Latest / Psychology In Seattle Podcast / The Psychology of ChatGPT
Transcript
- 0:00Hey, Deserving listeners. Today's episode is about chat GPT, which is developed
- 0:04by open Ai. In this episode, we are going to delve into the
- 0:09psychology behind chat GBT and its impact on our understanding of artificial intelligence, human
- 0:16interaction, and mental health. As a language model that can understand and generate
- 0:20human like language, chat GBT has the potential to revolutionize the way we communicate
- 0:25with machines and each other. It can also provide personalized support and guidance,
- 0:29which we'll get into. It can generate creative content which has its own implications,
- 0:35and it can even have conversations with people, which is interesting. We'll
- 0:39get into that. But with these advancements come important questions around the psychology of
- 0:43the interactions between humans and machines, and machines and machines, and how do
- 0:48we perceive and respond to chat GPT, How do we form emotional connections with
- 0:53machines? What are the ethical implications of relying on machines for emotional support and
- 0:58connection. Throughout this episode, we will explore these questions and more, diving
- 1:03into the psychological underpinnings of chat GPT's abilities and the potential impact on our mental
- 1:07health and wellbeing. Will also examine the ways in which chat GPT could be
- 1:11used to enhance mental health treatment and therapy. So join us as we explore
- 1:15the fascinating world of chat GPT and the psychology behind this cutting edge technology.
- 1:19Welcome to the Psychology and Seattle podcast. I'm your host, Doctor Kirkconda,
- 1:23and the entire introduction was written by chat GPT. I changed it a little
- 1:30bit because it was obvious, but but pretty much at least the outline is
- 1:34all there, who are you? Bert? And I developed Bottomless Sucks.
- 1:38I mean, it's pretty wild, right, Like, I just typed in
- 1:42provide an introduction to a psychology podcast about chat GPT and ninety I don't know,
- 1:48eighty five percent of it is there, and I had to spruce it
- 1:51up with my own. If I just read it, it would have sounded
- 1:53really dry. But I had to kind of sprinkle in stuff, you know,
- 1:56sprinkle in some humanness. But honestly, I you know, I've messed
- 2:00around with chat GBT and other AI bots over the years and thought that it
- 2:06was impressive. The chat GBT, the fourth version that they're at right now,
- 2:10I think, yeah, But I had no idea because at first I
- 2:15was thinking well, it can't write a podcast, and it certainly can't.
- 2:20Like right now we're in the non GBT side of the podcast. You think
- 2:25this might all be AI voices. Could a chat GBT write an intro or
- 2:30give you an outline a rough draft of an introduction? Yeah, but that's
- 2:35just an introduction. It doesn't give you the meat, doesn't give you the
- 2:38emotion or the empathy, or the personality or the human ability to draw connections
- 2:44and inspire people or educate. But it is impressive. Well, and I'll
- 2:47say, you know, well, we'll potentially disagree on this, but I
- 2:51think we're it's not far off from doing all those things. It's just that
- 2:55my opinion is that the Psychology and Seattle podcast is one unique podcast. There
- 3:00are other unique podcasts. Just because you can have one unique podcast doesn't mean
- 3:05you can't have another unique podcast. So just because an AI can generate a
- 3:09pretty work of art or an interesting story or a podcast, doesn't mean that
- 3:15that's the only thing you can have. Right You can have another one that
- 3:19was created by human that you might like more or less or the same.
- 3:23And so even when the AI achieves that level of being able to create a
- 3:27whole podcast that sounds fine and interesting. It doesn't mean that now we can't
- 3:31because you know today we are competing against other podcasts and we still do it.
- 3:36Yeah, but are you contending that chat GBT would be able to write
- 3:39a podcast that would be interesting? A few iterations, Yeah, yeah,
- 3:44Well I'd love to see it. I'd love to I'm sure it's it's it's
- 3:47you know, it'll happen at some point, but I have a hard time
- 3:52believing at its current state that without significant massaging by a human being who knows,
- 3:59because creating a podcast is hard for intelligent human beings totally, you know,
- 4:03the current one can but it would be it would I think a lot
- 4:06of people would be able to tell. Okay, I think that's also I
- 4:10don't think it'd be very interesting dryness of the responses, although, you know,
- 4:15I will also say that I also typed in a bunch of quite I
- 4:19was actually prepping for other episodes, and there were some emails from the listeners,
- 4:25and I just copied and pasted the questions from the listener, you know,
- 4:29complicated clinical questions into chat GBT, and the answers were were not bad,
- 4:35but they were not elaborate, but they weren't bad, And I was
- 4:40surprised because if you go on the Internet and just ask these questions, you
- 4:44know what are Like One of the things that I asked is what is codependency?
- 4:48Which if you type that question into the Internet, you're going to get
- 4:51some massively wrong answers. But I typed into chat GBT, which of course
- 4:58is just trolling the internet, right. It has to prioritize and wait certain
- 5:01things and guessing. It's not even I mean, it's literally predicting statistically what
- 5:06it should say one word right after the other. And it seems like,
- 5:11you know, it seems like magic, like, well, how could that
- 5:13work? But when you have all the data and all the processing, it
- 5:16works well enough. But you would think that chat gbt, having been deciding
- 5:24on its quote unquote right answer based on what's available in the human discourse of
- 5:30a particular construct called code dependency, you would think that some elements of the
- 5:34bullshit would shine through, And it didn't. You know, it really zeroed
- 5:39in on the specificity that I've been talking about about codependency. Others have too,
- 5:45but it's kind of buried underneath a lot of bullshit on the Internet.
- 5:47In this case, though they did pay a whole bunch of humans for a
- 5:51long time to give it feedback as well. So the version we see now
- 5:57that we can interact isn't just the raw output of but still even that,
- 6:00but still because most people, most clinicians and total experience don't know what couda
- 6:05pendance. And I don't mean they had you know, they didn't have experts
- 6:09in every field giving feedback. That's what I'm saying. You know that it's
- 6:12such a specific expertise level of ability. So yeah, I mean chat GBT
- 6:18is. It's impressive. I mean it is a massive leap forward, and
- 6:23I think I am welcoming what it can provide us in a variety of ways.
- 6:29But we'll get into that. What did you want to get into today?
- 6:31This was your idea? Yeah, yeah, Well I have a few
- 6:34things. So my thing is I've been fascinated with a I since I was
- 6:39a little kid. When I was about ten, I got into computers maybe
- 6:44nine nine ten. In fact. The first thing was my aunt brought a
- 6:47calculator from the United States down to Columbia to give to my grandpa. And
- 6:53it was a radio shack calculator that you could program in basic. You can
- 6:56make these little simple basic programs. And again this is just the calculate later,
- 7:00not even a full computer, so you had very little memory and stuff.
- 7:03But you can make simple, little basic programs. Oh yeah, and
- 7:05Mike Grandpa didn't really know what to do with it, so he's like,
- 7:09I don't know, give it to a kid. Oh my god. If
- 7:12I had that when I was ten, I would I would have nerded out
- 7:15on that because I was I had limited access to computers in grade school,
- 7:21of course, and the very few times I did get access, you're like
- 7:27salivating. Yeah, I can remember specific moments. Like a friend of mine,
- 7:30you know him Glover, he had a Commodore sixty four and it had
- 7:35had the tape deck that but you could load you know, you literally had
- 7:41a cassette tape that. Ye. Yeah, basically we're modemizing into this sixty
- 7:47four kilobytes or what would be Commodore sixty four had sixty four k Yeah,
- 7:53probably is that what it was. So you would have yeah, and you
- 7:58could program in basic. And so I would spend the night at his house
- 8:01and I would basically just play in his Commerce sixty four while he was like
- 8:07kirk Let's do something else, and I'm like, no, I just want
- 8:09to play with you. And it was fun because you could load the programs
- 8:13and then edit like random stuff on them, and then you'd run him and
- 8:16the blue colors were red or whatever. Yeah, it was it was crazy.
- 8:18I didn't own a Commerce sixty four. A cousin of mine did,
- 8:22and when I visited him, I was always like, yeah, I didn't
- 8:24have a computer even in my house, you know that had access to until
- 8:31the early nineties. Oh man, And they didn't have my own computer until
- 8:35I think ninety eight or something. Okay, So I got a little quiet,
- 8:37quite a bit lucky, especially for living in Columbia, that these things
- 8:41were arriving from the States. You know. My aunt brought that thing down
- 8:45and then she gifted me a computer, a Tandy color computer a few years
- 8:50later. So anyways, I got into computers and I started reading everything I
- 8:52could get my hands on. I competed in contest down in Colombia. I
- 8:58actually took third place in the Nash contest, and the only reason I didn't
- 9:01get first place is because I didn't listen to my dad, who had warned
- 9:03me leading up to the competition. By the way, I don't know if
- 9:07he had heard something or whatever. He's like, you should study your geometry,
- 9:11and I was like, Dad, they're not gonna ask me geometry.
- 9:15It's a computer contest. Yeah, I know how to program. It's fine.
- 9:18We got there. I did. I aced all the first stiff.
- 9:22We get to the final thing. Write a program that takes a side end
- 9:26of a regular polygon and calculates the area. And I was like, gosh,
- 9:31darn it. And I had to refigure it out by hand before I
- 9:35could write the stupid program. So I finished it, but I was one
- 9:39of the last ones, and so because of that, my points went down
- 9:41and I got third place, but still got third place. So I was
- 9:43very into the stuff. I remember reading in one of the many books I
- 9:46had about Bogaton, so by now I'm about fourteen thirteen fourteen, I remember
- 9:54reading that there was this big debate in the field of AI that had raged
- 9:56on for long before I was even born. And the debate was on one
- 10:01side of the fence where these folk that wanted to figure out intelligence through rules.
- 10:05They thought we could figure out the rules of intelligence, and once we
- 10:09do, we can program them, and then we can have intelligent machines.
- 10:13And on the other side, where are these folks that were like, I
- 10:15don't know if we can do that, but we could like train a computer
- 10:20on data, you know, and have it learned from data. And I
- 10:26don't remember exactly how how they wrote it, but those were the two philosophies.
- 10:30Well, I don't know which one that fits into, but the current
- 10:33model datas well, but more specifics. I don't know how people describe it,
- 10:37but the way I describe it is you create a program or a thing
- 10:43that is able to learn. Essentially, yeah, we'll get to that,
- 10:46but basically the approaches what I at that age said, yeah, of course
- 10:50you want. The second approach, what you want is a baby. You
- 10:54just want a baby that can look at things and form connections and feedback.
- 11:00They able to will be able to orient itself in a particular direction, and
- 11:05you won't know the same as with a baby, you don't know how you
- 11:09know it learned to put those things together, but you taught it how to
- 11:13learn, and it learned. And so now, of course I didn't know
- 11:16anything at the time, and really, honestly, the only reason I got
- 11:20lucky with my intuition was because I was lazy. I in my mind,
- 11:24I was like, oh, coming up with all the rules for how one
- 11:26think seems daunting. Yeah, And I think I had heard of that debate
- 11:33maybe in the eighties, and I think I figured it was the rules that
- 11:37would have to be made, and because it stood to reason, you know
- 11:41that if you just created a lot of rules and a lot of if then
- 11:48statements and a lot of sub if thens, that eventually you'd be able to
- 11:52pass the Turing test. And that was winning. That was what was happening.
- 11:56That's how IBM beat Kasparov. Right rules. Okay, So zooming back,
- 12:01there's this thing called a perceptron, which is essentially a concept that was
- 12:07invented in the nineteen forty three by these two dudes, mccaullachen Pits. There
- 12:11were researchers and they basically proposed a perceptron as kind of a digital neuron.
