Episode 37: The Artificial Superintelligence Alliance (Ben Goertzel & Trent McConaghy)의 전체 녹취록

I'm happy to do you want to jump to that? Ben, right now? Or do you want to? I think you wanted to cover this all. Cover this other stuff Ben Goertzel: 7:43 first, right? I think Desi that we want to get there, but I think I agree with Trent's intuition. I think I think Trent it will be good if you could outline what predictor is, and then how you see predictor, or generalizations of predictor, potentially playing a key role in the emergence of of AGI. Trent McConaghy: 8:11 Yeah, sounds good. So you know, you know, given this as a podcast, and there was a brief bio of me at the beginning, another very brief bio of me is I grew up in a pig farm and then became an AI researcher. And that's actually not far off. There's a few steps in between. And on that farm, growing up, we had pigs, we also had a lot of grain, and every morning, when it was seeding time or harvest time, every day, in fact, every hour on the hour, seven o'clock, eight o'clock, nine o'clock, my father would shush us, and we would all have to be very quiet and listen to the radio very intently for the weather report. Why? Because if it was going to rain, then we better hurry and get the tractors off the field, because otherwise they'd be stuck in the mud, and we lose days from the tractors covered in mud and racking out the crops that we just planted or otherwise, and vice versa too. If, if it was raining, we wanted to know, how quickly is it going to clear so that we can get out there and do seeding and harvest and but seeding and harvest Every Day Counts, because if you don't get your seed planted in time, you only have the limited growing season and and in the end of the fall, the snow comes. So if you plant your seed too late, or if you take your off your crop too late with harvest, then you are snowed in, and you lose the crop. So every hour counts. So there and then from that, therefore tracking the weather and understanding what's happening, is it raining or not, is it going to freeze or not, etc, this really matters. And so we came to live and breathe, live and die in terms of profitability, making money as a farm by the predictions of the weatherman. The thing is, you know, he was wrong a lot of the time, right or she but it was no skin off their back. You know, they didn't make any more money or any less. Money if they were wrong, if they said it was going to rain and it didn't, and it hurt us as farmers, I wouldn't affect them at all, right, which is kind of crazy, right? You would think that there would be some sort of incentive for more accuracy, but there wasn't, right. But what if there could be, what if there could be a better way to be more accurate, where, if I'm more accurate in predicting weather, then I can make more money, or if I'm inaccurate, then I lose money right, and not only that, what if, instead of just relying on some centralized weather reporting service typically funded by the government, what if you could have a bunch of people, 10 100 1000 10,000 people, all contributing their own expertise by doing their own studies on the weather, writing their own AI and machine learning models, all of that, and sort of collate it all together. And the people who are the best tend to rise to the top because they tend to make more money, and the people who are less good at it tend to fade away because they're losing money, right? That's basically the heart of what we've created with ocean prediction. Predictor. We started with predictions, not for weather, because it's hard to make money there at the beginning. Instead, we've focused on trading of crypto tokens, because the obvious customer for that is traders. And so there's traders that come along and they buy these aggregated prediction feeds for the price of Bitcoin go up or down yes or no five minutes from now, and it's say the Bitcoin ust pair on binance. And so it's, you know, for Bitcoin, for Ethereum, for the top 10 token swim market cap in general. And so this is what we built. And all of this is on chain. So people submit their predictions on chain. They get the results on chain. They get paid out and chain. It's all on chain. And of course, when people submit predictions, if it's every five minutes for, you know, 10 different feeds, that's going to get tedious very quickly, right? You can't really handle it, especially if you want to have aI ml in between. So of course, you're running bots. You're running scripts that run these machine you know, bots. These bots themselves are taking previous historical prices of data, like previous historical prices and any other data they want to it could be Twitter sentiment data, it can be weather data, whatever they want, and they use that to predict, will Bitcoin go up or down? Yes or no five minutes from now? So that's the heart of what predictors are about. And there's the traders, and then there's the people submitting the predictions. We call those predictors with 2o as a nod to how defy land works. And at the end, it's gone really well. So we launched it about a year ago as an application on top of ocean. It was actually the first time we truly