5

I am a computer science student new to philosophy.

I have been thinking about the Chinese Room Argument and its replies and have tried to think of it as a different, more physical analogy.

Imagine a rock falling from a mountain. The instructions are encoded in the paths of the mountain. The position on top from where the rock is set to fall is the input of the system and the position(or some other encoding) where it lands is the output. These paths are carefully carved just as the weights of a neural network are determined based on data observations. These paths can be arbitrarily complex, analogous to the syntactic manipulations in neural networks.

Assuming a version of pancomputationalism sufficient to enable the rock-mountain system to perform computations, I hypothesize that a sufficiently complex set of paths could produce outputs that simulate understanding.

My question is: can an arbitrarily complex mountain ever achieve understanding like of those in humans? If this mountain were capable of producing outputs that accurately mimic understanding Chinese, analogous to the room producing correct Chinese sentences, would the mountain itself be said to understand Chinese? Is there a fundamental difference between the syntactic manipulations in a physical system capable of computing like this and those in our brains?

Improve this question
5
  • 1
    This might inspire you. plato.stanford.edu/entries/computation-physicalsystems
    – J D
    Commented Jul 8 at 23:59
  • Your question implicitly assumes the mountain is dead/inanimate. Religions cultures around the world have loved and worshipped Sacred rivers and Magical Trees and The Sun. So why not a mountain?
    – Rushi
    Commented Jul 9 at 6:47
  • See Understanding: "To have episteme one must not only know a thing, one must also grasp its cause or explanation. This is to understand it: to know in a deep sense what it is and how it has come to be. (Jonathan Lear, 1988)"
    – Mauro ALLEGRANZA
    Commented Jul 9 at 7:23
  • To "compute" is not to understand: abacus computing was available well before AI.
    – Mauro ALLEGRANZA
    Commented Jul 9 at 7:24
  • 1
    I propose to approach the subject pointing to 3 distinct ontological levels. On the math level in theory every cybernetic phenomenon can be modeled. It will exist somehow without anything happening. The second level involves running models as computer programs or biological systems. Something is happening on another ontological domain. But the third level is phenomenological conscousness which can arise in conjunction with the second level in some but not all systems. This idea is developed in this very long conversation.
    – MichalRyszardWojcik
    Commented Jul 29 at 9:41

1 Answer 1

Reset to default
3

Producing outputs is not equivalent to understanding, which the Chinese Room Argument proves. In fact, generative AI has shown us a modern example of exact (or perhaps near exact) mimicry not being remotely close to true understanding.

Take the example of a child mimicking its parents, and then understanding them as part of the process. The reason they can do this is because of their sense of curiosity and capacity for self-growth, neither of which can be said to apply to a room or a mountain, no matter how complex the computations.

We see this today in the aforementioned generative AI models. They are trained solely on inputs and outputs, with no actual 'reasoning'.

In order to truly understand something, a computer/mountain/room would probably require programmed senses of curiosity and ability to grow naturally. This could eventually happen of course, in which case the question would need to be reassessed.

Taken as it is, your question begs a no.

Improve this answer
4
  • An ability to naturally grow doesn't imply understanding. A mountain with changing or growing paths will still be doing just syntactic manipulation. And as for the "programmed senses of curiosity" I am not sure what you mean. The mountain might be able to simulate curiosity, but does it possess the subjective experience of curiosity?
    – Muneeb Ur Rahman
    Commented Jul 8 at 23:21
  • Correct, but to grow in the way that I described (admittedly a bit vaguely) could perhaps lead to something we would regard as consciousness. The mountain would need a way to grow in the sense that its knowledge base or pool could grow on its own, without outside manipulation of its paths/computational abilities/etc. On curiosity, we don't know if that is programmable just yet. I was just saying that without a natural or programmed sense of curiosity and the aforementioned capacity for growth, the mountain cannot be said to understand anything.
    – Aibaahl
    Commented Jul 9 at 14:08
  • You are missing the point. Neural networks store "knowledge" in their parameters. Efforts are being made to make neural networks that can learn during inference. That wouldn't change anything in this context. The question is fundamentally about the difference between syntactic manipulation and genuine understanding or consciousness. The rock-mountain system, regardless of its ability to grow or simulate curiosity, is still just engaging in semantic manipulation.
    – Muneeb Ur Rahman
    Commented Jul 11 at 12:16
  • The reason those are relevant is because a sense of curiosity and ability to grow naturally would put the AI far closer to a human mind than any previous iteration. At that point, it would raise the difficult questions of understanding, knowledge, and consciousness. As it is per the question the mountain isn't complex enough to be said to understand anything. Only when it becomes alike to a human mind is that going to be relevant.
    – Aibaahl
    Commented Jul 11 at 14:39

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .