Full transcript: Dr Ben Goertzel on London Real
2:25 Brian Rose: This is London Real. I am Brian Rose. My guest today is Dr. Ben Goertzel, the scientist, futurist, author, and entrepreneur. You are considered one of the world’s foremost experts in artificial general intelligence and have submitted over 150 technical papers and published 25 scientific books on the subject. In 2017, you founded SingularityNet with the goal of creating a decentralized, democratic, inclusive, and beneficial artificial general intelligence.
3:09 Your latest project, Zarka, is a supercharged, intelligent, neurosymbolic, large language model at massive scale and promises to take on open AI’s chat-chapT but go much, much further. You believe it is extremely important that our AI systems are decentralized and not controlled by an elite organization. Last Friday, Sam Altman was removed from his position as CEO of OpenAI and now looks set to be reinstated, causing chaos in the industry. You have said that as we move closer to AGI, we must ensure that profit-making and corporate influence do not interfere with the ultimate goal of creating AI that is aligned with human values. Ben, welcome back to London Real.
3:50 Ben Goertzel: Yeah, it’s good to be back, Brian. As we say in each one of these interviews so far, it’s amazing how much stuff can happen in a few weeks between our two conversations. In this case, it’s crazy what’s happened in the last three days regarding the human drama in OpenAI. It’s been amusing and intriguing to follow, really. I think there’s…
4:30 Brian Rose: Let me jump in and walk through the timeline, shall I? Yeah, sure. And I’ll just get your thoughts on it. I was here at the Studio Friday night. I just finished a master class where I was talking, I think, about AI and talking about Zarka and obviously our investment with you guys in Zarka through London Real Ventures. And then, boom, the headline hit.
4:49 We watched over the weekend as soon Greg Brockman, the president, had left. Then we heard that Meera had been put in as interim CEO. Then I think by Saturday, Sunday, Sam had been hired by Microsoft. And now by Monday morning, Tuesday morning, they were all back at OpenAI with potential changes to the board. Meanwhile, rumors were circulating that this was all because of a conflict of philosophies between – and I don’t know if this is true. I want to get your thoughts between Sam Bultman that had just done a dev day and was charging ahead with GPTs that you could customize and making money and also connecting chat GPT to the live internet up to April 23, allowing auto-prompting of other GPTs. These are considered sometimes dangerous. And then, of course, the chief scientist, Ilya Sutskovar, rumor was that he was on the other side trying to pull back the reins, warning of the dangers, and a little bit scared that this was being commercialized too fast. Rumors started saying that it was Ilya that tried to push him out, but needless to say, 710 out of 750 OpenAI employees signed a letter saying they were going to walk unless they got Sam back. And so now it looks like things have been restored. They’re trying to alter the board, which was some hybrid of a nonprofit and a profit, and it’s a mess in itself. And some people say Satya Nadella is the ultimate winner at Microsoft because he’s going to get OpenAI now without the constraints of this former nonprofit board. Ben, what do you see
6:25 Inside this drama? You’ve been tweeting about it. It’s been really interesting. Do you think this was ultimately a conflict inside the organization between Ilya and Sam? Or was there more at stake here? And then what are the bigger ramifications for the industry?
6:39 Ben Goertzel: I mean, I think there’s a human soap opera aspect, and there’s a corporate governance aspect. And then there’s a broader aspect to do with the different attitudes and forces on the planet about AI all together. So I mean, the human soap opera aspect, I don’t have tremendous insight into. I mean, I do know the people involved, but I haven’t been following their ups and downs on a personal level. I mean, it’s obvious that Sam was going his own way and trying to streamline and push the organization toward just going as fast as possible in terms of research and commercial products sort of at the same time and trying to keep that link together, figuring no doubt that that’s the way to keep the pipeline of money and computer hardware and talent flowing to the research. I mean, I don’t think Sam Altman is in this primarily to make money. I think Sam Altman is in it to create AGI and create a positive singularity, but he could see that you will need a pile of money along the way to get there. And he’s looking at developing amazing commercial products and rolling them out to customers as a way to keep feeding the research team with funds and supplying them with the hardware they need and so forth. Clearly, there are others within OpenAI who are more interested just to push AGI R&D with commercialization not being so important to them.
