4 ms·

Honest question as someone entirely outside of any AI research: what's the I in AI then ? My life experience is very alike pattern matching the current situatio

Honest question as someone entirely outside of any AI research: what's the I in AI then ? My life experience is very alike pattern matching the current situation to everything I've lived before sprinkled with chaotic emotional reactions due to outside events - heat, hunger, etc. I really don't feel like there's anything more than that when thinking, just the brain going through paths that one can recognize with time.

mjburgess 4y ago
If I ask, "how did you like new york?" and you answer the question, sure, you can have you `I`. That requires having gone to new york, having tastes, having judgements, having memory -- having a subjectivity which is unique and composed of the history of your experiences which can be related to novel environemnts. It requres existing in a shared world we can both co-refer to with our words. It requires being embedded in a space of social and physical significance. It isnt google. "Language models" just zip the internet and search for patterns in text already written about the world, and reports it. It's one step above a tape recorder on playback.
- mpoteat 4y ago
  If the model responded, "I have not visited New York because I am a computer", would you consider that to be a good step? If you require physical embodiment, that is fair - but I don't think there's any principled reason to expect a multi-modal neural network could not be embodied with appropriate sensors or actuators. Transformer models are universal function approximators. They do not "search" - in the same way that a calculator does not "search" when you ask it to perform an operation.
  - mjburgess 4y ago
    Being a universal fn approximator is a pretty trivial property. The question is (1) does a fn exist to approx? (in most cases: No); and (2) how is the aprox constructed? With NNs, fns are approximated by compressing a sample of points along an empirical dataset which stands in for the fn. In most cases, however, the fn doesnt exist. There is no fn from "Image->Animal". The world is ambiguous, functions do not describe it. The different `x`s genuinely correspond to the same `y`, here: the same image (occluded animal) could either be a dog or cat. There is no function to approximate. Even when there is, compressing empirical samples along a trivial interval is a terribly fragile method: consider sampling the addition function (x,y)->x+y over anything less than the infinite domains of its inputs. Starting with an empirical distribution of "the solved problem" will never build an intelligent system. Intelligence is what animals have because the world is ambiguous and not already ready-to-hand. Intelligence is what you do when there isnt "functions lying around" to approximate. I doubt almost anything in CSci is even relevant to solving this problem. It's largely a bioengineering one. Being the agent in the world which manufactures functions by disambiguating it.
    - int_19h 4y ago
      "Animal" is an arbitrary definition that exists in one's head. Unless you assume an idealist position, the only way it can be defined is by a function that runs in one's head. It might be a very complicated function - sufficiently complicated that we can't really articulate it and resort to "I know it when I see it", and possibly self-modifying - but the same person will still produce the same output consistently when given the same input.
  - tsimionescu 4y ago
    > Transformer models are universal function approximators. You have to be very careful with the wording here. The class of Transformer models is a universal (continuous) function approximator. A single Transformer model can only approximate 1 particular function. Also, backpropagation/gradient descent are not universal function approximator constructors - they can't be used to construct an ANN that approximates an arbitrary (continuous) function even with arbitrarily many random input/output pairs from that function's domain. For example, given a function that is f(x) = floor(x) if floor(x) is even; x if floor(x) is odd, defined for any `double x`, there is no guarantee that doing back-propagation over a random sample of (x, f(x)) pairs will yield the correct function approximator for some arbitrary precision (especially if you don't start with the right architecture).
    - mjburgess 4y ago
      People confuse the "representational capacity" of `recurse(AW^X)` with the algorithm which sets `W`.
- jcelerier 4y ago
  > That requires having gone to new york, having tastes, having judgements, having memory -- having a subjectivity which is unique and composed of the history of your experiences which can be related to novel environemnts. It requres existing in a shared world we can both co-refer to with our words. It requires being embedded in a space of social and physical significance. are you saying that I would not be considered an intelligent being if I had a jar in a brain left with my thoughts since my birth ? I don't think that this is what humans mean by intelligence... also I don't really see the fundamental difference between "tastes", "judgements" and "memory" - making a judgment, having an opinion on something, etc... is just stopping the brain from reasoning past a certain step and internally saying "this is good enough" most likely because this is tiring to our frail physical bodies. Like, I don't like eating eggs for instance, but that's just because I don't make the effort of convincing myself that I like them. There's no intelligence in this, just laziness.
  - mjburgess 4y ago
    Yes, you wouldnt be intelligent if you were literally a brain in a jar.. you'd have nothing to say. You dont have any "thoughts at birth". A brain absent stimulation from the environment, doesnt do anything. The brain-in-the-jar example presumes that the whole nervous system receives identical stimulation as-if it were embedded in the world. Incidentally, this is what people mean by intelligence... by since no one has studied the relevant science, people do not know what's required to provide it.
    - jcelerier 4y ago
      > The brain-in-the-jar example presumes that the whole nervous system receives identical stimulation as-if it were embedded in the world. I should have chosen a similar sounding thing, I really meant what I said, not the original brain-in-a-jar thought experiment. > Incidentally, this is what people mean by intelligence... by since no one has studied the relevant science, people do not know what's required to provide it. I don't understand how these two sentences work together: the word's meaning is defined by its usage. If tomorrow most people in the world says that being intelligent is defined by waking up at 6AM and doing 10 jumping jacks every day, then that is what the word will mean whether you want it or not. In my "human frame of reference", "Yes, you wouldnt be intelligent if you were literally a brain in a jar.." really does not square with how intelligence is talked about for instance so... ¯\_(ツ)_/¯

mjburgess 4y ago

If I ask, "how did you like new york?" and you answer the question, sure, you can have you `I`. That requires having gone to new york, having tastes, having judgements, having memory -- having a subjectivity which is unique and composed of the history of your experiences which can be related to novel environemnts. It requres existing in a shared world we can both co-refer to with our words. It requires being embedded in a space of social and physical significance. It isnt google. "Language models" just zip the internet and search for patterns in text already written about the world, and reports it. It's one step above a tape recorder on playback.

