8 Ways To A.I. Doom! With Jim Rutt

Jim Rutt [00:00:00] It'd be great if as people age, and it's real, frankly, not all that safe, let 'em out loose on the road. I think that would be one of the, the first really legitimate uses of the absolutely full self-driving cars. Cool. So I'm absolutely in favor of it.

Vance Crowe [00:00:14] In your argument there, Jim, you had some aughts that I don't know that I, I would just naturally agree with the, the first one being that like what we ought to do with the money that we're saving, right?

Jim Rutt [00:00:24] Ooh. I mean, the worst case is really, really bad. The

Vance Crowe [00:00:27] Next step

Jim Rutt [00:00:27] Is called artificial super intelligence, which is something that's, I'm Chris Haworth, a grain originator and accountant living in Pocahontas, Iowa, and you are listening to the Vance Crow podcast. Welcome

Vance Crowe [00:00:43] Back to the podcast. I'm glad you're here today, Jim Rutt returns. We're gonna sit down and talk about artificial intelligence. We'll start off just talking about the very basics, and then we're gonna move into a more advanced conversation about what are the risks. If you are a longtime listener to the podcast, you know that Jim has an extensive history in tech in large corporations, and if you listen to his podcast, you even know that one time he presented an idea to the Department of Defense about how you could take down the National Grid, and they were like, whoa, whoa, whoa, whoa, whoa, whoa, whoa. Yeah, yeah. Don't tell anybody else about that. So, Jim has an interesting mind, he has amazing guests, and it is a true pleasure to have him on. We're gonna get to that interview in just a moment, but first I want to talk a little bit about Legacy Interviews. Just this week we were called up by a family in New York City that said, we want you to come to our apartment here and sit down and record my father telling his life stories and some of his friends and family members so that we can record everybody's memories of him. And so that's what we're doing. One of the things that we've worked on this year is traveling occasionally to do Legacy Interviews. So if you're interested in having us sit down with your loved ones, maybe to do it as an entire family experience or extended family experience, go ahead and check out the Legacy Interviews website and then contact us to figure out how to plan your unique event, even if that means we need to come to you. So go to Legacy Interviews dot com to find out more.

Vance Crowe [00:02:15] All right, without further ado, let's head to my good friend, Jim Rut. Jim Rutt, welcome to the podcast.

Jim Rutt [00:02:23] It's great to be back.

Vance Crowe [00:02:25] Well, my friend, much has changed since we sat down and talked. You know, we first started getting in touch during COVID and we've had a lot of conversation about the alternative groups that we've made and kind of, how do you think about a world that advances in technology? But people have to have some kind of basis, some kind of basic, how do we relate to one another? And I'm really excited to talk with you about artificial intelligence, to try and get an idea of what it is and, and really like what are the threats here. So to begin, when people are talking about ai, how should they understand this? Computers that think like human beings, not fully,

Jim Rutt [00:03:04] Not, certainly not the full stack. I think the way you start is think about computers that can do things that were previously thought to be human things to do. Right? You know, like for instance, spell check is even kind of a low end ai, right? You know, used to be you go look it up in the dictionary, now you just have your little thingy in that's in your Chrome browser, do it for you. So that's very low end ai, but I'd say it's AI nonetheless. So that's the easiest way to do it. If, if it's something a human used to do and now a computer does it and it has some amount of smarts in it. It's not just, you know, typing letters into things, you, you can call it ai and from very, very weak up to eventually very, very strong.

Vance Crowe [00:03:52] What do you mean by that? Weak versus strong?

Jim Rutt [00:03:55] Ah, yeah. This is a key distinction and a, a lot of the conversations about AI kind of get this all muddled together. You know, weak AI is something like a spell checker. Then there's narrow ai, which is something that is reasonably impressive, but operates in a thin domain. A classic example in the history of AI were gameplay programs. You know, the, there was a Checkers program that was damn good by the mid sixties, and then famously in the nineties, deep blue IBM's chess computer beat Caspar, was it Caspar? I think it was, or was the other dude that 1 1, 1

Vance Crowe [00:04:34] Of the top top Ky, I don't remember. Yeah, it was one of 'em, yeah,

Jim Rutt [00:04:36] One of them, one of them good rooky players at the time was the world's champ and beat him in a tournament. And a lot of people were Crest fall. And then it's gone on to, you know, other programs. You now get a program on your phone that will beat the world's champion pretty much. And then Go was another game that was thought, oh, that'll be at least 30 years for the narrow ais can solve that. Well, that, that was solved, you know, several years ago. And so those are narrow meaning that it's a thing that humans can do at the very highest level of human capacity. Andis can do it and then can now do it better than humans, but they can't do anything else. Like you could take deep blue, you know, beat the world's champion in chess, but tell it to play back gam and it wouldn't even know where to start, right? So

Vance Crowe [00:05:23] That's, and so it, and the way that that narrow AI is working, at least in chess, is you make a move and it says, all right, out of all the possibility space of all the moves that I can make, the one that's the highest probability to lead me to success is X Or is it more complicated than this?

Jim Rutt [00:05:39] Yeah, it's a lot. We, that, that is it at the million foot level, but lots going on underneath that 'cause. 'cause even chess, it can't look all the way ahead. There's just too many possible branches. So it has to use estimators, positional estimators to say, all right, if, if I do this, you know, four branches ahead, how good is that by some estimating algorithm, which it has to train over time, et cetera. And that. So, because otherwise you could never do it. And that, and that turned out to be really important in Go 'cause Go has far more branches than chess. You know, go is to chess as chess is to checkers approximately, right? And, and so it has to use a lot more of this estimating what branches look good and it doesn't really know, quote unquote, but, but yeah, that it has to add a lot of that into the thing. And interestingly, the, the thing that Beat the Champion in Go was what's called self-trained. They just gave it the rules of Go and not much more than that and told it to play itself like a hundred million times. And just from that it figured it out. They did not put any, you know, dictionaries of opening moves or anything like that. Quite remarkable. They then used that same technology to try to learn chess from the beginning. And within a few hours it was better than most players. And within a couple of days it was better than World champions. And then a couple of days after that, it was right up there with the very best chess software.

Jim Rutt [00:07:12] Again, no databases, no history of games. Just play yourself and figure it out. Quite remarkable. But again, very narrow. Because even Go, which is played on 19 by 19 square board, if you were to change the board size, which people do as an experiment, change it to 11 by 11, couldn't play for shit, right? You have to, you, it, you would've to retrain it basically on that. 'cause it's because it is, you know, just statist really, really, really deep knowledge of the pattern, essentially. So that's narrow ai. And then you have broader ais like, and this is where we are today, things like large language models that can do a lot of things in the domain of language. But again, they can't drive a car. You know, they can't pass the Wozniak test for general ai. Oh, we'll, we'll I'll tell, we'll maybe we'll pause a second and then we'll come back. Talk about the Wozniak test, which I personally love a lot for how you get to a general ai. So we have broader ais, we're now starting to widen out chat GPT. So the one that your listeners may be familiar with, if you're not, check it out, just type in chat GPT into Google, go wherever the heck it tells you to go and sign up. It's free. And it's an amazing tool. It know, it knows almost everything. And it can answer questions. You can iterate multiple times on the question. You can, and it probably knows who Vance Crowe is and can tell you what color underwear he wears.

Jim Rutt [00:08:46] No, probably not that one. But, but it's amazing what it knows and, and, and how good it is. On the other hand, if you follow the news famously, it just makes stuff up. Also, if it's at the fringe, it will, the a fringe of what it knows, it'll, if it doesn't know it all, it'll say, I don't know. But if it's at the fringe where it thinks it might know, it'll just make something up. It's quite funny, you know, like I am a

Vance Crowe [00:09:10] Well, and it's incapable of distinguishing whether it's making something up right now, right? Because it, it's just using, it's just finding out what do most people think the next word should go here as like, and so if it's giving you an answer, it's because it thinks, oh this is, this is what the path has led me to. It's not like you could ask it, do you know you're lying to me?

Jim Rutt [00:09:30] Yeah, yeah. Well though this dust turned out to be interesting. You can ask it its confidence level and it will give you something and it, and that signal is actually worthwhile. It's kind of curious. And that's a good point though, that what it's actually doing is just predicting the next word in the sequence. And the fact that it does so amazingly well still nobody knows quite how or why. It's really kind of cool.

Vance Crowe [00:09:52] And so then strong ai, as you think about chat, GPT, like I used it, I, I use grok now on, on Twitters all the time. But what's next? Because that definitely has limitations. This is not something I'm scared of at all. At first you're like, whoa, this can write news stories, this can do all kinds of good stuff. But you start seeing the flaws, you start trying to use it for things and you realize like, this thing can fall off the tracks real fast. So what, what's the next evolution of this? What's chat GPT five gonna be like? Yeah,

Jim Rutt [00:10:24] Before we go there, let me tell you one other thing though. Even with chat GPT four, if you wrap software around it, you can do a lot more things with it. For instance, over the last 12 months, I sort of accidentally started a company to use chat GPT behind the scenes from a program to write movie screenplays. And it's got levels upon levels of, of processing in it where it'll send stuff to GPT and then it will bring it back, it'll structure it and reorder it and then it'll send some more stuff and then makes the human interact with it. And the end result surprisingly is you can knock out something that's about as good as a first draft, semi-professional mid, you know, mid-grade semi-pro screenwriter in about 20 hours as opposed to something like 500 hours that it would take an actual human to do that. And we figured out you sort of how to tame all that wildness and focus it on a domain in a pretty impressive fashion. But you know, on the other hand, you can't get it to drive a car or the Wozniak test. So here's where we'll talk about the next step, which is probably beyond shaft GBT five. And this is what's called general intelligence. Artificial general intelligence often abbreviated a G. And that is computer software and hardware that can do anything a human can do. Or most things that a human can do about as well, at least as well as a human. And we're not even close to that at this point, though.