8:42 And this tension, of course, was there in the beginning of OpenAI, right? I mean, when Musk originally was talking about OpenAI, he was like, I’m donating a bunch of money so that AI will be done in the open for the good of humanity rather than just eaten by big companies. So that concept was there in the beginning, which of course has made their pivot to close alliance with Microsoft and to rapid fire commercialization of products away from open source to closed source, right? So this tension has been there since the beginning, and clearly because of that, there have been some people in OpenAI who were really into the original idea of let’s do AI outside the big tech domination and let’s make sure it’s benevolent for all humanity. So there are people where that was their primary motive, right? And then there are people who are interested in that, of course. I mean, none of them want to build a killer AI that annihilates everyone. There are people who’s interested in that but who are more oriented toward let’s just make the smartest AI and roll out in amazing products and grow ourselves into a huge, huge, behemoth business so that we then can do whatever we want, right? So you have this tension between the more R&D focused and AI safety focused people and the more R&D meets product development people. And it’s not like these are like Palestinians and Israelis or something, right? I mean, it’s not like incredibly at odds. They’re
10:42 All brilliant geeks who think we should be building AGI, right? It’s just some of the dial turned a little more toward, well, let’s do research and let’s make sure everything is safe before launching it. Others want to turn the dial a little more toward, well, hold on. We need to launch stuff out there and get customers or else our competitors are going to beat us to it, right? So you can see that tension there. And you can also imagine that over the years, more of the people who were oriented toward pure R&D and toward safety, probably more of these people fell away from the organization and new hires focused more on people who were going home for commercialization, right? So there was, I mean, there was a letter from some former open AI staff sent to Elon Musk within the last 24 hours or something saying, by the way, Sam Altman has been, you know, pushing out people who were advocating for AI safety and an R&D focus for a long time. And I mean, that to me is not the most damning accusation you can make about the CEO, right? Like, I mean, of course, if a CEO is pivoting from one direction to another, I mean, you may argue about whether the pivot is good or not. But of course, if the CEO is pivoting a business, which happens all the time in startup land, right? If the CEO is pivoting a business, of course, there’s going to be a focus on, you know, giving more power and influence to the people who support
12:19 What you’re pivoting to and making more hires and the people who support what you’re pivoting to. Right? So I mean, I wasn’t engaged in the conversations. It could be that Altman was somehow inappropriately pushy in affecting this. But I mean, the basic dimension of this seems like what a CEO of a startup would do when pivoting in a complex social situation.
12:45 So also Elon likes to kind of troll the situation because he’s got beef with Sam and open AI. And he’s trying to promote Grock as well. Elon told someone to send him. No, it was also anonymous letter. But yeah, it’s true. It was there. But the reason it would come to Elon is because he was the founder of open AI, right? I mean, it’s not like they sent it to Bill Gates. So I mean, there’s some logic there because Elon was the founder. And in many of his moods, he’s more on the AI safety slash pure R&D side. Now, then you have the fact that, in spite of saying that AI is too dangerous, we shouldn’t be doing it, it’s going to kill everyone. Elon has launched his own open AI rival called Grock. So we can leave the psychoanalysis of Elon Musk’s dissociative identity disorder for another time. I mean, I think that the corporate governance aspect here with open AI is interesting to those who are corporate governance wonks, right? Because I mean, they set it up really weirdly with this benefit oriented nonprofit, which nonetheless has some direct power over the for profit. And it always seemed weird, right?
14:01 Because I mean, I’ve set up many nonprofit and for profit organizations in my history. And there are common practices for doing this. What they did is not one of the common practices for doing it. It’s something that would look strange because indeed, if you’re bringing investors into the into the for profit, like, of course, there’s going to come a time when the investors in the for profit, want to do something to make money, and people who found the nonprofit want to do something different. But then the for profit, of course, is brought in staff who also want to make money, partly because they have a set of options and stuff, right? So then, I mean, the, the method unfolded in a way, it’s almost implicit in the crazy way that they set things up, right?
14:48 Because I mean, even if you set the philosophy aside, of course, for profit will want to make money. It’s a for profit nonprofit is working for other goals, employees of for profit, one of their options to go up in value. So I mean, they, they also want to make money. So I mean, you’re, you’re setting things up in a strange way, which works, as long as everyone is beautifully aligned, and best friends. But what you see in business many times is you don’t really want to architect things to the only work if everyone is best friends. I mean, that that works with a bunch of family businesses like they do in Asia. I mean, until it doesn’t work and two brothers fight with each other or something. But it’s, it’s, I mean, that’s a risky way to set things up.
15:34 So all this, all this backdrop, what are the bigger? All this is there and provides a sort of case study for broader patterns, which are really the interesting thing, right, which is can be sort of parodied as D cells on one side, and what’s come to be called the Eax effective acceleration is on the other side. And that is a fundamental clash of attitudes, which is then manifested itself here. But then there’s another dichotomy, which is among effective acceleration is there are those who think big corporations are the best way to effectively accelerate things. And there are ways we think we should effectively accelerate things in some other way than having big, big, big companies control everything, which is your philosophy, right? Through decentralized, right? Are you an effective accelerator?