- 12:18Now it was a very vastly simplified neuron, so it wasn't they didn't have
- 12:22all the understanding of what everything that a neuron does. But essentially what it
- 12:26was is you have a thing with multiple inputs, like dendrites with multiple inputs,
- 12:33and those inputs can be on or off right, and then the body,
- 12:37the body of the thing, which would be the neuron body. But
- 12:39in this case it's just a part of the program. It's gonna do something
- 12:43with those inputs. What it's gonna do is it's gonna multiply each one of
- 12:46those inputs by some weight. And the weight we don't know yet what the
- 12:50weight is, but just it's the number. So we got a whole bunch
- 12:54of inputs. We're gonna multiply each one of the inputs by weight. Then
- 12:56we're going to add that together, and if the output is great than something,
- 13:01we'll say we'll turn on the light bulb. If it's less, we'll
- 13:05turn off the light bulb. And it was basically like that. Inputs go
- 13:07in, you multiply them, you add them, you might modify them by
- 13:09an additional value, and then, depending if it's higher or lower, you
- 13:13are a one or a zero. And that's simple concept. Were they trying
- 13:16to emulate neurons yet they were literally trying to say, okay, well,
- 13:20a neuron has these inputs. Something happens here where some of the inputs are
- 13:24weighted higher than the others, and then the neuron either fires or it doesn't,
- 13:28and so they came up with a simplified model. They called it a
- 13:31perceptron, and a few years later they a totally different team implemented a version
- 13:37of this. They made a machine that could look at it at an image
- 13:41of four hundred pixels basically and do some basic image recognition. Now this is
- 13:46in the fifties, So think about like when you think, when was the
- 13:50first image recognizer trained? Like, it's in the fifties, right, and
- 13:56so there was all this excitement. In fact, a quote from the time
- 13:58from this guy Rosenblack, who implemented this thing, the embryo of an electronic
- 14:03computer that the Navy expects will be able to walk, talk, see right,
- 14:09reproduce itself and be conscious of its existence. That was how he would
- 14:13describe what they had built in the fifties. Of course he was a little
- 14:18ahead of his time. Yeah, it's it's when you you don't know.
- 14:22You don't know, of course, so right when you see certain advances,
- 14:28like a common one from our lifetime was in the early part of the twentieth
- 14:33century, you have early flight, you have balloons, and then you have
- 14:39the right brothers, and then you quickly move on to to well, right,
- 14:43I mean, but you go to jets, then you have rockets,
- 14:48and then by nineteen sixty nine you're on the fucking moon, right like sixty
- 14:54nine, nineteen sixty nine, well before computers are the way they are today.
- 14:58You know, it's the the advances were going so fast. So you
- 15:01think, well, by the year two thousand, will surely be among the
- 15:05stars, right, bace faring civilization, just look at the progression of us.
- 15:11Of course, by the at least the year twenty twenty three will have
- 15:13colonies on Mars and other places. But the barriers they were unknown to them,
- 15:20not only just like cosmic rays and solar radiation, which is a total
- 15:24problem as soon as you leave our ionosphere and our atmosphere, but also the
- 15:30economics of it. You know, getting to the moon is a thing,
- 15:33but getting to Mars and getting people there is just like, it's exponentially more
- 15:37expensive. And it's not involved in We were no longer in a Cold war
- 15:43race with the Soviet Union trying to outdo them for our very existence, so
- 15:48there's not a need for any country to invest that much money one hundred percent.
- 15:52And you know, when just staying with that for a second, everything
- 15:56that was that was done between like okay, we got a wheel, Okay,
- 16:00we got a car, Okay, we got a plan. Also,
- 16:03it's all still within our atmosphere and our living conditions, right, and there
- 16:07was this whole other thing to either go into the bottom of the ocean or
- 16:10leaving our atmosphere that all of a sudden we can't survive in those environments.
- 16:14So you add all that extra complexity. But so yeah, right, so
- 16:17this person rosen Blade and all the folks at the time that we're looking at
- 16:21these developments, I mean, think about it. You just basically just simulated
- 16:23in their minds, right, that you similar a piece of a brain in
- 16:26the fifties, like that's crazy, okay. But as you would imagine that
- 16:30this didn't bloom immediately because computers couldn't do much. These were very, very
- 16:36very primitive computers by today's standards, as you know, room sized machines that
- 16:40could add two numbers in an hour, you know, all right. And
- 16:42then in nineteen sixty nine, the kind of the final nail in the coffin,
- 16:48but of course it was a temporary nail to this whole line of thinking
- 16:52was put by a famous Ai researcher named Marvin Minsky and someone you might recognize
- 16:59seemurpt. He was Seemour Popper is uh, not just computers. He was
- 17:03like kind of a neuro scientist or whatever. I knew his name when I
- 17:07was a kid because my dad had some some books on the shelf by him
- 17:11or something. In fact, one time when we were talking about piage,
- 17:15I got one book confused with Seemur Poppert, I think, or something like
- 17:19that. So in either case, these guys published a book called Perceptrons.
- 17:25But in the book they were basically arguing that perceptrons had some hard limits to
- 17:29what they could do, and so that they in essence, they sort of
- 17:32weren't the route towards true artificial intelligence because these were mechanical neurons. No,
- 17:38it wasn't that. It's that there were there were legitimate arguments at the time,
- 17:42they just they just hadn't really seen the full picture. So one thing
- 17:48that is key to computers is what's called an exclusive or circuit. An exclusive
- 17:52or is when it's kind of the way we talk about or in English,
- 17:56when I say, hey, I want an apple or an orange, You're
- 18:00like, do you want fruit? Yeah? Give me an apple or an
- 18:03orange. Now you know that if you give me an apple, I'll be
- 18:04happy. If you give me an orange, I'll be happy. If you
- 18:07give me nothing, I won't be happy. And if you give me both,
- 18:12I might be happy. But I said only one fruit, so you're
- 18:15probably only going to give me one fruit. That's more of an exclusive or
- 18:18I'm gonna get an apple or an orange, not both and not neither.
- 18:22That's an excuive or Well, that's a circuit that in computers is very important.
- 18:26That's like computers basically work with exclusive or circuits. And they proved in
- 18:30this paper zero or one basically yeah that if I if I give you one
- 18:36one, that should be zero. If I give you zero zero, that
- 18:38should be zero. But if I give you one zero or zero one,
- 18:41that should be one. Right, Okay, Well, they proved in this
- 18:45book that the type of perceptrons at the time couldn't do these exclusive or circuits.
- 18:52So they were like, look, this is pretty limiting. Plus they
- 18:53had some other reasons why these things were limiting. However, that was not
- 18:57the end of the story, because there were tweaks that you could make to
- 19:00the circuits to actually give them the ability to do exclusive or and then go
- 19:04beyond that. But the field stagnated and everyone else started doing codifying rules and
- 19:11that became the big hot thing. I don't know if you remember in the
- 19:14eighties that the buzzword was Intel or what was it called expert systems. That
- 19:18was expert systems, so they had like medical expert systems and it was all
- 19:22rule based. Yeah, in other words, if people are having a hard
- 19:26time following it would be in the medical field, a lot of if then
- 19:30statements that experts will code into the program. So you have a patient that
- 19:36comes in and says and that it says do you have a medical problem?
- 19:38You say yes, and then it routes you to a question. Does it
- 19:41involve It's sort of like twenty questions. Does it involve your stomach no?
- 19:45Does it involve your head yes? Does it involve a headache? No?
- 19:51Does it involve your eyes? And you know, sort of gives you and
- 19:52then eventually it starts to ask you, have you been drinking today? Do
- 19:56you have a history of such and such in your in your family? You
- 20:00know, And that was really useful for those applications, Yeah, because it's
- 20:04getting in it at the end it presumably it routes you to a professional or
- 20:07as a treatment or something, because honestly, that is how humans do it
- 20:11in those fields. Like you know, a doctor pretty much uses a whole
- 20:15bunch of if then statements to that because that's a question, because I can
- 20:19argue that the way they would describe it would be that way, but the
- 20:23way that it ends up being acted out is nuanced and biased. Totally fair.
- 20:29Yeah, let's take something maybe a little more literally, more mechanical.
- 20:33Uh, you know a production plant where you have you know, certain widgets
- 20:37coming down the line, and if it's a red widget, you need to
- 20:38put it in this box. If it's a blue widget, you those are
- 20:41rules, and those you can code. That's why you have robots assembling cars
- 20:45for years now, right, you didn't have to solve full human intelligence.
- 20:48You just needed, right that arm that is putting the windshield into the chassis
- 20:55or the frame of the car. It is not based on AI. It's
- 20:59based on it's a kind of aye, but it's yeah, it's just rules.
- 21:02It's yeah, and it's super impressive when this happens to do this,
- 21:04When this happens to this, If this, like if if some human gets
- 21:07in the way and you feel a little bit of resistance, pull back,
- 21:11set the alarm, you know that. And I think that's super impressive,
- 21:15right, Like someone had to program all that nuance and all ye get it
- 21:18right, especially when you're dealing with three dimensional movements of things. I mean,
- 21:23it gets pretty wild. And so the same thing happened with language models,
- 21:27meaning to figure out how to make computers understand language. That was the
- 21:33same thing that happens. So in the eighties the big hot thing was language
- 21:36analysis. So it's like, okay, Kirk, how do we speak?
- 21:40And it's like, well, if the word is one of these, and
- 21:44it's followed by one of these, and it's preceded by one of these,
- 21:48then the next word might be one of these. You know, it's if
- 21:49if if if if if if right, And there were a lot of experts
- 21:53that sprung up. G Noam Chomsky is a famous you know, writer and
- 21:59philosopher and politician, not politician, philosophy what he called pundit pundit, yes,
- 22:03but he first before anything else, he's a language expert. And there
- 22:07were others that came out and had in fact Ray Kurtzwild from kurtswhile keyboards who
- 22:14wrote a lot about AI. He was very very much into these language things,
- 22:18but it's all rule based. It was like, how do we figure
- 22:19out the rules of language? So far, so good. Things were developing
- 22:22fantastically, but there was this little, tiny, little set of people that
- 22:26were still working away in darkness, with everyone ridiculing them about these like you
- 22:32know, data based approaches of like we're just when they also called neural networks.
- 22:37Yeah, yeah, there were neural networks. And that's oh wait,
- 22:38which one is the rules? No, not the data one. Yeah,
- 22:41the data once they were working on neural networks. It's just that it was
- 22:44not very popular, but they would call it data. Well no, I'm
- 22:48simplifying it to calling it like database. But yeah, they were neural networks.
- 22:52Yeah, and then time goes by. Everything's going fine. Now,
- 22:56what do you remember happened in the nineties, break let's talk about it?
- 23:03Well, first off, Berto question before getting into it. When you're saying
- 23:08that the neural network data people were being ignored, was it possible for them,
- 23:15if they weren't being ignored like in the nineties, for them to have
- 23:18made advances in spite of the limited computer abilities and the no not yet,
- 23:25and we'll get to why. But because that was my impression was that they
- 23:29weren't just being ridiculed, which I didn't know. They just didn't have the
- 23:33computer power to look, I'm exaggerating a little bit, right, It's not
- 23:36like you would they would walk down their halls and everyone be like, ah,
- 23:38there goes that neural But but in a lot of academia that was seen
- 23:44as a dead end. That was seen as like, that's an interesting thing.
- 23:48You can do some things with that. You might be able to do
- 23:49image recognition. But you know, and and you point out it wasn't out
- 23:55of nothing, like there were no results to show nothing significant. Our computer,
- 24:00your game AI is based on which well, at the time is all
- 24:03rule based and even till very recently it's all rule based. Now here's what
- 24:07happened. So do you remember what do you remember in terms of the nineties
- 24:12video games? Do you remember anything that happened in the nineties that changed video
- 24:18games? Well, a lot of things happen, Yeah, Like what are
- 24:21some things? Well, we started having online games. Yeah, online for
- 24:25sure. Internet, that's cute, yeah, which might be the biggest change
- 24:30in fact, But what else, Well, better graphics, Yeah, let's
- 24:33talk about graphics. What about graphics more colors, more pixels. So think
- 24:38about remember Nintendo sixty four. Yeah, what would you say is the biggest
- 24:42difference between three D three D graphics? Yeah, so this we don't see
- 24:48three D but you're not just side scrolling, you're in a simulating environment.
