Mr. Grice's progeny

 The automatic writing of the Surrealists was miles behind the true weirdness of certain crochets of the analytic philosophers in their prime, between the 50s and the 90s. A whole career, back in those glorious days of wine and Sputniks, could be spent pondering some aberrant sentence in English.

It is  this set of “suspect examples” that wash through the papers and (rare) books of Paul Grice.  In Studies in the Ways of Words, his magnum opus, he considers quite a few before getting to the gist of the book.

Among them are phrases involving the English verb try.  If I say, for example, that I tried to write that last sentence, I would score one hundred on the truth meter. I did try. I even succeeded. Which is the oddity of the case. As any parent knows raising any child in the primary school stage, the word trying is often used to denote – something not done. As in, I tried to do my homework! Hearing a child say, I tried to do my homework, one can be dead certain that said homework was not done. But that inference can’t be based on what we know, definitionally, about trying, which is neutral on the success of failure of the thing tried.

Or – to take a “suspect example” that I simply love: what about remember? If I sign a document, I am entirely entitled to say: I remembered my name. But if I did say so, it would make one immediately suspect that the ‘signing the document” scenario is a little off. Am I senile? Have I suffered from amnesia? Nothing in saying I remembered my name tells us, on a semantic level, that such suspicions could be raised. Every time I sign an email I can justly say that I remembered my name; but in saying it, I put into question some linguistic “instinct”. I am giving information that is, though true, inferentially overtrue, so to speak.

Grice, being English, calls this odd. And it isn’t just an anomaly. Rather, it points to a whole structure of inferences built into our discourse, or as he calls it our conversational use of language. These suspect cases are the entering wedge from separating language as the logician conceives it from the ordinary language studied by the socio-linguist: the pragmatics of language.

The key here is that language is used to imply. Grice devoted his philosophy of language to tease out the dimension of implying – of implicature – and to give it a canonical form. “Echoing Kant, I call these categories Quantity, Quality, Relation, and Manner.” All of which are utterly dependent on the big C – circumstances. Scenarios, stories, uses.

The rules that compose these categories are very Britishy, and one might phrase them differently – but they do a good job of making us feel implicature

Take for instance Quantity. “The category of Quantity relates to the quantity of information to be provided, and under it fall the following maxims:

1. Make your contribution as informative as is required (for the

current purposes of the exchange).

2. Do not make your contribution more informative than is required

 

This then takes care of the suspect example of remembering your name when you sign a document. The excess of information leads to other problems in the big C. It leads, I think, to literature.

I won’t go through Grice’s table. But a remark about implicature is relevant to my larger purpose – which is to link AI to its basis in computing. The table and its maxims are very much structuring principles – and not logical deductions. Implicature seems to resist the computerization that we can accord, say, sorting, or any conventional set based logic.

This gets us to the Large in Large Language Models. The way to force the computerization of implicature, it was seen in the eighties, by people working on Expert systems, was by sheer expansion of the corpus of data. That expansion is, of course, energetically costly. Even as, under the watchful eye of Moore’s law, chip size got smaller, the energy for creating a C large enough to accommodate implicature still seemed pragmatically difficult.

Well, of course, it turned out it is. Which is why we have this amazing energy suck that consists of building data centers to contain the universe of C – which consists of all human representations, audio, scriptorial, pictorial, etc. It is a matter of implicature without experience.

 Can experience be mimicked through recursion? My own feeling is that the universe of C can’t expand that much.  The apocalypse does not, to my mind, mean robot overlords – it means the exhaustion of planetary energy.

And so it goes in our odd faux-Gotterdammerung times.