Whatever way we demonstrate it, it isnt via Q&A. This is the worst form of pseudoscientific psychology you can imagine.
You're likewise neither high on cocaine just because you can type out an aggresive twitchy response.
Questions and Answers dont form the basis of empirical tests for physical properties, being neither valid tests nor reliable ones. The whole of psychology stands as the unreporducible example of this.
You can, in any case, replace any conversation with a dictionary from Q->A, turning each reply in turn into a hash table look-up.
Hash tables dont think, hash tables model conversations, thef. being a model of a conversation is not grounds to suppose consciousness. QED
Because we're interested in the underlying properties of a physical system, eg., people, and this systems happens to be able to provide extremely poor models of itself (Q&A).
We're not interested in people's extremely poor self-modelling which is pragmatically useful for managing their lives, we're interested in what they are trying to model: their properties.
The same is esp. true of a machine's immitation of "self-reports". We're now two steps removed: at least with people they are actually engaged in self-modelling over time. ChatGPT here isnt updating its self-model in response to its behaviour, it has no self-model nor self-modelling system.
To take the output of a text generation alg. as evidence of anything about its own internal state is so profoundly pseudoscientific it's kinda shocking. The whole of the history of science is an attack on this very superstition: that the properties of the world are simply "to be read from" the language we use about it.
Every advancement in human knowledge is preconditioned on opposing this falsehood; why jump back into pre-scientific religion as soon as a machine is the thing generating the text?
Experiments, measures, validity, reliability, testing, falasification, hypotheses, properties and their degrees....
This is required, it is non-negotiable. And what we have with people who'd print-off ChatGPT and believe it is the worst form of anti-science
> The whole of the history of science is an attack on this very superstition: that the properties of the world are simply "to be read from" the language we use about it.
This is something where I agree with you. Interestingly, non-naturalistic analytical metaphysics supposes it can do just that.
Philosophy is continuous with science in my view and hence in what words express
That is, in the use of words. Not in words as objects nor words as mirrors --/
this is the road to non-realist spiritualism
I don't have a problem with a person who maintains a non-scientific world view and with electrical AGI mumbojbo
But
of course, few do. They think that they're empiricists, scientists and in the side of some austere hard look at human beings
This is just anti-human spiritualism , it isn't science
what makes me vicious on this point is the sense of injury in what these ideas should be about. in my own mildly aristotleian materialist religion
How awful to overcome one long veil of tears, only to drape another one on -- these people are capable of seeing past human folly, but
fall right into another kind
it's disappointing--- we're animals which are both far much more than electric switches and far
less
this new electic digital spiritualism is a PR grift which i'd prefer dead
On the one hand, you express strong support for empiricism and the scientific method, but on the other, you express strong beliefs on how things must be without offering any empirical justification for them.
>We're not interested in people's extremely poor self-modelling which is pragmatically useful for managing their lives, we're interested in what they are trying to model: their properties.
>Experiments, measures, validity, reliability, testing, falasification, hypotheses, properties and their degrees....This is required, it is non-negotiable.
Whoa, whoah, whoah, hold on there.
Who says that Q&A in Psychological Research doesn't involve "Experiments, measures, validity, reliability, testing, falasification, hypotheses, properties and their degrees...."
?
Where are you coming from? Your responses don't sound very scientific. You don't sound like you're even aware of the different research methods within neuroscience and cognitive psychology. Your responses sound like someone who wants to be perceived as supporting a scientific approach, but doesn't understand how to actually do these things.
This is why I quized you and gave you the chance to respond about your issues with Q&A in psychological research. You just came back with surface level platitudes. which doesn't lend much confidence to the ideathat you have anything other prejudice.
Go talk to a neuroscientist, a cognitive psychologist, you need to catch up and quick if you want to speak on these topics.
> Hash tables don't think, hash tables model conversations...
They don't think, and they don't get very far in modeling conversations, either. Even the current LLMs are strictly better at modeling conversations than any hash table.
Actually since GPT is a function that takes a few thousand tokens as input and produces a deterministic probability distribution for ‘next token’, you could in theory just enumerate all possible input sequences, and precache all the corresponding results in a lookup hashtable.
That hashtable would be ridiculously larger than the weighted model that constitutes GPT itself. But it at least theoretically could exist.
GPT, for all its clever responses, could be replaced with that hashtable with no loss of fidelity. So it is not ‘better’ than a hashtable.
It is much better than a hashtable of equivalent size.
You could, in principle, use a hash table to implement what an LLM has 'learned' about token occurrences, but LLMs do not return the same response to a given prompt, while hash tables do. Consequently, Mjburgess' attempt to dismiss LLMs as mere hash tables is a flawed argument by analogy.
And even if it were a justifiable analogy for LLMs, it does not follow that it applies to Turing-equivalent devices in general. A hash table is unequivocally not a Turing-equivalent device, even though it can be implemented by one.
In fact, the more one argues that LLMs are ordinary technology, the greater the challenge they present to the notion that things like conversational language use are beyond such technologies. The most interesting thing about these models, in my opinion, lies in figuring out what their conceptual simplicity says about us.
I extended my reply before seeing your response, and my response to your reply is largely contained within that.
LLMs do not just return a probability distribution (and their prompts are not questions about probability distributions.) It would be a category error to conflate something that is part of how they work for the totality of what they do.
When you say "GPT, for all its clever responses, could be replaced with that hashtable with no loss of fidelity", where does one get all the prompts that will be given to it, in advance? A hash table does not respond to any input for which it has not already been given a response to return.
I don’t think this is obviously true… We don’t know for certain the nature of reality, and cannot positively claim that a hash table would give the same results as a sufficiently advanced predictor, because it supposes the predictor is deterministic. Assuming the predictor is advanced enough to be considered intelligent, that might not be the case.
Is your point that we can only demonstrate consciousness through pharmaceuticals or surgery?
reply