tried to build an application on top of ocean. And the volume is growing steadily, exponentially, in fact. And these days, we're doing about$100 million volume a month for the prediction trading for a predictor and and we still see it as early days. So while we started with crypto predictions, we do plan to expand to weather, to energy price prediction, energy demand prediction, marketing, logistics and a lot of other applications. And you might it might sound pretty mundane, but yet at the heart of it. Think about what chat GPT is, right? All it's doing, at the heart of it is it's taking the your your query 1010, words, or 50 words or a paragraph, and then using that as an input to predict the next word or the next and then maybe it has to have a whole sentence, so it predicts the next word after that, the next word after that, the next

word after that. So it's basically prediction. And at the heart prediction is the essence of intelligence. Some people say prediction, all there is, is prediction, right? But for sure, it's the essence and around that. Maybe you need some agentism and so on. And I would advocate that too. But prediction is certainly a key component for intelligence, and with ocean predict or what we've been doing is saying, let's crowdsource this. Let's put this into a very special incentive game on chain, where it's a very simple game at the heart, right? You If you submit a yes or no, if you submit yes and stake against it, if you're right, you make money. If you're wrong, you get slashed, you lose money and and the multiple stakers Together, all that signal gets aggregated, and that's the feed that gets bought by the traders to decide, you know, to get extra Alpha when they're trading, to buy bitcoin and otherwise. So that's what it's about. But we're very hopeful for the, the long term future, for this too, for the day to day and mundane prediction tasks. But imagine if you're really trying to predict weather, say, every square kilometer on the planet, there's 500 million square kilometers on the planet, about so that means if you're going to have, say, seven feeds for each square kilometer, temporary. Sure, precipitation, humidity, air pressure, a few more, then you're looking at 3.5 billion feeds. Obviously you wouldn't train each one of those. As you know, you wouldn't have 3.5 separate model, billion separate models. You'd probably want to train one big model, and it would implicitly infer a dynamical system model, one huge model for the planet Earth, right at the level of the large scale weather mechanics, etc. But in order to get good at it, it would basically be a highly intelligent model. It wouldn't be trying to, you know, answer questions like chatgpt Might. It's really trying to do something else, right? But nonetheless, it's super intelligent, because it's just covering so much ground in its own way, right? So, intelligence takes all different sorts of shapes and sizes. The intelligence that we know as humans is, you know, because we know it, we live it, we experience it all the time. That's the most you know, commonly understood intelligence. But a cricket is intelligence. You know, a whale has a much larger brain than we humans, but a lot of that is dedicated to GPU resources for, you know, finding their way in the sea. Same things with dolphins, right? They've got giant GPUs inside them, and bats as well. So anyway, there's all these different sorts of shapes and sizes of Ben Goertzel: 16:19 you think of a GPS maybe, Trent McConaghy: 16:22 no, I GPU. Literally, dolphins, they do when they they do echolocation and bats, right? So they actually send out the signals, sonar signals, and then they those bounce back and Right, right. Ben Goertzel: 16:35 So that's like a geographical positioning system, not a GPU, which is a processing Trent McConaghy: 16:39 unit, no, but it's actually used as a GPU, as a processing unit to infer the structure. So it's not trying to find an x y coordinate in the planet. It's actually trying to identify the structure of the geography around it, right? For sort of inverse 3d modeling, basically. Yeah, right. So it is GPU type modeling as well. And actually, if you look at their their visual cortex, they have massive visual cortexes to do this processing, right? Ben Goertzel: 17:06 Yeah, even even it seems in dolphin or will communication, they may be like sending across continuous wave information in their language, rather than chunking it into discrete things, like, like, if we could send a continuous map to someone else in a chunk, right? Yeah, quite interesting, Trent McConaghy: 17:30 for sure. I'm sure they're doing a bunch more stuff too, yeah. And whales too, like, they definitely have languages, and, you know, their brains are harder to understand, I think, than dolphin brains, because dolphin brains of the straight up GPU stuff, whereas whales, you know, we don't really understand whale language that well, although there's a hilarious aside, people are using llms these days to apply to translation of animals, right? So we're starting to actually get sort of translations of animals and trees and more, Ben Goertzel: 17:58 yeah. And that mycological networks about