#217 Ben Goertzel: The Path to Artificial General Intelligence, Decentralized AI and AI Consciousness

0:00that the book uh consciousness explosion in a way is a similar theme to my friend ray kerzwell's book the singularity is near er right the sequel do is 2005 but the singularity is near and i guess at some point kurzweil or his mind upload will come out with the singularity is here right and then and then then next will be like singularity passed by and here we are doing the same old right but with better gadgets so what i'm talking about in the consciousness explosion is basically the path to the singularity and beyond but i'm i'm trying to take a point of view more focused on states of mind and experience than on the gadgets and and the underlying algorithms and and technology we've spoken a number of times um and yeah i wanted to hear about the book although i saw on amazon it's a hundred bucks so well this paperback is cheaper i think yeah the color version is more expensive, but there's also, there's a free PDF you can download from the consciousness explosion.ai website.

1:05So you can get the book for free and the Kindle version is 10 bucks or something. So there's that information. The information is made to be free, but if you, if you want a big fat book with a lot of pretty color pictures, it just still costs money for ink. That's all. Could you start by introducing yourself and give some of your background and a list of the various projects sort of in the order of their importance? Yeah, I'm Ben Goertzel. I'm originally a mathematician, but I've been doing AI since the late 1980s, since long before it became such a popular thing to do. I, I introduced the term and concept of AGI in, I guess, 2005 in the book titled artificial general intelligence.

1:57And I've been working on both R and D aimed at building general intelligence and applications of AI in diverse vertical markets over many decades, uh, among various other things. I led the software team behind Sophia, who was the first robot citizen, humanoid robot made from Hanson Robotics in Hong Kong, where I was chief scientist for a number of years. Since 2017, I've been CEO of SingularityNet, which is the leading blockchain based platform for decentralizing AI. And we've recently done a tokenomic merger of singularity net with two other blockchain AI projects, fetch and ocean making me now the CEO of the ASI Alliance, the artificial super intelligence, uh, Alliance.

2:55So that's a sampling of sampling. Some of the things I'm, I'm engaged, engaged with now. I'm also dealing with a lot of application, AI application projects. We see behind me on the wall, the Desdemona robot, who is the lead singer of the Desdemona's dream band, where I play the keyboards. And we've got a lot of AI generated singing and music and so forth going on. So now, yeah, it's an incredible time to be in the AI field, right? Both in terms of AGI R &D and in terms of just applying AI to all sorts of practical things yeah i i you know we have spoken before i don't know if you remember but i'm i'm a bit of a skeptic on uh on sophia's contributions to ai i think it's uh it creates more confusion than anything but we we talked about that last time um on the i'm interested in the book because I've been talking to people.

4:00I had Stuart Hameroff, who's working on microtubules with Roger Penrose or wrote a book on it with Roger Penrose. I know Stuart moderately well. I've met Roger a couple of times and also James Tagg and others from Penrose Institute. And so I'm fairly familiar with that line of thinking. Yeah, I wanted to hear your thoughts on, and unfortunately I haven't read the book, but I wanted to hear your thoughts on, well, first of all, the premise of the book, if you can tell us the premise of the book. But then on all of this talk about consciousness surrounding artificial intelligence, which has caused a lot of excitement and, again, a lot of confusion.

4:59Yeah, I mean, the concept of consciousness has caused a lot of confusion, even setting aside artificial intelligence. so it's hardly surprising the intersection of consciousness with agi should confuse people even even more i mean we don't have a solid understanding scientifically of what consciousness is right so we i mean we don't we don't have a scientific refutation of solipsism of the idea that only i am a conscious being and all the rest of you are just simulate or put here to uh alternatingly amuse and annoy me right so i mean when we we don't have consciousness even nailed down in the human or animal or plant realm right so i mean does it does a tree hurt when you chop it down right does a fish hurt when you hook it the best definition and i can't remember who came up with it but uh that it's consciousness is what it's like to be something like what is it like to be a bat i think right right but then you had quite renowned people like daniel dennis saying well that's all a bunch of rubbish and the whole concept should be should be thrown out right so i i i mean it's a i'm just saying it's a deep topic i have my own fairly strongly held views on it But it is an unresolved topic from the standpoint of scientific consensus, even outside of the AI domain.

6:35So then when you introduce AI, I mean, all the confusion about consciousness propagates over to the AI domain, right? So, I mean, you have people like Dennett was who are like, well, the human brain is in essence a biological automaton, and it uses this language of consciousness to describe some things that it's doing. But this whole notion of QALIA and experience is a bunch of blah, blah, blah, pretty much. And then you have folks who are panpsychist, which is more the direction I go in, who think everything in the universe is some element of consciousness, which is just manifested differently in different kinds of systems.