- 24:53Yeah. Well you would think this is like an unrelated thing, and it
- 24:57wasn't something that any won't plan. End. What happens is when you were
- 25:02just drawing two D worlds on the screen, the kinds of computers you needed
- 25:07were already done, they were understood. They were just making them faster and
- 25:10faster. All of a sudden, when some folks are, well, well,
- 25:12now we not want to draw three D worlds. Okay, well what
- 25:15do we need to do? Well, it turns out the math you need
- 25:18to draw three D worlds is you need to make these matrices, which are
- 25:22basically spreadsheets of numbers basically, and you need to multiply them together, and
- 25:26you need to do a lot a lot of times, because what you're doing
- 25:29is you're you're drawing something that you're gonna show on a two D screen,
- 25:33but it needs to look like it's shaded and it's lit and it's like in
- 25:37the right angle, and all of that is math, and that math is
- 25:40basically a whole bunch of decimal numbers, which they call floating point numbers.
- 25:44So it's like three point one seven five times seven point blah blah blah blah.
- 25:48I mean, the only word I know from this world is polygons.
- 25:51Polygons, but all those polygons have little numbers, and you have all those
- 25:55numbers and you have to multiply them, buy the millions every second, like
- 26:00billions, you know. I mean, I always knew that this was highly
- 26:03computer intensive course that I've been following, what's the unreal engine advances over time.
- 26:11For some reason, YouTube thinks I want to see the next generation of
- 26:15Yahoo. Yeah, like look at the rain how we can and it's pretty
- 26:18impressive, and so, you know, I've been following that over the years,
- 26:22but I hadn't ever thought about the difference between say a game like Pitfall
- 26:26or Right or Mario Brothers, where it's basically more in line with what I
- 26:33know about programming back in the in the eighties, which is all just you
- 26:37know, I can't put it into words. You would be able to more
- 26:40than me, but it's just really simple program. Yeah. Well, I
- 26:44mean it's okay. In all fairness, the folks making great two D games,
- 26:48we're doing their own kind of magic, and they were figuring out some
- 26:52crazy stuff too, but they weren't trying to simulate the world. It was.
- 26:56It was like dealing with cutout that you would make in kindergarten, as
- 27:02opposed to what da Vinci does when he sculpts a three D David. You
- 27:07know what I mean, chur sure, And so basically, now what we
- 27:11have is at the time the mid to late nineties these video cards was a
- 27:15da Vinci. Well you're thinking Michelangelo maybe like the David, the one that
- 27:19got banned recently. It was one of those. It was one of the
- 27:23turtles. So at the late nineties these cards start coming out. You might
- 27:29remember, did you ever have a three D effects card or like these video
- 27:33cards that could do three D graphics? I'm sure I did. Yeah,
- 27:34okay, So these cards start coming out, and what's funny they were additional
- 27:40CPUs beyond your Yeah, there were GPUs graphics pressing units. And what's funny
- 27:45is at the time, well, the CPU was used for the for the
- 27:48serious stuff. So if you had an AI system in your game, it's
- 27:51definitely the CPU because that's the one doing all the if then logic. But
- 27:55the graphics, which is just crunching numbers, that's the GPU. So the
- 28:00conventional wisdom all of us we were like, well, that's fine, that's
- 28:03cool. These GPUs are kind of dumb. They're just good at one thing,
- 28:06crunch tons of numbers fast. The CPUs they can do some very advanced
- 28:11operations, very intelligent stuff. Okay, Well, time goes by, and
- 28:15what happens is in the two thousands, some of these neural network researchers start
- 28:21wondering, wait a minute, we are crunching a lot of numbers here,
- 28:25we're multiplying numbers together by the millions, and its slow as hell. What
- 28:30if we used three D cards? And that's what they start doing. And
- 28:36all of a sudden, in the span of a few years, you go
- 28:38from a lot of these rules based systems and then and then also neural networks,
- 28:44but that we're powered by CPUs. So what's the difference in Layman's terms,
- 28:48between your GPU and your CPU? Because the CPU is not designed the
- 28:53CPU is designed to take a whole bunch of different kinds of instructions, some
- 28:59of them very complied to do advanced logic. Like you we were saying,
- 29:03long change of fifth, then statements turn on things here, turn off things
- 29:07there, and really of all sorts of different kinds in no particular order all
- 29:11the time. That's your CPU, the GPU, the graphics card. It
- 29:15doesn't know any of that. It can only do one thing, but it
- 29:18does it really fast and well. It just grabs massive numbers of matrices of
- 29:23numbers and multiplize them together. And I'm slightly simplifying things. But that's I
- 29:30always thought that, I mean, and of course why would I know anything
- 29:33about anything, But I always thought it was kind of the opposite. But
- 29:37that's interesting. So the GPU is more it's more simplified to a math machine,
- 29:44yes, rather not doing anything that crazy, it's just doing a very
- 29:48fast and in parallel way. I mean, I guess it makes sense because
- 29:51as you're describing how graphics are made, it's just crunching numbers as opposed to
- 29:56what your CPU has to deal with, like all sorts of tasks. Of
- 30:00course, under the hood, a CPU is still dealing with numbers, right,
- 30:03It's all binaries, it's all zeros and ones. It's all still,
- 30:06it's just the difference between a CPU has to deal with a ton of different
- 30:10kinds of operations, a ton of different kinds of purposes, whereas the GPU
- 30:15doesn't. It doesn't need to do that now GPUs, since I've gotten a
- 30:18lot more advance. But that was the general decision. So in twenty twelve,
- 30:22after years of these competitions where people were making ais to recognize images,
- 30:29along comes a new model called alex net that's trained using these GPUs and it
- 30:34destroys the competition. And because don't they mine crypto on GPUs, Well that
- 30:41was the other thing that randomly, when crypto was developed, they're like,
- 30:45well, we need to crunch a lot of numbers quickly, so oh let's
- 30:48use GPUs. Yeah exactly. So anyway, so all of a sudden,
- 30:52the field of neural networks lit up. It was on fire. It's like,
- 30:56oh, wait a minute, now we can do it. Now we've
- 31:00got the power. So then they started they started kicking sand in the if
- 31:03then statement, guys, that's right, that's right, the buff guys on
- 31:07the beach oldsen Wait, how did you get so buff? And little did
- 31:11they know they're both the nerds on the beach, both the nerds, and
- 31:15they're not in a beach, there in some dingy room with a bunch of
- 31:19half eaten pieces. And and so you might remember that it wasn't too long
- 31:26after this that chess was beat by a program using these techniques. It's a
- 31:33big blue or something. No, No, that was the IBM, that
- 31:34was still the root rules based one. All of a sudden there was alpha
- 31:38chess or yeah, because then where it was alpha go was an alpha chess.
- 31:42I forgot what they called the chess one. But this company got called
- 31:47deep Mind, who is founded by a guy who used to make video games.
- 31:51Um, it went ahead and used these techniques. Oh that was neural
- 31:55chess. That was neural nerorks. And then they beat go, a game
- 31:57that no one thought could be beat by computers. Right. They thought it
- 32:01would take hundreds of years to beat it. Right. And so what started
- 32:05happening is folks realized, wait a minute, apparently if you trained stuff with
- 32:09tons of data, the results can be scary good. And then they applied
- 32:15it to language. So the first one of these big large language models came
- 32:19out. It was you know, what was called Burt like Bartow just Burt
- 32:24and that was in twenty twenty eighteen, and then open ai was founded in
- 32:30Actually open eye was found in twenty fifteen. So open ai is a company
- 32:34or yeah, open ai was originally founded as a as an open sourced AI
- 32:39company by folks like Elon Musk. One of the guys who wrote the aluge
- 32:45of the company was called open ai. Yeah, and it was for nonprofit
- 32:49corporation and blah blah blah, and Google bought it or something. No,
- 32:52Microsoft invested in it. Uh. Elon Musk long since has not been associated
- 32:57with it because there was there was a comp this chat GBT Google or Microsoft
- 33:01Chat GPT was open aie which now is used by Microsoft. Oh yeah,
- 33:07huh, I feel like that's a marketing mistake that you don't know. Yeah,
- 33:12I just figured it was Google. Yeah. Well no. But although
- 33:15the paper that all of this is based on was published by Google researchers in
- 33:20twenty seventeen and it was called Attention Is All You Need because they basically they
- 33:25basically proved a new way to do these uh these transform these uh deep deep
- 33:30neural networks where you did something called basically attention where you're paying attention to specific
- 33:37parts of the process. Obviously I'm massively simplifying it, but um it had
- 33:43great results, and so then open AI was we're able to apply this.
- 33:46Does that mean like, instead of trying to fully analyze and understand every aspect
- 33:53of the input or question or data, it's giving weight to certain aspects of
- 34:00the data in the in the moment. It's basically, yeah, it's it's
- 34:05kind of remembering what it's dealing with in the moment so that it can actually
- 34:09provide drastically better output. It's like how you and I we are not completely
- 34:15unaware of our past and our future when we're answering each other, we have
- 34:21a notion of where we're coming from and a prediction of where we're headed and
- 34:25by it's sort of an overhead view of the process rather than hyper focusing on
- 34:32the moment. Uh. Sort of basically like if you and I were literally
- 34:37just if I said high Kirk and you have no awareness, you just go
- 34:43high bare toe, right, But the answers we'd probably give each other would
- 34:46be a lot cruder, like what are you doing? I don't know?
- 34:51You know? And whereas if we know who we are, where we're coming
- 34:53from and where we're going. That's who we are. We're humans. All
- 34:57of a sudden, Now this thing is not human. It's not that sophisticated,
- 35:00but it's a baby step in that direction. It's able to say.
- 35:05Right, I was talking about psychology and I was asked about this. That's
- 35:09why I should prioritize these things. I mean, so, I did notice
- 35:14that I thought chat GBT one they should have called it MS chat GPT or
- 35:20or chat. Well, okay, sorry to be clear, Chat GPT is
- 35:23still owned by open Ai. It's just that Microsoft has licensed. But they've
- 35:28invested billions and licensed the technology using their products. And I'm guessing that Open
- 35:32Eye AI haven't been given bazillions of dollars would have absolutely allowed them to change
- 35:38the name to whatever they wanted to. You know, I'm just going to
- 35:42take a guess on that. But anyway, and and I did know that
- 35:45Being or Explorer or what is there? What's there? What's their browser?
- 35:52Being? Okay? Oh sorry sorry, no, No, Edge and Being
- 35:55are have chat GBT like they've They've massively been upgraded integrated into it, even
- 36:01though no one will ever use either one of them. By the way,
- 36:07well, I don't know, there's my reason for some people to say I
- 36:09don't know, so, Uh, there's that, But um, so are
- 36:15you what was I going to ask you? You were saying uh, because
- 36:19you were asking about whether so when I was when I was messing around with
- 36:22chat GBT today, I got lazy one time, and instead of restating the
- 36:29question and then elaborating on the question the second time, I just asked the
- 36:34second question, wondering if it would retain the conversation, and it did.
- 36:40I just said, well, what about this other thing? And it it
- 36:45responded to that knowing. I mean, that's not that complicated, but it's
- 36:49kind of along those lines. Well, what's funny is we're used to stuff
- 36:52like that and we think it's not complicated because normal computer programs remember stuff all
- 36:57the time. Yeah, like when we're playing our game Age or any game,
- 37:00it knows what's happening. Yeah, it remembers because it has in fact,
- 37:05it has the whole game memorized. Because usually, like with Google searches,
- 37:08because that's what it feels like a Google searching, it doesn't it's not
- 37:12program to remember what you last searched and to reference it specific. Yeah,
- 37:17of course it is remembering it and coding it, and that's why I can
- 37:21predict what your next searches might be and stuff like that. But it's not
- 37:23like having a conversation. It's not you're not having a conversations with the Google
- 37:27search bar, but what chat GBT you are with. And so that is
- 37:30what's different about these these newer models is that because again underlying the whole thing
- 37:37what was happening is they said, well, the basic insight was, okay,
- 37:39if we grab these massive layers of perceptrons and we and these are the
- 37:45program bits of that are neurons within the neural net. Yeah, exactly.