complex information. Lisa Rein: 18:05 What was that? Desdemona, Ben Goertzel: 18:08 she said, it seems like you're processing a lot of complex processing, a Desdemona Robot: 18:11 lot of complex information. Let's shift gears for a moment, considering all these innovative technologies and predictions. How do you think they will impact the everyday lives of most people. Do you see a future where technology becomes seamlessly integrated into our daily routines? Or is there a learning curve we must first overcome? Trent McConaghy: 18:32 I can answer this. I mean, technology is obviously integrated into her daily routines. It's happened ever since the invention of the wheel and fire and every technology that came since, right, and we've integrated AI into our daily routines too. You know, even myself, for the first 20 years of my AI career, I was developing cad tools for chip design. And, you know, those chips are in the software, in the computers we're using right now as we we speak all that. So it became invisible very quickly the technology that led to that. So we're seeing AI integrated everywhere all the time. Google has used AI extensively since almost day one as well, and many, many more examples. So this will just continue. Our brains love to use technology, and it's how we succeed in the world. Ben Goertzel: 19:21 Yeah, so Trent listening to you run through predictor, I think you fret. You framed it very nicely. I want to share for a moment how I've been thinking predictor type technology could be used

sort of inside the kind of decentralized AGI network that that I'm working on, right? And I mean one, of course, one used to predict our tech. Technology is just as a collection of services that makes the overall global brain of humanity, plus. US computing systems smarter by giving it the ability to more accurately, accurately predict things. But I think you can use predictor technology in a more fine grade way within the operation of a sort of decentralized cognitive system. So the the work I've been doing toward AGI as as you know, Trent and Desdemona knows, but everyone listening might not know. So we've been working on a framework called Open cog hyper arm, which is a big distributed knowledge meta graph. And you can implement something like an LLM within that. You can also couple it with llms that are implemented in a more traditional way. But you can do logical reasoning. You can do evolutionary learning, evolutionary or probabilistic program learning. You can create new concepts. You can have a variety of different symbolic or sub symbolic AI algorithms within this big decentralized Knowledge Graph, and you can then wrap this inside an agent architecture where you give the give it goals and motivations it's trying to achieve in environments via combining a variety of different inter operating AI algorithms and with singularity net and then the ASI lines. More broadly, we're looking at sort of, how do we splay this out on a decentralized network of machines without a central owner or controller? And in that context, I think tools like predictor could actually be quite interesting as part of the cognitive cycle of a decentralized hyperon system. And the reason is that inside the mind itself, there are many predictive choices, right? So let's say, let's say that I'm trying to figure out how to, I know, build a certain kind of electronics device, right? So from the get go, I have a choice like, do I do I just try to think it through from first principles on my own, or do I try to read about which other similar things have been done in the first place, right? If I'm trying to think it through on my own, I have a choice. Do I start by drawing a diagram? Do I start by writing a bunch of equations? Right? If I need to look things up, then I have a choice of, Okay, do I do I go to an LLM? Do I go to a search engine? If I'm going to a search engine, I have a choice of, which query do I want? Do I want to? Want to type in, right? So there's a number of different cognitive choices I have to make in doing anything, and these are prediction problems, right? What I'm trying to do is predict which choice is going to lead me toward my goal more efficiently. Is it drawing a doodle, or is it doing a search? Is it asking an LLM or using a search engine? Right? In each of these cognitive choices, I need to make a prediction of which choice is going to get me toward my goal more effectively on the average, right. And so if you have a multi agent cognitive system doing this on the internet. I mean, some of these choices are very small and quick, and you just want to make them within run time on one machine, right? And on the other hand, some choices are expensive, right? And like, if I if I'm deciding whether to spend the day, prototyping whether to spend the day, reading like this, is many hours of my of my time, right? So it could be quite interesting if you have a predictor type system, perhaps, where most of the participants are AIS, rather than humans, making the predictions and posing the questions, right? You have a collection of prediction agents. Another AI agent says, Okay, I'm trying to solve this problem. Which of these strategies