[ExI] LLM's cannot be concious

Jason Resch jasonresch at gmail.com
Thu Mar 23 23:09:23 UTC 2023


On Thu, Mar 23, 2023, 6:39 PM Adrian Tymes via extropy-chat <
extropy-chat at lists.extropy.org> wrote:

> On Thu, Mar 23, 2023 at 1:02 PM Jason Resch via extropy-chat <
> extropy-chat at lists.extropy.org> wrote:
>
>> Others had argued on this thread that it was impossible to extract
>> meaning from something that lacked referents. it seems you and I agree that
>> it is possible to extract meaning and understanding from a data set alone,
>> by virtue of the patterns and correlations present within that data.
>>
>
> With the caveat that referents are themselves data, so if we include
> appropriate referents in that data set then yes.  Referents are often
> referenced by their correlations and matching patterns.
>

I don't understand what you are saying here.



>
>>
>> I am not convinced a massive brain is required to learn meaning. My AI
>> bots start with completely randomly weighted neural networks of just a
>> dozen or so neurons. In just a few generations they learn that "food is
>> good" and "poison is bad". Survival fitness tests are all that is needed
>> for them to learn that lesson. Do their trained neural nets reach some
>> understanding that green means good and red means bad? They certainly
>> behave as if they have that understanding, but the only data they are given
>> is "meaningless numbers" representing inputs to their neurons.
>>
>>
>>
>>> Just because one type of AI could do a task does not mean that all AIs
>>> are capable of that task.  You keep invoking the general case, where an AI
>>> that is capable is part of a superset, then wondering why there is
>>> disagreement about a specific case, discussing a more limited subset that
>>> only contains other AIs.
>>>
>>
>> There was a general claim that no intelligence, however great, could
>> learn meaning from a dictionary (or other data set like Wikipedia or list
>> of neural impulses timings) as these data "lack referents". If we agree
>> that an appropriate intelligence can attain meaning and understanding then
>> we can drop this point.
>>
>
> I recall that the claim was about "no (pure) LLM", not "no (general)
> intelligence".
>

My original claim was for an intelligent alien species.


> Also there  is  a substantial distinction between a dictionary or
> Wikipedia, and any list of neural impulses.  A pure LLM might only be able
> to consult a dictionary or Wikipedia (pictures included); a general
> intelligence might be able to process neural impulses.
>

In all cases it's a big file of 1s and 0s containing patterns and
correlations which can be learned.


>
>> There is no task requiring intelligence that a sufficiently large LLM
>> could not learn to do as part of learning symbol prediction. Accordingly,
>> saying a LLM is a machine that could never learn to do X, or understand Y,
>> is a bit like someone saying a particular Turing machine could never run
>> the program Z.
>>
>
> And indeed there are some programs that certain Turing machines are unable
> to run.  For example, if a Turing machine contains no randomizer and no way
> to access random data, it is unable to run a program where one of the steps
> requires true randomness.
>

Randomness is uncomputable. And I would go so far to say say true
randomness doesn't exist, there is only information which cannot be guessed
or predicted by certain parties. This is because true randomness requires
creation of information but creation of information violates the principal
of conservation of information in quantum mechanics.

In any case my point wasn't that everything is computable, it's that the
universality of computation means any Turing machine can run any program
that any other Turing machine can run. The universality of neural networks
likewise implies not that every function can be learned, but any function
that a neutral network can learn can be learned by any neural network of
sufficient size. Our brains is fundamentally a neural network. If our
brains can learn to understand meaning then this should be in the scope of
possibility for other neural networks.

Much has been written about the limits of psuedorandom generators; I defer
> to that literature to establish that those are meaningfully distinct from
> truly random things, at least under common circumstances of significance.
>

I am quite familiar with pseudorandom number generators. They are a bit of
a fascination of mine.


> One problem is defining when an AI has grown to be more than just a LLM.
> What is just a LLM, however large, and what is not just a LLM (whether or
> not it includes a LLM)?
>

That's a good question. I am not sure it can be so neatly defined. For
example, is a LLM trained on some examples of ASCII art considered having
been exposed to visual stimuli?

Jason
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230323/30872669/attachment.htm>


More information about the extropy-chat mailing list