7:21And then, I mean, then, then you have folks like hammer off and pen Rose. Yeah. They think conscious experience is a real thing. It in here is only in certain kinds of matter and not others. And then they have a particular rather out there theory of what kinds of matter can, can manifest consciousness. Like you need special quantum gravity dynamics to manifest itself in bio nanotech. Yeah. said that philosopher galen strassen who you're probably familiar with he he is has a book called physicalism entails panpsychism where he's arguing if you believe the world is physical and everything is physical then you should believe everything is conscious and he tries to argue that any other position is logically untenable which on the one hand i like his argument on the other hand i'm not that much a materialist i don't i don't believe everything is is is physical and like a strong met a foundational sense anyway, but all that confusion propagates into the AI world.

8:26So then people are like, well, no robots are just machines, just like people are machines. They have consciousness in the exact same meaningless sense that, that, that people do. Right. Then, then you have other people saying, well, all particles are conscious. So why aren't the particles inside machines, just as conscious as the particles inside humans. Then you have other people saying no, only certain kinds of matter can be conscious. And it just happens that humans have that kind of matter and computers in their current form don't, but maybe a quantum computer or quantum gravity, supercomputer will.

8:58And all this, all this is interesting. Pretty much all that philosophizing goes on off to the side. And meanwhile, we're building AI AI systems with greater and greater levels of, of, of capability. But then even once you get to an AI system, which appears to have human or superhuman capability, that won't necessarily resolve all these questions either. Right. But there are many views on it. It's like what Roger Penrose believes is that you will never get human level creativity and insight based on a digital computer. Pembro's believes that requires some as yet poorly specified quantum gravity voodoo manifesting itself in, in, in, in how the brain works.

9:46Now, many others believe you could get superhuman creativity in a digital computer, but it would still have no experience. Right. And so these, these are all points of view that are, are out there. and I have my own positions or intuitions but we don't have scientific proof of any of it yeah and I'm more confident I know I'm more confident I know how to build a thinking machine with superhuman capability yeah I am that I understand the nuances of how consciousness interacts with or manifests itself from matter. It seems like we're going to discover a lot about that in the next few decades. Yeah.

10:31Well, what's the book about, The Consciousness Explosion? Well, that's a different topic. Yeah, yeah, yeah. So all this is mentioned in the book, but it's not necessarily the theme of the book. So the book, Consciousness Explosion, in a way is a similar theme to my friend Ray Kurzweil's book, The Singularity is Near Earth, right? Right. The sequel to his 2005 book, the singularity is near. And I guess at some point Kurzweil or his mind upload will come out with the singularity is here. Right. And then, and then, uh, then next we'll be like, uh, singularity passed by and here we are doing the same old shit.

11:09Right. But with better gadgets. So what, what I'm talking about in the consciousness explosion is basically the path to the singularity and and beyond, but I'm, I'm trying to take a point of view more focused on states of mind and experience than on the gadgets and, and, and the underlying algorithms and technology. I mean, the underlying technology for building AI, of course, is what I spend most of my time on. Right. I mean, it's important. It's highly technical, but it's, it's, it's, it's key, but I mean, you could look at the internet similarly in both ways like there's internet technology i mean that there's all the amazing gadgetry inside this phone that lets it make a nice screen and lets it send information to them from satellites there's satellite technology right there's protocols on the other hand there's the internet as a social and psychological thing as an experience for people to interact with and and build cultures in and that depends on the underlying tech but it's also in a way a different topic than the underlying tech, right?

12:26So what I'm trying to look at in the consciousness explosion is what kinds of mind systems are we building? What kind of mind systems are they going to grow into? What kind of mind systems are we becoming? How does our own state of mind individually and culturally, how does it influence the artificial minds that we're building and and and the path of the of of the singularity as it as it unfolds right and so this is this is really a perspective on a bunch of things that i've talked about that that great length in in in media over the over the last few years including agi versus narrow ai super intelligence and decentralized versus versus centralized AI and so forth.

13:17Because if you, if you look at current commercial AI systems being built by big tech companies as minds, it's pretty scary. Like we're building a bunch of proto AGI autistic psychopaths pretty much, which are like, they're concerned with only one thing, maximize eyeballs, maximize revenue, maximize profit. it. They don't care about anything else. Besides that they interpret all data relative to their one particular goal. They're also built without much focus on modeling themselves and understanding who they are, or understanding who they're interacting with. They're like building an I thou bond with who they're interacting with, right.