mpoteat 4y ago
If the model responded, "I have not visited New York because I am a computer", would you consider that to be a good step? If you require physical embodiment, that is fair - but I don't think there's any principled reason to expect a multi-modal neural network could not be embodied with appropriate sensors or actuators. Transformer models are universal function approximators. They do not "search" - in the same way that a calculator does not "search" when you ask it to perform an operation.
- mjburgess 4y ago
  Being a universal fn approximator is a pretty trivial property. The question is (1) does a fn exist to approx? (in most cases: No); and (2) how is the aprox constructed? With NNs, fns are approximated by compressing a sample of points along an empirical dataset which stands in for the fn. In most cases, however, the fn doesnt exist. There is no fn from "Image->Animal". The world is ambiguous, functions do not describe it. The different `x`s genuinely correspond to the same `y`, here: the same image (occluded animal) could either be a dog or cat. There is no function to approximate. Even when there is, compressing empirical samples along a trivial interval is a terribly fragile method: consider sampling the addition function (x,y)->x+y over anything less than the infinite domains of its inputs. Starting with an empirical distribution of "the solved problem" will never build an intelligent system. Intelligence is what animals have because the world is ambiguous and not already ready-to-hand. Intelligence is what you do when there isnt "functions lying around" to approximate. I doubt almost anything in CSci is even relevant to solving this problem. It's largely a bioengineering one. Being the agent in the world which manufactures functions by disambiguating it.
  - int_19h 4y ago
    "Animal" is an arbitrary definition that exists in one's head. Unless you assume an idealist position, the only way it can be defined is by a function that runs in one's head. It might be a very complicated function - sufficiently complicated that we can't really articulate it and resort to "I know it when I see it", and possibly self-modifying - but the same person will still produce the same output consistently when given the same input.
- tsimionescu 4y ago
  > Transformer models are universal function approximators. You have to be very careful with the wording here. The class of Transformer models is a universal (continuous) function approximator. A single Transformer model can only approximate 1 particular function. Also, backpropagation/gradient descent are not universal function approximator constructors - they can't be used to construct an ANN that approximates an arbitrary (continuous) function even with arbitrarily many random input/output pairs from that function's domain. For example, given a function that is f(x) = floor(x) if floor(x) is even; x if floor(x) is odd, defined for any `double x`, there is no guarantee that doing back-propagation over a random sample of (x, f(x)) pairs will yield the correct function approximator for some arbitrary precision (especially if you don't start with the right architecture).
  - mjburgess 4y ago
    People confuse the "representational capacity" of `recurse(AW^X)` with the algorithm which sets `W`.
jcelerier 4y ago
> That requires having gone to new york, having tastes, having judgements, having memory -- having a subjectivity which is unique and composed of the history of your experiences which can be related to novel environemnts. It requres existing in a shared world we can both co-refer to with our words. It requires being embedded in a space of social and physical significance. are you saying that I would not be considered an intelligent being if I had a jar in a brain left with my thoughts since my birth ? I don't think that this is what humans mean by intelligence... also I don't really see the fundamental difference between "tastes", "judgements" and "memory" - making a judgment, having an opinion on something, etc... is just stopping the brain from reasoning past a certain step and internally saying "this is good enough" most likely because this is tiring to our frail physical bodies. Like, I don't like eating eggs for instance, but that's just because I don't make the effort of convincing myself that I like them. There's no intelligence in this, just laziness.
- mjburgess 4y ago
  Yes, you wouldnt be intelligent if you were literally a brain in a jar.. you'd have nothing to say. You dont have any "thoughts at birth". A brain absent stimulation from the environment, doesnt do anything. The brain-in-the-jar example presumes that the whole nervous system receives identical stimulation as-if it were embedded in the world. Incidentally, this is what people mean by intelligence... by since no one has studied the relevant science, people do not know what's required to provide it.
  - jcelerier 4y ago
    > The brain-in-the-jar example presumes that the whole nervous system receives identical stimulation as-if it were embedded in the world. I should have chosen a similar sounding thing, I really meant what I said, not the original brain-in-a-jar thought experiment. > Incidentally, this is what people mean by intelligence... by since no one has studied the relevant science, people do not know what's required to provide it. I don't understand how these two sentences work together: the word's meaning is defined by its usage. If tomorrow most people in the world says that being intelligent is defined by waking up at 6AM and doing 10 jumping jacks every day, then that is what the word will mean whether you want it or not. In my "human frame of reference", "Yes, you wouldnt be intelligent if you were literally a brain in a jar.." really does not square with how intelligence is talked about for instance so... ¯\_(ツ)_/¯