[ExI] GPT-4 on its inability to solve the symbol grounding problem

Giovanni Santostasi gsantostasi at gmail.com
Mon Apr 10 22:24:08 UTC 2023

Could it be that we become conscious when we actually started to use
language and had to identify and separate I from you?
If so then it is not a surprise that from a LLM consciousness can emerge
because language needs subjects and objects. LLM had to come up with
grammar and semantics (that are not programmed in and not sure people get
it) and therefore it is not impossible at all that a sense of self could
emerge in the same manner from a deep understanding of how language works.
In fact, this is probably the secret of what is going on here, language has
so much structure, information, and meaning that training a system with
billions of parameters on such an information pool gives rise to
understanding, intelligence, and a rudimentary form of consciousness.


On Mon, Apr 10, 2023 at 3:11 PM Giovanni Santostasi <gsantostasi at gmail.com>

> It is obvious that there are several levels of consciousness. Tononi (I
> actually worked in his lab for some time) and Kock have a theory of
> consciousness that can actually attribute a number to how much
> consciousness there is in a system (of course if you agree on their
> definition of consciousness). I think it is one of the best (even if not
> complete) approaches to understand scientifically what consciousness is.
> According to this theory, consciousness is present in everything even a
> rock. It is just the amount of consciousness that matters. By the way, what
> they try to calculate is something called Integrated Information which is a
> quantity that measures the level of information present and how it is
> connected between the parts of the system. Too much connectivity doesn't
> seem good for consciousness and too little is also not good. It seems there
> is a perfect spot in between that creates the highest level of
> consciousness (that the authors equate to this quantity). I don't think a
> scalar (a single number) can capture what consciousness means but maybe it
> is a good proxy for it and gives us some interesting insights. The beauty
> of it is that you can use it to calculate the Integrated Information
> indirectly (because it is very complicated to calculate all the connections
> in a brain both theoretically and from an experimental point of view). But
> you can use proxies like blood flow or EEG activity to measure Integrated
> Information indirectly. This quantity is low when measured during sleep as
> you would expect, in different states of coma. It has been used to measure
> the awareness in patients with locked-in syndrome and establish some
> subjects were still aware and then they tried to communicate with them with
> success.
> Instead of prolonged philosophical discussion this type of research is
> what is needed to understand what consciousness is all about.
> By the way, I do have a common position with Gordon.
> I think there is something special in human-level consciousness. I don't
> think consciousness is a continuum. Yes, in theory, we could associate a
> value of consciousness to anything but the true consciousness that we
> usually refer to is something relatively unique. It is probably something
> akin to phase transitions. I'm not sure Integrated Information behaves in
> that way where there is room for phase transitions, I need to look into it.
> Also, I would like a theory of consciousness that was
> multi-dimensional, and maybe a tensor would be a better way to represent it
> mathematically.
> There are many clues that consciousness comes both as a gradation (think
> about what happens when you drink alcohol to the point to passing out) but
> also it is phase shift (for example when we go from slow wave sleep and
> wake up all of the sudden). Slow wave sleep is an interesting state of the
> brain where a lot of things are happening in the brain (making memory long
> term, deleting memories we don't need anymore, like my exes) but we are not
> aware, in fact, it is very similar to a coma. It turns out that the same
> type of brain waves we see in human slow-wave sleep is the type of waves
> (in terms of frequencies) in reptiles. Lower forms of animals have lower
> brain frequencies and humans have higher gamma frequencies that are not
> present in many other animals. There is a lot of evidence that indicates
> consciousness is supported by these higher frequencies. I like to think
> that a lizard is basically in a slow-wave sleep state even when it is
> awake. It is not really conscious and its actions, like responding to the
> presence of a mate, running away from danger, and so on are basically
> automatic actions that do not require a higher level of consciousness. It
> is some form of awareness but I agree with Gordon here, that if we could
> find ourself in a lizard brain even for a few seconds we would not
> recognize it as consciousness (basically it would feel like we are in deep
> sleep).
> Not sure what level of consciousness these AI have but I agree it is not
> quite yet human level, maybe they are more in a REM state, they are kind of
> dreaming.
> Giovanni
> On Mon, Apr 10, 2023 at 2:46 PM Giovanni Santostasi <gsantostasi at gmail.com>
> wrote:
>> Jason,
>> I'm going to look for the experiment about the rat memory chip. I may
>> have conflagrated two memories in one (slow wave reactivation of memories
>> in a maze and Berger's work on memory on a chip).
>> But really the details of what was remembered and reproduced by a model
>> are not that important (even if I admit remembering the layout of a maze is
>> more impressive).
>> Even if it was just a memory associated with pushing a lever is still
>> incredible that we have a model for memory and by reproducing this model on
>> a digital chip we can reproduce the memory.
>> Again, this example alone shows that brain stuff is in the pattern of
>> information and that information doesn't care about the physical medium,
>> just the functional connections between the parts. From what I remember
>> Berger calls it spatiotemporal code. Memory seems to be a code in space and
>> time (what I call an activation pattern). This is it. I think everything
>> else that is important in the brain, including consciousness, is written in
>> the same code.
>> Giovanni
>> On Mon, Apr 10, 2023 at 12:53 PM Jason Resch via extropy-chat <
>> extropy-chat at lists.extropy.org> wrote:
>>> On Sun, Apr 9, 2023 at 10:50 PM Gordon Swobe <gordon.swobe at gmail.com>
>>> wrote:
>>>>> On Sat, Apr 8, 2023 at 10:13 PM Jason Resch via extropy-chat <
>>>>> extropy-chat at lists.extropy.org> wrote:
>>>>>> Using the phrase "only software" suggests to me you are not familiar
>>>>>> with the implications of the Church-Turing thesis.
>>>> I am quite familiar with the thesis. I differ with some people,
>>>> especially here, about the implications.
>>>> I have shown the converse, denying their awareness , leads to a logical
>>>>>> contradiction.
>>>> I think you have only mangled the language. My smarthome doorbell does
>>>> no more than mindlessly detect motion, yet you call that "awareness." Even
>>>> if I allow for that generous use of the term "aware," you take it a
>>>> step further and equate this mindless process to consciousness. That is not
>>>> what I or think most people mean by either awareness or consciousness.
>>>> It looks to me that you are merely anthropomorphizing the machine I
>>>> call my doorbell. You would probably say that my thermostat in the hallway
>>>> of my home is also "aware" and "conscious."  But as with my doorbell, I
>>>> doubt my thermostat has any private first person subjective experience of
>>>> its existence or surroundings, which is what I mean by alive and aware and
>>>> conscious.
>>> What is the simplest possible conscious state that you can imagine? What
>>> are its contents?
>>> Jason
>>> _______________________________________________
>>> extropy-chat mailing list
>>> extropy-chat at lists.extropy.org
>>> http://lists.extropy.org/mailman/listinfo.cgi/extropy-chat
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20230410/49081985/attachment.htm>

More information about the extropy-chat mailing list