hot take: introspection isn’t really real. you can’t access your internal state in any meaningful sense beyond what your brain chooses to present to you (e.g visual stimuli, emotions, etc), for reasons outside of your direct control. when you think you’re introspecting, what’s really going on when you think you’re introspecting is you have a model of yourself inside your brain, which you learn gradually by seeing yourself do certain things, experience certain stimuli or emotions, etc.
your self-model is not fundamentally special compared to any other models you have. it works the same way as your model of anyone or anything else, except you have way more data on yourself, and also you directly experience your own emotions and sensori stimuli, as opposed to having to infer them for other people. often your emotional brain sabotages your ability to understand yourself, but also it sometimes sabotages your ability to understand other people too (e.g groupthink, tribalism).
your self-model can diverge arbitrarily far from reality. when you’re emotionally unintegrated, you have a model of yourself that fails to understand how your emotions truly work, so you will systematically mispredict how you will actually behave, or try to fix yourself in ways that don’t work because you are misunderstanding the causes of your actions. further evidence comes from some of the split brain experiments which show people confidently hallucinating causes for their actions which are demonstrably untrue.
(for emotional integration in particular, misprediction is only half of the thing; the other part of the thing is sometimes you can get detached from stimuli and stop perceiving them entirely. this other part is not neatly explained by this hypothesis)
your self-model contains a nested self-model, but this isn’t special in any way. your models of other people have a model of yourself contained within (what does this person think about me?). your nested self model can also be arbitrarily wrong—it is very common to fail to understand the ways in which your top level self model is wrong. you probably don’t have a third level nested self model because that’s not very useful and very costly to maintain.
your inner monologue is not most of your thinking, and doesn’t give you faithful representation of your true thoughts. people without internal monologues do just fine. also, it’s very very common for people to lie to themselves.
I dont think this is just speculative/unfalsifiable. I claim that thinking of your “introspection” as self modelling will lead you to make better decisions irl. you can apply techniques for learning to model other kinds of knowledge. you can realize that cognitive biases that apply to other kinds of modeling also apply to modeling yourself.
there’s no clean boundary between your self model and your model of other things. if it’s useful to model e.g your phone, or gut bacteria, or glasses, or partner/close friends, in close interaction with your stimuli and actions, there is no sharp self boundary.
a regular computer program can truly introspect on itself in ways that humans cannot, but this is fundamentally not that interesting. it definitely doesn’t mean computers are conscious.
corollary: “introspection” is not in any way related to consciousness or moral patienthood, and it is uninteresting to ask whether AIs or nonhuman animals are capable of introspection for the purposes of determining things about consciousness and moral patienthood.
What experiences have you had that lead you to call this a ‘hot take’?
[I rephrased a few times to avoid sounding sarcastic and still may have failed; I’m interested in why it looks to you like others dramatically disagree with this, or in what social environment people are obviously not operating on a model that resembles this one. My sense is a lot of people think this way, but it’s a little socially taboo to broadcast object-level reasoning grounded in this model, since it can get very interpersonally invasive or intimate and lead to undesirable social/power dynamics.]
I heard that when people are in therapy, their self adapts to the school of psychotherapy. For example you start getting Freudian dreams if you are in Freudian therapy, but you start getting Jungian dreams instead if you are in Jungian therapy.
This seems to support the hypothesis that when we think we have discovered something deep inside us, often we have actually constructed it to fit our preconceptions.
(I suspect that Buddhism also mostly works this way. When Buddhists say that they can verify the truth of all Buddha’s words by introspection… on one hand, yes they can; on the other hand, if they instead believed in Jesus, they could verify that just as well. Asking yourself is like asking an LLM: whatever you believe is true, it will confirm.)
in my worldview this is very easily explained. if you do Jungian therapy your self model starts incorporating Jungian concepts for explaining your own brain. You didn’t change the way your brain works fundamentally, you just changed your own model of your brain. The same way that if you read a book on the biology of plants you’ll start viewing them in the lens of cells, and if you read a book on the ancient spirits associated with each plant you’ll start thinking of plants as being animated by the ghosts of our ancestors.
The big mistake happens when people think of their self model as actually genuinely introspection. Then, you might think that you’ve changed the shape of your mind instead of only changing your understanding of your mind.