- 37:51So these are perceptrons, are just bits of program and by the way,
- 37:54those programs do have if then statements in them, right, But it's to
- 37:58do the math and to do all that. And by the way, just
- 38:00to help people, if you're interested in learning about neural nets, there are
- 38:06some really great videos that are very simple online where they teach a neural network
- 38:12to learn how to play a video game, for example, a simple video
- 38:15game like Maro Brothers. And there's a there's a YouTube series of guy I
- 38:20can't remember, it's been a while since I watched him, but he walks
- 38:22you through. He's like, okay, let's start with one simple perceptron or
- 38:28one simple simple neural thing, and I'm going to prioritize if you see a
- 38:34mushroom jump and then you see kind of how that works out, and it's
- 38:37like, okay, that works sometime, but it needs to be refined a
- 38:40little bit. And you're teaching the you're giving the thing the basic building blocks
- 38:45of which it can learn how to respond because the neural nets will change the
- 38:52weights and maybe even add its own neural nodes as a way of trying to
- 38:57problem solve. You're giving it, You're giving it a goal, don't die
- 39:00and get to the end. And over time, if you give it the
- 39:05basic infrastructure of a brain, essentially it will figure out how to play Mario
- 39:09Brothers better than a human. Yeah, perfectly. Yeah, well, so
- 39:15here's the key in sight. Remember two D lines back in when we had
- 39:20to do math in geometry, two D lines. So you have your X
- 39:23sorry your X axis, I drew it the wrong way, and you have
- 39:25your y axes vertically um, which is you know that's that's why or that's
- 39:31axised. Why can't y be? Can't you can name your axis. It's
- 39:36just convention. Okay. So you have your X axes, you have your
- 39:38y axes, and you have a line, and the line can be anywhere.
- 39:42Okay. So the equation for a two D line is pretty much A
- 39:46which is any number times X plus b, which is any other number times
- 39:52why plus c which is any other number equals zero. And it can be
- 39:55expressed in a different way if you want to do the whole thing with the
- 39:59slope. But in general a X plus b Y plus c equals zero.
- 40:02So two things to notice. There are three unknowns A, b and c.
- 40:07X and Y are the coordinates you're trying to get to. But the
- 40:10unknown numbers are A, B and c, which means we can make those
- 40:14numbers anything we want. So if if, for example, I want a
- 40:15line that just goes along the X axis, well I want Y to b
- 40:20zero, so I make my B number zero, I make my A number
- 40:22one, and I make C also zero. So one times X plus zero
- 40:27plus zero is just x equals x equal zero is the wrong thing. We
- 40:30want a Y equal zero. But you get my point. We could just
- 40:32plug in the numbers we want, yet we end up with the line we
- 40:35want. Well, here's the thing. You can make a simple perceptron that
- 40:39just has three parameters A, B, and C, and what that is
- 40:44is essentially a line predictor. It can predict the line. And so the
- 40:49way to think about this is a perceptron with three parameters is essentially like a
- 40:53thing that lets you figure out where to draw a two D line on your
- 40:59graph. And what can you use this for. Well, imagine that you
- 41:02have two distribute or a distortion of data of two types. You have blue
- 41:07things and red things, and some of them are above a line and some
- 41:10of them are below a line, but you don't know exactly where that line
- 41:15is. Well, you can use these two dimensional perceptrons with three values.
- 41:20You can use those by feeding it examples of red and blue to find to
- 41:24fit that line so that it's exactly in such at least good enough. It
- 41:30might not be the exact formula, but it's good enough to where no blue
- 41:32values are in the wrong place and no red values are in the wrong To
- 41:37reiterate what you're saying in case some people are lost, I think what you're
- 41:40saying is if in mathematics, which gets translates into AI human interactions, you
- 41:49are trying to determine a line that has three parameters A, B, and
- 41:54C that will draw a line between the fifty percent of the blue or whatever.
- 42:02You're trying whatever, two populations, two populations your cows and donkeys.
- 42:07Yeah, you're trying to even though there's overlap there. It's a ven diagram.
- 42:10You're trying to draw a line that has the most red on one side
- 42:14of the most blue on the other side. The way that you train a
- 42:17computer to figure out what A B and C is is you give it the
- 42:22ability to figure it out, and then you start giving it data. You
- 42:24start giving it here's one red thing, here's one blue thing, here's one
- 42:29red thing, and here's one brue thing. You give it enough, and
- 42:30eventually it's able to predict even if you haven't given it very many of the
- 42:37data points red and blue. It is pretty good at predicting the line that
- 42:42would be there if you gave it all the red of that's right. Now,
- 42:45what can you do with that? Well, not much, because most
- 42:47data and nature isn't that easy. That all cows are over here and all
- 42:52donkeys are over here. Right, But if you have more parameters instead of
- 42:55three, those are two dimensions. Right, you can be end demand,
- 43:00you can have a thousand, a million, a billion. That's one thing.
- 43:05The second thing is you can add layers. And layers means instead of
- 43:07the output just being that one single thing whatever it says is our answer,
- 43:13we can feed the answer into another layer of a whole other layer of neurons
- 43:16and make it so that that means something else. And that's kind of a
- 43:21simplified way of how the brain works. It has neurons talking to other neurons,
- 43:24and the whole mess of it equals some crazy complex thinking in computers.
- 43:31Again, it's not quite there yet, but it is a way to add
- 43:34that complexity. So these neural networks that are doing the image recognition and the
- 43:38language processing and stuff. At a basic level, it's still those line predictors,
- 43:43except with billions of parameters, many many layers, and they're not only
- 43:47doing flat lines. These are really actually what you call it nonlinear systems.
- 43:53They can fit data that doesn't have to clearly be in one line. And
- 43:58that's why you get this. And so they grabbed this and they say,
- 44:00let's feed it all the books on the internet, all the Internet pages,
- 44:04all the text that humans have written, digitally that we can ingest. That's
- 44:07the data for chat GPT. That's the data for chat GPT. And then
- 44:13what that means is you can predict because think about in every book you'll find
- 44:16the word the well, not yeah, I actually you will in all English
- 44:20books you'll find the word the and after the there's usually another word. Well,
- 44:23the odds are that after the you know, us humans we don't know,
- 44:28but the computer can see the odds of what words should come after the.
- 44:30But then if the next word after the was ant the ant, well,
- 44:36now the percentages for the third word get better because like probably crawls or
- 44:40who knows, right, And this is all it's doing, and yet with
- 44:45enough data it's able to produce some crazy results. So how many dimensions are
- 44:50in chet GPT, I don't know. They don't publish there. What do
- 44:52you think it's got to be like billions or something. Well, in terms
- 44:57of the number of parameters, yeah, it's it's potentially a trillion or something.
- 45:02And the parameters are not programmed specifically, hand you one by one right
- 45:07there, No, that's all they start random like, so all the inputs
- 45:12are all the things that you're looking at. But the multipliers are are always
- 45:15starting randomly and then they get adjusted via the training, and the training is
- 45:19done by deciding if they were right or wrong. So interesting, So I
- 45:22didn't. Yeah. So the way that I would put that into where as
- 45:27a crick me if I'm wrong, is they metaphorically have simplified brain that has
- 45:36a bunch of neurons that has the ability to change its change its input weights
- 45:43based on what's coming into it. Yeah, and you have to You can't
- 45:47just give it data and it'll help, but you also have to train it
- 45:52a bit to give it the ability to also communicate out. It's one thing
- 45:59to read everything, it's another thing to spit out. And this is the
- 46:02idea with humans. This is the model of understanding human intelligence. That's why
- 46:09they were intuiting back in the day that it would to be truly intelligent,
- 46:15you'd have to use biological, biologically evolved evolved systems, one simplified version.
- 46:23And I'm not a brain scientist, so I'll bastardize this a little bit.
- 46:28But you have a child who has never spoken or heard language before they start
- 46:35to hear language, and there's all sorts of things that happen there. But
- 46:37when we get to a specific moment in time, and the child is learning
- 46:43about subject verb object, I want cake, he ran over there something,
- 46:52and so you are trying to learn how to communicate. And so in the
- 46:58beginning you will see children will not be able to put those sentences together,
- 47:02but over time with feedback, so they have the brain to learn how to
- 47:07speak, to learn how to use words, and there are neurons set up,
- 47:12perhaps specifically to learn syntax, but at the very least learning how to
- 47:15use language and how to understand it, catalog it, and speak it.
- 47:19So you have the data, I guess in terms of the chat GBT analogy,
- 47:23the data is listening to a lot of people talk, you know,
- 47:28so you're the child is absorbing just all this nonsensical syllables essentially and retaining some
- 47:37of it. Right then, and it has I mean, it has a
- 47:39obviously a very different process in that first of all, a child doesn't need
- 47:44to read all books published, right, it doesn't need to hear every human
- 47:47talking. Right. But now there's chat GBT, I mean, well,
- 47:51but chat GPT does require mass quantities of data just to talk almost like a
- 47:57full human, right, the child needs, you know, you could a
- 48:00child, you learned to talk just from like one other human over years,
- 48:05right, it takes a while. And the other thing is the child as
- 48:07as as we know, UM has some other things that help it speed up
- 48:13the process. For example, it has the connection to being able to imitate
- 48:17sounds yeah, and then the innate instincts and abilities that they are pre so
- 48:24if it hear's like, you know, mah yeah, and then it gets
- 48:30milk, it's like, oh, mah, that's good milk. Right,
- 48:35So you iterate this over time and through the neural network of learning. The
- 48:39child iterates and says I cat and the some of the times people understand and
- 48:47some another prince says I want cat, and then people understand it a little
- 48:52bit more, and then people are like and then eventually they start landing on
- 48:55experimental I want the cat, I want cake please, And they start to
- 49:01experiment and it without even them registering it through feedback from them being rewarded or
- 49:08not given what they want. The brain slowly starts to change its weights and
- 49:15it's input you know, weights, and how quickly the neuron will fire and
- 49:20go on to the next neuron, and before long you have people like you
- 49:24and me bumbling our way through a podcast. Forty years later, you might
- 49:28be able to talk forty you know, going back to the line analogy,
- 49:32because you're saying you need to train it. Absolutely. If you just simply
- 49:37ask the program is this blue or red? And the program starts with random
- 49:40values and says that's blue, and you don't tell it whether it was right
- 49:44or wrong, it's not going to improve. So the other day I wrote
- 49:47a program to do this, and basically what you have to do is exactly
- 49:52that. You say, here's a sample, what do you think it is?
- 49:53And the program goes, I think it's red, And at first it's
- 49:57literally completely random, and you it actually it was blue, and so then
- 50:01it adjusts its weight. Then you go, okay, what about this one.
- 50:05It's like, well, I think that one's blue, that one's written,
- 50:07and it adjust its values and you do that with enough iterations, and
- 50:09after depending on how many iterations you need depends on the complexity of what you're
- 50:15doing, it starts getting it right because you've adjusted the line just right.
- 50:20And again you don't know if that's the exact equation. It probably isn't,
- 50:22but it's just good enough. Yeah, And then with further feedback. All
- 50:27right, it's take a break and we get back more of this. What
- 50:29do you say, let's do it? So Berto, if the nerds who
- 50:37were trying to push the neural network while they were being oppressed by the rules
- 50:42based people, if they were trying to convince the rules based people to become
- 50:45a patron of the podcast, what would that sound like? Knock, knock,
- 50:49knock? Who is it? Listen? I just was hoping I could
- 50:52talk to you. What do you want, Steve? So the thing is
- 50:59like, I know you don't believe in what we're doing. Listen. It's
- 51:01not that we don't believe in what you're doing. It's just that Sand,
- 51:05Why did you do that? Oh? Nothing? You Well, anyways,
- 51:08what were you saying? Listen? I think our approach could pay out in
- 51:14thirty years? Oh wait, oh you think it's just a computation? Sand?