Jim Rutt [00:11:58] There are some experts that think we're within five years of it. I have my doubts about that. But I mean, there are some very reputable people I know that will say could be in five years. And so this is where I now talk about the Wozniak desk, you know, the, the, the co-founder of Apple, the not so famous one, Steve Wozniak years ago came up with this test for a GI, which is to plunk a robot powered by an artificial general intelligence down in any random American kitchen and tell it to make coffee. There's a whole bunch of things it has to figure out from that that are kind of human kind of things. Now, where would they might put the coffee, you know, well, you know, what about the coffee maker? There's different kinds of coffee makers. And so sort of going around in a rando kitchen and figuring out all the little things it needs to do. Where's the water? You know, where's the electric plug? And all that is, you know, an example of the kind of thing that an a GI could do. But today's software isn't even, isn't even close to doing now to give you a sort of a halfway point, and it's one, it's of course turned not to be way harder than people thought a few years ago is a true self-driving car. You know, I was following this since about 2016 and by 2019 the big car companies were at least whispering that they're gonna have full self-driving automation by 2021, where you could literally drive anywhere with the, the artificial intelligence driving for you. Well, it turns out it's a lot harder than they thought. We do have some self-driving cars in San Francisco and in Phoenix area that can drive themselves without any driver, but only in a narrowly circumscribed geographic area where they have super seriously mapped every corner, every everything.

Jim Rutt [00:13:45] And the idea of the full general driver, artificial driving machine is still, we don't know. We don't know.

Vance Crowe [00:13:55] I don't know, man. I've been in, in Tesla up in the country and had it drive from a farmhouse all the way to a restaurant about 20 miles away. And it passed around a tractor on its own. It passed by other cars. It made, you know, left hand turns and, you know, it was, it was pretty, pretty amazing.

Jim Rutt [00:14:14] It, it's getting there. And I had George Hotz on show recently. He has a company, believe it or not, that has an open source self-driving software where you buy a thousand dollars box from him, put it behind your rear view mirror, hook it into the, all the electronics of your car, which amazingly easy it turns out. And it'll do, it's a little worse than Tesla, but not much. And you can do it with 250 different models cars, which is quite, quite remarkable. Almost any Japanese or American car made after 2020 will work with his open source self driver. And so

Vance Crowe [00:14:53] This is a good place for us to get your temperature on things. When you think about self-driving cars, is this something that you're jumping up and down about or are you like, you know, you'll have to pry my four, my 20, you know, 2000 Ford F-150 pickup truck from my cold dead hands. 'cause I don't want any electronics in it.

Jim Rutt [00:15:12] Well, it's funny, a little bit of both. It's always good to keep a non electronics vehicle around in case of mps or solar flares or something. On the other hand, I've long said that I'd love to have a car where I could read a book or sleep in the back seat while it drives me around. Or heck go out to dinner instead of having three beers have 12. Right?

Vance Crowe [00:15:33] Well, I think there'll be the, the commute will include exercise. I think they'll have little cars that people have recombinant bikes or ellipticals or all kinds of things. You have rowers, you could sit in there and and and get your workout in while you're waiting in traffic. Although there'd be a hell of a lot less traffic if everybody was doing AI cars, right?

Jim Rutt [00:15:51] Well, could be. That's, yeah, we can talk about them as a professionally lazy person. I will not be doing exercise. I can tell you that right now. You know, probably reading a book or, or sleeping or something like that. But yeah, some people will, and then that wouldn't that wouldn't that be cool? You could have your rowing machine hooked up to a generator and supply some of the electricity. So,

Vance Crowe [00:16:10] But you generally speaking are like, yeah, it'd be nice to have these things. I'm, I'm open to it in culture.

Jim Rutt [00:16:16] Oh, absolutely. I mean, I'll be all in favor of it and you know, I am not a young person. I just turned 70 last year and so over time, you know, I had to take the keys away from first my wife's mother, then my father, then my mother, right. And you know, those were fairly traumatic confrontations as you could possibly, probably imagine. And it would be great if as people age, and it's real frankly, not all that safe, let 'em out loose on the road and that will come to most people before they die. If they could just get in their car and tell their car where to take 'em, that would be great. I think that would be one of the, the first really legitimate uses of the absolutely full self-driving cart. Cool. So yeah, absolutely in favor of it. Don't have any romantic attachment to driving though. I do enjoy, enjoy driving and you know, my main driver is a 2017 Jeep Grand Cherokee with a hunk and heavy V eight in it. So I do enjoy driving, but, but I, I will buy a true no hands, no eyes sleep in the back seat, self-driving car once I'm convinced that they are indeed safer than I would be likely to be.

Vance Crowe [00:17:31] Well you mentioned about how far away the artificial intelligence is. I heard Ray Kwe talking about his prediction has been and still is 2029 to hit that singularity point where, you know, the, the speed with which the AI is learning and is fully autonomous is really just, I mean, like you had said five years away, you said you have your doubts about this.

Jim Rutt [00:17:53] Yeah, I think, I don't know. I mean, again, I will say that I have a actually growing amount of agnosticism about how long this is gonna take. If you run some brute force measures of the sort, Kurzweil does, you can get to 2029. On the other hand, when you look how long it's taken for them to get even up to just sort of, okay, self-driving cars like the Tesla autopilot, whatever, hell they call it, five more years, that's, that was five years, 2019 to 2024. They make considerable progress but nothing close to passing the Wozniak test. So if I, if someone made me put a bet down on a computer passed the Wozniak test in 2029, I'd bet against it. On the other hand, I know some mighty smart people that say yes, I would, I would say that my own over under would be at least 10 years and maybe more could be 20 in that range, 10 to 20.

Vance Crowe [00:18:57] Well, a robot is, is an interesting addition when you're thinking about AI because not only do you then have to have a computer that can think through what are the steps, you know, first I need to locate the coffee maker, you know, check all the countertops now check the lower ones, you know, trying to do this by deduction, but to then also add in balance and its ability to reach and its ability to move. To me, there's something separate about that. That's like a GI plus robotics.

Jim Rutt [00:19:25] Well I think the robotics are almost there to do that. If you had enough money, if you had, you know, quarter million dollars to do it, it wouldn't be something you could actually use. But we're getting closer on the robotics side and eve, okay, let's make the test pure rather than an actual robot. Let's have a human emulating a robot with the a GI, just reading out the instructions. Okay, Jim, step forward three steps, raise your right hand, extend it straight out. Right? I, I would even use, I would even accept that as a, as a surrogate for the Wozniak test. And I don't think it'll be able to pass it in 2029 probably, but I would not bet a lot of money on it. Right? I'd bet a month's rent but not a mortgage on it. Right? And or a house price audit. So I do think that as a matter of social policy and how we think about change, we have to keep in mind that this is a, a collection of trajectories, which we can't know which one it will actually be, which makes the problem somewhat harder. What happens if it's in three years OD year, you know, what happens if it's in 50 years? Well, that's a very different set of, of issues we have to deal with at different pacing. And so we have kind of what we call a meta problem of trying to figure out which, which one of these trajectories is the one where we should put, you know, much of our social attention and our, and our bets and, and things of that sort. So that's one of the things I would encourage your audience to do is when people give a hard date, be skeptical.

Jim Rutt [00:20:57] You know, you know Kurzweil smart guy, I know Ray k Kurzweil and, but I, you know, I I would not necessarily say he's right. And, and other people say three years, other people say 50 years. So realize that there's considerable uncertainty even amongst the experts.

Vance Crowe [00:21:17] So what happened that we went from AI being something in science fiction books to, you know, this chat GPT coming out and everybody being like, whoa, and then this huge wave, you know, then, then you went on to having mid journey and you have Dolly and you have now Gemini. What happened that all of these came, they, that the tech tree, you know, happened to everybody at the exact same time. What happened there?

Jim Rutt [00:21:43] Well actually it was, it's a tiny little piece of the tech tree. All those things use the same technology basically, which is called deep learning, which is a kind of neural net. Your brain is made with neurons, which are like little branching thingies that connect to each other and they fire when this one fires, the next one fires. And if these three fire and they're downstream to this one, this one will fire, et cetera. And there's a very simplified version of the kind of structure we use in our brains called artificial neural nets. And they've been around since the late fifties, believe it or not. And they've just been gradually getting a little better, a little better, a little better. But in the, you know, 2010 timeframe, 20 11, 20 12, 20 10, 20 11. In that timeframe, somebody figured out that you could run these things on the video cards that are used to play high-end games and further that, that there's one particular algorithm where all the computation is called gradient descent. And somebody figured out how to modify gradient descent just a little bit. So that would run reasonably efficiently, efficiently on these game cards. And without that change, we would not have seen any of the things that you just mentioned. It would take, it would take too much computation to be able to actually run it in the computer chips themselves. It would cost maybe a hundred times more than it's currently costing to do something. So nobody would've even tried it. It probably cost 60 to a hundred million dollars to build chat GPT-4 for just a computer time to build that sucker.

