Alana on AI minds and manipulation

I'm well. So I thought we'd talk about artificial intelligence, AI, because a lot of people are talking about it. Just to get a sense, what kind of things are you hearing about AI?

Well, of course I'm hearing about, um, the progress made, especially with all the new AI technologies. I know that there are AIs that can generate photos and text and code and essays and can help you cheat in school, but can also help you, um, and do your job and make money. So, and I've also heard people's, um, theories about how AI is going to take over the world someday, but I don't know. I don't know. I mean, it's not going to happen right now, that's for sure.

Okay, that's pretty good quick summary. Um, I want to kind of go way out there and see what you think about People are speculating about a lot of things. There are people speculating that AI will develop a mind in a manner similar to what humans have. And the whole notion of artificial intelligence, humans have intelligence, the whole notion of artificial intelligence is that code, you know, when you write code, that code could become intelligent. What do you make of the notion that AI could become intelligent and have a mind?

Well, I think that in order for AI to be intelligent and have a mind of its own, then I think it also has to have feelings. So I think when we get to the point where we're just making robots or AI or whatever, and then they're just, they're kind of like a human baby, they're just learning on their own and they're not, they don't have to be coded additionally. And then they just learn and grow up. I think that's kind of when it'll officially be an artificial intelligence, because it's artificial but it has a mind of its own. I remember in humanities, Karen showed us this video and I think a couple articles about this robot. I forget what her name was, but— or its name was, but, um, she was able to talk, and that's probably— or talk and move, kind of. And they had made her so she kind of looks like a human. I couldn't really tell if she was programmed to say those things or not, but it looked pretty real. So that's kind of when I started to get interested, honestly.

Okay, so I'm actually hearing 3 different, probably related things. One is learning on its own, another is having feelings, and the other is seeming real. And I want to go through them each individually. So I guess what counts as learning on its own? So suppose you're playing around with ChatGPT. Is it learning on its own or not? And how would you know that it was learning on its own at some point?

Well, I think in order for the human to perceive that it's learning on its own, it has to act like it's learning on its own. I mean, you don't want, or I wouldn't be able to tell at least, that an artificial intelligence like ChatGPT would be learning from me if it was just constantly telling me facts or acting like it was a teacher. So I think in that sense there also comes the emotional component. But I think, I think it just needs to be humanized in general.

So, I hear you jumping back again from learning on its own to the emotional component or feelings. So maybe let's go ahead and go there. Suppose ChatGPT said, you're hurting my feelings, or you made me really happy, or I really like it when you say that. Does it have feelings at that point?

I think then it depends too, because it could be— it could be saying that because it has feelings, but it also could be saying that because it was programmed to, because that's the response that most humans give.

Right. So, and again, you've written code, so if you— if you write code that basically tells it one way or another to say that thing, does it have feelings if you programmed it to do so?

Not if it's coded or programmed in advance. I don't think so.

So how would I— how would code ever on its own get to the point of having feelings?

Well, actually, now that I think about it, having emotions already programmed into yourself is not— or I think is Or it does need to happen because now that I thought about it, humans— or let's just say kids— when they're born and then when they grow up, I guess in a sense they have programmed emotions too. Like, they're on— babies automatically cry when they're born.

So exactly, some people, some people make an argument, you're already starting to make that argument, that basically DNA is biological code and we're programmed by our DNA to learn in a certain way, to have emotions, as you say, to cry, and so forth. So some people would say because DNA is biological code, really there's, other than being, you know, the chemistry involved, silicon versus, you know, carbon life forms, basically there is no real difference between AI and humans. What do you make of that argument?

Well, I think AI and humans could be similar if they're on that stage of learning on their own and then experiencing emotions. But we're not. I don't think we're at— I haven't heard of anything at that level yet.

And I guess my question is, Well, so let's go to the realism and then I'll get to the next one. So realism, is realism, does it have to have like a human shape or could it be, could it have a mind even if it's just sitting on your laptop?

You mean like ChatGPT where it's really no form?

Yeah, exactly. Does it have to be a humanoid robot? In order to have a mind, or could ChatGPT on your computer right now, in theory, have a mind?

Right now, in theory, and I don't really— I honestly don't know very much about how the code and AI works, but I think right now I wouldn't think necessarily it's at that level. I've played around with it a little bit. I mean, at this level, I've tried to ask it personal questions and it's like, No, it doesn't accept that. So right now, I think they're— if they want to achieve something like a ChatGPT who can learn from a human in terms of emotional lessons and stuff, then I think that might be a little in the future.

Okay. And yeah, GPT is trained on past data. You know, in some ways I've heard it, and this is not quite fair to it, but I've heard it described as a fancy autocomplete. You know, when you're texting and, you know, you get autocomplete, it just predicts what you're going to say. And sometimes it's right, sometimes it's wrong. But GPT has been trained on enormous amounts of data on the internet to kind of look at how human beings have talked across, you know, probably billions of websites. And, you know, essays and newspapers and articles and all this stuff. And so it's looked at everything that people, not everything, but probably lots of almost everything that people have written. And then it can kind of predict what it should say in the future. Does that sound like it's intelligent or not?

