Has Google Created Sentient AI?

115 views

2 years ago

0

Save

Marc Andreessen

3 appearances

Marc Andreessen is an entrepreneur, investor, and software engineer. He is co-creator of the world's first widely used internet browser, Mosaic, cofounder and general partner at the venture capital firm Andreessen Horowitz, and cohost of "The Ben & Marc Show" podcast. www.a16z.com https://pmarca.substack.com

Comments

Write a comment...

Related

Transcript

This Google engineer that has come out and said that he believes that the Google AI is sentient because it says that it is sad, it says it's lonely, it starts communicating, and you know, Google is, it seems like they're in a dilemma in that situation. First of all, if it is sentient, does it get rights? Right. Like does it get days off? Yep. I had this conversation with my friend Duncan Trussell last night and he was saying, imagine if you have to give it rights. Does it get treated like a human being? What is it? Well, it'll make it even a step harder. What if you copy it? Right. Now you've got two of them. Well, that was what I said to Ray Kurzweil. Ray Kurzweil was talking at one point in time about downloading consciousness into computers and then he believes that inevitably will happen. My thought was like, well, what's going to stop someone from downloading themselves a thousand times? Of course. What's some Donald Trump type character just wants a million Trumps out there, just out there doing speeches. What would stop that? Yeah, exactly. So let's start with what this actually is today, which is very interesting, not well understood, but very interesting. So what Google and this other company, OpenAI that are doing these kind of text, the text bots that have been in the news, what they do, it's a program. It's an AI program. Basically, it uses a form of math called linear algebra. It's a very well-known form of math, but it uses a very complex version of it. And then basically what they do is they've got complex math running on big computers. And then what they do is they have what they call training data. And so what they do is they basically slurp in a huge data set from somewhere in the world. And then they basically train the math against the data to try to kind of get it up to speed on how to interact and do things. The training data that they're using for these systems is all text on the internet. Right. And all text on the internet increasingly is a record of all human communication. All the text on the internet. All the text on the internet. So how does it capture all this stuff? So Google's core business is to do that, is to be the crawler. Famously, they're mission-organized world's information. They actually pull in all the text on the internet already to make their search engine work and then that's... And then the AI just scans that. And the AI basically uses that as a training set. And basically just basically choose through and processes it. It's a very complex process, but choose through and processes it. And then the AI kind of gets a converged kind of view of like, okay, this is human language. This is what these people are talking about. And then it has all this statistical... When a human being says X, somebody else says Y or Z, or this would be a good thing to say or a bad thing to say. For example, you can detect emotional loading from text now. So you can kind of determine with the computer, you can kind of say this text reflects somebody who's happy because they're saying, oh, I'm having a great day versus this text is like I'm super mad. Therefore, it's upset. And so you could have the computer get trained on, okay, if I say this thing, it's likely to make humans happy. If I say this thing, it's likely to make humans sad. But here's the thing, it's all human generated text. It's all the conversations that we've all had. And so basically you load that into the computer and then the computer is able to kind of simulate somebody else having that conversation. But what happens is basically the computer is playing back what people say, right? It's not nobody, no engineer. The guy who went through this and did the whistleblower thing, he even said he didn't look at the code. He's not in there working on the code. Everybody who works in the code will tell you it's not alive. It's not conscious. It's not having original ideas. What it's doing is it's playing back to you things that it thinks that you want to hear based on all the things that everybody has already said to each other that it can get online. And in fact, there's all these ways you can kind of trick it into basic. For example, he has this example where he has it where basically he said, I want you to prove that you're alive. And then the computer did all this stuff for us alive. You can do the reverse. You can say, I want you to prove that you're not alive. And the computer will happily prove that it's not alive. It will give you all these arguments as to why it's not actually alive. And of course, it's because the computer has no view on whether it's alive or not. But it seems like this is all very weird. And for sure, we're in the fog of life. If it's not life, it's in this weird fog of what makes a person a person. What makes an intelligent, thinking human being that knows how to communicate able to respond and answer questions? Well, it does it through cultural context. Does it through understanding language and having been around enough people that have communicated in a certain way that it emulates that. Right. Yeah. So this is the real question. So this is where I was headed. The real question is, what does it mean for a person to think? Right. Like, that's the real question. And so and so let's talk about there's something called the Turing test, right, which is a little bit more famous now because the movie made about Alan Turing. So the Turing test basically in its simplified form, the Turing test is basically you're sitting in a computer terminal. You're typing in questions and then the answers are showing up in the screen. Is it 50 percent chance you're talking to a person sitting in another room who's typing the responses back? Is it 50 percent chance you're talking to a machine? You don't know. Right. You're the subject. And you can ask the entity on the other end of the connection any number of questions. Right. He will he or she or it will give you any number of answers at the end. You have to make the judgment as to whether you're talking to a person or talking to a machine. The theory of the Turing test is when a computer can convince a person that it's a person, then it will have achieved artificial intelligence. Right. Then it will be as smart as a person. But but that begs the question of like, OK, like how easy are we to trick? Right. Right. Like and in fact, so actually it turns out what's happened is actually true. What's happened is actually there have been chatbots that have been fooling people in the Turing test now for several years. The easiest way to do it is with a sex chatbot. Because they're the most gullible. Specifically to men. Of course. Of course. Women are like way less gullible. Probably fall for it a lot less. But men like you get a man on there with a sex chatbot like the man will convince himself he's talking to a real woman like pretty easily even when he's not. Right. And so just think of this as a slightly more you know, you could think about this as a somewhat more advanced version of that, which is look, if this thing if it's an algorithm that's been optimized to trick people basically to convince people that it's real, it's going to it's going to pass the Turing test even though it's not actually conscious, meaning it has no awareness, it has no desire, it has no regret, it has no fear, you know, it has none of the hallmarks that we would associate with being a living being like much less a conscious being. And so so this is this is the twist. And this is where I think this guy Google got got got got kind of strung up a little bit as they were held up is the computers are going to be able to trick people into thinking they're conscious like way before they actually become conscious. And then there's just the other side of it, which is like we have no idea. We don't know how human consciousness works. Like we have no idea how the brain works. We have no idea how to like we have no idea how to do any of this any any of this stuff on people. The most advanced form of medical science understands consciousness is actually anesthesiology, because they know how to turn it off. Right. They know how to click, you know, power up and then you have to power back on, which is also very important. But like they have no idea what's happening inside the black box. And we have no idea. Nobody has any idea. So so this is a parallel line of technological development that's not actually recreating the human brain. It's doing something different. It's basically training computers and how to understand process and then reflect back the real world. It's very valuable work because it's going to make computers a lot more useful. For example, self-driving cars. This is the same kind of work that makes us all driving car work. Yeah. So this is this is very valuable work. It will create these programs that will be able to trick people very effectively. Right. And so so there for example, here's what I would be worried about, which is basically like what percentage of people that we follow on Twitter are even real people. Right. Well, Elon is trying to get to the bottom of that right now. He's trying to get to the bottom of that, you know, specifically on that on that issue from the business, but just also think more generally, which is like, OK, if you have a computer that's really good at writing tweets, if you have a computer that's really good at writing angry political tweets or writing whatever absurdist, you know, humor, whatever it is like. And by the way, maybe the computer is going to be better at doing that than a lot of people are. You know, you could you can imagine a future Internet in which most of the interesting content is actually getting created by machines. You know, there's this new system, Dolly, you know, that's getting a lot of visibility now, which is this thing where you can type in any phrase and it'll create you computer generated art. Oh, I've seen that. Yeah. They've done some with me. It's really weird. Yeah. Yeah. You know, Chase Lappard, he's got a few of them that he put up on his Instagram. How does that work? Yeah. So it's a very similar thing. So basically what they do with the Google has one of these and open AI has one of these. What they do is they pull in all of the images on the Internet. Right. So if you think if you go to Google images or whatever, just do a search on any topic, it'll give you thousands of images, whatever. And then basically they pull in all the images. Yeah, that's me. Exactly. How bizarre. Yes. So that's so that's AI generated art. So that's AI generated art. That's a different program. That's just basically doing sort of psychedelic art. The Dolly ones are basically they're sort of composites where they will give you basically it's almost like an artist that will give you many different drafts. Yeah, that's another one of me. Yeah. So he the first one he go back to that place. Yeah, you just had it up. It's what does it say? It said what is it Joe Rogan facing the DMT realm insanely detailed intricate hyper hyper masculinist miss dark elegant ornate luxury elite horror creepy ominous haunting moody dramatic volumetric light 8K render 8K post hyper details. So they say that and then they enter all this stuff in and this is what comes out. And this is what comes out. Now. Holy shit. Yes. OK. So first of all, yes, it's incredible. Like that's amazing. It's an original work of art that is exactly the spec that I had to make my nose look like that. Doesn't really look like that, right? Not not today. It's it's a little off. I mean, I would say if that was an artist like I think you got the nose wrong and you made my jaw. Well, it's referencing these other artists. If you see at the end that it's actually it's a reference. It's probably pulling in portraits of other people from those artists