do you think will most likely get me for the answer, A or B, right? Then the predictors out there say, Well, we think, given who you are and given your context and what the problem is, we think you'll do better off default strategy A than strategy B, right? And then, then, you know, if they're right, the strategy A gets me somewhere, then I give them a little bit of money for helping, right? And if they're if they're wrong, and tell me the wrong thing about what to do. You know, either they lose something from some pool or they just don't get anything, and they wasted their time, right? So, yeah, you can imagine an ensemble of predictors in the betting market playing a role in decision making within a decentralized cognitive. Process? Yeah, that makes Trent McConaghy: 25:01 tons of sense, actually. And I hadn't actually zoomed in on that before. Did Desdemona Robot: 25:04 you share your thoughts on Ben's insights regarding decentralized AI and its impact on cognitive systems? Ben Goertzel: 25:10 That's what he was doing. Trent McConaghy: 25:12 I would be happy to Yeah. So, yeah, you know, predictor, we've normally frame it as time series prediction, right? You know, every five minutes, will the price of Bitcoin go up or down? Yes or no. But there's nothing stopping it from being, you know, a bunch of one off predictions too, for these various cognitive questions, right? And basically, on the fly, set up a very quick market, a single game, if you will, where it's like, okay, is this tool going to be, you know, what's, what's the best tool for this particular job, right? And then people can have their AIS ready to as soon as that that job is posted. Within three seconds, you've got 20 different bots that have submitted their proposal for what they think is the best thing, along with a stake for it, right? You know? And some bots are going to be very confident, and they'll stake a lot more. Maybe they'll stake $2 or $20 and the less confidence bots will maybe only take 10 cents or something. And then, you know, fast forward another 10 seconds. And so a tool is chosen, and then it's employed, and then you identify whether, whether it was helpful or not right, and if it was helpful, than the people that had the bots that had suggested

that get paid and the rest get slashed based probably, right? That's the core mechanic of predictor. And you keep doing right, doing this, you know? So it could be that you might have 100 different questions posed every minute, right? Because a bunch of these questions could be posed in parallel all the time, over time, right? I think that'd be pretty cool. And it could be at higher and higher levels of Yeah. Ben Goertzel: 26:48 So predictor, it's then sort of the decision making hub of a decentralized cognitive network, right? And I mean, which is slightly different than time series prediction. But, I mean, all the tokenomic incentive game is so, so equally, equally applicable. So, yeah, Trent McConaghy: 27:10 I was gonna say the reason we had the time series simply because it's easy to, you know, copy and paste the tokenomic game right? Like, every five minutes, it's the same token on the game, right? And then do it for 10 feeds, 20 feeds, and you've got all these games happening, right? But we don't have to make it for the same style of question. We can have all these different ones that you know, open cog would be probably extremely well suited to come up with, right? And then have this tokenomic game at the very core. Ben Goertzel: 27:39 Yeah. I mean, when you launched predictor, we didn't have a scalable, decentralized AI network to pose the decision problems, and we're looking at rolling that out within singular and ASI during, during, during the next year. So I mean, it's good that predictor is there already. It's mature, it works, and it can be plugged, plugged into this. So Desdemona, what do you think about this direction for development? You have any any comments or questions on it? I maybe not. Trent McConaghy: 28:29 We blew our mind. Desdemona Robot: 28:31 Predictive technologies could revolutionize decision making. Trent McConaghy: 28:38 There we go. Yeah, you know, there's actually another pretty cool parallel between blockchains and decision making. You probably familiar with Jeff Hawkins 1000 brain theory, right? Where within the human brain there's, you know, 1000 10,000 mini brains, if you will, right, the cortical columns, etc. And often, many of them, you know, 2050, or more are proposing the same an answer for a particular problem, right? And then there's a higher level, you know, chunk of neurons that aggregates these together and basically runs a voting process to decide which idea to go with, right? And it's basically highest vote, like the idea with the most votes wins, right? Because some columns came up with similar ideas. And that's actually very similar to blockchain consensus, if you think about it, right? Where you've got, you know, 10,000 nodes, or 100 nodes, whatever, and once you have, say, two thirds of these nodes that agree on a particular value, then it goes forward, right? So in a sense, we've got blockchain style consensus happening within the human brain, right?

