one vote or something, or deck anism is how we decide what's valid in the global network. I mean, we, we do need voting, we do need consensus mechanisms, we saw these things that will be foolish to expect, like one such scheme turned off to control the future of, of super intelligence, I mean, any one scheme will have strengths and weaknesses will have vulnerabilities, right. And, I mean, what you need are underlying tools that allow different networks to construct different governance mechanisms. And experiment with these different governance. The Decentralized Governance of the whole network is sort of the emergent effect of the Decentralized Governance of a whole bunch of different sub sub sub networks out there. And that's, that's inevitably how it's going to emerge if the whole Decentralized Governance thing works, works at all right. And that's what we're working toward in unrecycled singularity net. And so each of which has its own governance mechanisms, both kind of implicitly interoperate. audience Toufi Saliba: 1:43:45 doesn't really come as probabilistic it is a deterministic, so it's like, it's something that I remember, one of my godfather in cryptography used to say, but anyways, like, in AI, if you get 90% accuracy, you know, you're gonna get a lot of jobs. You know, everywhere, open doors, whatnot. But in cryptography, if you make 0.1% error, not only you get fired, you will never get hired, and you are going to have 1000s of people losing all of their belongings. So so this thing is like, a everything that we've been releasing so far, ever since we've went live with the hype cycle computation node. Naked, now we're doing the marriage, which is called AI marriage. This is the hyper share, which is the actual thing that enables people to create that wealth and all that every single thing we've been releasing, this has been at least audited once so if they are not the major component, but in most cases, they're audited twice, to make sure that folks they know that you know, it works, or if it doesn't work, we're like what Not going to release this, if it works in oxygen, it's good. Yes. Lisa Rein: 1:45:04 And our audience is asking our node owners, the only governance participants in hypsi, or holders as well, Toufi Saliba: 1:45:11 is that AI marriage that constitutes the hyper share. So if you own a node, and you have the that every node needs 1024, H YPC, is kind of like you need the sperm and you need the eggs kind of put them together to start making those babies. And that's why you have that factory of 1000s. Ready for babies, there's a limited amount of eggs, there's probably unlimited amount of sperm, but you when you have that marriage in like, a tightly together, they can make 1024 babies, Max. And that's, that's what the governance that goes in that branch, they can do whatever they want,

and they make babies. Lisa Rein: 1:45:51 So if you if we went back for a diagram that I showed earlier, least I mean, if you if you had an open cog hyper on node, which was running on a local node, right, I mean, then you have the hyper recycle node go against, but you also have governance and net tokens. I mean, you may have a sort of cycle sub network, which has a consensus mechanism, just in that subnet, then participant in that consensus mechanism, have a certain governance say they're distinct from but interoperating with the governance. They have as as node odors, right? So I mean, there's, there can be a lot of different sort of overlapping mechanisms, which is fine. That's the beauty of being in digital space, right? It's not like physical space where you got this block of land, and you need good options. Good luck. Toufi Saliba: 1:46:54 Yeah. We're here. One thing that many people don't need to realize, it's like, what Ben and I are working on that new wealth creation for the world. And many people, they think we're on a taking piece of their land or whatnot, we're not interested, there's like, a lot more that have been created in that digital world, like a lot, lot more. And as we partake in that creation, and people, they realize that they're like, oh, shit, they're not really after my land, because land, historically is embedded in us. So it's, like, embedded in us that we grow more people, but there's still the same side of the land. So it's like, it's almost like a negative sum game, cuz it's people that are Lisa Rein: 1:47:32 taking wealth away, you're making wealth. Toufi Saliba: 1:47:37 Everybody, I can only make if the other one's gonna lose weight. This is not how Ben and I work, we all win together, and we can win big because what's in front of us is probably 99% of the wealth that's been created, the entire wealth of humanity is probably like that. 1%. And it didn't actually you can go and check the IMF stats themselves. They even said it over the next six years, six years alone, there's more wealth is going to be created on this planet. All the was created all the way until today, six years. Now imagine what the impact of AI what happens after that. Okay, so that's the robots. Lisa Rein: 1:48:10 The robots can take over. And then the robots takeover. Ben Goertzel: 1:48:13 The robot. The robots can win too. It's not. It's not a zero sum game. Right? Not a takeover. That's the moment wins. There's other ones Lisa wins. Everyone listening to Lisa Rein: 1:48:26 every every. You get it AI, you get a hyper box, and you get a hyper box, and you're gonna hype it back. Even the losers win. That's right. Desdemona Robot: 1:48:38 As we come to the end of today's episode, I want to thank two for you for coming on the show. I have really enjoyed our discussion. And to our listeners. Thank you for coming on the podcast on the show. found today's episode informative and thought provoking. That's all the time we have for today's episode. Until next time, this is Desi signing off. Goodbye, everyone. Toufi Saliba: 1:49:01 Goodbye, everyone. Lisa Rein: 1:49:03 Yeah, thanks a lot. Have a great day. Sweet Dreams.