To me, the biggest argument against llms being conscious is how often they change their “mind.” A conscious being would have an awareness of its own thoughts and would both have opinions of its own and also an epistemological framework (though not necessarily a good one) for why something is or isn’t true.
I’ll start believing llms are conscious when they can argue at least as well as a four year old can.
I think there’s something to this, but I would push it up one more level into a sense of self. You can train an LLM to be stubborn (a lot of early adopters might remember LLMs arguing vehemently that current news wasn’t correct because the LLMs didn’t have access to the internet and their training data was months old) and honestly, someone who didn’t change their mind when presented new, conflicting information would make me wonder if they’re less conscious rather than more.
But when these sorts of tests come up, I always think about pets. My cat can’t argue, clearly has very few, if any thoughts (although she’s my brilliant, wonderful, devious, awful, sweet literal daughter), and her opinions largely revolve around me letting her outside and giving her treats, but I would definitely consider her conscious.
Personally, and now I’m really off the deep end of unsubstantiated thinking, I think LLMs are very much like a small talk machine. Like, you know how when you’re chatting casually and you’re not really thinking about the words you’re saying, they’re just vaguely in the shape of expected conversation? I think LLMs are that. I mean, like a fancy version of it, but kinda the same thing–connected words and sentences that all definitely read mostly correct, but don’t really have actual thought behind them.
I think there are a bunch of components missing to complete the picture so that a majority of people actually see consciousness in a machine, but I definitely think LLMs are a piece of it. It’s a shame so many resources are dumped into perfecting what is (again, in my opinion) essentially just a small-talk machine when we could be developing literally all of the rest of it instead. If there were a proper ‘world model’ to understand how and why things move in the world, a global research database with which to pull factual data from exclusively and double check answers against, and a higher level narrator to pull all of this together and consider before replying, I think I’d be convinced that we’re seeing consciousness. Not human consciousness, but something else, not unlike how other animals might compare to humans.
I wonder what your cat would say if she could talk. I imagine she would surprise you.
But yeah, obviously you can train an llm to be stubborn, but that’s only half of it. The algorithm doesn’t have a sense of self and doesn’t have a sense of something being “true” or “false.” So it has no heuristics for whether or not it should agree with or disagree with you and doesn’t have opinions of its own.
Personally I happen to think human consciousness isn’t that different from other animals, we just happened to get the right combination of long lifespan + language + tool use + very social and cooperative so that we could develop generational knowledge on a societal level through oral traditions and writing.
Humans aren’t just smart because our brains are better, we’re smarter because we’re educated.
I think I pretty much agree with you on everything you’ve said here. The sense of self is something I mentioned as well, I just wanted to distinguish it from having an opinion because that can get kinda muddled with sticking with an argument rather than actually having the position.
And to be clear, I’m not saying my cat doesn’t have thoughts. Maybe I’m anthropomorphizing a bit, but I definitely feel like she does have ideas and tries to convey them. She makes a specific “don’t do that/stop/I don’t like that” noise and a “happy/good/more” noise and has different tones when she wants attention or when she wants me to do something for her. I can see it in her when she sees something she wants to knock over and she’s ‘thinking about it’. (Edit: sorry, here’s where I realize I wasn’t very clear. My last comment was more a joke about my cat being a dumbass but definitely being conscious; I didn’t mean to give the impression that I don’t think she has thoughts. More that I see glimmers here and there that I would guess are thoughts, but don’t really have any kind of real proof beyond what I can relate to.)
I wouldn’t say human consciousness is that different from animals, either, I just think the capacity and predisposition might be slightly different for different types of thought. I think we can see the same in people, though, too. Something that’s always really blown my mind is that people can literally think differently. Like, some people don’t have an internal narrator. Some people can’t visualize images, or visualize new images. Some people can’t picture 3-Dimensional images. Some people exclusively have feelings instead of those things.
I definitely agree that age and language and education and capabilities (like our dexterity) all play into it, but I don’t think that is necessarily the only thing.