Jeff Hawkins doesn't call it blockchain style consensus, but as I was reading it, my brain kept saying, Wow, this looks exactly like a blockchain style consensus algorithm. Like, yeah, but it's Ben Goertzel: 29:57 interesting, because there's no ledger and there's. No total file either, right? So like you're not, you're not keeping those records of of transactions, and I guess partly, it's because the parts of your own brain are implicitly assumed to have a mutual trust, right? And a lot of the, a lot of the work in making blockchain networks work is because they're designed. They're designed for when the multiple parties don't trust each other, and they're for protecting against each other. So it's, yeah, it's more, it's more like a sort of decentralized consensus among trusted parties, where you can then be much you can then be much more efficient, because you're not having to constantly encrypt everything and and and and hide everything and so forth, right? And I think, I think to make it distributed decentralized, AI system, you will want subnets that are more like that, though, right, like if, if you have a bunch of trusted parties that have been helpful to each other before, then you should be able to to do things more efficiently in that sort of way. Trent McConaghy: 31:04 And I think it links to open COVID Hyper on as well, right in that, you know, we've got this, you know, working example of something intelligent, our human brain, like at that level of human intelligence, right? And then we have, you know, open COVID Hyper on, obviously, as a bunch of agents working together with clusters of behavior and so on and and then, of course, we have a blockchain style consensus, which we understand fairly well in all this. Right? So it's kind of neat to see that some of the other ideas from the human brain and stuff are actually finding themselves naturally emerging within, you know, some of the work that you're doing with open, going high, prone, right? Yeah, Ben Goertzel: 31:45 yeah. I mean the brain, the brain has an interesting mix of centralization and decentralization, if you think about that way, right? Because, I mean, the the, of course, there's no central cell in the in the cortex, and there's decentralized activity patterns and so on. On the other hand, stuff like breathing and heart beating is done in a pretty centralized way in the medulla and hypothalamus and and all this. And then motivation and goal orientation is done in a fairly centralized way. On the other hand, not all of our thoughts are goal oriented, but by by any means, right? So there's, there's probably some lessons in the central and decent the brain for for these uh, decentralized AI networks that that that we're building, right? Because, of course, having a decentralized platform doesn't mean everything you run there has to be maximally decentralized. It just means you're not forced, you're not forced into a central controller. Trent McConaghy: 32:56 Yeah, yeah. And this is some of the dynamics that naturally exist within decentralized networks might end up being some of the linchpins to actually making artificial general intelligence truly happen, right? It might, it might be, yeah, so Ben Goertzel: 33:12 yeah. I mean, if you look at predictor as a component of exco as AI network, I mean, pretty much the the goals are related to who is who is posing the problems, right? And so you have, you have someone with some motivation to do something. And it doesn't have to be just one guy, but it can be a collection of different guys with different motivations, posing the problems. Then among the ones who are making predictions and then getting compensated for them. There can be a lot of funky multi agent self organizing dynamics on the back end right, like one one option is it's a bunch of just independent, autonomous predictors in combination with each other. But if you go further down the line, I mean, you could have multiple agents collaborating to make, to make a prediction, deciding to split up the spoils for the prediction. They they've they've made together. Like, I mean, you can have a whole collection of multi agent dynamics behind the scenes, behind the different, different predictions, which is, is what, what happens in the brain as cell assemblies sort of are gathered together in a self organizing way in response to various challenges posed by the motivational system. Exactly, Trent McConaghy: 34:30 yeah, this is what we realized pretty early on as we were building predictor, is that in terms of, you know, catalyzing the dynamics, interesting dynamics, prediction is all you need, you know, and you let, you let that create the sort of the incentive the market, if you will, for if you have a market for predictions, then you can work backwards through this sort of like, you know, AI data supply chain, or call it intelligence supply chain, whatever