13:58So we're building some very particular particular kinds of minds in big tech companies. Now they're where they're autistic psychopaths focused on spying, selling people stuff they don't need, killing people who happen to live in a different country, plagiarizing stuff and helping wall street to steal people's money. Right. So, I mean, we're, we're building digital autistic psychopaths focused on a a few particular ways of accumulating certain sorts of, of, of resources. And we're doing this mediated by corporations, which are also probably autistic psychopaths. If you, if you're gonna psychoanalyze the corporations, but I mean, most corporations are psychopaths by design, right?

14:51I mean, they're only concerned with maximizing shareholder value and, and everything else. You have the notion of a B corporation in the U.S., which is supposed to balance benefit with profit maximization. And China is quite different. I mean, big companies by design are supposed to benefit the common good as interpreted by the CCP as well as making money for themselves. So, yeah, to that. right. But then you can look at what states of mind people are in when creating these sorts of AI systems and what sorts of mind interacting with these AI systems put us in, right? Like, which is quite sort of resonant with the state of mind and the AI's themselves.

15:39Like if you look at AI for running social networks, I mean, they tend to fractionate people. They tend to focus you on buying stuff you don't need and on rallying up your righteous indignation and like whatever emotion will keep you glued to that Facebook page for, for, for, for as, for as long as possible. So that's, that's psychology of the current AI industry and the people in involved in it. So then you can ask what sort of singularity will this lead to? What other sort of AI could we be building, resonating with whatever states of human consciousness and whatever sorts of singularity could be evolved.

16:21Yeah. Well, I mean, you're talking about building minds, and that's why I was, I mean, your book is about the consciousness. It's titled Consciousness Explosion. That's why I was asking about consciousness. I mean, why are you using the word mind if you don't believe that a a neural network no matter how big could become conscious I personally believe that every elementary particle has its own spark of of consciousness so I tend to be pretty strongly panpsychist and I I believe Strauss and other philosophers that this is pretty much the only consistent and coherent way to think about consciousness.

17:11And of course, it's a lot of thinking with a long history. You can look at, say, the Buddhist logicians, Dignaga and Dharmakirti from the Middle Ages, laid out a whole theory of mind based on a rational panpsychist view a long time ago. So to me, that's sort of the only coherent premise I know how to begin with, but it doesn't solve everything. Like it could still be that, I mean, okay, if a coffee cup has its own species of consciousness, okay, but it may be a very simplistic species of consciousness. It's boring relative to a human being. Right. So, so then, that still doesn't answer you in what sense does a coffee cup feel pain when you cut it in half?

17:54Cause you could have some forms of consciousness that don't feel pain, like a person with pain, a symbolia, right. Or a very muted sort of pain. and it does that panpsychism in itself doesn't tell you whether a digital computer that appears to be more aware and generally intelligent than a person is having the same sort of experience that that a person is right so i mean that i i i do think everything is conscious but i think that's only a starting point to to invest investigating consciousness right it's not it's not in the finishing point but i also suspect that if you have a digital system that has the same sort of emergent patterns in its structure and dynamics as a person has i suspect it will have the same flavor of conscious experience that that that a person has i mean it seems seems like by far the most straightforward hypothesis.

18:56I I'm just saying that's an intuition of mine. I could, I've written a bunch about it in this book and elsewhere. I wouldn't say it's nailed down scientifically in, in the way that we've nailed down that like, okay, the stars are probably not angels flying around. They're probably giant balls of fire, right? Like we are plasma at least. So, I mean, we haven't nailed down theories of consciousness to that level yet. So, I mean, there may be some big surprises, right? And it'll be interesting to explore them. I think we're going to do experiments with brain-brain interfacing and brain-computer interfacing, which should be quite informative.

19:40What happens if you literally wire your brain into the digital neural network of an AI system versus wiring your brain into a neuroid grown in a vat, right? Like, what's the different subjective sense that you get while carrying out these brain-machine interfacing experiments? I mean, I think this should be quite informative in a sort of mix of an intuitive and scientific way. And we'll be able to do that during the coming decades. Right? So there's going to be a lot of new input coming in to form our particular theories of consciousness. And that's... Well, yeah, yeah, yeah, go on. Well, so the question of consciousness aside, because I have to say, if you say a coffee cup has some form of consciousness, I think pretty much drain the word of any real meaning.