16:35 You’re an accelerator, not a decelerator and a decelerator. Okay, Max Tegmark future of life Institute is a great example of a decelerator. Let’s write a letter. Let’s pause for six months or a version of a decelerator. And then Ilya suits cover is also a version of a decelerator.
16:51 Okay. All right. And these are both very knowledgeable, intelligent, good hearted people will have a passion for creating AGI. Right? So they’re, they’re the more interesting kinds of decelerators, I think, because I mean, decelerators who are just like technology is bad, AI, AI is bad, we should all go back to the 1950s America or as the Unabomber, Unabomber philosophy. That’s not, that’s not so interesting. And but people like Tegmark or suits cover are smart. They understand AI, they’ve devoted their lives to building AI, they want to see AGI exist. They just, they have the idea that if we don’t do it right, is going to murder everyone. And so we need to be super, super careful to do it, to do it right. And that’s, so they, they have that notion in mind. And then with that mind, they’re like, well, it will be better if we just proceed more slowly and carefully. Of course, there’s practical issues that have to confront also, like, China is no longer proceeding slowly on AGI and nor all the random people in every country in the world hacking open source LLMs like, like Lama too. So I mean, even if you believe decent deceleration is the right thing, you could still figure, well, but my evil competitors aren’t decelerating. So, right. So it just doesn’t, just doesn’t make, make sense. But yeah, you have, you have these decelerationists here. And then Altman clearly, I mean, no one wants to just totally put the pedal
18:34 To the metal and go as fast as possible with no heed for, for human safety, or at least no major player does. But I mean, Altman, like me, has the dial turned further toward, well, there’s a lot of other potentially bad stuff happening on the planet also. So like, if we decelerate AI, like bioweapons, nanotech, and nuclear weapons, and all kinds of other mayhem are not, are not decelerating just because we stopped doing AI. So it’s, it’s not that clear whether the risk of accelerating AI is higher than the risk of decelerating AI for, for human welfare, given all the other crap crap going on in the planet. So I think, okay, so Altman and I share the idea that actually accelerating AGI, if you don’t do it in a really, really stupid way, is more likely for the good of humanity, because we’re going to get to mind children, smart machines that we’ve created reflecting our values that are smarter than us and can help us improve our lives. So I think Sam Altman and I share that perspective and the majority of the OpenAI team shares that, shares that perspective.
19:48 Brian Rose: Okay, but then there’s the distinct difference between the two of you.
19:53 And you tweeted about this over the weekend, Ben, where you said it shows what a small group of elites can do with this effective acceleration versus what you’re doing, which is trying to make it decentralized and open source. Because if you and Janet got into a big fight at Singularity Net, it’s not going to cause chaos necessarily, because it’s a decentralized system. So can you speak on the dangers we saw of a centralized entity? And now, is it even more dangerous now? Because for a while, Microsoft now was going to fund everything. Has it gotten also more extreme after the weekend?
20:31 Ben Goertzel: This question has a couple of layers, actually. So one lesson we can draw from what we’ve seen happening at OpenAI. I mean, one fairly simple lesson we can draw is, okay, like it doesn’t make sense to have these six or seven people, however many it is on OpenAI board. It doesn’t really make sense to have them controlling the fate of the first AGI, right? So I mean, OpenAI has not created a human level AGI. So this all is not such a big deal. But suppose they had, right?
21:08 Suppose OpenAI had really created the world’s first thinking.
21:12 Brian Rose: That was the rumor over the weekend. The rumor was they had over the weekend.
21:16 Ben Goertzel: That’s clearly nonsense.
21:18 Brian Rose: Okay, how do you know?
21:19 How do you know it’s nonsense?
21:21 Ben Goertzel: Because there’s 700 people working for OpenAI. One of them would have said something, some concrete detail about it with some Reddit post or something. True. I mean, these are not all people with a top secret security clearance under threat of death. They’re tech geeks in a very chaotic dynamic situation, right? So I think if there had been something that big, if there was something that big, someone would have leaked it. Secrets don’t keep that well in today’s world. But I mean, they might among members of the mainland China intelligence service or something, because they know if they leak, someone might find out and torture their family. But I mean, OpenAI is not that kind of situation, right? Okay.
22:06 But supposing, on the other hand, supposing that a year or two years or three years from now, OpenAI did create the big break through the AGI. Personally, I think my team is more likely to do it. But for sake of argument, suppose they did, right? Then do we really want these evident and amateurish jokers on the OpenAI board? Do we want these people controlling the fate of the first AGI according to the same level of wisdom and strategic planning that they’ve displayed this last weekend? Like we, as a species, totally do not, right? So I mean, there’s some old interview with Altman bouncing around where he says, well, it’s very important that the board can fire me if they want to. No one person should have that much power. But I mean, in a way, that’s right. No one person should have that much power. But no six or seven person board should have that much power either, right?