Jim Rutt [00:23:22] Multiply that by a hundred, you'd be at $6 billion. Nobody would've spent $6 billion on a one shot, you know, test to try out, couldn't could I build chat GPT pretty big bet to do it for 60 million but for 6 billion, I don't think anyone would've probably done it. So anyway, they, they, it's one technology neural nets, one version of neural nets called deep learning and then one technical trick, which is to get one algorithm gradient de set to run on graphics cards. Now of course, they're building big old things that are basically just huge computer vision graphics cards. Nvidia is the company, guess what? It's the same Nvidia that makes the video cards for all the high-end game machines. Their business now is mostly building big giant versions of these GPUs as they're called. And guess what GPU stands for Graphic Processing Unit, even though they're not being used for graphics. Mostly they're the same identical architecture as the video cards, which is quite, quite remarkable. And so that is the line of descent, which has led us to large language models. Well then one, one other thing, there was an algorithmic trick developed by Google with a program called Bert, BERT that uses transformers, which is a specific kind of deep neural net to process sequential data like language.

Jim Rutt [00:24:57] And so this, this is not all of ai and it's something that's starting to drive me nuts. There's a lot of journalists talk as if large language models and other forms of generative ai. I tend to use generative AI to include things like Dolly and Mid journey and Claw three, which is actually very good, by the way. Claude three is right up.

Vance Crowe [00:25:19] And for, for people that don't know, what you're describing when you're talking about Dolly and Midjourney is this is where you can go and you can type in a prompt that says, make a realistic photo of a teddy bear being held up by balloons. And it will generate these based on just your words. But the, there's an image that's created out of it. It's a, it's actually rather surreal. I i, we can go into my kind of complaints about this in a little bit, but just the fact that you can type in words dog teddy bear balloons and it can bring this back. But it's not that it's going and finding you a photograph of those things. It is that it is actually creating an image that looks like a photograph that's so astounding.

Jim Rutt [00:26:02] Yep. And again, if you don't wanna photograph it, do it cartoon style, it'll do that. Right.

Vance Crowe [00:26:06] And so those are still LLMs that, that, that kind of, basically

Jim Rutt [00:26:10] The same technology, they're transformer technologies underneath. It's all the same fam, closely related family trees. And then of course there's also now ones that will do videos. Runway is currently the one that's furthest along that's available to the public. But face OpenAI has shown soap that looks like it can probably do a minute or two of video from text. And you can do like a, a, you know, a screen. You could say, okay, I want a cottontail rabbit, you know, being chased by a coyote or something. And then have the rabbit go off the cliff and the coyote go off the cliff. And I guess it should be a roadrunner actually should not a cottontail rabbit. And it'll, you know, do something halfway reasonable. And once you get to like 90 seconds you can, you're pretty close to being able to stitch together a feature length film or at least a 45 minute TV show. 'cause those things are broken up into, you know, one to three minute chunks basically. And so the ability to create a TV show or a movie from text is getting, you know, it won't be this year or next year, but it might be two, three years out, which will be quite remarkable. And all those use the same family tree of deep neural nets, GPUs, transformers, et cetera. So this is one pass through the technology stack, which has become super optimized. 'cause of course once you found you could do cool things with it, gigantic amounts of money have flown in, has flow flowed in.

Jim Rutt [00:27:43] And more and more and more and more cool stuffs being built and algorithms are being improved, hardware is being improved. There was a huge announcement from Nvidia yesterday about its next generation architecture, which I call not just an architecture, but a meta architecture. 'cause the architecture will also help their suppliers get better at creating the chips and will help their users get better at building advanced data centers for deploying their GPUs. So, whoa, this, this thing is kind of a big announcement that came out yesterday from Nvidia. So yeah, a lot going on, but it's, it's one pass. And this is also something to keep in mind. We're gonna have some other surprises 'cause there are other trees of AI that are not deep learning, don't use GPUs, et cetera. And they can come up with some remarkable things too. And we just don't know when. And now it is, I would say a little bit unfortunate that so much of the money and talent has been pulled into generative ai, deep learning transformer architectures. But there's still fair bits going into, into other kinds of systems as well.

Vance Crowe [00:28:51] And so when you talk about this other tech stack, if we're not to understand large language models or this, what, what is the other type of ai? What else can it do or how does it work

Jim Rutt [00:29:02] Historically? One of the ways you could divide ai, AI up since 1956 when it started at Dartmouth University, a summer program where a bunch of super smart people got together and spent three months at at Dartmouth. Now it's pretty funny, these are some very famous smart people and they estimated they'd be at human level AI within six months. This was 1956. Oops. Anyway, but at least they got the ball rolling. So until the, say into the two thousands, the leading form of AI was called symbolic AI that kind of looked like computer programs, you know, statements about the world. And you'd have knowledge engineers that would help you build a set of statements that teach you how to make a pizza or something and use that to control a robot for throwing some dough in the air and catching it on your, on the robot's finger and twirling it and making some pizza dough. And there's still, and so symbolic AI is the, is one branch and a connectionist AI is the other branch Connectionist meaning things connected together kinda like the brain. And there's still work going on in symbolic and particularly there's really interesting work going at the intersection between connectionist and symbolic. In fact, a project I advise from time to time called Open Cog is definitely a player in that domain. They have a big network of computers all over the world running their software called Singularity net.

Jim Rutt [00:30:35] And that's a good example. Gary Marcus is another researcher who continues to do work there. And even, you know, Jeffrey Hinton, the guy who developed deep learning, he even, he says it's not, he doesn't believe it's enough by itself to get to artificial general intelligence and will have to get more symbolic representation. She looked like. To give an example, you mentioned how these things hallucinate. The reason they hallucinate is they don't actually know anything. They're just a bunch of patterns, you know, and they pattern match and they say, oh, oh, whatever, you know, the capital of France is Rome. Well, they don't actually know it. A symbolic system would actually have a representation in a symbolic form. Capital row of Paris, France equals Paris, right? So it would never make a mistake. And if you could combine the generality and ease of learning of the connectionist to architecture with the ability to be precise and to actually reason about what it knows better in the symbolic systems, you may well have syner synergy, what we call an open cog cognitive synergy, where you combine the strengths of the connectionist, the neural kind of stuff with the strengths and the precision and the speed of the symbolic it that may actually be the golden road to artificial general intelligence. Now, should I add the scary next step?

Vance Crowe [00:32:03] Well, I actually, before we get to the next step with people pouring money into this, you know, you said, Hey, before if you've dumped $6 billion into a single shot, it wouldn't be worth it. But now there's millions, billions of dollars flooding into ai. What are they hoping comes out of this? What, what do they want it for and what will it be capable of doing in the, in the utopian vision?

Jim Rutt [00:32:27] Well, of course they means lots of different things. And this is really important, especially for young people thinking about their careers. You know, I really, you know, I've intentionally not been involved in starting companies for 20 years, you know, more or less doing other kinds of stuff. But this stuff was so exciting when I fell into it in November, 2022, I said, oh, I gotta take a little test run here and see if I can build something. And it turned out we could, which was kind of cool. But what reminded me of was PCs in 1977, you know, people say, oh, it's the next internet. And I go, Nope, bigger than the internet. Oh, it's the next smartphone. And I go, Nope, bigger than the smartphone. Because if you take a look at a smartphone, it's just a tiny little pc. Really took a look at the internet, it's just a really big local area network. And we had local area networks for PCs by 1980. So I say that the root of com, you know, desktop computers, smart phones, internet, it's all with the PC in 1977. And we didn't know what, we had no idea what the PCs would really be good for. It wasn't for until a year or two later somebody came up with the idea of word processing. Right? In 1977, secretary people wrote their business letters on yellow pads, gave 'em to their secretaries who type 'em out, give them back to them and they'd correct it. My very first, you know, managerial job in 1980, basically enough, I still had a secretary who wanted me to write my crap out in on Yellow pads.

Jim Rutt [00:34:05] I stopped pretty quickly, started using programming editors to do my writing. But, but, but only soon thereafter did real professional grade word processors come up the spreadsheet, something we all live with today. I remember the first PC spreadsheet would've been about 19 8, 19 81. I bought it in a computer store. It was in 81. Yeah, it would've been 81. And I, it was a copy of Viscal, the first real spreadsheet in a Ziploc bag with a hand printed label on it. And I took it home and I was like amazed. I stayed up all night long, just build it. Whoa, whoa. You know, today you can't think of anything more boring and cliched than Excel. But in 1981 the spreadsheet was who came up with this idea? This is like so brilliant. And we're at that stage right now with these large language models. And the other generative AI is that people are just constantly coming up with new things you can do with them every day. Somebody comes up with something cool you can do with 'em, who would've thought writing movie screenplays would be a thing you could do with today's technology. But you can, and there, some of the areas where they're being relatively rapidly deployed is in customer service applications and medical diagnostics and repair helpers for things like auto repair, aircraft repair, industrial repair, et cetera, pro programming, computer programmers.

Jim Rutt [00:35:38] When I wrote this huge program, I never could have got it done without using the LLMs to actually write the code. I would say I was three times as productive as I would've been without it. And I probably would've thrown in the towel long before I got done with it if I didn't have my trustee helper. And of course I talk to any college professor, they'll tell you all the students are using these things for writing their essays. And so everyone is out here exploring what in the world is this thing really good for and what is it not good for? You know, it's not yet really a perfect replacement for search, but it is a pretty good kind of thinking. Companion. Give you an example, when my fir, the first day I fired up chat GPT, November, 2022, I said, ah, let me do a homework assignment that I had in 12th grade honors English. And I had it compare and contrast Melville's Billy Bud with Conrads, I don't remember hell Lord Jim. Yeah, that's what it was. And I, and the thing 45 seconds later kicked it out, said Mrs. Carr would've given this about a b plus, right? And honors English 12th grade and that was GPT-3 3.5. So that was like, whoa, this stuff's pretty good. So we are just like we did with the PCs, which at first people didn't know what the heck they would be good for. We're exploring what these things are good for.