Instead, I think the right way to figure out if your self model is correct is to make predictions about your future behavior and see if they come true; act based on your self model and see if you become more successful at life, or whether you mysteriously repeatedly fail in some way.
hot take: introspection isn’t really real. you can’t access your internal state in any meaningful sense beyond what your brain chooses to present to you (e.g visual stimuli, emotions, etc), for reasons outside of your direct control. when you think you’re introspecting, what’s really going on when you think you’re introspecting is you have a model of yourself inside your brain, which you learn gradually by seeing yourself do certain things, experience certain stimuli or emotions, etc.
your self-model is not fundamentally special compared to any other models you have. it works the same way as your model of anyone or anything else, except you have way more data on yourself, and also you directly experience your own emotions and sensori stimuli, as opposed to having to infer them for other people. often your emotional brain sabotages your ability to understand yourself, but also it sometimes sabotages your ability to understand other people too (e.g groupthink, tribalism).
your self-model can diverge arbitrarily far from reality. when you’re emotionally unintegrated, you have a model of yourself that fails to understand how your emotions truly work, so you will systematically mispredict how you will actually behave, or try to fix yourself in ways that don’t work because you are misunderstanding the causes of your actions. further evidence comes from some of the split brain experiments which show people confidently hallucinating causes for their actions which are demonstrably untrue.
(for emotional integration in particular, misprediction is only half of the thing; the other part of the thing is sometimes you can get detached from stimuli and stop perceiving them entirely. this other part is not neatly explained by this hypothesis)
your self-model contains a nested self-model, but this isn’t special in any way. your models of other people have a model of yourself contained within (what does this person think about me?). your nested self model can also be arbitrarily wrong—it is very common to fail to understand the ways in which your top level self model is wrong. you probably don’t have a third level nested self model because that’s not very useful and very costly to maintain.
your inner monologue is not most of your thinking, and doesn’t give you faithful representation of your true thoughts. people without internal monologues do just fine. also, it’s very very common for people to lie to themselves.
I dont think this is just speculative/unfalsifiable. I claim that thinking of your “introspection” as self modelling will lead you to make better decisions irl. you can apply techniques for learning to model other kinds of knowledge. you can realize that cognitive biases that apply to other kinds of modeling also apply to modeling yourself.
there’s no clean boundary between your self model and your model of other things. if it’s useful to model e.g your phone, or gut bacteria, or glasses, or partner/close friends, in close interaction with your stimuli and actions, there is no sharp self boundary.
a regular computer program can truly introspect on itself in ways that humans cannot, but this is fundamentally not that interesting. it definitely doesn’t mean computers are conscious.
corollary: “introspection” is not in any way related to consciousness or moral patienthood, and it is uninteresting to ask whether AIs or nonhuman animals are capable of introspection for the purposes of determining things about consciousness and moral patienthood.
What experiences have you had that lead you to call this a ‘hot take’?
[I rephrased a few times to avoid sounding sarcastic and still may have failed; I’m interested in why it looks to you like others dramatically disagree with this, or in what social environment people are obviously not operating on a model that resembles this one. My sense is a lot of people think this way, but it’s a little socially taboo to broadcast object-level reasoning grounded in this model, since it can get very interpersonally invasive or intimate and lead to undesirable social/power dynamics.]
the experience that led to calling it a hot take is i was arguing against someone who disagreed with this right before i wrote it up
What was their position? (to the extent that you can reproduce it)
I heard that when people are in therapy, their self adapts to the school of psychotherapy. For example you start getting Freudian dreams if you are in Freudian therapy, but you start getting Jungian dreams instead if you are in Jungian therapy.
This seems to support the hypothesis that when we think we have discovered something deep inside us, often we have actually constructed it to fit our preconceptions.
(I suspect that Buddhism also mostly works this way. When Buddhists say that they can verify the truth of all Buddha’s words by introspection… on one hand, yes they can; on the other hand, if they instead believed in Jesus, they could verify that just as well. Asking yourself is like asking an LLM: whatever you believe is true, it will confirm.)
in my worldview this is very easily explained. if you do Jungian therapy your self model starts incorporating Jungian concepts for explaining your own brain. You didn’t change the way your brain works fundamentally, you just changed your own model of your brain. The same way that if you read a book on the biology of plants you’ll start viewing them in the lens of cells, and if you read a book on the ancient spirits associated with each plant you’ll start thinking of plants as being animated by the ghosts of our ancestors.
The big mistake happens when people think of their self model as actually genuinely introspection. Then, you might think that you’ve changed the shape of your mind instead of only changing your understanding of your mind.
Instead, I think the right way to figure out if your self model is correct is to make predictions about your future behavior and see if they come true; act based on your self model and see if you become more successful at life, or whether you mysteriously repeatedly fail in some way.