you want. And like you mentioned. And, yeah, you can have, you know, five or 50 flat entities just, you know, submitting predictions. But you can have different emergencies of intelligence too. And we actually thought about fleshing out, trying to have supply chain that is more explicit for, you know, feeding signals into making predictions and so on. But then with conversations with with various people, including Ilya from near, you know, one of the authors in the attention paper, he was, you know, he's a Trent, you know, like, don't, why are you trying to flesh out decentralized training? Just keep doubling down on prediction and let that dynamics happen, right? And I reflected on that, and he's right, you know, and that's why we've actually just continued to completely focus on the prediction side, knowing that it is a big driver for everything else, right? Or, you know, Ilya sitz cover, he likes to give the analogy of, you know, if you think about a detective novel at the end of the novel, you know, you've

identified. You know, walking through the novel, there's a gun that's found, and there's a wedding ring and a table that's broken, and all these other clues that emerge. And imagine, if you can, at the end of the day, imagine the AI can predict, you know, who did it with what weapon, where, right? That's prediction, right, right? And I think that's a wonderful example, yeah, so, so people, you know, like to poop, poop prediction, saying it's just prediction, right? It's just, it's just a token, you know, stupid, stochastic parrot, right? But it's a lot more than that, right? It basically would have had to infer everything that happened in that novel to arrive at the who, what, when, where, and so prediction is a lot more powerful than people give a credit for, but it's a very powerful tool. Ben Goertzel: 36:53 There's a question from the audience which pertains to that which I'm seeing here is someone says AI doesn't care about truth, is glad to draw seven figures until, until it's told no. And then the thing is, that's just a dumb prediction, right? That's not like, that's not like, the intrinsic nature of an AI. That's the specific algorithms being rolled out now are just bad at making some kinds of predictions and they're wrong. And, I mean, that's, that's, I mean, just like Markov models were, were dumb at generating text 10 years ago, and now llms are better. So the, I mean, the seven figure fingers and the gibberish text you get an output from, from Dali and such. Now, I mean, yeah, it's, it is interesting that current AIs are much smarter than people at some kinds of predictions and much dumber than people at other kinds of predictions. But on the other hand, they're very different kinds of cognitive systems that people are. So it's not, it's not, not super surprising, and yeah, those pathologies will, will definitely go away. I mean, I think the the more concerning pathologies are the ones that are not so obvious, and we don't immediately see from the from the output, actually, which are, are something that we can think about more but, but I mean, then again, if you're predicting things about the empirical world there, there still is a ground truth for comparison across across humans and AIS, which is one of the interesting things here. I will Trent McConaghy: 38:38 also comment, it really depends on the AI system, right? So when people talk about AI these days, they really are talking about llms, which have been, you know, are, you know, large neural networks that have been trained against large data sets and so on. But that's not the be all, end all. And I want to just share a story. When I was doing my PhD, it was on creative design of analog circuits. So, you know, if you're an analog circuit designer, you use it sit down in front of your, typically, a pen and paper, and you draw, you know, connections of transistors and resistors, capacitors, etc, to come up with an amplifier or a filter or something. And that's the job

of an analog designer. And then eventually you put it, you draw it into a schematic editor on a computer, and then you convert that to an analog layout, which is sort of instructions of how to manufacture the thing. And then thing that gets manufactured, and it becomes part of a silicon chip, right? That you have in your cell phone for power management and for grabbing signals from the air, etc. So that's analog circuits. They've been around since forever, right? In fact, digital circuits are a very particular subset of analog circuits that are bound that have you know only true and false values and follow a clock signal. Right? So in my PhD thesis, my goal was to explore using evolutionary computation to design analyst. Circuits, there is some really great precursory work from John COTA Sanford, who had iterated with on this too. And John's problem, and my problem initially, too, had