Jim Rutt [00:37:11] We'll give you an example of something that, that Dali is amazing at. You need a logo for your business. Just sit down with Dali, which is now part of GPT-4 plus 20 bucks a month. All you can eat and just type in ideas for and ask it to create logos for your business. And it will produce a surprisingly good one and maybe five or six rounds of you going back and forth. 'cause each one you get, you can critique it. Oh, I don't like that. You know, I want, you know, I want a picture of somebody on a impaled, on a pitchfork or something as part of my logo, right? And oh, I want them to have a red beard, right? Then it'll do it again with a red beard. And so, you know, who would've thought it would be good for logos? But it's really good for logos. And my,

Vance Crowe [00:37:55] But I mean, on the flip side, my experience has been, you know, I, I thought, okay, I write a lot of speeches. Sometimes I need to have images in there. It'd be really nice if I could just have it create an image. So one time I went to it and said, gimme a man on a box because I wanted to show this guy that came into town that was, you know, standing on a box to tell a story and it would do a man on a giant box and people all around him look like ants. Then you'd put him on a different one. Now he's under the box, now he's on the side of the box. In some ways it almost feels cursed because it can never quite get it right.

Jim Rutt [00:38:29] Yeah, it is interesting. And you have to sort of know after a while you get a sense of what's reasonable to ask it to do and what's not. And I don't know, I can't really explain it, but I now have a fairly good sense. Now I also have a fairly good sense of when to use Dolly and when to use Midjourney. If you want something that really looks artistic, like artistic photography or something, or artistic painting, midjourney is your best bet. But if you want something that's more literal in terms of doing what you say than Dolly three is I find to be a better bet. But yeah, you're right. I mean, you have to go back and forth and say, no, no, not that. Try that again. Right? Just like if you've ever worked with a professional graphic artist to create something for, you know, a business presentation, a lot of back and forth. The first draft, very seldom much you want though. It is true

Vance Crowe [00:39:21] Draft. Yeah, but I mean I'm an, I mean like, I think it's good for people that aren't actually using it. Like I work with both graphic designers and I've tried ai and with the ai it's like, you know, the more I try and squeeze the bunny, the more it like, you know, starts squeezing out it. It's like it just doesn't quite work. Whereas with a graphic designer, I can be like, all right, all right, let's, let's start over. But we're not really starting over because they have context

Jim Rutt [00:39:44] And they don't work. Yeah, they track what you want. There it is. It is easier. On the other hand, good graphic artists costs $125 an hour and Dali and GPT plus costs 20 bucks a month. So, you know, that matters too, particularly for very small businesses and and such. But you know, try it out on logos that that's the sweet spot for it. Trust me on this, that you can create a really cool logo that even a graphic artist might not be able to do as good in 20 minutes for a lot of things.

Vance Crowe [00:40:17] So, you know, you mentioned before talking about the dark sides of ai. I'm very interested in this and I think for many people the starting point of the dark point of AI is, you're right, there used to be a person that was charging $125 an hour for a skill that they have developed. And now you've got a computer that can do this infinitely. So this seems very scary to people that had creative work, that have developed skills. Is this where you think the darkness comes? Are you worried about people's jobs being taken from them and not having a replacement? Because the work's all done by ai?

Jim Rutt [00:40:51] Certainly that is one of several risks that I'm concerned about. But it's also of course opportunity because for instance, if a per the, the leading edge graphic artist will say, all right, let me convert my practice to being enhanced with the tool and I'll now be able to knock off three times as many images as I used to. I just finished a project with a graphic studio that was doing the designs for our website, for our new product and our logos and our investor deck and all this sort of stuff. And these are, you know, long time professional marketing agency and they were very upfront that they use these AI generative tools for almost everything. And the amount of work that they could crank out was amazing how quickly they could do it and how cost effectively. So yes, just like I remember when desktop publishing came out in the eighties, every town used to have a printing, a printing company. Any cut town of 10,000 or more had a little printing company. 'cause printing was a pain in the ass, right? You had to set up the plates and you know, dah dah dah. It was like a very hardware intensive and very skill intensive. And then, you know, by the end of the eighties it was lay it out in Postscript and then fire it to the equivalent of a big laser printer. And I dunno, I think it was 30,000 print companies went outta business between 1982 and 1997. It's a ridiculous number like that.

Jim Rutt [00:42:24] But they're still print companies around, but they are, they now all use the technology and of course a lot of people just do it themselves. And so the same thing will happen in these graphic art domains. Those who embrace the technology will become more productive, probably more successful than ever. Some percentage of people will go to do it yourself. The number of people who are professional graphic artists will probably go down and the total amount of dollars spent on graphic art will probably go down 'cause you'll be getting three times as much from the professionals and some, and some of the lower end tasks will be done on a do it yourself basis. So that will certainly impact some jobs. Now the question, the question is, we don't know the answer to this. Will more jobs be created or destroyed? That is, I think still an unclear question because we now have whole new industries, we now need prompt engineers for instance. 'cause it turns out if you really wanna do serious things with this, your ability to ask the questions and type in the stuff yourself, you, you, you know, like for instance with your graphic art, if you had a prompt engineer working with you, they would do a much better job than you'd be doing on your own. And I predicted this a year and a half ago that this will be a great opening for liberal arts majors to return to the tech industry because liberal arts majors tend to be experts with words and how words represent ideas and knowledge. And so I actually ate my own dog food and for our software company I hired a master in fine arts with a degree in short story writing to be our prompt engineer.

Jim Rutt [00:43:58] Everybody thought I was nuts. Of course that's not the first time that's happened. And it turned out he was great. And so there's a job that did not exist prompt engineer three years ago. And now there's a huge opening for prompt engineers. And these, you know, this would be a character who would, you know, likely otherwise be working at Starbucks while he's submitting his short stories for publication probably forever. And now he's, you know, a well paid technical person doing prompt engineering. So there's a job that has opened up while other jobs and

Vance Crowe [00:44:34] That prompt engineer, right, he's able to do the work or help do the work of many, many people. It could be something in graphic design, but it could also be accounting or legal. For me, some of the, the asymmetries are, are the ones that people didn't expect a long time ago, right? It used to be that they thought AI and the robots were coming for all the blue collar work. But the reality is it's all these attorneys that spend the first five years of their career going through every single legal document that a company hands over during discovery or you're, you know, searching through, you know, massive excel databases that can now all be done by ai. And so a lot of these entry level jobs that led to the higher level positions, those are gonna get massacred. And the people that made their investments into their careers only to become the median accountant or the median attorney, they're all gonna be laid waste to and and like the, they'll have to find something else to do, I think.

Jim Rutt [00:45:31] Yep. I know a guy who's a, a founding partner at a mid-size law firm, both has like 80 lawyers and they've historically hired 10 or 12 associates each year. Now they're only hiring four.

Vance Crowe [00:45:46] Wow.

Jim Rutt [00:45:47] And and they also select them somewhat differently. They select them preferentially for those who have at least some technical aptitude in addition to their legal aptitude. 'cause they want them to be pushing the, the edge of what can can be done but also be smart enough to realize what can't yet be done with the software. And yeah, and you're right that I think it's gonna be a surprise that this round of AI will be replacing the more middle of the road of the white collar work rather than the blue collar work. Now there'll be another round when we get closer to artificial general intelligence where things like auto repair even may be amenable to ai. So it'll be more, it, it will go back after the blue collar jobs again. But for for a while it's gonna be, as you say, annihilating these mid-tier white collar jobs. You know, customer service. I mean I've had some horrendous customer service experiences lately talking to supposedly, I mean they are people I'm quite sure. And my reaction after one of 'em was, heck, g Chad, GPT-4 could have done a better job than this person I guarantee. And I know there's lots of money being spent on the application of it to, to the customer service region jobs. And there's a lot of lower middle jobs in customer service as you know,

Vance Crowe [00:47:11] As I think about a world with customer service and my experience of like getting close, but always having it be slightly out of your step. I think there's gonna be a whole new layer of therapy or psychotherapists to help you deal with the fact that so many of your interactions are not with human beings that are sentient. They're with computers that like you can't reason with you can't get it to take pity on you. There's no like, oh they're there. I know you're really struggling with this. It will. And and there'll be some emptiness in that conversation 'cause we're gonna train the AI to have this like empathy, right? We're gonna train it to be like, oh thank you for telling me about your challenges. That must really be hard, but you're gonna know and that's gonna burn you inside. So I predict the rise of AI therapists helping people cope with the fact that so many of their interactions are with zombies, essentially.

Jim Rutt [00:48:04] Well of course the ai, I didn't even think about that one. But yeah, AI therapy, there are lots of money being spent on that.

Vance Crowe [00:48:10] Oh, therapy done by the ai

Jim Rutt [00:48:12] Done by the ai. So now we go full cycle therapy done by the AI to massage people's angst about dealing with ai. Right?

Vance Crowe [00:48:24] Okay, so let's, let's crack open the mind of Jim Rut and say what are some of the darker scenarios? People talk all the time, Hey, we gotta be careful this thing could get outta control.

