They mean different things and have different truth conditions.

(a) is true if and only if the cat is on the mat
(b) is true if and only if I assert that the cat is on the mat

(b) can be true even if (a) is false. — Michael

So when someone says, "The cat is on the mat," they are not asserting that the cat is on the mat?

There is no programmer out there, for example, that went through and intentionally answered whatever question you might pose to ChatGPT. — Hanover

I see your point. It's a tough nut. Do we need to try to find some limit cases where we could speak of a programmer "intentionally" doing something via a program? And do we agree that the idea of a program doing anything intentionally is a non-starter? (just leaving Davidson out of all this for the time being)

Notice, BTW, that I'm trying to push back against what I'm calling the "impersonation" by speaking only of "the program" and not personifying it with a name such as ChatGPT, or implying that one could pose a question to such an entity. This is part of the very clever way that the programmers encourage the illusion that the program could have intentions or express meanings, etc. And I'm not saying this is nefarious in some way -- creating this illusion is vital if we're going to get along in cyberworld, where icons stand in for 0s and 1s, etc. But it needs to be resisted in philosophical examination of the kind of questions posed here.

I see your point. It's a tough nut. Do we need to try to find some limit cases where we could speak of a programmer "intentionally" doing something via a program? And do we agree that the idea of a program doing anything intentionally is a non-starter? (just leaving Davidson out of all this for the time being) — J

I'm with you on this. I think we all talked about langauge with a sense there was something special that occurred in the conscious that would make it impossible to communicate without it. What protected this view was the Turing Test barrier, where no interaction with a computer could be remotely confused as human. Then all of a sudden (so it seemed) ChatGPT dropped, and while you can decipher it as not being a human, you can't really argue it doesn't perfectly appear to understand the questions you are asking based upon any inconsistent behavioral manifestation.

What this means to me is that the ability to engage in langauge games does not require an inner state. What this does not mean is that we can ignore what the conscious state is or that langauge does not provide us a means for that conversation.

What this means to me is that the ability to engage in langauge games does not require an inner state. What this does not mean is that we can ignore what the conscious state is or that langauge does not provide us a means for that conversation. — Hanover

Good. Just out of curiosity, has it been shown that an AI program can pass the Turing Test? The examples of bot-talk that I've seen cited in TPF wouldn't fool me for a minute, but maybe there are better ones. And it does depend, as you say, on whether we should see the Turing Test as a standard for whether we can be fooled (I'm saying we can't, yet) or as a standard for revealing "no inconsistent behavioral manifestations." Thus, the program might perform perfectly in that regard, but when paired anonymously with a human who answers the same questions, it might nonetheless fail the "fool me" test. Leading to the intriguing question: why? Does a human exhibit more than consistent question-answering behavior, even in a test designed for question-and-answer?

What this means to me is that the ability to engage in langauge games does not require an inner state. — Hanover

This seems like a surprising conclusion to me, as I would say that both humans and LLM's require going through a lot of complex inner states in order to engage in language use. Would you elaborate on your reasoning? Also, is a distinction between conscious states and subconscious states of relevance, and if so how?

So when someone says, "The cat is on the mat," they are not asserting that the cat is on the mat? — Leontiskos

They are asserting that the cat is on the mat. And they're speaking English. But just as "the cat is on the mat" doesn't mean "I am speaking English", it also doesn't mean "I assert that the cat is on the mat".

I see the cat on the mat!

LLM's require going through a lot of complex inner states in order to engage in language use. — wonderer1

My 2 cents -- and @Hanover may see it differently -- is that by putting it this way, we're succumbing to the illusion that "an LLM" could have any states at all. The computer on which it runs could, I suppose, but I don't think that's what you mean.

Having said that, I should also say that I'm not very familiar with how computer programmers talk about their work. Is "inner state" a common term? If so, do you know what they're meaning to designate? Could there be a distinction between inner and outer, speaking strictly about the program?

I see the cat on the mat! — MoK

No, that's just my avatar. :wink:

No, that's just my avatar. :wink: — J

Yes, I know! :wink:

Yes, I know! :wink: — MoK

Ah, but then you don't actually "see the cat on the mat" . . . my avatar is a digital entity, to put it generously.

Which doesn't mean you haven't asserted doing so, of course! Assertion doesn't depend on the truth of what is asserted, as we were all taught.

Assertion doesn't depend on the truth of what is asserted, as we were all taught. — J

Strange. Google defines assertion as: An assertion is a speech act where a speaker claims or declares a proposition to be true.

Right, but it's still an assertion even if the speaker is mistaken.

But the intention of its programmers, as best we know, is to impersonate intention on the part of the program. — J

Perhaps. But what I'd like to emphasis is that Davidson's theory of meaning is not dependent on intent. It assigns a truth value to an utterance. It can be used to infer an intent, but does not derive meaning from intent.

In the case of ChatGPT the sentences can be interpreted, given a truth value, and yet no intent be inferred.

I’m a little confused. If malapropisms “by their very nature run contrary to the conventions of language” then there are conventions of language. So the very existence of malapropisms is proof that there is a (conventionally) “correct” way of speaking (else nothing could be a malapropism). — Michael

Yep, there are conventions in language. But Davidson argues that they are not what give our utterances meaning.

I understood Tim to be arguing that it is convention that explains meaning. If that is so, it is hard to see how going against a convention, as in the case of malapropism, can be meaningful.