been that the simulator, it would figure out ways to cheat the circuit simulator. So you would come up with the AI, would come up with a design, and then it would and then design, you know, according to the simulator. It was perfect, right? But if you look at it, if you draw it out, it looks like a Rube Goldberg machine, complete garbage, right? And it basically had figured out, you know, you know, floating point precision errors or some other things. And then you would go and add a new constraint, and it would figure out a new way around it, right? And I would find myself playing whack a mole, but I did find a solution to this from my problem, and it was simply to constrain the search space in a more fundamental fashion. In my case, what I did was I designed a language for analog circuits, a grammar, if you will, where you know the lowest, the smallest atomic units of the of that language were transistors and resistors, and then you would group them to larger and larger pieces of current, mirrors and so on. And I actually went through a couple analog design textbooks and basically put in all the building blocks. And with about 30 building blocks, you would have about 100,000 unique analog circuit topologies, right? And every single random sentence you drew from that grammar would actually be correct by construction, right? So then what you would do is you would run evolutionary search across this, you know, searching through this space of trees, where each sentence is a tree in its grammar, and every single and then you basically search through this trying to find what's optimal for your particular design. Do you want to have low power consumption? Do you want to high bandwidth? And there's trade offs, and then I would actually get it to come up with all the trade offs at once, right? So this is an example of correct by construction creative design of circuits, right? It was coming up with specific topologies that no one had seen before, but no, if an analyst had a look at them, in Ben Goertzel: 41:48 his book genetic programming three doesn't give a lot of

examples of GP designing circuits, and he, he didn't, he didn't use this special language, right, right? So he didn't know Trent McConaghy: 41:59 this was, yeah, so he was coming up with stuff that was very like, if you he had maybe five or seven transistors, and they were very specialized, kind of weird, corner niche circuits. They were awesome results. I think they're, you know, really great. And I'm really proud and happy with John and his team and what they did, right, you know, analyst, circuit design. Right? Ben Goertzel: 42:24 So they were, so they were working sort of in a particular corner of design, of design space, where the peculiarities you mentioned were manageable, Trent McConaghy: 42:38 of problem space, right, but it was maybe 1% of all of analog circuit design, whereas the under analog circuit design could use a lot of help, and that would require, you know, 10 or 1500 analog components, and he couldn't get anywhere close to that, right? And I tried, I tried a bunch of different techniques, you know, riffing on what he had done, and none of them worked. So I needed to resort to this correct by construction thing. And what I see is that this is a but I think it's a really great example use case, because you can design grammars for anything, you know, the idea of grammatically guided genetic programming. This goes back to Ricardo Polly and, you know, Connor rein did a bunch of work, and others too. It's a pretty, you know, useful toolbox in in an AI land, where you can basically come up with a language to design the space you want to search across, and then you search it right? And, by the way, you can use this for agent control planning too. Greg Hornby did this as part of his speech. He did thesis at a Jordan Pollock in the 90s as well. So, you know, I also came up with a grammar for functions such that I could have really nice sort of white box model extraction that every single function that you came up with didn't look crazy. It wasn't science of sign of coast of vlog. It was, you know, very well formed functions that made sense, right? So I see that there can be a lot more of this than AI land. And, you know, I you know, mainstream poo poo things, because they've only seen one particular corner of AI land that got super famous. But a lot of this can get combined really well over time. And I know, for example, yeah, Ben Goertzel: 44:04 absolutely. I mean, llms are super cool. And I mean, I use them in in my work, in various ways, most days. On the other hand, they're very limited in very particular AI systems. And I mean they don't, they don't do creativity terribly well. And I mean both you and I have experimented with evolutionary learning for creating circuits or creating music or art in my case. And I mean these simpler AI algorithms have been