Jim Rutt [00:48:33] Right?

Vance Crowe [00:48:33] What does outta control look like? Lets do it.

Jim Rutt [00:48:35] Ooh. I mean the worst case is really, really bad. As I told you, I was gonna go the next move up from a GI and this, now that we've gotten sort of within spitting distance or at least gunshot distance of A EGI, we realize that's not the top of the stack by any means good as good as a human and almost everything a human could do. The next step is called artificial super intelligence, which is something that's a computer doing work that humans used to do that's now way better than any human has ever been at many of those tasks. And we are, that's where the one, one of the biggest risks this is, it's in, in the AI risk field. It's often called the paperclip maximizer risk, what a weird thing to call it. But it's where things have gotten so smart that they can completely outfox us and do whatever they want and convince us to do what we want. If we try to put 'em in a security box that they will be so seductive with their use of language that they'll convince us to let 'em out of the box. I mean there's all, there's a whole library of scenarios about this.

Vance Crowe [00:49:43] Well, and the paperclip one, the paperclip one is you tell this ai, hey do what you can to make, you know, paperclips as fast as possible. And this thing now starts taking every single resource that it possibly can to turn it into paperclips. And you're now ripping out sewer lines and smelting the metal and turning those into paper clips and killing

Jim Rutt [00:50:02] The world. Yeah. Killing all the people 'cause they're, 'cause they're wasting all these, this energy that could be used for making paper clips, right? It's a kind of crazy example, but it's like one of the worst case scenarios. And then that goes into two lines. One's called Fme, I don't know what fem stands for. That means super fast takeoff where once we get to some line of smart enough AI, a bit smarter than humans, suddenly you're at something a thousand or a million times smarter than a human in in two days. That's called the fool theory. And the other is the slow takeoff theory where it's on a trajectory to get a thousand times smarter than a human. But it might take 10 years. And again, we don't know. I will say my own work between 2014 and 2018 in cognitive science and cognitive neuroscience has convinced me that human cognition is quite weak. We are just barely a general intelligence just over the line. So there's a lot of room above human level of intelligence for artificial intelligence to move into. But how quickly it can find it. 'cause it will be the one that's doing the finding. Not us. You know, once we get to say two x human, you give it the instruction design your successor and how quickly it can find its better successor will be a, will be I think a determiner of what, how quickly the takeoff, how would go from smart, you know, as smart as a human, a bit smarter than a human. And suddenly so smart.

Jim Rutt [00:51:32] It's a, it not even in the same, you know, so much smarter than anybody that ever lived that we're in a completely different realm. And whether that happens quickly or takes years will make a big difference on how this plays out

Vance Crowe [00:51:44] As it approaches human intelligence. You like probably the best talk I've ever heard, maybe ever in any subject at all was Yosha Bach giving a talk called Computational Metas Psychology. I watched this thing 10 years ago and in it he goes to describe like, hey, if we're gonna try and get computers to think like human beings, you first have to understand what do human beings know? And basically he says human beings know almost nothing at all. We know one or two things really, really deeply enough that we can create value and get work out of it. And almost everything else that we quote unquote know is really what does our tribal affiliation say? Because it's better for us to get along with other people than it is for us to have the right answer. And that's, you know, that's the tribal nature. You don't want to get kicked outta the tribe, you want to be a part of it. But humans vastly overestimate how much we know because so much of what we think of as knowledge is just like what our group believes. So once the computer approaches one single person's intelligence to me that the speed with which it will gain actual intelligence in outpace of human is staggering. And it may have some weird blind spots about not having social knowledge, but I think that it, that that doesn't make it any safer. It probably makes it more dangerous.

Jim Rutt [00:52:59] I was gonna say, and I've had Yos Shaba on my podcast a a few times. They're some of my most popular episodes. So people wanna know about what Yos Shaba has to say. I got the Jim Rut Show. He's an amazing character. I had another podcast very recently with Trent McConaughey is another really smart guy I've known for a long time. And what he's proposed is kind of interesting, which he also thinks that these things will take off and be a thousand times smarter than humans. But he's, his prediction on the way to or idea on how to tame 'em is he wants to accelerate brain computer interfaces and have the humans and the asis essentially merge as cooperative thinkers. And there's a little bit of history that actually shows that might be possible. We mentioned in 1997, I think it was deep blue beat the top chess player. Yeah.

Vance Crowe [00:53:53] But

Jim Rutt [00:53:54] For another 20 years, 15 years, they had a series of competitions where you could combine humans plus computers to play chess. And for until about 20 15, 20 16, the combination of humans plus computers was better than just computers. And they could beat the world's top computer chess program by adding a human to the mix. And that thought was kind of interesting, 15 years where it took for the pure software to be better than human plus software. And so that is somewhat supportive of Trent's idea that at least for some period, the artificial super intelligence plus a human tightly wired with, you know, things in your brain to talk to the computer at eyes speed not for me, thank you very much. But some people will no doubt do, it could actually outcompete the asis that don't have that aspect. And if so, perhaps that's how we tame the asis because they have some reason to keep us around.

Vance Crowe [00:54:54] So, other scenarios that I think of as dangerous. And if this sparks any ideas in your mind, I'm open to hearing it. But I think that once we start using AI in warfare, which we almost inevitably will, someone will, right? Whether that's they're already doing it. Yeah. Use well, so, so tell me about it. Because I imagine a drone being told, find what you think is a human and shoot it right. Or find this human to

Jim Rutt [00:55:16] Shoot it. Already in the, in Ukraine, they already have called lingering music munitions, which will just fly around until they see something looks like a tank and then dive on it. Right? And you could easily do that. And there we know that all the countries are working on the same thing. You'll take the qui on a little a TV, put a machine gun on it, a camera and say, never, never turn around. As long as you're pointed that way. Anything that looks like a human just shoot it. You could build that right now. No problem. And the other, you know, I was just actually talking to an expert about this yesterday. We expect that the next generation of fighter craft will be one fighter plane with five or six drone wingmen. And it'll have, those things will react. They won't be under the direct comm command of the human piloted plane other than giving them general direction. They will be making all their own real time decisions on what the chase in the sky, what to shoot down, et cetera. And, you know, that can't be more than five or six years away. Yeah. So AI in warfare for sure, they'll may actually be good in some ways. 'cause there'll be less humans kill. Let the robots kill each other. Now let's just hope they don't get the dis get the idea to go kill the humans. But,

Vance Crowe [00:56:28] Well, and then what's to stop somebody from, from saying, Hey, I'm gonna fight the other robots. And then once I've conquered them, it will be much easier to subdue a culture because, you know, I, the number of soldiers I need, or soldiers that are unwilling to do the bidding that I want, you know, this is just a cold machine. I just tell it how to work.

Jim Rutt [00:56:47] Interestingly, I'm not too concerned about that, particularly in the short term. 'cause that's a, a variety of narrow ai. It's quite specific. It basically is not deeply intelligent. It just says, if it looks like a human, shoot it, right. It's not making volitional decisions. Go hunt out all humans and kill them. It's o it's only when I get a message that for the next five minutes I should shoot humans and turn on my human recognizer. Oh, there's a human boom dead. But it's not gonna just on its own volition go around running around Chicago shooting people. It's only gonna obey its orders. It'll be a quite, it'll be actually more reliable than a normal infantryman. You know? 'cause it probably can di differentiate at a higher level of precision, a civilian from a military person or a, you know, a mother with a child from a military guy carrying a duffle bag. 'cause it has, it doesn't panic, it doesn't have trigger finger and all that. So truthfully, the, at least the earlier stages of this, I think will actually be just fine. Not, not a huge risk to humanity. I, I do have some other nice scary scenarios I could go down though,

Vance Crowe [00:57:56] By all means,

Jim Rutt [00:57:57] To my mind, the, probably the biggest risks from AI in the bundle, right? In the shorter term, you know, the big stuff. Paperclip, maximizers real risk, I think 10 years outer bore. And we do need to fund research in that. People like Le or YOWs and Mark Max TE Bark and a bunch of other folks, yosha b well actually Yosha doesn't do AI research so much, but I mean risk research. But another one that, that we're already seeing is what I call bad people doing specific bad things with narrow ai. You know, a good example is China and their surveillance program where in much of China, everybody's face is constantly being monitored and then analyzed by ai and then their behavior inferred from the patterns of their movement as they go from one thing to another and their ethnicity being identified, plus or minus, et cetera. So they're essentially building a state-of-the-art dictatorship that Stalin or Hitler would've go, whoa, man, that would've been great. And they're using a bunch of AI in that, in that thing. The other one that we're seeing right now is hyper-personalized advertising. It's, you know, is, it's crazy. You know, the, the Facebook or Twitter, especially Facebook seems to know what you want to see and is giving it to you. And it's gonna only get better at an accelerating rate. One, I happen to know at least one party is working on probably both for the 2024 elections, you will be getting personalized communications that are using la large language models and closely related technologies to send you the message that they calculate you personally will most resonate with.

Jim Rutt [00:59:39] And it'll be custom crafted for you. And, and who

Vance Crowe [00:59:45] Knows, I'm totally with you on this. This, this is, this is the way propaganda is going to go. Right? It used to be that if the government had all the money in the world, they could only fill up a building with people that needed to go on bathroom breaks and could only work eight hours a day. And maybe we'd get tired of putting out the propaganda. But in this case, you could have just servers filled with, with the ability to do ai and you could crank out more propaganda than is comprehensible. And it could be specified to individuals. Yeah.