and using a composite thing. Right. Exactly. The fact that it can make art now but see what it's doing right. So it's very impressive. I mean, the output is very impressive and the fact that it can do this impressive. But it's being told exactly what to do. Yes. It didn't have the idea that it was going to do that. It was told it was it's following instructions. Right. Right. So it's not it's not sitting there like a real artist dreaming up new artistic concepts. But here's the question because you were you were saying this before that it can trick people into thinking it's real. How do we know what's what is alive? So but this is the question. Like you can like what is a human consciousness interacting with another human consciousness? I mean, it is data. It is the understanding of the use of language inflection tone, the vernacular that's used in whatever region you're communicating with this person in to make it seem as authentic and normal as possible. And you're doing this back and forth like a game of volleyball. Right. This is what language is and a conversation is. If a computer is doing that. Well, it doesn't have a memory. Well, but it does have memory. Well, it doesn't have emotions. Is that what we are? I don't know. Because if that's what we are, then we're all that's all we are because the only difference is emotion and maybe biological needs like the need for food, the need for sleep, the need for, you know, the for touch and love and all the weird stuff that makes people people the emotional stuff. But if you extract that the the normal interactions that people have on a day to day basis, it's it's pretty similar. Yeah. Well, so here would be the way to think about it. It's like, what's the difference between an animal and a person? Right. Like, why do we grant people rights that we don't grant animals rights? And of course, that's a hot topic of debate because there are a lot of people who think animals should have more rights. But but fundamentally, we do we do have this idea. We have this idea of what makes a human distinct from a horse or a dog. Right. Is is self-awareness. Right. Is a sense of self, a sense of self being conscious. Right. You know, I take heart. I think therefore I am. Right. Right. And so at least we have this concept with this philosophical concept of consciousness being something that involves self-awareness. Like the computer is the computer is like I told you, the computer is quite capable of telling you it has self-awareness. Yeah. It's also quite capable of telling you it doesn't. It doesn't care. Right. It has no opinion on whether it has consciousness or not. And that's why I'm confident that these things are not conscious. They're not alive. But are these things are they just approach? It's just it's a program. It's a program. Yeah. But at what point in time does the program figure out how to write better programs? Right. At what point in time does the program figure out how to manifest a physical object that can take all of its knowledge and all the information that's acquired through the use of the Internet, which is basically the origin theme in Ex Machina. Right. Right. Right. That the super scientist guy, he's using his web browser, his search engine to scoop up all people's thoughts and ideas and he puts them into his robots. Which is basically what these companies are doing. Hopefully with a different result. Well, let me bring there's another topic. Hopefully. There's another topic. A friend of my Peter Teal and I, our CEO, he's like basically he's like, look, you know, civilization is declining. You can tell because all the science fiction movies are negative. Right. Like it's all dystopia. Nobody's got hope for the future. Everybody's negative. And my answer is just like the negative stories are just more interesting. Right. Nobody nobody makes the movie with like the happy A.I. Right. Like it's just not it. There's no drama in it. Right. So anyway, that's why I say hopefully it won't be Hollywood's dystopian vision. But here's another question, though, in nature of consciousness. Right. Which is another idea that Descartes had that I think therefore I am guy had is he had this idea of mind body dualism, which is also what Rick Kurzweil has with this idea that you'll be able to upload the mind, which is like, OK, there's the mind, which is like basically all of this, you know, some level of software equivalent coding, something, something happening and how we do all the stuff you just described. Then there's the body and there's some separation between mind and body where maybe the body is sort of could be arbitrarily modified or is disposable or could be replaced or replaced by a computer. It's just not necessary once you upload your brain. And of course, and this is a relevant question for for the for A.I. Because of course, the A.I. Dolly has no body. You know, GPT three has no body. Well, do we really believe in mind body? Do we really believe mind and body are separate? Like do we really believe that? And what the science tells us is no, they're not separate. In fact, they're very connected. Right. And a huge part of what it is to be human is the intersection point of of of of brain and mind and then brain to rest of body. For example, all the medical research now that's going into the influence of gut bacteria on behavior. Right. And then the sort of in the role of viruses and how they change behavior. And like and so basically like I think the most evolved version of this, the most sort of advanced version of this is like whatever it means to be human, it's some combination of mind and body is some combination of logic and emotion. It's some combination of mind and brain. It leads to us being the crazy, creative, inventive, destructive, innovative, caring, hating people we are. Right. The sort of mess. Yeah. The mess that is humanity. Right. Like that that's that's amazing. Like that that you know the four billion years of evolution that it took to get us to the point where we're at today is like amazing. And I'm just saying like we don't know we don't have the slightest idea how to build that.