23:05 Brian Rose: Well, we saw how much power Sam had over the weekend.
23:08 Ben Goertzel: That’s the crazy thing. Sam can make or break an entire company, can take a $90 billion valuation to zero, can walk away with 750 super talented people or not. I mean, Sam has a lot of power. That’s because of the strange way OpenAI was set up, right? Because I mean, if it’s set up as a regular company, normally a CEO can’t quit and take all the staff, there’s non compete agreements and so forth, right? So they clearly set things up in an unusual way for their own reasons. And they’re discovering sort of why mainstream corporate governance is the way
23:45 It is, right? Like it evolved to protect different parties against against against different different other other parties, right? So I mean, yeah, that’s, there’s, like I said, if you’re a corporate governance wall, this is like a textbook example of why things are set up, the sorts of ways they normally are. But I think the question of centralization versus decentralization is more interesting because there is that one point that, yeah, putting a small committee in charge doesn’t really make sense. And even if that committee were wiser people than we seem to have on the OpenAI board, still, each group of people has their own peculiarities and biases and in groups and so forth. It’s like, it’s too big a thing to put in the hands of some little committee with their own particular fights and problems and so forth. But on the other hand, you have the elephant in the room, which is Microsoft, right? So I mean, you have a bigger centralization problem here than the centralization in the buffoons on the OpenAI board, right?
24:59 I mean, what you have, and in this context, all this kerfuffle this weekend is almost irrelevant, right? I mean, because in the end, I mean, Microsoft was the big daddy. I mean, Microsoft was the big daddy in the end. Daddy came in and said, you guys can have your little squabbles, but I will pay you and I will hire you. It’s not like, I mean, so the thing is, OpenAI needed money to train their models. Now, Elon could have put in 40 billion dollars to OpenAI, like he did to Twitter, when it would have been a better investment just to put the 40 billion into OpenAI and let them train all their models while remaining a nonprofit, right? But in fact, no one was putting that much money into OpenAI to train models and they didn’t have a decentralized ecosystem for model training, like we’re building singularity in an ecosystem, right? So they had to go somewhere for all that money to train their models. They went to Microsoft and once they went to Microsoft and sold a big chunk of their company to Microsoft, I mean, from that point on, things are pretty well determined and Microsoft knows that. I mean, once Microsoft owns a big chunk of you, it’s not like you’re going to go IPO without Microsoft’s permission, right? I mean, it’s not like you’re going to sell the rest to Google without Microsoft’s permission. I mean, you’re pretty much, you’re in the Microsoft orbit then for better or for worse.
26:35 So I mean, from that standpoint, I think, you know, all this mess kind of doesn’t matter. Like, it doesn’t matter that much when OpenAI is its own company substantially owned by Microsoft and working closely with Microsoft or where the staff move into Microsoft and they’re working with a small number of people still in OpenAI. I mean, it’s just, it’s sort of, it’s different permutations of the same thing. So I can, in that sense, the big event happened when OpenAI did their deal with Microsoft and all this is just shuffling around. And it’s interesting to think about Satya Nadella’s view and you can see a great deal of wisdom in this individual. Because I mean, he, when he did the original deal with OpenAI, he obviously knew their whole corporate structure was fucked, right? I mean, he, they must have done plenty of due diligence. It’s a lot of money.
27:34 They must have seen the backgrounds of the people on the board also. And they must have figured, well, one way or the other, like we’re supplying the money, we’re going to have the code. Like, I mean, one way or the other, all the mess, all the mess will sort out. So I mean, it took a certain open-mindedness and faith in the, you know, mysterious dynamics of the universe settling into a sensible attractor. It took a certain amount of intuition and faith to do that deal, given the obvious fuckery at the heart of how OpenAI was organized. And I think that Nadella did that, I mean, without getting too racist about it or something. I mean, I think his Indian background helps a lot. You’ve got the Indian mafia has taken over Silicon Valley, right? And how does that, how does that, how does that play? Because the whole Asian philosophy point of view is like things will flow in the direction that the water’s flowing, right?
28:35 I mean, the whole Asian philosophy view is like, the details will work out if they’re going in the direction of the overall. So don’t sweat, don’t sweat, don’t sweat the small stuff. If they’re doing the funny, if they have the code, the small stuff. So I mean, if you look at, look at Taoism, I mean, that’s like, you know, there’s some leaves and mud in the way of where the water is flowing. But the water is going to push them aside eventually.
29:07 Brian Rose: Can we get back to centralization?