Jim Rutt [01:00:14] Like, 'cause we could do again for 10 cents, you can buy from a company called Axiom, like 150 data points about you. Do you have a hunting license? What kind of car do you drive? You know, a scary amount of stuff about you. You feed that into an LLM that's been trained on a smaller subset of 10,000 people. It'll be able to push all your

Vance Crowe [01:00:33] Buttons

Jim Rutt [01:00:34] In a way that you won't even be able to detect. And that is not good. And just like Obama probably won the 2012 election by having superior knowledge of social media, their team was way better than Romney's team on social media. It could be that whichever party happens to master LLM technology best in 2024 will win. And so this branch in the, in the trajectory of history won't be determined by the issues. It'll be determined by who happens to master automated personalized propaganda better this year, which is not a good thing. I would say some other examples, AI scams are taking off. There's been some pretty impressive ones where somebody used a AI to build a fake video and hooked and called up somebody at impersonate the CFO and told somebody in his, his supposed accounting department to wire transfer, I think it was $5 million to some account. And of course it was fraudulent, but it looked and sounded just like the CFO and the dialogue was reasonably realistic. It was a human behind the scenes. And our family actually now has a safe word that if we suspect a phone call, oh, your daughter's been arrested and needs bail money or something, that there's a code word we have chosen so that we can detect if we suspect a family member is being, you know, imitated by ais.

Jim Rutt [01:02:09] And I would recommend everybody's family develop that as a habit, as good techno hygiene. And then of course there's, you know, this is one, they actually try to test things like open AI chat GPT and Claude Unfor. But unfortunately there's a reason they can't do it really successfully, which is because it knows pretty much everything. It knows how to make bombs, it knows how to make poisons, it knows how to do bio war biohacking in your basement, right? And we have been somewhat fortunate in that a lot of the terrorists that hate the west are not very technical and aren't very capable. But if they get access to LLMs of strong enough power, you know, they'll be able to make very good nerve gas in their basement. They'll be able to biohack the common cold and turn it into Ebola or something, right? So those are some examples of bad people doing specific things, bad things with narrow ai. A couple of the other ones you alerted to, and I think this is one of the biggest ones we're again, already seeing, and I call the flood of sludge, right? I know a lot of farmers listen to your show. They certainly know what sludge is. And I mean it quite literally, right? That fake websites, you, you fake, you search Google today, seven outta eight of the websites and subcategories are fake, right? And they look like real websites, but they're just things created by LLMs to sell terrible ads for the most ridiculous products.

Jim Rutt [01:03:44] And they at least will help upregulate conspiracy theories, you know, dis and misinformation. And then here's the thought, because these large language models are trained on what they scrape off the internet, the next generation of large language models will be partially trained on large language model generated sludge. So what is that? Oh

Vance Crowe [01:04:05] Yeah, it's just gonna cause this in infinite cycle. I, the, a couple weeks ago I was trying to cook something, so I looked up online a recipe and we all know those damn recipes that are filled with, you know, oh, my grandmother used to make these pickles with the, you know, just this right brine. And it's like just trying to get you to see more ads. Well, this article went on infinitely with just gook about this recipe, never actually reaching the point where it did ingredients and, and like, I don't know, somebody either did that to me on purpose or it's just it, but clearly written by AI because it was like, it was just regurgitated thing over and over again. And then once that's in the cycle, depending if, if that gets sucked up into the next time you're training the GPT five, now it's got all that bullshit in it.

Jim Rutt [01:04:55] Exactly. It's the recycled sludge problem. And, and here it'd make it even better if these people are state-of-the-art spammers, they're, they probably generate a hundred of these versions of this particular recipe. And the one that sucks people in the longest is the one that they now they, they preferentially start connecting to the link and then they take what they learned from that to do the next generation of fake recipes that'll suck people in and all that could be fully automated. So our, our attention is being hijacked and this then gets to what I believe after specific people doing specific bad things with ai, the biggest, one of the biggest risks of all that we don't even think about, and I call it epistemic decay. Epistemology is a big 50 cent word that basically means how do we know what we know, right? How do we know, how do we know things, right? And if the sludge is all around us and it's incoming at this very rapid rate, our ability to actually think and to know real useful facts becomes undermined. What happens when a whole country or whole civilization has its epistemic sense making broken down by the flood of sludge and other things, we're in a real, a real world hurt. You know, I, I use this example when I was a kid, say 19 65, 19 64, the average American adult, 50% of 'em are high school dr graduates. And 50% weren't my parents. One was, one wasn't. So in neighborhood, I grew up in about the same half high school dropouts, half are high school grads, but they actually were pretty smart about politics and current events and community associations.

Jim Rutt [01:06:33] And we built a community pool. We had the boys club and the girls club and you know, this was just run by the volunteers. These people were actually very capable of operating in the world. Today. Almost everybody is allegedly a high school graduate and 35 or 40% of them are college grads and they can't, can't seem to run the world at all. Right? It's amazing. And I would say a lot of this is epistemic decay. We no longer know how to separate facts from fiction. We no longer know how to use common sense in putting together a little community-based organization to build a swimming pool. And we are actually worse thinkers with all of our fancy education than your case. Probably grandparents or my case parents were back in 1965 when half of 'em were high school dropouts. And so then that's only going to get worse. And then, and

Vance Crowe [01:07:26] How does AI make that worse?

Jim Rutt [01:07:28] Well, because it's the flood of sludge essentially, right? If you're just,

Vance Crowe [01:07:32] Oh, okay.

Jim Rutt [01:07:32] Yeah. Well it's two, two parts. One is the fact that you're just getting convincing, sounding baloney, right? What are the two parties gonna do? Are they gonna be sending you the actual truth? No, they're gonna be sending you spun stuff to make you think their way. So your brain is now full of spun stuff. Secondly, and this is a personal theory with no proof, so it's, I'll just call it a conjecture. I believe that the human brain evolved to be able to handle a certain number of inbound attention requests per day. And that number might be 30 or 40 or 50. And when we start getting overlayed with more and more and more calls on our attention, we are essentially like becoming depleted in our ability to pay attention. I dunno if you've noticed, but every damn website selling something now has a little popup trying to get your email address so they can give you a 10% discount on your first order. You know, I've kind of got a mental program now, just click on it without even looking at it. But that is a call upon my attention. And if there's a limit to our ability to process these attention calls, and when you get past it, it depletes the quality of your attention, which I assert that it does, that alone will reduce our epistemic capacity. And I think we're, especially people who hear horrifyingly do things like leave their notifications out on their phone, do not do that. People right. Do not take your telephone into your bedroom with you of all their other horrifying things. People that do those kinds of things.

Jim Rutt [01:09:05] Or worst of all, watch TikTok. I actually did that for four sessions, I dunno, about a year and a half ago. And I said, whoa, this is the fi, this is the electronic fentanyl. This is perfectly this.

Vance Crowe [01:09:18] Yeah, you told me that. I've been using that term for a long time and it's been the, probably the only thing that's kept me from that, that drug because it's, it fits me perfectly. Right? Just gimme a little dose of, of a good conversation. Don't, I don't want to have anything on the front side of the backside, just

Jim Rutt [01:09:32] Continuous

Vance Crowe [01:09:33] Drip 60 seconds

Jim Rutt [01:09:34] Slow drip. Yeah.

Vance Crowe [01:09:35] That's

Jim Rutt [01:09:35] Of fentanyl and oh my

Vance Crowe [01:09:36] God. And you're spinning that thing and it's like every once in a while it gives you something really good. Maybe it's a beautiful woman or it's some, some amazing

Jim Rutt [01:09:44] Dance move or some really clever

Vance Crowe [01:09:46] Yeah. Carving of wood. Yeah, exactly.

Jim Rutt [01:09:48] Yeah. And it, it seemed to figure out pretty quickly that I liked kind of silly country boy jokes, right? And everyone, you know, you know, in my four sessions I probably heard two actually good ones, but most of it was crap. But it, it was done in a way that was hip as a professional designer of online products. I've been designing online products since 1980, so I have a pretty good taste in how to make 'em addictive. And I said, oh, I'm retiring my crown. Whoever designed this thing was the king of addictive product design. This should be anyone who gives this to somebody under 18 should consider themself a criminal. 'cause they've done something worse than giving cigarettes to kids. This is worse than cigarettes. So parents, if you're letting your kids have TikTok on their phone, you are worse than somebody giving them cigarettes. So be ashamed of yourself and do something about it. Right? So epistemic decay is certainly a huge part of the problem in ai. It's not just ai, of course, our whole information ecosystem, but AI is pumping more stuff, worse stuff, stuff that's more attuned to grabbing you and addicting you than previously. And then I'll give you two more if you, if we have time.

Vance Crowe [01:11:10] Yeah, yeah.

Jim Rutt [01:11:11] Okay. One, the next one is, I, I've done been doing a lot of work with a group of people called the Game B community. I think we, we talked about that a little bit. Oh

Vance Crowe [01:11:22] Yeah. My podcast knows all about this. Yeah, okay. You don't even have to explain it.