Correct. People think what they say is true even when the assertion is false. They are just not aware of it.

It can be used to infer an intent, but does not derive meaning from intent. — Banno

Yes, that needs to remain clear. You read a poem; you derive a meaning; it may or may not be what the poet intended, though it's often reasonable to infer that. You've heard of "the intentional fallacy" in lit-crit, right? Same issue.

intent is a necessary component in Davidson's triangulation theory. — Hanover

It explicitly isn't.

We can attribute an intent to someone only after we have understood what they are saying. Understanding their utterances is prior to attributing an intent. Understanding their utterances is not dependent on attributing an intent.

Should Davidson not hold that way, he would lose the foundational element for meaning to exist and he would blur into a "meaning is use" position. — Hanover

I don't see that you have explained why this must be so. Davidson is in line with Wittgenstein in saying that we should look at what is being done with an utterance rather then looking for any opaque intent on the part of the utterer.

It begins to look as if we disagree on the accounts given by both Davidson and Wittgenstein.

You've heard of "the intentional fallacy" in lit-crit, right? Same issue. — J

The problem occurs in the US Supreme Court as well, apparently.

The problem occurs in the US Supreme Court as well, apparently. — Banno

Which particular piece of poltroonery do you have in mind? Corporations as persons?

The question is whether it's entirely just a rules based language game or whether you're trying to find some other foundational structure. That's my point directly above related to Davidson's need to rely upon ascribing intent else he would just be a conventionalist. — Hanover

Again,

The interpreter surmises a sentence S such that the utterance of "p" by the speaker will be true if and only if S. S is confirmed or adjusted on the basis of ongoing empirical evidence.

There's no appeal to internal meaning or intention - doing so would result in circularity.

Intent might be inferred post hoc. — Banno

This account does not rely on speaker intent. Nor does it rely on setting out the intent of the speaker, although it might be used to do so.

So when someone says, "The cat is on the mat," they are not asserting that the cat is on the mat? — Leontiskos

So you just wrote "the cat is on the mat". Twice.

Did you thereby assert it?

Or can you do other things with the string of letters <The cat is on the mat>, apart from asserting them?

I would say that both humans and LLM's require going through a lot of complex inner states in order to engage in language use. — wonderer1

The inner state of a computer is usually described physically, while the inner state of a person is described using intentional terms - as believing this or that, wanting something to be the case, and so on. Two ways of speaking.

So the question is, do we attribute belief and desire to ChatGPT?

And the partial answer is that we do not need to do so, in order to give meaning to the sentences it produces.

Which is another argument against the idea that meaning is speaker intent.

Amongst others. I'm not a close follower of Supreme Court process, but the idea that we can discern some imagined shared intent amongst the authors of your constitution is absurd. We might play on "author-itarian" here.

the idea that we can discern some imagined shared intent amongst the authors of your constitution — Banno

Oh, that. Originalism. What the Framers intended. A bit like a literal reading of the Bible.

They seem to go hand in hand...?

The idea is that there is a correct interpretation.

The idea is that there is a correct interpretation. — Banno

And that what we started with is the key to such an interpretation. I know you're doubtful whether there could be a useful interpretation of holy books, but such an interpretation, if there is one, isn't likely to be the one the author(s) had in mind when they wrote about, e.g., Adam, Eve, floods, tablets of stone, etc. Those events, I suppose were "originally" meant to be accounts of true things. A better interpretation will not accept that.

I know you're doubtful whether there could be a useful interpretation of holy books — J

Not useful - folk do put the texts to various and varied use. But that there is one interpretation that is the correct one - that's, shall we say, undecided, perhaps undecidable.

Part of the reason is that giving an interpretation is not a single act, it's an ongoing process.

...A better interpretation... — J

Better for what? Again, no absolute scale is available.

So the question is, do we attribute belief and desire to ChatGPT?

And the partial answer is that we do not need to do so, in order to give meaning to the sentences it produces.

Which is another argument against the idea that meaning is speaker intent. — Banno

The answer is that Chat GPT uses parallel processing A.I. chips but its logic is linear, digital ( binary) and deterministic. In about 10 years we may have A.I. architectures that integrate complex dynamical systems (CDS) models, which will diverge radically from today’s parallel architectures (e.g., GPUs, TPUs) by embodying principles like self-organization and intentionality. CDS-based AI chips may blur the line between computation and biological processes, resembling intelligent materials more than traditional silicon. The interesting thing about complex dynamical systems is that they organize subordinate linear deterministic elements via superordinate recursive intentionality. If we reduce the higher order intentionality to lower order determinism we lose the meaning of their sentences.

In about 10 years we may have... — Joshs

Been that way all my time.

Heuristically programmed algorithmic computers, parallel processing, neural networking... each promised more than it delivered.

The missing bit is that a description of an intentional state is not a description of a physical state.

intent is a necessary component in Davidson's triangulation theory.
— Hanover
It explicitly isn't.

We can attribute an intent to someone only after we have understood what they are saying. Understanding their utterances is prior to attributing an intent. Understanding their utterances is not dependent on attributing an intent. — Banno

Charity is basically about attributing intent to the speaker.

If we assume that the speaker’s beliefs, at least in the simplest and most basic cases, are largely in agreement with our own, and so, by our account, are largely true, then we can use our own beliefs about the world as a guide to the speaker’s beliefs. — SEP

We're looking for the speaker's beliefs in order to understand the speaker's intentions.