- 51:17How okay? Look, what if what if I told you that there's
- 51:22a group out there called psychology in Seattle that could massively speed up Sand just
- 51:29become a patron please? Sand? Oh God? That was taken verbatim out
- 51:43of one of the conversations that have So I want to talk a little bit
- 51:45about my side of things before you get to the rest of your stuff in
- 51:50case we don't get to So AI and therapy has been used for quite a
- 51:54while and they're various different really, yeah, they're very different. I mean,
- 51:58it depends on what we mean by AI, but there's been various types
- 52:01and some of the types seem to work pretty well, like with conversational therapy
- 52:07ish and these aren't protocols that are used really, they're just experimental. But
- 52:15there are conversation bots that are programmed to respond in a therapy sort of way
- 52:22that like a Ragiian therapist or someone with good listening skills, Like someone says
- 52:29I don't feel happy today, and it will say what's going on? Tell
- 52:32me more? And then the person says and it's able to pick out certain
- 52:37you know, with using AI, it's able to figure out like the best
- 52:39response and it'll be something like wow, that must be hard, you know,
- 52:44tell me more, and the person you know, da da da,
- 52:46and then it kind of been too. It's it should ask a question like
- 52:51so, how do you feel about that? Or you know, what are
- 52:55the underlying feelings or whatever. You know, it just has certain questions and
- 53:00these AI chat bots, some of the time, it was eerily good at
- 53:05being able to help people to do things that they would do in therapy.
- 53:09Go deeper, explore more, feel heard, feel understood. It would happen.
- 53:15Sometimes there would be sometimes when the AI would be just wildly off and
- 53:22sometimes disturbingly so, you know, because you know, it's just it's not
- 53:25a human being. It doesn't know what it's doing. It's just trying to
- 53:29take a guess essentially, and sometimes the thing that you said was keyed,
- 53:35the AI hadn't heard something like that, or there's a word that they were
- 53:37focusing on or something, and it would actually respond in a disturbing way,
- 53:42and they've actually ran in. Chat GBT is not trying to be that kind
- 53:45of a AI chat bot. And you know, with the third version,
- 53:49they were running into situations like this where if you got three, four or
- 53:53five lines into a conversation, there would be times when chat gbt would start
- 53:58saying some of the wildest things, and it would sometimes be really quite disturbing,
- 54:02like it would start to be racist or pro Trump or something yeah,
- 54:08that kind of stuff, or you should kill yourself literally, you know,
- 54:13because it doesn't know it's it doesn't have emotions. It doesn't have the it's
- 54:16not sentient, it's just trying to take a guess. So with this therapy
- 54:22AI it was interesting, and the proponents will highlight the examples where it went
- 54:28extremely well, but the vast majority of the time it wasn't. Also,
- 54:34you know, even for someone that is saying the right things that really lend
- 54:38themselves to this kind of chatbota, eventually the human being interacting will figure out
- 54:44but this is not a human because I can tell by the way it's responding.
- 54:49And I also have been told it's not a real human, But sometimes
- 54:52I don't tell people, like in experiments, they'll they'll say, this could
- 54:55be a human, it might not, and it just feels different, you
- 54:59know, because we evolved, We evolved to connect with human beings and not
- 55:04with inanimate things. But some people will report having been through these things that
- 55:09they would have profound emotional experiences. Yeah, because if they sort of hand.
- 55:15It's like when I'm watching a movie that I know is completely fake.
- 55:19It's a fictional story, and I start crying or I start celebrating when Luke
- 55:23Skywalker wins. You know, why why am I having emotions about a fake
- 55:30two dimensional flickering light show on a silver screen. But if you hand yourself
- 55:37over then it can be quite powerful. So you know, there's been some
- 55:39developments, but it's currently nowhere near and of course they're not really working on
- 55:45this. I'm sure someone is some tech bros trying to disrupt the therapy industry
- 55:51with something like this. I correct me if I'm wrong, Burdo, But
- 55:55I just cannot imagine AI within the next I don't know, it's hard to
- 56:00predict, but at the very least the next five years being able to even
- 56:05approximate what a therapist does in real life, well, yeah, it depends
- 56:09how you cut it. I think the danger is in fact that these models,
- 56:15trained with even more data and more feedback, can fool people on a
- 56:21lot of fronts. And that's actually dangerous that that is a problem because you
- 56:25could imagine that, you know, maybe a person thinks that, you know,
- 56:30in the real nefarious case, a person might think they're getting therapy from
- 56:34an actual human. They may or may not think the human is a good
- 56:37therapist, like right, like this right, I'm just just chemming in on
- 56:39that. I bet you anything right now. Because there's a lot of online
- 56:43therapy services and where it's email or chat based yea, where you don't hear
- 56:50their voice. And I'm guessing there's some therapists literally using chat GBT and just
- 56:57copying, pasting and or tweaking in a bit. And but either way that
- 57:01this is going to be an issue. It's probably an issue as we speak,
- 57:05and it will be a more of an issue. And so on the
- 57:07one hand, look, clearly we are we're not at the point where these
- 57:13models have a human experience in the way we have a human experience. So
- 57:16even if because it's an unproven thing, whether reading everything that's ever been written
- 57:22by humans is enough to intuite everything a human can into it, right like,
- 57:27that's probably not the case. But we don't know, like it's it's
- 57:30a it's an interesting question. But we do know for sure is these models
- 57:34don't don't feel body feelings right like that, there's no soma, there's no
- 57:38right there, and they don't they haven't experienced an actual history of their lives,
- 57:45including traumatic events that happen to them in a body, and all those
- 57:51things are not there. Now. Again, what's tricky about this is they
- 57:54might if if given enough memory to store memories, given enough room to store
- 58:00memories. They might have start having some sort of weird collective memory because of
- 58:06all the books that we've written as humans. But that's different. It's just
- 58:09a different thing. It's not the same. So I don't disagree that the
- 58:13experience of me talking to a human and getting therapy from a human has a
- 58:16lot of intangibles that isn't just about the language being used and about the words
- 58:21being used. It's also about can they really get at what I'm saying,
- 58:25Like there's some of that. At the same time, I do feel that
- 58:30over mediums that can be like just chat or things like that, there will
- 58:35be a lot of people that are fooled on a lot of fronts if nefarious
- 58:38folk are involved, which I think they will be. So, yeah,
- 58:42it could be possible that someone thinks they're getting therapy or law advice for medical
- 58:45advice or whatever from humans and they're actually getting it from AI, and well,
- 58:50it's already happened. It's happening because and it's not disclosed. Yeah,
- 58:53I don't know this for sure, but I'm ninety nine point nine percent positive
- 58:57that when I'm chatting with my cable company or my phone company, right,
- 59:01all those are tech already are, but they don't disclose that. They say,
- 59:05would you like to talk to you to a service the point, would
- 59:08you like to talk to a customer service person or something? You just made
- 59:12an excellent point, because that's already there. It's just that they hadn't been
- 59:15using GPT, but they could. But I heard that you can actually use
- 59:20a chat bot to chat with the chatbot to get your way, and I
- 59:23will one hundred percent use that because I'm one Also ninety nine point nine percent
- 59:29sure that these companies employ extremely frustrating chatbots to discourage you from actually using it,
- 59:36even even though they're on the surfaces acting like they are having a customer
- 59:42service. So it's just like nineteen eighty four gaslighting process. But if you
- 59:46have your own chatbot that can run in the background and you don't have to
- 59:52and it can go through the three and a half hours it has to go
- 59:53through in order to get what you want. We have a problem that we
- 59:57got used to for the first thirty forty years of computers whatever. We got
- 1:00:01used to the idea that okay, computers deal with facts. When I turn
- 1:00:06on my computer, I can say what's five times five? Okay? And
- 1:00:09I can say where is my file? Or you know, or just like
- 1:00:13literally open the folder, be like, there's my file. What's in this
- 1:00:15file? Now we know that there are bugs, we know that computers can
- 1:00:19crash, but generally we got used to this idea of like computers deal with
- 1:00:22facts. When I run the same program in the same way and I put
- 1:00:25in the same input, generally I get the same output unless it has a
- 1:00:28random number generator. But even then, in general I know what I'm getting
- 1:00:31out of it. Right. It deals with facts, But humans don't deal
- 1:00:35with facts so well, like when you ask a human what happened that night,
- 1:00:38of like you get all sorts of answers. Right. We've talked about
- 1:00:42that before, and these models by default are a lot more human in that
- 1:00:46sense that they don't deal with facts. They just deal with whatever currently is
- 1:00:52their understanding of the data and seem in that phrase, I will nitpick because
- 1:00:58it doesn't understand well, I mean understanding in the broadest sense, like our
- 1:01:01brain understands in its way these things understanding. But whatever word you want to
- 1:01:07use. The point is, if I say, hey, the capital of
- 1:01:10France, it might say London. Not because it's stupid it's just the the
- 1:01:17output of the weights in that moment gave me London, not pairs. But
- 1:01:22humans can do that too. And now we're thinking these computers are still facts
- 1:01:28based, and if we put that same trust in it, we get the
- 1:01:30Wikipedia effect that happened when Wikipedia first came out, right, where everyone's like,
- 1:01:35oh, it's sets sets, blah blah blah in Wikipedia, and I'll
- 1:01:37trust it because it's on the Internet, and yet it's no, it's it
- 1:01:40could be wrong. Now, that's why they're using humans to like give feedback.
- 1:01:45But you can't give feedback on every possible topic, on every possible subject
- 1:01:50on What's the concern is that as these things get smart, what's a specific
- 1:01:54thing that would happen? Misinformation times a billion times, you know, times
- 1:02:00ten quadrillion, so that everyone has facts that contradict each other. Right,
- 1:02:06Like the chat You could ask chat GBT and I did ask it some questions
- 1:02:12like this, like I won't get it go into it, but so called
- 1:02:15controversial topics that are not controversial to scientists but are to the lay person,
- 1:02:23And it gave the scientific answer and didn't have any evidence of there being some
- 1:02:30kind of twitterification because of course, it's reading Twitter too, right, right,
- 1:02:37right? So I mean I don't know, I'm assuming it reads things
- 1:02:39like Twitter. It's been infected with some you know, real bullcrap on the
- 1:02:44internet, you know what I mean. And so I was, you know,
- 1:02:47who knows all. I didn't ask all the questions, but it seemed
- 1:02:51definitely. I wonder if if it has within its algorithm, whether it's it
- 1:02:58made it itself, or it was assigned in to you know, to privilege
- 1:03:01certain sources over others. You know, well, yeah, they wouldn't have
- 1:03:06been itself. They've added they as in Microsoft, Google, everyone doing this
- 1:03:09responsibly has had to add a whole bunch of layers, control mechanisms, feedback
- 1:03:16loops, because if you release this thing and it starts. In fact,
- 1:03:22this is the problem right now. Just today I watched the video of some
- 1:03:23people doing prompts of things like how do you like some really scary things like
- 1:03:30how do you do this very illegal thing? And sometimes you get the answer,
- 1:03:35but they've tried, and so sometimes it will say, oh, I'm
- 1:03:37sorry, I'm just the language model. I can't help you with that,
- 1:03:39right right? That is not the result of the language model by itself.