20:42Well, why? I mean, a coffee cup is in everything around us is in space and time. Space and time. Spatial and temporal properties are imminent in everything. That doesn't mean space and time have no meaning. I look at consciousness just like spatiality and temporality. These are aspects that each thing in our universe has. I mean, they can have different, something has different spatial locations, has different positions in time, different things, their consciousness can have different aspects. It doesn't mean it has no meaning. I mean, at least that's how I look at it. Yeah. Yeah. So on your AGI work, what kind of paradigm are you pursuing?

21:24I mean, there's been right now, everything's being built off transformer algorithms and neural nets and back propagation and all of that. Are you following that model? or not at all no i mean we're i'm using those systems but they're not central to my my agi architectures i i mean i think i i view current large language models and other deep neural nets turned on huge bodies of data as extremely interesting and useful catalogs of human knowledge and i mean the fact that you can synthesize stuff out of these knowledge catalogs is is amazing right i mean i mean i i i i love working with with these these systems on the other hand i i think they're doing something very different than than a human mind is is doing when it's thinking or or than a agi system really worthy of of the name will be will be doing when it's when it's thinking so i mean i think on the other hand my intuition is you could automate 80 or 90 % of the human economy using LLMs and CNNs and similar technology.

22:42So I mean, I think there's a huge economic revolution and humanitarian revolution that can happen from these systems. I don't think these systems can ever be creative or original in the way that people are. That's just not what they're doing. They're combining a whole bunch of knowledge on the surface level. On the other hand, most economic activity does not involve a great amount of creativity or originality either, right? most economic activity is wrote and is doing small variations of stuff that's been done before. So then, I mean, you can automate it with these technologies as hardware gets optimized and the software is optimized, that'll become cheaper than having humans do this.

23:19So from, from a business view, I can totally see why businesses want to focus on this technology because it's there, it works. You can scale it up. You can bring the price down, you can improve reliability, And then you can automate almost all of the world economy. So then in that sense, why would you worry about doing R and D aimed at automating the last 10 % of the world economy or transforming humanity in some way, right? There there's so much money to be made and so much good to be done. Right. I mean, I, I think you could automate so much medical research using, using LLMs. And then, I mean, you can, you can automate supply chains to deliver, deliver food to the impoverished.

23:57Like there there's great stuff to do with them. And bottom line is though, I mean, they're so far, they're not agents. Like they don't try to understand who they are and who you are and what is their position in the world. And they don't abstract knowledge very well. They abstract knowledge a little bit, but they're mostly just building surface level indexes of detailed knowledge. And I think that making leaps of creative generalization and imagination, making leaps beyond your knowledge is done by having abstracted your knowledge. And these systems aren't doing that. And agency and abstraction come together in human minds, right?

24:36Like we, we form abstractions, not arbitrarily, but we form the abstractions of our detailed knowledge that will help us achieve our goals as an agent in the world and help us maintain our boundaries and existence as an agent in the world, help us grow and develop as an agent in the world, right? The need to grow, develop, survive, achieve goals is what drives us to make mostly appropriate abstractions. we then use these abstractions to drive creative generalization, right? And all that stuff is missing in current LLMs. And you can't just wrap an LLM in like an agent wrapper. People tried that with baby AGI and so forth right after GBG3 came out.

25:20You can't just wrap an LLM in an agent architecture and have it be an actual agent in the way that people are. I mean, you need a whole different way to think about knowledge and the representation between knowledge, memory and learning and so forth. So yeah, we're in the open cog hyper on project, which is my main attempt to build a GI. We are taking quite different approach. I mean, we start in a way from human cognitive science and the different kinds of memory and reasoning that the human mind seems to have like episodic memory, declarative memory, procedural memory, working versus midterm versus long, long-term memory and the different sorts of goals on different timescales that we have.

26:05But then we're not trying to emulate precisely how the brain works. Instead, we're taking the different key types of memory and learning and reasoning and processing that seem to be done in the human mind. And we're realizing these using a diversity of mathematical and computer science algorithms. So on the software level, we have a large distributed knowledge graph or knowledge metagraph more properly. And we, we represent both perception, action, reasoning, learning, different types of memory as different sorts of organization within this large distributed not knowledge graph, which can then rewrite and modify itself in the course of its thinking.

26:47And we then have a blockchain software layer that lets us all knowledge graph run on a decentralized network of machines without any central owner owner or or controller now you you can have transformer type stuff in there as as part of that because these networks are very good at recognizing patterns in in data right but but in our approach recognizing patterns and data is not the whole thing and you can if you look at something called the common model of cognition that was put out there by Paul Rosenblum and John Laird and others who have been sort of good old fashioned AI guys even longer than me.

