AI and the existential question about language
As you are reading this, can you be sure that the mental states this sentence elicits in your brain do indeed approximate the ones in mine as I was writing it?
Perhaps, when we speak, our words carry vastly different meanings for each party, and we only appear to communicate because our misunderstandings just happen to make sense to each of us separately.
Natural human languages seem to only get meaning by relating to each other. A foreign language would make no sense to us, until we can relate it to a language that we already know. One famous example - the Rosetta Stone. The Voynich Manuscript, assuming it's not just a hoax and full of gibberish, has so far escaped comprehension because there is no existing, interpretable text that can be related to it.
How, then, were we able to learn our first languages? Well, we relate it to the physical world. We learned what "apple" means by seeing an apple, and what "sweet" means by biting into it.
So, here is the question: is all this necessary? In other words, can languages get meaning without being related to anything at all? Imagine someone who was deprived of all sensory inputs, and has only ever been fed language (speech or text) - would they be able to "understand" it?
I hope you can see the connection to AI now.
I could be convinced that the answer to the above question is yes by far less than what LLMs are able to do nowadays. And to me that is an absolutely mind-boggling result (that no one seems to be talking about). Does it mean our languages have some intrinsic statistical structure that creates meaning that is inherent to the language itself?
I don't know.