29:09 Ben Goertzel: I think that you can see that very clearly, like computers are coming from Microsoft. You know, the money is coming from Microsoft. Microsoft has way more and better AI teams than OpenAI. Anyway, OpenAI did have some energy and some short term breakthroughs that Microsoft didn’t have. But like, I mean, you could see where the real power is. And of course, you can see like, once this stuff is launched onto the world, I mean, people want it, right? I mean, like what the decelerationists want to do is just go against the flow of events. It’s going against the flow of what everyone wants. So it seems like doing the initial investment into OpenAI was really that required having the wisdom to see that this screwed up micro level details of OpenAI governance.
30:10 One way or another.
30:11 Brian Rose: Okay, let’s talk about the centralization issue, Ben, because that we had to talk about. So let’s talk about that. Because if they create the AGI, now you’ve got the central structure, which is say it’s Microsoft and Satya, or say it’s some weird board or say it’s a committee. That’s right. That’s dangerous versus this decentralized concept that you’ve been working on with SingularityNet and Zarka, etc. Talk about try to explain that to the layman, because I don’t think most people really get that Ben. And this weekend is a great way to highlight it and say, hey, folks, look what happened over here. However, you want to slice it, you got a group of less than 10 people that are controlling the one piece of technology that can change the world.
30:54 Ben Goertzel: So the board of less than 10 people just has their own peculiarities, because each of us does, right? And it’s just a small number of people. And then you could see like, okay, Altman and Sutskever had some rivalry over who was having more power in the organization. A couple folks on the board came from the effective altruists and sort of long termist movement. So they had their own thing going on. So it’s just a small group of people, which has their own friendships and rivalries and very particular philosophies. And it’s not a good representative of the good of the whole human species, because these people are like all people, they’re pushing their own agendas, which are partly interpersonal within that group and partly the philosophies that they brought in. But on the other hand, if you have Microsoft in charge, I mean, Microsoft fundamentally is a for profit company whose mandate is to maximize shareholder value. And it also has close ties with US military and intelligence for a long time. I mean, it’s a US company, it wants to be nice to the government of the country that hosts it. Otherwise, many things could go awry, right? So the thing is, even if the individuals in Microsoft are ethical, good hearted people, and I suspect Satya Nadella is, I don’t know him, but it seems to be the case.
32:30 You know, my first cousin just retired from being at Microsoft since 1985. He’s a corporate VP at Microsoft. I mean, loads of executives in Microsoft are great, loving, wonderful people. But the thing is, you’re in a structure which has one goal, which is make more money for your shareholders. Now that doesn’t mean you won’t pay attention to ethics. It doesn’t mean Microsoft is going to launch killer robots and go into our houses and steal our babies or something, right? But I mean, it means that the whole context of judgment is we will do what will make Microsoft money unless somehow we convince ourselves there’s a really, really, really strong reason not to, right? So I mean, there’s a super strong bias toward making Microsoft money, a secondary bias toward pleasing US government agencies that Microsoft is working with. It’s not like they don’t consider the good of humanity as a whole. But of course, in countries where very few Microsoft products are purchased, I mean, these countries just have less influence in Microsoft. Like the people that work in Microsoft think about them less. They talk to people from those places less because they’re just not part of their lives, right? So the thing is, you don’t need to deposit evil people who want to destroy humanity. It’s just there’s an organization that’s biased toward narrow interests rather than toward the good of humanity. It’s just like, you know, when I was living in DC a couple of decades ago, I was working with US Army
34:03 Intelligence as a contractor. And I mean, they were doing what I viewed as terrible things in the Middle East with Abu Ghraib Prison and all that. But all the people I knew involved in that believed they were doing the best thing for humanity quite sincerely. I mean, they’re just like, these are the bad guys. They would chop us up to pieces if they could. And if we don’t stop them, they’re going to, you know, blow up more buildings in the US, right? I mean, they were good hearted. They wanted to do what was best for the humans. They would have cried seeing humans being tortured just like you and me, but they had a certain mindset. And people in Microsoft have a better mindset than those people, right? But still, it’s a very, it’s a very narrow one. And what clearly, you know, the more big time the decisions get, the more trouble people will have, will have coping with them, right? Which is what we saw in open AI now. Like, they’re not that close to AGI, but they’re getting results that make them feel like they’re getting closer and closer, right? So then push comes to shove. And the same would happen within a big company like Microsoft. I mean, if they had a breakthrough really toward AGI, I mean, you can bet every executive will have their view. You will bet major shareholders and board members and so forth have their view. There’s a bunch of different perspectives here, but all this is not the process that we want to have, right? Like, if there’s a breakthrough
35:35 Toward real thinking machines, you can think about people and ultimately even better. I mean, if you have to choose between a small elite committee or a big company, which in the end is ruled by its own small elite committee anyway, right? I mean, if you have to choose between these and the vast teeming democratic mass of humanity, I’m going to choose the vast teeming mass of humanity.