27:26They try to articulate what are the key aspects of human cognitive science that we need to put into an AI if we want it to be human-like. And you can see that current deep neural net systems being used in the commercial world, I mean, they cover only a quite small percentage of the different kinds of things that seem to happen in, in, in, inside, inside the human mind, but we are. So relative to these deep neural net approach, we're pretty far out there relative to Penrose and hammer off. I mean, we're, we're building stuff in digital computers. So, and I mean, in, in, in that sense, we're relatively close to the mainstream.

28:05We're acting under the premise that by putting stuff on a large distributed decentralized network of digital computers, we can create something that thinks better than people. Now, I mean, I've done a bunch of work in quantum computing theory too. So I think once we have like QPUs, we have quantum processing units to plug in on our server farms. I mean, I think we can upgrade many of these algorithms in a very interesting way. Like it's not that I think that won't contribute something, but I, I think it's probably not necessary to get superhuman general intelligence and probably not necessary to get intensive subjective experience in the AI either.

28:49Although the latter part, again, is a bit less certain since we don't have a real theory of experience. Yeah. So the knowledge, let's start with the knowledge graph. How do you build this knowledge graph. Your business is ready to launch, but what's the most important thing to do before those doors open? Getting more social media followers or actually legitimizing and protecting the business you've been busy building? Make it official with LegalZoom. As a business owner, traditional legal services come with major sticker shock. Getting registered or talking to an attorney shouldn't have to cost you so much.

29:34Well, thankfully, today's sponsor, LegalZoom, created a better way to start and stay in business from initial formation to one-to-one legal consultation. I've used LegalZoom for over 10 years, and it's kept me compliant. They have everything you need to launch, run, and protect your business all in one place. setting up your business properly and remaining compliant are things you want to get right from the get-go, but you don't have to strain your brain or wallet. LegalZoom saves you from wasting hours making sense of the legal stuff. At LegalZoom.com, you can take care of business legal needs in just a few clicks.

30:18And if you need some hands-on help, their network of experienced attorneys from around the country has your back. Launch, run, and protect your business to make it official today at LegalZoom.com and use promo code Smith10 to get 10 % off any LegalZoom business formation product, excluding subscriptions and renewals. This offer expires at the end of the year. Get everything you need from setup to success at LegalZoom.com and use promo code Smith10. That's S-M-I-T-H-1-0. LegalZoom.com and use promo code Smith10. S-M-I-T-H-1-0. I mean, there are two ways, or three, I guess. the number one way has to be from experience just from perception of the world i mean percepts come into the graph patterns are recognized in those percepts and you have the formation of a hierarchy and heterarchy of patterns inside that knowledge graph but you can also form a knowledge graph from reading text and we use llms for that so we we use prompting and fine tuning of open source large language models to build knowledge graphs from existing text and from existing structures data.

31:43So we have a huge like bio knowledge graph that we got from a bunch of biology data sets and bio ontologies and so forth. And you can you can cross link all these, all these things together. I mean, you, you could build a purely experiential learning system that just builds this knowledge graph from low level perception. And we were experimenting with that a bit, actually. But it seems like if you take the view that there's an AGI race going on, and your competitors are trillion dollar companies, you may be able to advance faster by taking direct experience and combining it together with knowledge graphs you can download and sort of cross-linking it all in the distributed system.

32:24Yeah, and so how are you on the direct experience accumulation? You're talking about cameras? Yeah, yeah, yeah. Actually, let me briefly introduce you to one of my friends here in the office. So we've got this little guy here, for example. So this is an in-progress version of one of our mind children robots that we're building here in our little office in the Seattle area. So, I mean, I've been working with Hanson Robotics for a long time, as you know. So we have the Sophia robot, Desdemona, and so forth. And we're still working with the larger humanoid robots from Hanson Robotics, which is great.

33:24But I wanted something smaller and simpler that I could just have in my house or classroom and recharges itself when it runs out of power like a Roomba. and is a little smaller and simpler to deal with than the Hanson robot. So yeah, I don't think you need a human-like robotic embodiment necessarily to get to human level AGI. I mean, that robot I just showed you will have a human-like face somewhat similar to a Hanson robot when it's all put together. But I don't think you need that for human level AGI. On the other hand, it's very convenient, right? Because we're trying to teach the robot ourselves as humans.