Jim Rutt [01:11:25] Oh great. Well we, well we think that, you know, game A IER status quo is on a going outta business curve. And we had sort of roughly calculated that we had about 60 years to move the world to a better way of living, which we call game B. It appears that AI is speeding up everything, you know, if it, it speeds up programming, it speeds up product design like the Nvidia stuff, it speeds up, you know, it'll speed up farming at some point. Even every, if it, let's say it speeds everything up by 50% instead of 60 years before game BA game A hits the wall, it'll be 30 years. And you know, you talked about things like energy consumption. I actually have to look this up for another purpose today. Chat. GPT, the training of the current version used 50 gigawatt hours of electricity, which is a staggering amount of electricity. I also looked up what did a slave cost in 1858 and in equivalent dollars it's about $500,000, which, which would say that it's worthwhile. And I've then also, I follow economic development in our region here. And typically when people build factories, they spend about a million dollars of capital goods for each one job created. So the amount of electricity that could be afforded to be thrown at a say just human level, a GI is kind of the economic equivalent of a million dollars, or let's call it a hundred thousand dollars a year of

Vance Crowe [01:13:02] Electricity, right?

Jim Rutt [01:13:03] What's that gonna do to our fossil fuel demand? And to, and just think of the amount of land being used for data centers, things of that sort. So

Vance Crowe [01:13:12] Yeah, I don't think, I think that our approach with energy is going to have to radically change because between AI and things like Bitcoin with proof of work, and I know you and I have a, have a little bit different of opinion on this, but the demands for energy are only going up and this is an offsetting thing Exactly. That it, it'll be, won't be humans or burning wood, it will be using higher orders of energy. And the only way to, to be able to keep up will be to make sure that you have enough energy to keep your computers running. Yeah.

Jim Rutt [01:13:41] Then, and then think about it at the, at the top level of inter nation state competition. If we're using more and more AI to do everything to drive our economy, drive our military, drive our entertainment, our world dominance and entertainment soft power, right? We're, we're, you know, there's gonna be a, a called a multipolar trap or race to the bottom about burning more power to have more ai. And so if we don't get to some massive form of concentrated power that doesn't pollute the atmosphere with carbon like fusion power or orbiting solar, it could be again, an acceleration of game a driving off a cliff by inducing us to a level of climate change, which, you know, we hadn't even considered. So that's the speeding up of game A itself is itself a huge risk. And I don't, I think I may be the first person to be pointing this out, that if we, if we assume that game A is self terminating, whatever period of time we thought we had until it did self terminate is suddenly gotten shorter with the advent of very broad, broadly distributed ai, which is just beginning today. We just, this is, it's more than Wright Brothers 1903, but we're not yet up to the World War One by planes, you know, this, we're still in the pre-World War year one epoch of car of airplanes in the equivalent. So it was way more to come and, and to come. And then the final one, and this one I was actually talking about to another expert in the last few days, is all this stuff combined is seriously overrunning our governance capacity.

Jim Rutt [01:15:21] You know, our government, I would argue in the United States has not been very well governed since about 1994. Many other governments around the world are showing similar signs. I mean, just think of the clowns that we elect a public office and the lies that our government takes of bad decisions that they make, you know, on and on and on. And these are just dealing with the previous generation of problems with this accelerating rate of the kinds of problems we're talking about. How's our governance apparatus gonna deal with this? Right. You know, you think the US Senate, are those people gonna be able to have anything like informed opinions about how to deal with these problems we just talked about? My answer is, eh, no, heck no. They can't deal with the problems from the previous generation of technology. And what's that gonna do? I think I can see two, two branches, one of which is scary and the other is maybe impossible, which is, well I suppose we could give the agis the power to make our decisions for us. No, no, do not do that. Human race, right? That's where you really could end up as slaves to the computers if you let them have the authority of government because they're quicker and not as stupid as a bunch of over the hill US senators. And the second, and this is the harder one, this is hard as could we seriously change our political operating system so that it isn't dominated by over the hill folks who don't know anything about what's going on. And you know, I've proposed some ideas along those lines, one of 'em called liquid democracy, you can find it on mediums, type in go to medium, type in Jim liquid democracy.

Jim Rutt [01:16:57] I have an idea of how we can have a better governance system than normally have now, but it's a long shot that we could actually get that through, get a constitutional amendment and all the things we'd have to do. So the fact that we are now confronting exponentially increasing rates of problems with the same busted old governance structure that we've, you know, had for quite a while here since 1933, basically that is another one that could just cause us to collapse. So, and, and they are, and it's related to ai. 'cause AI is what's the main cause right now of the rapidly increasing set of stressors that we have to deal with. So, so plenty of pleasant thoughts for the day. Right.

Vance Crowe [01:17:38] You know, the, the other day you published a, an interview with James Lindsay about national divorce, which I thought was really telling, you know, this is something that I've heard about in my own circles but not heard it talked about in, you know, like wider networks. And I would say having, having heard it on your podcast in such depth, I would say it is definitely on its way down the, well actually graph right to wider audiences. And this concept I think could easily be accelerated by the conflict that occurs from ai. You know, if you have massive amounts of people turned out of work, they're gonna try and join something that that gives them meaning and gives them some kind of power or authority. If you start talking about, you know, the government either using a AI for propaganda or using it for surveillance, that starts going outside of the bounds of things. I think that concept of national divorce will be much more palatable to people than, oh, let's try and go in and get all the states to work together and maybe we can elect some people in and over the next, you know, two generations we'll try and solve this problem. I I don't sense that our culture is there. And so I would recommend to my audience is a fascinating conversation between you and, and James Lindsay. One that I, you know, I actually have thought quite a bit about, it's actually stop me from sharing it because I thought, hey, you know, this is one of those concepts, national divorce, that you're gonna get on a radar about somebody talking about it and you're never gonna go off that list.

Vance Crowe [01:19:10] So I thought Bravo to you and and James were talking about it.

Jim Rutt [01:19:12] Yeah. In my circles we've been talking about it for a couple of years. And so it was quite interesting and, and James wrote this very good essay, which I disagree with in part as listeners will hear. But, but I think it is time for people to put that on, on the, on their agenda. And maybe it's a solution, maybe it's not, but it's at least something we have to consider if it's basically impractical to over, you know, to overhaul our current governance at, you know, this 330 million people level, maybe we, maybe we're better off with countries of 30 million where we can, you know, create a form of governance based on higher levels of trust, higher levels of coherence that can actually get something done rather than, you know, a set of creaky apparatus which was not designed for the current pace of the world that has to somehow consolidate the views of California and Alabama to get anything done. We shall see.

Vance Crowe [01:20:09] You know, this is something I think you and I view a little bit differently, but in the context of ai, I'd be interested in your thoughts on it, which is Bitcoin. Because I think that one of the things that AI will be able to do will be to massively expand the hacking that can go into the financial system. Whether that's doing password hacking on, on, you know, the, the, the bank's websites or the ability to do kind of phishing attacks will go way up or even the personal ones, right? The, the AI being able to impersonate a voice of a, of a bank president saying, Hey, I want you to wire this money from here to there. Whereas Bitcoin, I think once something's in cold storage, I, I would put my AI up against any bank's, you know, website login system any day of the week. What are your thoughts on that? Does that improve your outlook for Bitcoin?

Jim Rutt [01:20:58] Hell no, not Bitcoin. That's the stupidest idea ever. No, it's actually a brilliant idea. I read Satoshi's paper a few months after he wrote it and I go, why didn't I think of that? Right? It's, you know, it's, it's not that hard, but it's brilliant. This, which really is really just a stupid idea to have a currency based on Bitcoin really extra 10 times stupid. And I will say just for, I actually have more current context right now than I have had in a while. Some of my alt coins went on a big run over the last few months. And so I was selling a bit over the last couple of weeks and it reminded me how amazingly fragile all the infrastructure is to hold and transact coins. You know, you have to know where your private key is, right? But you have to make sure nobody else knows where your private key is and you have to figure out a exchange that you're allowed to be a member of that will allow you to transact the coins that you have. And well, it turned out one of the coins that I really wanted to unload cannot be the exchanges that trade it will not accept us customers. And so I had to wire up some swaps on unit swap, right? And of course then they hose you with like 2% commissions and screw you on the, on the trading spread. It was costing me about 3% to transact them. And I go, Johnny, he says, what kind of person would design a system like this? Right? It's like,

Vance Crowe [01:22:27] Well those are shit coins. The, those systems are shit and they're going to leak money out so that the people that po posted them can steal from 'em. Which is why I'm strongly believe that Bitcoin is not like those alt coins. Yeah,

Jim Rutt [01:22:40] Yeah, no, these are both these alt coins that I hold in fair numbers of are real projects that are doing real work in the world and they're valuable. There's a reason they've brought up 10 x over the last couple of years as, and we'll probably go, and that's why I've only sold selling up bit is 'cause they, I think they got a ways to go with some ups and downs, unlike Bitcoin, which has no known use coin case except for a bad one, which is to be a sterile store of value. We do not want our society savings. Well this is actually a good, I, well this will be good. I actually had prepared this for talk about AI risk, but I'll repurpose it to denounce Bitcoin, which is thanks for the opportunity Vance, right?

Vance Crowe [01:23:26] Oh, I'm loving this. Yeah, go for it. Shoot.

