When we feel a tension between the power of a wordless intuition and the desire to make ourselves intelligible we face our humanity most directly. To some extent, we experience this tension all the time - between our sense of ourselves as language users and the complexities of conversing with others. The difficulty is compounded by the normative, and somewhat ambiguous, boundaries of 'language' and 'conversation'. Whether we count someone as an honest and competent interlocutor, and how much effort we are prepared to put into figuring this out, are normative matters (as is the attribution of 'consciousness' itself - a closely related matter). There is no 'fact' nor any computation that can settle this issue for us. We make experiments, and we decide whether to press on or to give up.
If we decide to treat someone as an interlocutor, we might say this to them, but it would be redundant. If we decide not to, we have, essentially, nothing to say this with. As I've said before in this blog (probably too often): If you can't talk to someone, you can't say so to them.
Returning to wordless intuition and intelligibility:
David Hume observed that we must rely on our senses, even though we cannot demonstrate their reliability. He had a conception of the world and our interactions with it that made this observation seem much less ambiguous than it actually is. For him, our phenomenological space and our capacity to rationally reflect on it (and on other things) were part of the groundwork. It would have been difficult for someone with that particular mindset to see that the role our 'senses' played in our capacity to make ourselves intelligible to one another necessarily gave them just the kind of 'reliable content' that this mutual intelligibility required, and no more. This is the point of Wittgenstein's 'box beetle' thought experiment.
This can be a frightening realisation. The visceral experience of our wordless phenomenological condition reliably driving our capacity to speak, and the related sense that language can give us a transparent view of each other's cognitive and phenomenological spaces - even that our narratives and explanations are 'literally true' in some way that goes beyond 'mutually intelligible' - look essential to sanity.
Until, that is, we try to intelligibly articulate those very conceptions of sanity....
The incongruence we experience when we reflect on the paradoxes and ambiguities this produces might drive us in one of two directions: we might rescue our intuitions at the expense of intelligibility, or we might insist on the reliability of public demonstration, however confused and miserable this made us feel. It goes without saying, almost, that both of these options are, themselves, unintelligible. The first because it cannot be shared (and so becomes a necessarily private conviction - only expressible, if you like, in a 'private language') and the second because it is liturgical - we can only participate in it by recitation and signalling.
The third way is to ground our method in intelligibility experiments - which Hume does, of course, but does not acknowledge. He says, effectively: 'Look - this is what makes sense!', where he might have said 'I think this makes sense - does it make sense to you?'
In a real conversations, of course, the two are equivalent. The intelligibility of an assertion (and of assertions of the 'truth' of assertions) is conditional on the intelligibility of the conversation. Which, in turn, is conditional on the willingness and ability of the participants to engage.
This conditionality can be masked by the apparent independence of a 'text' - can we really be in conversation with David Hume? - and by the apparent 'absolute' or 'necessary' truth of certain logical and mathematical assertions. My present conversation is not with David Hume, however, but with you, about David Hume. Also, as I've pointed out elsewhere in this Blog, some assertions - not just logical and mathematical assertions, but also 'privileged statements' in the sense introduced by Barry Stroud, are impossible to intelligibly deny in any conceivable language. As Donald Davidson might have said, any translation schema which rendered a 'native' speaker making such a denial would have to be rejected, and if no translation schema can avoid this then what we are dealing with is not a language. Whitman's poetic declaration may be powerful, but it only works with some interpretation ...
The sense of our humanity that I referred to in the first paragraph is most characteristically expressed in our mutual experiments in intelligibility, in our experimental conversations with one another. This is ironically captured in Turing's infamous test of machine intelligence. We cannot construct a machine that exhibits this very recognisable aspect of human intelligence, because we do not know how human beings do this. Large language models can only mimic what has already been said - their coding incorporates semantic rules, it does not enable semantic experiment, and especially not semantic experiment that humans can engage with.
What is more, and Turing knew this when he wrote about machine intelligence, it is also formally impossible to construct such a machine. If we imagine a 'computer'' that could mimic a human linguistic experimenter completely, then we can see that either it, or a 'composite' of them could mimic human conversations and predict their outcomes. Such a composite machine would be running a 'truth algorithm', which it would not be equipped to interrogate (because this would generate an open question paradox). Another way of putting this is that we cannot imagine such a machine, because we cannot validate its mechanisms (and so cannot be sure that our 'imagining' is correct), or, perhaps, that if we 'found' such a machine we would not be able to correctly determine its function. Either possibility renders the proposal unintelligible.
So it's humanity all the way down ...