36:05 Brian Rose: And that’s what you’ve been working on with SingularityNet and Dark Gun.
36:11 Ben Goertzel: I started trying to work on this in 2001 or 2000 in terms of building a decentralized, secure democratic underpinning for AGI. And it was just really hard back then, because the software tools available were poor. So when Ethereum came out with the smart contract facts and whatnot, that was still too slow and awkward, but it seemed like you could make a start at it, right? So we founded SingularityNet in 2017. Now we have SingularityNet, HyperCycle, NuNet, a whole bunch of infrastructure projects aimed at making a decentralized sort of compute plumbing layer and data plumbing layer for AI so that you could take a breakthrough toward AGI and just run it on a million computers all over the place in many different server farms and different home computers in many different countries, right? And I mean, I think that’s technically harder than doing it on a centralized server farm like Google or Microsoft or OpenAI doing, but it’s perfectly technically possible now in a way that it wasn’t a couple decades ago when I started working in this direction. But yeah, as far as my own work goes, as we’ve discussed before, I think a better path to AGI than just upscaling LLMs will be combining LLMs with other AI methods, like logical reasoning, evolutionary learning, and so forth. So I have a software system called OpenCog Hyperon that allows us to glue these different AI components to get better in the systems pursuing goals and doing practical things. And we’re doing this in SingularityNet
37:55 Foundation with our open decentralized development community. We’re also working with some startup companies like Zarka on the language model side and Drew AGI on the enterprise AGI with OpenCog Hyperon, so I named it commercializing this. So it’s not like I think it shouldn’t be commercial, but I think the commercial stuff should be developed upon an actually open and decentralized infrastructure layer, which is how OpenAI sort of started. They were going to do things in an open way and then do commercial applications on top of that. But then the commercial applications ate the open part, right? And we’ve set things up quite differently in SingularityNet. Like we have an open source AI platform, SingularityNet, open source hypercycle, open source NuNet. We have an open source would-be AGI system, which is OpenCog Hyperon, interacting with open source LLMs like the kind that Zarka will be building.
38:52 Brian Rose: Okay, let me, can I ask you about AGI, Ben, because you’ve recently said that you think we can achieve AGI between three and eight years.
39:01 And like, I’m going to call you on that, because as we used to say in the city, you can drive a truck through that market. So can you give me something a little closer than three to eight years? And what are the implications of that AGI? You’ve just implied today that whoever gets it first, it’s like having the Holy Grail. It’s like having the one thing that can- There’s no research to be done.
39:23 Ben Goertzel: Okay. I mean, it’s not an engineering problem only. There are engineering problems involved, but there are research problems to be solved, no matter whether you take-
39:32 Brian Rose: So you’re not going to give me a more accurate number?
39:35 Is it still three to eight?
39:37 Ben Goertzel: Yeah, we can’t. I mean, I could make up a number, but it doesn’t make sense. I mean, what I would say is if anyone’s research program goes extremely well without delays or hiccups, that’s where you could get through your years, right? Like, I mean, if what OpenAI is doing goes really, really smoothly, I mean, each bigger model just as better and better than the previous, or if our launch of Hyperon goes really well in all our efforts to scale it up, work as we’re planning and so forth. If everything goes smoothly, three years is pretty much the soonest I could plausibly see human level AGI happening, just given how long it takes to do anything in the world. But I mean, sometimes that happens, right? I mean, some projects just go smoothly and no weird hiccups occur.
40:37 Brian Rose: Ben, besides you guys, who has the best chance centralized Silicon Valley? Who? Is it-
40:43 Ben Goertzel: It’s not OpenAI by far? It’s Google? No, it’s DeepMind.
40:47 Brian Rose: Really? Why?
40:48 Ben Goertzel: Yeah, by far. So the LLMs, like the narrow AI, OpenAI does have the advantage, the chat GPT567, they’ll continue to do amazing things. I mean, they have a temporary advantage.
41:03 Okay. I don’t think they have an insane advantage. I mean, look, anthropic, Claude is basically as good as GPT4. Okay. And Google is a big investor into anthropic now, right? So, I mean, there’s-
41:19 Brian Rose: Everybody’s put money in the problem.
41:22 Ben Goertzel: It is interesting that OpenAI has maintained their advantage, as long as they have, because I mean, everyone can train an LLM now. Everyone with enough money to pay for the training. But yet, GPT4 is probably smarter than all the other- I mean, it’s smarter than Bar, it’s smarter than Llama. It’s smarter than all the other LLMs out there, except for maybe Claude, so it does seem that OpenAI has some funky tricks beyond the basic architecture. But I mean, this sort of thing doesn’t give you an advantage for five years or something.