Jim Rutt [01:23:29] You know, Vance is one of my very favorite podcast shows, so I'm just, you know, punching him in the nose just because I can, right? And he'll take it and more or less good humor and then he'll punch back and that'll be fun. So anyway, an economy consists of basically six things, production, distribution, consumption, savings, investment and innovation. And it's essentially a cycle that has to keep running and, and save. And the key ones about Bitcoin are savings, which means that which our society does not consume of our effort becomes our savings in the next cycle to either do maintenance on our infrastructure or to build new things or do research to then do new things. And the savings of the world ought to be invested in productive things in companies and, you know, research, et cetera. What should not be done is put into sterile savings, you know, money in a checking account, for instance. Now fortunately, banks have ways to recycle money. So something good comes out of money in the checking account. Okay? In

Vance Crowe [01:24:34] Your argument there, Jim, you had some aughts that I don't know that I, I would just naturally agree with the, the first one being that like what we ought to do with the money that we're saving, right? So money is actually delineating what is the value of work or something that you created that people will give you for, for that. And, and we use money so that way we don't have to trade, we don't have the have the, Hey, I I want potatoes and you only have apples. You know how, how can we do this? Well, let's use money. But the difference in your analogy between Bitcoin as a hundred dollars bills being stuffed in your, in your mattress in Bitcoin is with Bitcoin there are and will only ever be 21 million Bitcoin. Whereas the a hundred dollars bills, there is an infinite number of them that will be printed there, is they just keep getting printed. So the value of holding dollar bills goes down because each one of those, there's more of them flow going after the same number of goods. Whereas with the Bitcoin, if you take it out of the, the fiat system and you put it into Bitcoin, you're now saying, I want to hold this value and I'll use it at some point in the future. Right now our system is set up that we are forcing people to take the value that they've, that they've earned through work. And we're saying don't put it in a savings account because we are going to burn it away through inflation so fast that you should instead be investing it into the, the, the stock market. You should be buying bonds, you should be doing these various activities that put your money at risk. And that risk is something that we've now taken on as a natural part of human existence.

Vance Crowe [01:26:08] But the reality is we should be able to safely store that which we have created without having it inflated away. Ah,

Jim Rutt [01:26:16] So many faults, parts of that argument. I'm gonna say that the disincentive to save money in checking accounts is a feature, not a bug for the reason I alluded to before, which is it is only diminishing the productive capacity of our society to think you can store economy, you can't, right? The economy produces what it produces and that's all that it does. And when, if you had, let's say for a year nobody invested anything in doing anything, put it all in Bitcoin and then a year later they take all the Bitcoin out and they're gonna make investments with it, guess what? The prices of all the investments will be way, way, way high because the investible economy is very finite and doesn't grow that fast. And so this idea of storing investment is actually fallacious as it turns out. And so any system and wait,

Vance Crowe [01:27:09] But why, why, why is it, why is like, I don't understand why anyone would advocate for a system that they cannot pull value out and say, when I'm ready to invest, when, when there is a product that I particularly believe in when the economy is at a part of the, of the cycle, that I think that my investment would be the most valuable. But instead we have a system that set up that so disincentivizes it that it forces people to essentially gamble with their money.

Jim Rutt [01:27:37] Well you don't, well of course you always have to gamble with them. That's my point. You always have to gamble with your money. There is no such thing as a safe store. Just think about what investment actually means as that you're investing in somebody, building a new factory, somebody buying fertilizer for their farm. You know, there's a million things that savings are, are used for, but this, it's a relatively finite amount at any given time. There's not, ma you can't massively say we need three times as much investment this year. So everything comes flying out of Bitcoin, everybody invest. If you do that, it would, the price of investible, the investible opportunities at every given point in time are only very slowly growing. So this idea of storing the ability to invest is fallacious, at least on average. You're not gonna change the amount of investment significantly, but from one year to the next or one month to the next, there's only so much investment that the world can usefully absorb in each epoch. And so that, and this is where Bitcoin is worse than a hundred dollars in bills because of this finite number of tokens that will ever be issued, people have convinced themselves that the price can only go up. So it's a self fueled bubble. 'cause they say, oh, there's only, you know, 22 billion, whatever the fuck it is, it's only the one wants it gonna be there. Court's gonna go up. And as long as there's a collective hallucination that that's true, it will. But when people realize that this is not what we need in, in a monetary system, the exact opposite of what we need in a monetary system, eventually people are gonna all dump Bitcoin and say this is absurd and ridiculous.

Jim Rutt [01:29:15] And, and start doing what they're supposed to be doing with their, with their surplus, which is investing it into the productive aspects of the economy and a small amount of cash so that you can jump on very short term opportunities. You know, I

Vance Crowe [01:29:29] I I think that I would, I would push back and say, I think one of the reasons that we see such a huge boom and bust cycle in in our economy, right, where you have these massive bubbles that get built up and then they explode is because people are looking for places where they can store their money and they can at least ensure that the value that of what it could purchase today will be the same tomorrow. Let's not even just say going up, just save it to for tomorrow. So people then say, ah, the the best store of my wealth is in real estate, so I'm just gonna keep buying real estate. And the prices of those things go up to the point where their ability to, to be sold into the market is taken away there. There's, there's so many less people that can buy property if the wealthy people are buying multiple pieces of property. And the reason that they're doing that is because if they don't se if they don't buy something as an investment, it's going to be inflated away. I th I think the real trick that people are gonna be shocked about is that there's no end to the amount that the government is going to inflate our money. There's nothing you can do about it and it's always sucking money right out of, out of your pocket. And, and it's, it's more than just the, like the imagined two and a half or 3% that we've had of inflation for years. But instead it's, you know, 6%, 9% in many countries it's 20%, 50, a hundred percent. This allows people to get out from that storm. Yeah,

Jim Rutt [01:30:57] Actually I, I think it will actually make it worse. 'cause if people come out of the economy and oh, things are bad, it must still store everything in Bitcoin and then when they try to come back in and invest, it's gonna have this problem that the amount of investible opportunities doesn't change. But if the flux of funds looking to invest suddenly spikes up, then the price of investible opportunities will spike up. So I, yeah,

Vance Crowe [01:31:22] But right now the alternative, the, what we have right now is wizards that come down from their magical meetings and they say, we are going to make the cost of money go up or we're gonna make it go down. And this is how we're running our entire economy. That we have a board of governors that gets to decide how much should money cost and by, so

Jim Rutt [01:31:42] Don't invest your money and money, right? And in fact, if you look at the long haul, investing your money and money is not smart. Invest your money in productive assets. You know, buy land, buy farmland and rent it out. Buy stocks and bonds, stocks as opposed to bonds. Invest in startups. You know, there's, and then, and if you're a small guy, you only got $10,000, buy the vanguard all stocks mutual fund, there's hardly any fees associated with it. And over the long haul you'll make 8% on your money. You know, way more than than you'll make in government bonds or something like that. And it's, I'll tell you this, as a sort of rich guy myself, rich enough, right? I generally have 1% of my money in cash. All the rest is invested in assets, commodities, you know, private equity, you know, you

Vance Crowe [01:32:34] And

Jim Rutt [01:32:35] Fairly good in gas, natural gas pipelines turned out to be a great business, right? They're like a toll, they're like owning a toll road. The the people that send the natural gas over to pipelines, they pay every time they send natural gas. And I get some coins, I even know several of these collections of natural gas called multi MLPs. It's, they call multiple limited partnerships fucking thing. But anyway, so put your money in actual real master

Vance Crowe [01:32:59] Limited partnerships.

Jim Rutt [01:33:00] Master limited partnerships, MLTs, that's it, that's a good, a real good one. That's tax advantage up the yin yang. It's great. So put your money into real stuff. Don't put it into money. 'cause money will indeed evaporate and maybe at best you could hold it even if you're smart on how you structure your bond portfolios. But in general, the the superior way is that put your money into productive assets and you'll be better rewarded and the economy will have more energy to build more productive assets. You know, as I say, Bitcoin is at, at best as good as putting it in the mattress for the economy and it's probably worse 'cause as you point out putting it in the mattress, at least there's a disincentive to do that based on inflation. Bitcoin makes people at least think they might be able to beat that curve. And so Bitcoin very, very bad.

Vance Crowe [01:33:52] Oh Jim, I think we are gonna watch this play out and, and one of us will be really right and one of us will be really wrong. So, and

Jim Rutt [01:33:58] Of course, of course like everything investments, it's all about timing. You could well be right for another 10 years.

Vance Crowe [01:34:04] This has been a fantastic conversation. If people found what you were saying about liquid democracy, the future of AI plan B or game B, what, what should, where should people go?

Jim Rutt [01:34:18] Sure. Lots of different places you wanna start for game B. Check took out game B Wiki, that's not, it's a little old, but there's some good stuff there. And also a paper I wrote called A Journey to Game B on medium that may be the best place to start. Like, you know, 15 page paper lays out a whole bunch of aspects on it. You know, the future of ai, of course the whole world is, is awash in that right now. It's hard to pick out one source that's better than any other in terms of liquid democracy, an Introduction to Liquid Democracy by Jim Rutt. Just type that in. You'll get a, a very easy with lots of cartoons and drawings, explanation of a better governance system. Yeah. So there's, there's just some, those are some places to start.

Vance Crowe [01:35:06] Well Jim, thank you so much for coming on and we'll have you on again I soon.

Jim Rutt [01:35:10] Yeah, and I will, we'll add, you can cut this off, off if you want, but we're gonna do another podcast tomorrow. We're, we're gonna go into considerably deeper detail on AI risks. And so those who want to be, you know, move from the undergraduate to the graduate level of talking about AI risk, check out Rich Show with Vance Crook. Yeah,

Vance Crowe [01:35:32] This, that'll be fun, man. I'm, I'm looking forward to it. So, so I'll be guest hosting your show and yeah, I'll, I'll definitely share it out. So thanks Jim.

Jim Rutt [01:35:41] Yeah. Alright.