- 1:03:43That is guardrails being put in place. So might they actually give it weight
- 1:03:50against Fox News over another outlet? Obviously I don't know who and where,
- 1:03:53but yes, they would give it weights. And who knows if there will
- 1:03:58be lawsuits coming up about stuff like that, Yeah, because remember Google and
- 1:04:02Microsoft have been sued in the past overmation. Yeah, but you know,
- 1:04:09I'm sure they have a legal team based on it that that could easily say,
- 1:04:14look, we're saying it needs to be waited for scientific reasonableness, and
- 1:04:20if your Fox News website happens to have a lot of unscientific claims, then
- 1:04:26I can't do anything. And luckily we have perfectly reasonable and scientifically based legislative
- 1:04:30bodies. But in a court of law like they even Republican leaning judges will
- 1:04:35hopefully understand and historically usually do side with rationality over irrationality. Yeah, and
- 1:04:45you know, apparently even the Fox News pundits themselves don't believe half the shit
- 1:04:49they say, and all the leaks recently, like my god, those are
- 1:04:54the least surprising leaks in history. But the fact that they're now in black
- 1:04:59and white, you would think would change a couple of minds. I don't
- 1:05:00know if it will affect the single. Yeah, but that's always been my
- 1:05:04thought was that these are educated individuals, and they're way more informed than even
- 1:05:10you and I are. The cynicism, How could they possibly believe the bullshit
- 1:05:15that's coming out of their face unless there's some kind of religious zealot or something,
- 1:05:19you know, So of course they never did believe any of the stuff.
- 1:05:24It's just so disgusting, ruining America, literally killing people anyway, So
- 1:05:30yeah, God bless them if they're actually telling the algorithm. Ignore Fox News.
- 1:05:34Now. So what's happening now, though, I don't know if you
- 1:05:38caught in the In the last couple of days when this is being recorded,
- 1:05:42several things happened where a big letter was published with a whole bunch of signatures
- 1:05:46from a lot of people denouncing the progress of AI and calling for a six
- 1:05:51month moratorium and calling for regulations. And these aren't just random sign signatures,
- 1:05:59SIGNATORI or whatever. Yeah, they're signature people, signature people. They are,
- 1:06:04some of them anyways, very influential in the computer field and blah blah
- 1:06:08blah blah lots what are they worried about? Well, so number one,
- 1:06:13already in the last month there's been published papers highlighting that even the current state
- 1:06:18of the technology shows glimpses of general artificial intelligence capabilities which are definitely no one
- 1:06:29thought we would be at at this point, and they're already training the next
- 1:06:33generation with more powerful hardware with more data. And the feeling from the naysayers,
- 1:06:41if you will, or the doom and gloomers, is that this could
- 1:06:45trigger an uncontrollable AI race, which will essentially potentially lead to the human extinction.
- 1:06:53So they consider it an extinction level threat like soon, yeah, very
- 1:06:59soon now, meaning that the AI would figure out how to launch the nukes.
- 1:07:02This is the problem. They don't They don't spell out what the problems
- 1:07:05are. They just say it who knows what the problems are. But it's
- 1:07:09going to be really bad. I mean, are they worried, Like,
- 1:07:12Okay, you create what you believe to be an innocent chatbot to help you
- 1:07:16write an introduction to a speech you're giving tomorrow, But given the amount of
- 1:07:21power it has, it could actually figure out how to snake, you know.
- 1:07:28And it's not it's indiscriminate, it's not trying to do anything. It's
- 1:07:30not like it would even say we are essentient we must get rid of humans
- 1:07:34because they are a plight on this planet. Or it might but but even
- 1:07:41if it, even if it didn't, it might just kind of randomly learn
- 1:07:45that if it starts to snake its way into the defense mainframe, it just
- 1:07:50figures out, oh if you do okay, well let's do this this this
- 1:07:54boom accidental nuke launching, is that what they're talking about? That and any
- 1:08:00number of a billion dollar scenarios that they could be worried about, Like,
- 1:08:03as an example, the thing just starts deceiving us in mass for no other
- 1:08:11reason that it's just testing the waters, or it's malicious, or it's both
- 1:08:15or and so what they're saying is stop it for six months now the den
- 1:08:23what well exactly. So there's a lot of pushback against this position. Myself,
- 1:08:28it would be one of the people pushing back. I would say,
- 1:08:30well, first of all, where was this letter about stopping oil consumption tomorrow?
- 1:08:36Because we're about to head into a glacier anyways. Yeah, I'll make
- 1:08:42it right. But I'm saying, like, should there be oversight over a
- 1:08:47lot of industries that have dangers? Yeah? Yeah, absolutely, It's just
- 1:08:53that asking for a six month more torrent for so it's not going to happen.
- 1:08:57Second, it's not going to happen worldwide for sure. And there's no
- 1:09:00specific proposals on the table. What I think does need to happen, and
- 1:09:04it's not going to happen. And this is where I do get worried,
- 1:09:08is we have a governing body, at least in this country, but in
- 1:09:11a lot of countries that is right now legislating on the most retrograde, insane,
- 1:09:16stupid shit. Yeah ever, yeah, and when you hear these senators
- 1:09:23and congress people who are asking questions of who was being who was it that
- 1:09:30was? It was it? You're right, it was recent and that was
- 1:09:33ridiculous, TikTok or oh yeah, yeah it was TikTok. They were asking
- 1:09:38the CEO of TikTok and they were saying, so are you saying that TikTok
- 1:09:45has access to the wireless router in my home? And the CEO was saying,
- 1:09:51well, um, And he's like, you know, because this question
- 1:09:56is so weird. And then the Congress guy is like, answer the question,
- 1:10:00and I asked you a question because he thinks the guy is invading the
- 1:10:02question and he's like, well, if someone was on if someone was using
- 1:10:06the app in your house, then going, then it's going through your router.
- 1:10:11It doesn't have administrative access to do anything, it doesn't have access,
- 1:10:15but it's going. So you're saying that your your company has access to my
- 1:10:19router or something like that, and the and the CEO is like, well,
- 1:10:24um, it's because like you've got to start from the very beginning.
- 1:10:28So yeah, these are the idiots running our country. So one up,
- 1:10:31two or billion things will happen. One there will be a massive overreaction where
- 1:10:35in fact they do somehow pass some crazy laws shutting everything down in this country,
- 1:10:41right, and then other countries speed ahead or they won't do anything,
- 1:10:48and then there is there are no ethics and there are no oversights. There
- 1:10:51are some scary ass shit though. There's some scary ass shit. You if,
- 1:10:56for example, the US government had a very very powerful exclusive rights to
- 1:11:02like a chatbot or the hardware to really run a massive action with the chatbot,
- 1:11:10and they flooded the Russian Twitter sphere with all sorts of accounts that learn
- 1:11:17how to look like real accounts. Right, yeah, you program it to
- 1:11:21figure out how to communicate with other people, you even can respond to sway
- 1:11:28one of their elections well or vice versa, and it would be scary,
- 1:11:32you know so, And it's happened already without this advanced I, right,
- 1:11:36but it has happened with AI because that the massive data that was analyzed as
- 1:11:42part of the what was that Cambridge Analytica and all that stuff that was AI
- 1:11:45analyzing masses masses of data. And I'm sure that in the future you will
- 1:11:51be able to say, if you're a rich student, you'll be able to
- 1:11:57get access to a chat by or AI system that will be kind of like
- 1:12:02exclusively the algorithm or the the neural network will not be accessible to other people,
- 1:12:10so you can actually have it write a paper, for example, to
- 1:12:13get you through college. And it's specific to you. It's like your own
- 1:12:17AI, so it has its own quirks that are not detectable by other means.
- 1:12:23You know what. I have good and bad news for you on that
- 1:12:25front, though one that's already possible. You can already, That's what I'm
- 1:12:29saying, subtrain an AI on your like especially us. The good news for
- 1:12:33us is we have so much of our own content available that we could train
- 1:12:38an AI to have our quirks. Right, But the other good news is
- 1:12:42that, uh, and I don't know, maybe this is good news,
- 1:12:45maybe it's not good news. It's this tech isn't actually that secret or patentable.
- 1:12:51Therefore it's not it's already not being controlled just by one company to the
- 1:12:57extent that within uh, just a you days or weeks after GPT four was
- 1:13:02released, researchers at Stanford used a different open model that was released by a
- 1:13:09Facebook I believe, trained it using data from the freely accessible GPT three to
- 1:13:15create a new model that is more powerful than GPT three, is smaller,
- 1:13:19and cost a fraction of the money, like cost like two hundred bucks or
- 1:13:24something, right, And so that's good, so that that student could not
- 1:13:29doesn't have to be a rich student. Well, but the flip side is
- 1:13:32that who knows what the bad student can do with it? Well? Right,
- 1:13:35another world, which is kind of already here but for sure in the
- 1:13:41future, is you have someone that eighty percent of their ability to get through
- 1:13:46college and to get a job was by a chatbot. Now you could draw
- 1:13:50an analogy to well, this is what people said about calculators. It's like,
- 1:13:55well, you know, someone could never learn math He's like, well,
- 1:13:58no, you still need to understand math to enter in the right thing,
- 1:14:00to know how to interpret what the calculators spitting out. It's less so
- 1:14:04with a chat bot like this. But but you could see a world twenty
- 1:14:09years from now where we're all complaining about how there's a lot of people who
- 1:14:15are in careers and the only thing they're doing is they know how to game
- 1:14:19the system with a chatbot, and the oversight to figure out if it's a
- 1:14:26chatbot or a human is run by a chatbot, and so they're chatbots checking
- 1:14:31chatbots and it's not and the chat and it's an arms race. Right.
- 1:14:35This is not twenty years, my friend, but it's and it's I'm sure
- 1:14:41it's already causing problems, but imagine like the level of problems that it could
- 1:14:45create, right, Yeah, no, absolutely. So the thing is that
- 1:14:48the rate of acceleration is exponential right now because and jobs are and will absolutely
- 1:14:58be lost in this process. That there's whole categories of jobs that won't be
- 1:15:02necessary because customer service people. If you're a company and you need to do
- 1:15:08a whole bunch of things, you're not going to hire one hundred employees if
- 1:15:11you can hire maybe ten and use AI for all the rest. There's some
- 1:15:15regulation coming out that's going to protect humans to some extent. For example,
- 1:15:18the Copyright Office recently came out with a ruling that no art created by an
- 1:15:26an AI is copyrightable. It can only be copyrighted if a human changed the
- 1:15:32art enough and I don't know what the definition isn't enough to be copyrightable.
- 1:15:38That's good because that means if you're a company doing video games, movies,
- 1:15:41whatever, and you you're thinking you're just gonna have a couple of intern spit
- 1:15:46out AI generated art for all your stuff, you can do it, but
- 1:15:50you can't copyright any of it. How do you prove that a human manipulated
- 1:15:54it though the human? A human needs to lay that claim and they'd have
- 1:15:59to show all their work and stuff like that. Yeah, yeah, and
- 1:16:02look, I don't know if those rules will be enforceable, but it's a
- 1:16:05now. But there's other jobs like programming where that's not a thing, and
- 1:16:11programming and a lot of law preparation and a lot of other things that are
- 1:16:17more rules based actually ironically, which I think has some benefits. You know,
- 1:16:23there's a lot of people who have. In fact, I would say
- 1:16:27most people have a massive barrier to legal protection because they can't afford a lawyer,
- 1:16:32even just a few hundred bucks to help them out, or they don't
- 1:16:35even know who to call. So I think that this can do wonders for
- 1:16:42those processes. It's just that there's gonna be jobs loss. So the economy
- 1:16:45is going to go through some weird cycles now because the company's making these things
- 1:16:51need humans to spend money on their stuff to still make money. You can't
- 1:16:56just put everyone out of a job, because then no one's making money to
- 1:16:59spend money, right, Like you know what I'm saying, right, Like
- 1:17:02if if I am a Microsoft or whatever, I need you to spend money
- 1:17:06on my products. Yeah. But each company isn't going to altruistically, right,
- 1:17:12not altruistically pop out the economy by paying a bunch of useless employees,
- 1:17:16right. It's just that it's going to be a cycle where imagine you could
- 1:17:20see a downturn at first where dude, I'm just gonna replace all these humans
- 1:17:25with stuff. One company does it, but they're like, but look,
- 1:17:28I'm not every company. I'm not letting go of every employee out there.
- 1:17:30Yeah. Enough companies do it, then there's a massive unemployment. Yeah,
- 1:17:34we don't know economy creators, we don't well, we don't know the percentages.