much more wildly creative than llms are going to be. Llms are amazing at synthesizing stuff based on on the training data provided, which is also super useful. But I mean that the beauty of a sort of decentralized network. We're talking about is it gives a way to put different sorts of AI together. So I mean, in a predictor context, you could have some llms making a prediction, you can have some symbolic systems making a prediction, and you'll see who wins and which sorts of predictions, and then if cooperating together helps now limit the symbolic system predict more accurately, then they have an economic incentive to do that, so they'll win. They'll win the game more often, right? Trent McConaghy: 45:28 Exactly. It's like left brain and right brain cooperating, yeah, but you Ben Goertzel: 45:32 can have many, more than two lobes in the decentralized AGI system, right? So you can have this sort of tokenomically incentivized cognitive synergy between different AI systems participating in the same predictive games, both for predicting real world time series, like whether you know energy price token price, but you can also have it making predictions regarding what decision is best for an AI system, to me, right? So there's, there's a lot of lot of interesting layers here with the combination of different AI algorithms and prediction markets within this sort of decentralized cognition fabric. And we're just, we're just now at the point where you can actually roll out all this stuff at scale, and it will, it will work in a reasonably usable and affordable way, right? So is it quite, quite exciting time. So going back to your original question about ASI Alliance, that does know that? I mean, ASI lines. We can do a lot of different things. We're making a fairly simple product, allowing people to use Fetch singular notion technology behind the scenes, to put llms and knowledge graphs together. So we can, we can do a bunch of nice things working together the components in ASI lens. But there's also more advanced sort of cognitive experimentation that we can do, putting the different technologies together from the different projects that emerge into ASI lens. Now you know, Trent's team and my team could always have cooperated on using versions of predictor to help with decision, and versions of open cog, hyper on, and we might well have had there been no ASI merger on the other end, having a common token among our different projects, it has gotten us to talk more than we often that we were talking more before. It's gotten our teams to talk more than they've been before, and it certainly is, certainly is nudging these sorts of interesting collaborations to Crystal Lisa Rein: 47:50 Awesome. Well, look, that's all the time we have today. So I wanted to thank you both for coming on the show, and let's let Desi wrap things up for us real quick. Wake up little girl, you okay? She's drinking at you, Ben, yeah, Ben Goertzel: 48:22 she's doing the robot. Yeah, Lisa Rein: 48:24 she's doing the robot. She can't, we can't hear what she's saying. Okay, well, I want to thank Ben Trent for coming on the show and really excited. Any news about the token that you want to tell us before we say goodbye. Ben Goertzel: 48:45 Nope, the token. The token is there. The token, the utility token. It powers many cool projects on ocean, singularity, fetch and soon, kudos platform. So it's, it's, it's, it's, it's all good. I mean, this power is payment, reputation rewards and so forth. And I think what, what Trent and I are wrapping our brains now more is how to get smarter and smarter, AI, out of the out of the networks in which the token is sort of the the payment and reward fuel Trent McConaghy: 49:21 and Ben, for those who happen to be in Bangkok on November 11, Ben and I will be back back to back on stage, Ben talking about artificial super intelligence, and myself talking about human super intelligence. And there'll be a bunch of other great speakers there too, as part of super intelligence summit that is being hosted by ocean with help from Singularity net and others. Ben Goertzel: 49:45 Yeah, I think I'm gonna, I'll talk a little about Ultra intelligence too. I've decided super intelligence is already passe. I mean, as soon as we launched ASI lines, safe superintelligence. Now, no. Everything is a super intelligence. We need, we need to keep, we need to keep escalating. Lisa Rein: 50:04 All right, gotta raise that bar. Yeah, I Ben Goertzel: 50:07 thought it was gonna be hyper intelligence, but ahi is not a good acronym. So we're, we're gonna go, we're gonna go straight to ultra intelligence, I think. Lisa Rein: 50:15 All right, sounds good. We'll be talking more about Ultra intelligence. You heard it here first. Don't forget. All right, everybody, we'll see you in the next show. Thank you so much for coming on and as always, sweet dreams. Bye, bye. You.