The phenomena of inner experience (2007) [pdf]
hurlburt.faculty.unlv.edu
hurlburt.faculty.unlv.edu
There is a danger that people assume that the only way to do something is the way we do it.
This is the "Your flying machine won't work because it doesn't have feathers" argument.
Or more formally
Entity X has property Y.
Property Y is utilized to achieve property Z
Entity A does not have property Y
Therefore A does not have property Z
It is possibly worth pointing out that the word by word construction aspect of a LLM also cannot be used to eliminate the possibility of any form of consciousness. It would merely indicate that the generation of a response may be a series of very short term transient consciousnesses. Later tokens would just be new conscious 'instants' but with the context of the previous response. This is easy to see because the dividing line between instants is clear. If you turn it around and ask how do we know that we are not doing the same thing ourselves, it's too unclear to come up with an answer. Is you with a memory of yesterday the same as yesterday you? These are not new questions and rather than coming up with satisfactory answers we have managed to make the puzzle deeper by asking awkward questions like "what about transporters?" Sci-fi has explored these ideas quite a bit(Will Riker and Thomas Riker).
Given the Star Trek reference I'll also do a mini rant about Discovery.
The episode where they pick an arbitrary criteria, dreaming, as being a special property to be accepted as being an allowable person, is almost precisely this fallacy wrapped in a model of an established order dictating that 'what we have' is right. That they tried to deliver this as a parable of a minority being 'seen' makes it tragically ironic.
It strikes at the heart of the thinking that makes up 99.9% of my life but seemingly most people never talk about it or are completely unaware of it.
The example does not say it out aloud that it assumes you can do "wondering" without symbols. An example like that implies that what it describes as happening actually happened, therefore proves that that is how the world works.
I think that is a logical fallacy even though I'm not sure if it has a name or official definition. The fallacy is that because in an imagined situation something is imagined as having happened, it would prove something about the actual world, without us also proving that such an imagined thing can actually happen in the real world.
Whatever you may call them, that is in fact what is meant by “symbols” by the people using that term in this context. Essentially, an identifier that represents an abstraction. It’s an incredibly basic feature of all cognitive systems we’re aware of.
Does Chat GPT dream of electric sheep?
We can only trust that it’s telling us the truth.
Just like you can only trust someone else that they are telling you the truth about their own experience.
Our language is essentially just and LLM hallucinating words to try to explain our non-symbolic experience.
In other words, all experience is non-symbolic, but we are trying to symbolize it all the time.
I subscribe to panpsychism, everything is conscious/ness. Consciousness is part of the fabric of reality/universe.
Crazy times.
[1] https://press.princeton.edu/books/hardcover/9780691122946/on...
I don't trust them. I think they have inner experience because they're the same kind of thing I am and I have inner experience.
There's no reason to think that an AI couldn't have non-symbolic modes of thought. Symbolic systems can be used to implement non-symbolic systems, which then have the constraints and limitations of the non-symbolic system. Neural nets are widely considered to be sub-symbolic systems. This was a core debate between the connectionists and those that argued symbolism was required for cognition and that neural nets were a dead end.