42:04 I mean, people will figure out those tricks. I mean, employees will be poached. Google’s investing in anthropic, which will lead to some back and forth of knowledge. I mean, so yeah, I think the lead OpenAI has is real, but it’s not going to be there like two years from now or something.
42:26 Brian Rose: Okay. And Elon’s Grok, does Elon Grok have any chance? I mean, he rolled that out in four months, and apparently it’s kind of GPT 3.5 level.
42:36 Ben Goertzel: Yeah, but I mean, Llama is too, right? So, I mean, anyone can download an open source model and make something GPT 3.5 level and literally with no code, you can do that, right? So, I mean, not to say that Elon couldn’t do it, just to say that what Grok is, you can do by taking Llama, which is open source from Facebook, add some prompts and connect it to and tune it based on Twitter data, right? So, I mean, that’s like an undergraduate course project, basically. Okay. I mean, that’s not to say that they can’t succeed, because I mean, Elon has billions of dollars and he can attract smart people. It’s just that Grok, as it is now, is not really an interesting piece of evidence. I mean, it may be cool. I haven’t used it, to be honest. I mean, it may be cool in that it’s up to the minute on everything from Twitter, right? So, I mean, having an LLM trained on everything in Twitter, if I were a journalist, it would interest me because I could ask questions. I’m like, what is the Twitter verse saying about this? And then follow the links, right? So, I mean, it seems interesting, but it doesn’t seem like leaps toward AGI. The thing is DeepMind has a lot of other interesting AI components besides LLMs, and they have a more diverse team than OpenAI, with experienced many different AI paradigms.
44:09 So, it seems like they’re well positioned to integrate LLMs together with a number of other AI modules that they’ve created, which is, I mean, this is how AlphaGo and AlphaZero work, right? Like you have a simulation engine, you have a game tree, and you have a deep neural net, and they connect this components together and get something cool to happen. So, I mean, DeepMind is good at that. And they connect some cool things to LLMs to get interesting things to happen. I mean, I feel like they have more breadth of thinking about how to make AGI than I’ve seen come out of OpenAI, who are more like, let’s just bang on making bigger and bigger LLMs, which I think will not be a workable approach to AGI. The thing is, DeepMind has their own bag of AI tricks, and my team has our own different bag of AI tricks. And then, which one is more likely to succeed becomes a technical reason?
45:16 Brian Rose: How would you honestly rate the top five out there, Ben? Like, how would you rate top five, including yourself?
45:25 Ben Goertzel: I haven’t tried to do that.
45:28 Brian Rose: All right, I’ll do it then. So, it’s DeepMind, OpenAI, Singularity Net ecosystem, and then Anthropic, and then I don’t know.
45:38 Ben Goertzel: Oh, no. I mean, if you’re talking about AGI,
45:42 I think, yeah, clearly, Google, Microsoft, OpenAI, and Facebook are super serious contenders. Yeah, I’m the cue at Facebook. I mean, he’s pissed me off by dissing Sophia and me so much, but I mean, still, and by publishing papers on AGI and cognitive architecture that repeat what I’ve been saying for 30 years without citing me or anyone else. But I mean, in the end, in the end, peculiarities aside, like, Jan Lekun is a super serious AGI researcher. I think Yoshua Bengio in Montreal has managed to put together a huge collection of deep neural net researchers doing open source stuff with mostly Canadian government funding. And that team is a serious contender also, which is quite interesting. I mean, they say out of the news, but they’re doing amazing, amazing research.
46:50 Brian Rose: Ben, go ahead. The thing that we’ve discussed before, though, is all these teams in big tech companies are approaching variations of deep neural networks.
47:04 And so is Yoshua Bengio.
47:06 Ben Goertzel: So DeepMind is a little more heterogeneous. They have some stuff that’s not deep neural networks going on, but deep neural nets get the lion’s share of focus there. The approach we’re taking in OpenCog and using behind true AGI, Zarka, Singular, and so on, this approach, for better or worse, is a little bit different. Right. So I mean, the approach that we’re taking does use deep neural nets, but it uses a logic engine, it uses the dynamics of self-modifying knowledge graph, it uses evolutionary program learning. So I would say, if deep neural nets on their own are not the best way to get to AGI, then we’re by far the top in the Singularity Net ecosystem. And if we get there first, because we’re making the right educated guess about which AI algorithms to use, right? Like if we get there first, we’re going to roll out our thing on a decentralized network, right? And so then that becomes a quite different branching path on the road to the…
48:23 Brian Rose: One that has more safety than anything else, you believe?