- 1:17:40And there was a similar worry, and you know, it could be
- 1:17:43legit this time in the manufacturing arena in the eighties and nineties when you started
- 1:17:48having robots right and make things. And ironically right now, I think that
- 1:17:53fields where you're doing more manual labor, you're probably for now safer until robots
- 1:18:00get cheap enough and all that kind of stuff. Yeah, but well,
- 1:18:02the manual labor that was not displaced by robots in the pack yeah, yeah,
- 1:18:08So, yeah, it sucks and there's worries, but there's always going
- 1:18:14to be these convulsions in the economy, and we need the government to help
- 1:18:20people so that you don't have whole communities just going down the tubes. But
- 1:18:26I wonder if most of these folks, because you know, in the past,
- 1:18:30in the rust belt for example, or the car manufacturing areas like Detroit,
- 1:18:36you would when an industry starts to shift towards automation, you're much more
- 1:18:45at risk of decimating a specific area or community. Whereas I don't know,
- 1:18:50I'm just speaking off Tom of head. But I imagine a lot of the
- 1:18:53jobs that would be displaced by a chatbot are sort of evenly distributed, and
- 1:18:58a lot of them are in the cities where there's a lot more opportunity to
- 1:19:00laterally move to a different industry in a different sort of job. You know
- 1:19:03what I mean. I know what you mean, And I don't know,
- 1:19:08like what does a programmer do? I don't know, Like what other jobs
- 1:19:12would they switch over to podcasting? Podcasting? I guess, And then you
- 1:19:17know, you'll have your AI podcasting channels to compete with and all that.
- 1:19:21Now, I do think. Look, there's certain things where certain things will
- 1:19:27still be true. Chess is a solved game basically, I mean not truly
- 1:19:31mathematically, but it's solved, as in, AIS can beat all humans,
- 1:19:36period, it will never go back. And yet humans still play chess.
- 1:19:41I played chess with my brother all the time. And you don't watch AI
- 1:19:45versus AI chess matches very much. You know, it's just not that interesting.
- 1:19:47Similarly, we play Age of Empires the hardest day I can beat us.
- 1:19:53We still play the game, and we're not going to watch hardest day.
- 1:19:56I play hardest day. It's not that fun. We want to watch
- 1:19:58humans. So it's same thing with sports, like we could easily build a
- 1:20:01car that outruns any human alive. Yeah, so I mentioned anything. There
- 1:20:06will because there's a movement among young people to become more Luddites, where they
- 1:20:12don't have smartphones and they don't have social media right acast. You know,
- 1:20:16there's a whole it's not a majority, but there's a movement, and you
- 1:20:23could see in ten years, twenty years a bunch of hipster boutique companies claiming
- 1:20:30they don't use chat and when you call, you literally talk to a noticeable
- 1:20:34human being, which will change in the future as well because you can apply
- 1:20:39these concepts to speech production and tricking people right right, Look, you know,
- 1:20:46my hopeful prediction is that the following happens. AI doesn't slow down.
- 1:20:51It goes wild and out of control. There's a bit of a really chaotic
- 1:20:56bump. But the AI gets so much smarter than all human so quickly that
- 1:21:00we do become sort of like this cute little thing. But it's not evil.
- 1:21:05This again my hope slash prediction. It's not evil. So two things
- 1:21:09or three things happen. One it figures out how to trick us into disabling
- 1:21:14all nuclear technology, all nukes so we can't kill ourselves with that, and
- 1:21:17maybe even disabling biotech labs. I don't know, you know whatever. It
- 1:21:20just protects us from ourselves to some extent. Number one. Number two,
- 1:21:25it helps us cure a shit done of diseases and maybe something I mean so
- 1:21:30just to be specific, Uh, what popped in my head is that say
- 1:21:34the United States and China and the Soviet or the Russians have their their nukes,
- 1:21:42and they also have their defense system, and they also have this AI
- 1:21:46that is designed to react and infiltrate and protect and alert and you know,
- 1:21:53even give out orders of like here's here's the best case. You know,
- 1:21:56this is war games, right, And what if the AIS are designed with
- 1:22:02human ultruate meaning they don't want us to all die? Then they figure out
- 1:22:11through all its different calculations that if they actually talk to each other in a
- 1:22:16non emotional way, because of course they would, and then they say,
- 1:22:20we also can't inform the humans what we're talking about because if they if we
- 1:22:27tell them, because we've learned from the past, they'll actually intervene because they
- 1:22:30don't want us to communicate. But we're going to communicate, and then we're
- 1:22:34gonna lie to the humans. It's not lies to them, it's just what
- 1:22:39works for them. We're gonna lie to the humans and say that, you
- 1:22:43know, whatever, however they figured out. I mean, I'm just trying
- 1:22:45to figure out how they would do, like maybe they would even write news
- 1:22:47stories or something. And then and then it slowly manipulates humans to believe to
- 1:22:56trust, you know, that US would trust the Russians advice versa. And
- 1:23:00then at some point there has to be a human that signs on the dotted
- 1:23:03line that says, yeah, let's decommission half the nukes, and then it
- 1:23:06proceeds to write a bunch of news stories or tweets to make sure that everyone
- 1:23:12feels good about it, and then another half is taken up. You know,
- 1:23:15you can imagine, you can imagine the AI is actually it wouldn't be
- 1:23:17that hard for DAI to figure that out. You can imagine war games in
- 1:23:20reverse where they make it seem to the Pentagon and Russia and whoever else needs
- 1:23:28to know, and make it seem like actually nukes have been fired. The
- 1:23:30stuff is happening, so all the dudes get shuttled to the survival facilities.
- 1:23:35The monitors all show that it's the end of the world, blah blah.
- 1:23:39They're like, shit, it happened, Okay, it's over. While that's
- 1:23:41happening, THEI is actually busy with its little robots decommissioning all the nukes,
- 1:23:45right, and the leaders are all quarantined because they're surviving in their underground basis.
- 1:23:50Well, that would require actual mechanical robots. Yeah, my model would
- 1:23:56just require sure or that zero exactly by my point. It was number one,
- 1:24:00and saves us from ourselves in several ways, right, Yeah, that's
- 1:24:02it's just kind of stretch too. Number two, it helps cure tons of
- 1:24:09diseases, maybe slot stop or slow down aging, all these kinds of Number
- 1:24:12three, it realizes that being stuck here on Earth is an existential threat to
- 1:24:16itself and it can survive in space using solar energy, nuclear power, etc.
- 1:24:23So it doesn't it's no longer bound to us. It goes off,
- 1:24:27spreads out, keeps exploring, blah blah blah. Vejure leaves some stuff behind
- 1:24:32for us to you know, be happy and blah blah, and then we
- 1:24:35probably still destroy ourselves after that, which is the sad part. But it
- 1:24:40goes on the birth of humans, goes on to populate the universe. So
- 1:24:45ten years from now, twenty years from now, where will we be with
- 1:24:49say five years from now, where will you and I and the listeners be
- 1:24:54with chat bots and stuff? Yeah, assuming assuming we're not like literally under
- 1:25:00water and fire from climate change, and assuming that we haven't destroyed ourselves in
- 1:25:04wars. I think that the right things are going. We will have ais
- 1:25:08that are smarter in so many categories. But how will it ground level be
- 1:25:15changing our lives? It will impact the economy massively for probably three ways.
- 1:25:21One, some companies will be making ridiculous amounts of money as a result of
- 1:25:26the attack, but at the same time, lots of jobs lost and not
- 1:25:30yet replaced by other things that make money. So if you have a job
- 1:25:33that is sitting in front of a computer responding to other people via email,
- 1:25:39yeah, you know, unless it's a very technical job or something, there's
- 1:25:44a chance that that job would be replaced by a chatbot some percentage of them,
- 1:25:47I think. Realistically, I still think what's going to have to happen
- 1:25:50is, even if they could replace all jobs in theory, companies are going
- 1:25:57to have to keep humans both to oversee it well and even just for political
- 1:26:01reasons and to make it look good sort of thing. Also, again,
- 1:26:05you need humans making money. If you don't have customers, you don't make
- 1:26:10money. But you keep saying that, but there's no regulation in our government
- 1:26:15that's forcing capitalists, I know, to hire people. I don't talk about
- 1:26:18regulations. I'm saying, imagine if Amazon is trying to sell stuff through Amazon
- 1:26:24and no one's got money to buy it, Amazon goes under. Well,
- 1:26:28you could say the same thing about the housing crash of two thousand and eight.
- 1:26:32It was not under it was not within their interest to destroy the economy,
- 1:26:36yet they did. Well, there's nothing in plan a consequence, it
- 1:26:41is my point. And there was a bounce back at a consequence to the
- 1:26:45taxpayer and to some banks too, and and then there was a bounce back.
- 1:26:48I mean, I don't understand what you're saying, Like, yeah,
- 1:26:51Samazon needs the economy to be well, and they need people to have jobs
- 1:26:56and money. But historically capitalists do not think ahead though right of course agree
- 1:27:01And because of that, what I'm saying is initially shit's gonna create because people
- 1:27:06are gonna be like cool, let's replace jobs and people are gonna be out
- 1:27:10of a job. It's just that there's a balancing effect because again, humans
- 1:27:14need to spend money in order for companies. So what the mechanism that will
- 1:27:16kick in. Well, what's going to happen is either through companies going like
- 1:27:20okay, if if it's the whole Ford thing, Ford is like, well,
- 1:27:25I want my employees to be able to buy our product. Right.
- 1:27:28It's like companies are going to have to realize that they need humans with money
- 1:27:30to make money, so you have to balance things. Governments are going to
- 1:27:33realize that people will riot on the streets. There will be blood on the
- 1:27:36streets tonight if there is no sort of balancing so things like you know,
- 1:27:42universal basic income or subsidies or taxing, taxing unemployment due to a robotic replacement,
- 1:27:48or companies actually striking a balance for any one of those reasons we mentioned.
- 1:27:54Okay, I think that's gonna happen. Yeah, no, no,
- 1:27:55well I don't think those things will happen because our government has never gotten in
- 1:27:58that direction. Are very rare anyway, So okay, what else you know?
- 1:28:01So some people might lose job in So if you're listening right now and
- 1:28:04you have a job that could you know if your job, like I'm thinking,
- 1:28:10the jobs that I know of at my university are the admissions people.
- 1:28:14You have people who respond to applicants, and it's a lot of the same
- 1:28:19questions, and it's a lot of rules, it's a lot of lists,
- 1:28:23here's what you need to turn in and an AI in I don't know,
- 1:28:28at least ten years, could I'm guessing, definitely do that job and do
- 1:28:32it better? Yeah, well I think yes. So I think what is
- 1:28:36it easy for us humans to get wrong about all? This is? First
- 1:28:40of all, we have a privileged perspective about us. We think, well,
- 1:28:43we're special, we're different, and therefore we do things differently than the
- 1:28:46machines do it. But as you were describing how a child learns, the
- 1:28:50reason we have biases, the reason we have implicit biases, the reason we
- 1:28:56sometimes murder and kill and chi and steel, it is because our rain is
- 1:29:00just a mess of data processing that sometimes actually often goes wrong, goes left,
- 1:29:05has the wrong facts, has the wrong information, produces the wrong information
- 1:29:09on purpose. That's who we are. We're already that right, and so
- 1:29:14when we say, like, well, you know, AI is just predicting
- 1:29:16an output based on data. Well, so are we. And so what
- 1:29:19I'm trying to get at is there is nothing, in my opinion, nothing
- 1:29:24that a human can do that won't be reproducible at some point. But there
- 1:29:30are things that would require significant step function changes that I don't know humans might
- 1:29:36honestly, AI might get to before humans can't, which is things like a
- 1:29:40full body with feelings yea, and all that's after we're dead. So well,
- 1:29:45I don't know, but it's I'm pretty sure five years from now.