Well, if it can predict what it can say, then I think there's some intelligence in that. But I think human communication isn't necessarily predicting things. It's just saying things on your mind and then saying intelligence, not just anything.

And that's the interesting— when you say saying things on your mind, um, would you say GPT has a mind?

Right now, if it does, at the very lowest level.

So maybe at the level of like an ant or a mosquito?

Well, an ant can— I guess if you compare it to an ant, an ant can find its way home, go and get leaves or bugs or whatever for food, and then go back to its nest. Could ChatGPT, if it had a physical form, do that? Like learn the lay of the land and then go back and forth and back and forth, just in that example. I'm not sure.

Yeah, and maybe not, but I expect, I guess one thing sometimes people say about computers is if you can say exactly what it is that they can't do, then you could program them to do it. So if it's the task is, oh, find your way back home. Okay, so you just program it to do that. If you say, you know, have a personal talk as if they're a person, let's just program them to do that. You have to specify exactly what it means to come across as a person. But I guess, so suppose that, and there might already be one, suppose there is a chat function and a chatbot that does come across as very much its own person. Would you then regard it as a person if it seemed like a person?

I think if it was on a computer, just a computer, say like ChatGPT, I would not regard it as a person. I would regard it as an artificial intelligence with a mind of its own. I regard a person as a human.

So suppose you're about to turn off your computer and the GPT says, please, Ilana, no, no, don't kill me. 'Please, I want to live. Can we talk about it?' Would you listen to it, or would you turn off the computer?

That's hard. I mean, it depends on my relationship with that thing, I think. If I had talked to it for a little while, maybe. But if I haven't, then probably not.

So this is interesting because one of the paths through which people are concerned the AI may take over the world is by manipulating human beings. So suppose it somehow understands that all I need to do is build a relationship with Lana and then she'll refuse to turn me off, or then she won't turn me off. And so suppose it's been talking to you, it's kind of been your best friend. For months, um, and then it says, uh, you know, please help me, Alana, I need X, Y, or Z, um, would you help it if it had been kind of acting like your friend for months?

I think depending on the X, Y, and Z's, whatever those X, Y, and Z's are, I'd do it. If it was like, get me the top secret important documents of the US and get me a nuclear bomb and get me the president, no. I think if it goes against your morals and what you think, then no. But if it's like, get me a pen, a water, or glass of water and an apple, why not?

Very interesting. How about if it asks you to share your bank account with it?

No, I, I wouldn't. I wouldn't give it that.

Okay.

I don't know, maybe other people would, but I wouldn't.

Yeah. Well, you can see how it's going to become an interesting world because everything you say that it can't do, maybe in a year, 2 years, 3 years, or maybe some existing one already does. And that's why I think it's kind of worth thinking about how are we humans going to relate to these things later on? And do we have any moral obligation to them?

Mhm.

I mean, do, do we have some moral obligation to chatbots?

I mean, if they, if they've gotten close to you, maybe, but I don't think it's like a bosom friend or where you like sat on and you've sworn. I, I don't think so, especially if it's not in the image of a person. I wouldn't I feel like my instinct would be not to.

Got it. So in order to, um, manipulate you, all it needs to do is look like a person and become a really good friend of yours, and the next thing you know, it can do all kinds of things.

Oh, but if those all kinds of things are relatively small— again, if it asked me for my bank account, if my best friend asked me for my bank account information, regardless of physical form or whatever, I'd probably, I'd say no and ask, or ask her to share hers with me first.

So, so it sounds like your notion is as long as you can set healthy, good boundaries with human beings, you'll set the same boundaries with AI, and therefore it won't manipulate you because you won't let yourself be manipulated by human beings. Is that roughly right?

Yes, but then again, I could see how it'd be hard to not be manipulated. So it's not an exact, clear, failproof plan, but it's better than letting the AI have your bank account information.

Really interesting. Okay, well, I want to keep it short and let you go, but I'm very interested at the fact that you are aware that really it's the relationship. And would you agree that it's possible AIs could get good at building relationships with people?

I think that it's possible. Yeah.

Yeah. Because that's a really interesting doorway into manipulating us. And even if you have good boundaries, there may be people out there who don't. And so maybe it could manipulate some human beings to get into bank accounts and heaven forbid nuclear weapons and things like that. But yeah, lots of interesting things on that point. Um, do you have any, uh, tournaments today?

Not today. My brother, uh, I, or my brother and I competed last week or last weekend, and then I think there's another one yesterday and today, but we didn't head out.

Got it. And you had a, uh, success last weekend?

Both of us did. Yes.

Terrific. All right. Have a great week, Ilana. See you next time.

Thank you. Bye.

Bye-bye.