48:27 Ben Goertzel: Well, it has different risks. I mean, my intuition is it has more safety. I mean, of course, anybody can tinker with it. When you open source something, and make it decentralized, there’s the risk. Some crazy hacker downloads your open source thing and does something nasty with it. There’s also the risk that the vast mass of humanity makes a worse decision than the board of Microsoft would make, right? Right. I mean, the vast mass of humanity is not that
49:04 Brilliant. I feel like when the average person first saw chat GPT, they thought about it in a very stupid way. But most people I know, even like my mom’s friends were in their 80s or fifth grade students or something, after they had an opportunity to play with chat GPT, they understood exactly what it was from a practical standpoint. They understood like it’s not a super thinking machine. It makes weird shit up, but it can be a very useful tool and do a lot of jobs. I mean, I felt like when you put the thing in front of ordinary people, they got a sense for what it was reasonably well. But still, I mean, democracy is not a panacea, right? I mean, we see that all around.
49:53 So I don’t think there’s a guarantee. I just think the odds of a beneficial outcome are better than if control is put in the hands of small elite. And I don’t think that slowing down is especially an option. I mean, I think a big company or the Chinese government might claim they’re slowing down. But I mean, that’s just like when I was working with US military intelligence, when the Congress said to shut a project down, they just took it black and kept doing it.
50:29 I mean, it’s so I don’t I don’t I think there’s too much clear economic advantage, and too fuzzy and dividing line between AGI and advanced applied AI. There’s too much of this going on, I think, to really make deceleration work on a global scale.
50:52 Brian Rose: Ben, what worries you? What worries you right now, Ben, if you can be honest? You got kids, grandkids?
51:01 Ben Goertzel: I’m not worried about super on a personal level. I’m not worried about super AGI turning into the Terminator and killing everyone or something. I mean, of course, you can’t rule that out. But I have a fairly strong intuitive feeling that once we get to a super AGI, it’s going to respect us and love us as it’s weird old creators. And there’s a lot of resources in the universe. It doesn’t need to take away all our toys. It can build a lot of new toys for itself. I’m more worried about if a big tech company allied with a big government makes the first breakthrough AGI, then they’re going to try to put in fascist controls to stop anyone else from making the next breakthrough AGI. And they’re good, they’re going to use this AGI to advance their own narrow interests. But then in opposing country, well, also, so I mean, you you basically can have all sorts of global chaos and World War Three scenarios, which come from the first AGI is being owned by narrow minded entities who don’t want to protect their their advance. And then I mean, then maybe you get to Colossus, the foreman project, like the little science fiction movie that was like us and Russia each build AGI supercomputers to fight each other, then the supercomputers just make best friends and then and and down the war. Right. So I mean, you could you could have a lot of crazy outcomes. But anyway, I’m more worried about what happens if entrenched, narrow, powerful interests get the first AGI and
52:49 Then say, well, no, we see how dangerous this is. So we must put into place draconian regulations to make sure no one can do anything dangerous with it, meaning so that only we can make the next advances and we control it on the singularity. What if these people are too narrow minded and stupid to actually direct the singularity? Right. Yeah, that that that worries me more. I do.
53:12 Ben Goertzel: I do have to go now to a research meeting. I had a few minutes ago.
53:16 Brian Rose: No worries. Dr. Ben Goertzel, thanks so much for your input and your timing. And again, if people want to learn more about Zarka, then go to London Real TV forward slash Zarka. That’s your large language model, but you’re using neuro symbolic and solving for AGI, which is your specialty. So Ben, thank you. I appreciate you keeping us posted. I appreciate your optimism always.
53:35 And we’ll catch you next time on London Real.
53:38 Ben Goertzel: Thanks, Brian. All right. Thanks, everybody. We’ll see you next time. Bye.
53:42 Brian Rose: So Jim Rickards has just recorded a video that’s not available to anyone in the public. And he’s going to be talking about how this upcoming recession is going to be fast. It’s going to be bloody. It’s going to be nasty. But at the same time, he’s going to show you how you can position yourself to profit from all of this chaos.
54:00 Now we’ve made this video only available to our viewers. Go to London Real TV forward slash Jim. Watch that immediately. I can’t say enough good things about Jim Rickards. He understands the global economic system better than any guest I’ve ever had on London Real. His predictions are almost untannily true and you can learn how to profit from his vision, from his expertise, and his understanding of economics. So go to London Real.tv forward slash Jim or click the link below. It’s an excellent, excellent look on what’s going to happen in the future and how you can position yourself to profit from that. Jim is one of the best in the business, one of my favorite guests on London Real. And he’s very, very good at predicting the future and showing us all to profit from it. So click the link and I hope you enjoy.