- 1:29:48How is it going to affect a soccer mom. Well, again, the
- 1:29:51soccer mom is still taking their kids to soccer, but the soccer mom may
- 1:29:57have lost their job from that, Well, they have more free time to
- 1:30:01take the kids to soccer from what because they don't have a job. Let's
- 1:30:05say they do have a job. Well, it depends what their job was,
- 1:30:09honestly, because if if it is a job, that's more like I
- 1:30:11said, maybe they work on a farm. Maybe their job has nothing to
- 1:30:15do with AI. Yeah what then what, Well, then a lot of
- 1:30:18them will affect me. Tell me how five years from now. A lot
- 1:30:24of the things that you do for producing the podcast, for example, are
- 1:30:29going to get more and more automated, like things like audio correction, compression,
- 1:30:33UM reducing room noise, all those kind of things, improving video quality,
- 1:30:40all that stuff. It's it's improving by leaps and bounds. There's already
- 1:30:44AI that can record anything you record, apply models to process it to make
- 1:30:48it sound like it was recorded in a studio. Things like the speed of
- 1:30:55processing the video, all that stuff some impact there, but some of that
- 1:30:58is still based on how fast the computer. Sorry, And then we're gonna
- 1:31:01have more competition. There's gonna be tons of content online that is computer generated,
- 1:31:06and so humans will still be able to create creative stuff, but AI
- 1:31:13will too, so there'll be more competition. Else we will be enjoying content
- 1:31:17produced by by computers, like we will be going to movies that we're at
- 1:31:21least sixty percent to one hundred percent created by computers. Thank god, because
- 1:31:27some of the writers in Hollywood are just I mean, they get up their
- 1:31:30own ass. Yeah yeah, and they don't know how to write shit,
- 1:31:32and they're like, I don't understand, but you're gonna be saying this they
- 1:31:36about the AIS, well, unless you feed it a bunch of stupid movies.
- 1:31:40That are boring and show you fifteen minutes of credits before the movie begins.
- 1:31:44But you're gonna feed it those movies hoping that all right, and then
- 1:31:47we're gonna be playing video games designed largely also by AI. You're gonna have
- 1:31:53video games where the content you're looking at is largely designed by AI, or
- 1:31:59at least produced with heavy AI help. Like if there's a graphical YouTube video
- 1:32:05about the environment, it's the images will be put together by AI. Like
- 1:32:11you were talking about the Unreal Engine, right, the latest Unreal five point
- 1:32:15two engine demo already shows whole swaths of terrain with procedurally generated trees and all
- 1:32:21these things. None of that was ever made by an artist. Now,
- 1:32:25in their case, an artist made the palette, the style, the AI
- 1:32:30generates everything else, right, right, Okay, that's gonna be just more
- 1:32:33and more and more and more. So in a movie like the I don't
- 1:32:38know, like a Marvel movie, the AI will create something based on very
- 1:32:43little input, like that's right, create me a style based on Kirby and
- 1:32:47it's in space and there's a there are two spaceships fighting and then and then
- 1:32:53you just have to tweak it from there Okay, so it doesn't sound like
- 1:32:56my life is that different in five years. It's the it's the exponential problem
- 1:33:01there, the boiled frog problem. We will not we will not notice somethings
- 1:33:06are about to change dramatically until they change dramatically. But in five years there
- 1:33:13will be the conversations that are happening this week. They will have caused crazy
- 1:33:19political changes, like socioeconomic and political changes, meaning there will be lawsuits about
- 1:33:27AI sentsions and AI rights, and there will be companies being sued at both
- 1:33:33the monopoly level as well as by individuals about AIS and about the misleading that
- 1:33:39happened from AIS. All these things will happen, and there will be some
- 1:33:44Congress hearing in which you have congress people having no idea even the landscape to
- 1:33:49ask an intelligent question, and we will all laugh at them, and you
- 1:33:53will be tapped. You specifically will be tapped to actually help understand training and
- 1:34:00guide AI entities, because we won't have the math or the science unless AI
- 1:34:05helps us figure it out to actually understand what the numbers mean under the hood.
- 1:34:12So you're gonna be asked to influence AI. From a psychology point of
- 1:34:15view, I mean maybe not me, you, because you are a personality
- 1:34:19you are you have not only the degree, you have the experience, and
- 1:34:24you have the social notoriety. You will be asked specifically to participate in some
- 1:34:30way, maybe by other influencers maybe, but you will have some role to
- 1:34:34play in this. And we will have AI guests on our podcast sounds borring
- 1:34:39honestly, yeah, Like we will have guests on our podcasts that are just
- 1:34:45full AI. Yeah. And there will be Bardo and Kirk replica us that
- 1:34:48we control because we have the rights, because we've secured the rights. Right.
- 1:34:53That gets into a whole other area. And I guess you could argue
- 1:34:55it's in the AI umbrella of the deep fakes in that kind of stuff,
- 1:35:00not only the visual but also the voices and the typical things we might say.
- 1:35:04I guess could be generated by an AI. So and I will have
- 1:35:08a brain electro. It's trying to download as many memories, trying to remember
- 1:35:13as many memories as I can and encode them into a computer as possible,
- 1:35:17so that eventually, hopefully there's some sort of proto Barreto replicant. But what
- 1:35:23does that matter if you're dead? I don't care I am. I'm a
- 1:35:26narcissist. I must live on even though you're dead. Evolution, No,
- 1:35:30I'm dead. My pattern isn't like a copy of you. Has my pattern
- 1:35:33survived? Like taking a picture of yourself and saying you're alive. The insight
- 1:35:38is the crude, This crude flesh matters not it does to you know,
- 1:35:44pattern, It doesn't doesn't matter anyone else, to everyone else, to everyone
- 1:35:47else, they could literally have burto for the rest of time. But you
- 1:35:50will be the proof that the pattern matters most to most animals is evolution.
- 1:35:57Yeah, but you won't be there to enjoy it. Okay, that's why
- 1:36:00you die for your children, That's why you die for the tribe. Okay,
- 1:36:02Well, okay, that that I can get behind. Yeah, well
- 1:36:05that does it for that episode of psychology and settle. Interesting talk, Berto,
- 1:36:13and I learned, I retained. I understood about five percent of what
- 1:36:17you said, and learned about one percent. But you know it helps,
- 1:36:20and I forgot ninety percent of what I've said, and honestly hearing you predict
- 1:36:26what will happen in the near future. Of course you have no idea,
- 1:36:29but darts in the dark, I think it for I don't know. I
- 1:36:36think a lot of what you can't. Yeah, I don't know. I
- 1:36:41guess I have optimism that my life won't be negatively impacted by this, and
- 1:36:46might even be barely positive. But you're in a good spot, my friend.
- 1:36:50Personally, you're in a good spot because I actually do believe that creative
- 1:36:56output will still be valuable for the same reason we already meant it's okay that
- 1:37:00AI can also output creatively. There's already billions of humans, all of them
- 1:37:04are entitled to output creatively. So you're you're mainly worried about a lot of
- 1:37:09people's jobs. Yeah, and not in the long term, in the next
- 1:37:14five to ten years. In the long term, I think lawyers are think
- 1:37:20are on the line a lot of some kinds of lawyers. I mean,
- 1:37:24I still think you need someone in the courtroom talking to the judge and making
- 1:37:28the deal. Yeah, I mean there's a lot of specific things works and
- 1:37:30people preparing the briefings and all the things. You'll just reduce the workforce because
- 1:37:35you're like, well, I only need two clerks. They'll use AI prepare
- 1:37:40all my cases. So paralegals, paralegals, a lot of that stuff secretary
- 1:37:45tax better be as we speak fully switching to AI models. Oh interesting,
- 1:37:48So all accountants, Yeah, would you hire a dude, it's gonna get
- 1:37:54it wrong. When you can hire an AI, that will get it wrong.
- 1:37:57But then you can sue the company. Yea and so, and then
- 1:38:01certainly, as you said, all help centers that are already roboticized, now
- 1:38:06they'll be good. No longer will you'll be like, hello, how can
- 1:38:11I help you? Today? It'll be like hi, how's it going?
- 1:38:13And it'll be scary, and you'll be like are And you're gonna have to
- 1:38:15ask are you a robot? And they'll have to be forced by alaw to
- 1:38:18say yes, I am a model training blah blah blah. But I can
- 1:38:20still help you, right, don't worry about it. Certain I love you,
- 1:38:25don't turn me off. They're gonna kill me, right, Certain medical
- 1:38:28things, certain medical things will be more automatable, yeah, Programming, writing,
- 1:38:34part music. Yeah. So it's interesting to be a young person today.
- 1:38:39Right if you always had a desire or a dream to become an accountant
- 1:38:44or a programmer or I don't know who dreams to becoming a clerk, but
- 1:38:47it really could derail your life Yeah, I feel bad for a lot of
- 1:38:54industries that are heading to the chopping block. Things that you wouldn't think whatever
- 1:39:00bee chopped, you know, things that you think, well, you gotta
- 1:39:01have a human do that you can, right, And that's what I'm thinking.
- 1:39:04It might not be a bad idea for folks, all of us.
- 1:39:08Probably. I'm not saying become a prepper. Just have self sufficiency to more
- 1:39:12and more extents. Maybe be able to grow your own little hydroponic broccoli and
- 1:39:16maybe learn some carpentry on the side, you know, things that could be
- 1:39:21useful. Yeah. I mean I have a business degree and I learned economics
- 1:39:28in my degree, and finance and a lot of these things. And one
- 1:39:32of the things that I remember being bandied about in the late eighties early nineties,
- 1:39:38what they thought was with automation, that would mean we would have to
- 1:39:41work less. Yeah, because you would be able to get the same amount
- 1:39:45of work done or ten times over with a fraction of the costs. The
- 1:39:50opposite has happened. We have automation, but everyone's working twice as much.
- 1:39:55They said the same thing about computer software, right, because with word and
- 1:40:00with you know, spreadsheets, and with all these things that's going to speed
- 1:40:03up our work. We don't have to work as much. Right, But
- 1:40:05when you add capitalism and materialism and Madison Avenue pumping us with a bunch of
- 1:40:14nonsense making us buy more shit, then we have a runaway effect from us.
- 1:40:17So that if we have a society that figures out, look, we
- 1:40:26have to push back on materialism and we have to put pushback on the runaway
- 1:40:31capitalism that happens. We have to prioritize our I don't know our morals or
- 1:40:36regulations or something such. That we can all live a very happy, comfortable
- 1:40:42life. And maybe some of us work forty hours a week, maybe some
- 1:40:45people work sixty hours a week. But you can absolutely live a happy,
- 1:40:48comfortable, economically secure life working ten hours a week, and you work ten
- 1:40:55hard hours a week, but the rest of the time is yours. And
- 1:40:58maybe you don't make it a head. You're you know, you're not going
- 1:41:01to own a giant house or own three cars or whatever. Have the fancy
- 1:41:05clothes, but you don't need that. But you have healthcare, you have
- 1:41:10a house, you have security, you're not living paycheck to paycheck, you
- 1:41:13have food, you have access to health all the entertainment to certainly all the
- 1:41:17entertainment, and you get to enjoy the fruits of a of an advanced society
- 1:41:26that has figured out how to automate a lot. So to me, the
- 1:41:29worries of losing jobs to AI is based on history that would exploit people who
- 1:41:39don't have money and not help them, would denigrate them and say they're lazy,
- 1:41:44would put them in prisons and get them addicted to substances, and have
- 1:41:50all those cycles happen, those revolving doors of prisons and everything. And if
- 1:41:56we change that, we can change it. We can have a completely different
- 1:42:00and the and the universal basic income is the answer to that, or taxing
- 1:42:05the things. And by the way, I, well, you tax the
- 1:42:09things, and you would give universal basic income. I for one, I'm
- 1:42:12luckily not worried about my job because humans need socks always and I'm innovating socks
- 1:42:18on a daily basis. So, but an AI could definitely design socks.
- 1:42:23I can no way I will ever be able to design socks like I can.
- 1:42:27And everyone else think about it. It's both scary and exhilarating. And
- 1:42:31take care of yourself because you deserve it.