You should check out the button things that let dogs and cats talk. Some of those animals are surprisingly opinionated and I think you’d enjoy seeing it for yourself
To me, the biggest argument against llms being conscious is how often they change their “mind.” A conscious being would have an awareness of its own thoughts and would both have opinions of its own and also an epistemological framework (though not necessarily a good one) for why something is or isn’t true.
I’ll start believing llms are conscious when they can argue at least as well as a four year old can.
I think there’s something to this, but I would push it up one more level into a sense of self. You can train an LLM to be stubborn (a lot of early adopters might remember LLMs arguing vehemently that current news wasn’t correct because the LLMs didn’t have access to the internet and their training data was months old) and honestly, someone who didn’t change their mind when presented new, conflicting information would make me wonder if they’re less conscious rather than more.
But when these sorts of tests come up, I always think about pets. My cat can’t argue, clearly has very few, if any thoughts (although she’s my brilliant, wonderful, devious, awful, sweet literal daughter), and her opinions largely revolve around me letting her outside and giving her treats, but I would definitely consider her conscious.
Personally, and now I’m really off the deep end of unsubstantiated thinking, I think LLMs are very much like a small talk machine. Like, you know how when you’re chatting casually and you’re not really thinking about the words you’re saying, they’re just vaguely in the shape of expected conversation? I think LLMs are that. I mean, like a fancy version of it, but kinda the same thing–connected words and sentences that all definitely read mostly correct, but don’t really have actual thought behind them.
I think there are a bunch of components missing to complete the picture so that a majority of people actually see consciousness in a machine, but I definitely think LLMs are a piece of it. It’s a shame so many resources are dumped into perfecting what is (again, in my opinion) essentially just a small-talk machine when we could be developing literally all of the rest of it instead. If there were a proper ‘world model’ to understand how and why things move in the world, a global research database with which to pull factual data from exclusively and double check answers against, and a higher level narrator to pull all of this together and consider before replying, I think I’d be convinced that we’re seeing consciousness. Not human consciousness, but something else, not unlike how other animals might compare to humans.
I wonder what your cat would say if she could talk. I imagine she would surprise you.
But yeah, obviously you can train an llm to be stubborn, but that’s only half of it. The algorithm doesn’t have a sense of self and doesn’t have a sense of something being “true” or “false.” So it has no heuristics for whether or not it should agree with or disagree with you and doesn’t have opinions of its own.
Personally I happen to think human consciousness isn’t that different from other animals, we just happened to get the right combination of long lifespan + language + tool use + very social and cooperative so that we could develop generational knowledge on a societal level through oral traditions and writing.
Humans aren’t just smart because our brains are better, we’re smarter because we’re educated.
I think I pretty much agree with you on everything you’ve said here. The sense of self is something I mentioned as well, I just wanted to distinguish it from having an opinion because that can get kinda muddled with sticking with an argument rather than actually having the position.
And to be clear, I’m not saying my cat doesn’t have thoughts. Maybe I’m anthropomorphizing a bit, but I definitely feel like she does have ideas and tries to convey them. She makes a specific “don’t do that/stop/I don’t like that” noise and a “happy/good/more” noise and has different tones when she wants attention or when she wants me to do something for her. I can see it in her when she sees something she wants to knock over and she’s ‘thinking about it’. (Edit: sorry, here’s where I realize I wasn’t very clear. My last comment was more a joke about my cat being a dumbass but definitely being conscious; I didn’t mean to give the impression that I don’t think she has thoughts. More that I see glimmers here and there that I would guess are thoughts, but don’t really have any kind of real proof beyond what I can relate to.)
I wouldn’t say human consciousness is that different from animals, either, I just think the capacity and predisposition might be slightly different for different types of thought. I think we can see the same in people, though, too. Something that’s always really blown my mind is that people can literally think differently. Like, some people don’t have an internal narrator. Some people can’t visualize images, or visualize new images. Some people can’t picture 3-Dimensional images. Some people exclusively have feelings instead of those things.
I definitely agree that age and language and education and capabilities (like our dexterity) all play into it, but I don’t think that is necessarily the only thing.
Oh ok 😅
Yeah with that clarification I think we agree.
You should check out the button things that let dogs and cats talk. Some of those animals are surprisingly opinionated and I think you’d enjoy seeing it for yourself