The "hidden state" has nothing to do with noumena. But that confusion is where this thread has wandered. — Banno

Easy to assert: can you explain the difference?

By the way; you're jumping to conclusions as usual: I haven't claimed they are the same; I'm asking about the difference.

Can you explain the similarities?

One is a mathematical simplification, the other a philosophical confusion.

Can you explain the similarities?

One is a mathematical simplification, the other a philosophical confusion. — Banno

Yeah, I didn't think you could explain it; just a tendentious characterization, which is the sort of thing I've come to expect from you. It's a shame; you could probably do so much better.

This doesn't seem to be saying anything cogent; can you explain further? — Janus

It's the difference between a representational relationship ("the kettle" symbols and perceptions of the kettle <-> the kettle), exemplified by a neural model, and the kettle as treated as an experiential object ("imagination", "appearance", "fiction") in a conceptual scheme ("environmental story telling device"). The former is a practical and representational relationship with the states of the environment, in which environment states' content historically+reciprocally determines word semantic content and perceptual content (without necessarily being one to one), the latter is an experiential relationship of what is produce in/by a model to an agent. Is a lens's image in the lens or of what it pictures?

The "hidden state" being "hidden" doesn't necessarily make the claim that the hidden state's "content", whatever it is, is "hidden" from perception or symbolisation since it's used in those processes AFAIK (that needs to be demonstrated or interpreted out of it). That's like placing a semantic or perceptual veil over reality.

From my perspective, what makes @Banno and @Isaac able to agree on a surface level is that they're able to agree that there is a semantic or perceptual veil of some sort; just for Banno it's transparent to the point of non-existence (see sub discussion of the world being English shaped or events taking propositional form) vs for @Isaac the veil is totally opaque, if constraining, to hidden state content in both cases - negative tight constraints ('can't be this...) rather than transmission of content (red approximately equals (these wavelengths for us in this context)).

In my one liner, the cognitive grasp of an object is the interior of a model's state, rather than an environmental one interacting with an internal one (it's the bit in your head), and thus it's not the kettle as it is in the environment (that's the bit in the world). And the limit on possible thought bit corresponds to the claim that neural models place tight constraints on what environmental behaviours are like rather than transmitting a summary of their content.

Thanks for the chuckle.

The "hidden state" being "hidden" doesn't necessarily make the claim that the hidden state's "content", whatever it is, is "hidden" from perception or symbolisation since it's used in those processes AFAIK (that needs to be demonstrated or interpreted out of it). That's like placing a semantic or perceptual veil over reality. — fdrake

How we model whatever we are sensorially affected by is hidden, since there is no way to definitively link our conceptualizations with what is pre-conceptual. There would only be a "veil" if we assumed that our models are somehow distorting what they are modeling; which would be an entirely unwarranted conclusion, since we have no way of comparing our models with what is being modeled.

Edited the post a bit to be more explicit, if you're got anything further to add, please do.

You have nothing to say to my response?

As to whether or not the hidden states are hidden from perception, I would say that depends on how you define perception. If we are affected pre-cognitively would those affects count as perception? If you say yes, then surely you would have to then draw distinction between those pre-cognitive "perceptions" and conscious concept-mediated perception, no?

It's the difference between a representational relationship... exemplified by a neural model — fdrake

To be sure, neural models are not representational... So that doesn't look right.

From my perspective, what makes Banno and @Isaac able to agree on a surface level is that they're able to agree that there is a semantic or perceptual veil of some sort; just for Banno it's transparent to the point of non-existence — fdrake

So for me there is a veil that doesn't exist...? Again, that doesn't look right.

Hidden states are inputs in recurrent neural networks, not things-in-themselves or some other philosophical notion. Confusing the two seems to be the source of the present impasse.

As to whether or not the hidden states are hidden from perception... — Janus

See?

↪fdrake You have nothing to say to my response? — Janus

I do, but I'll do it when I'm less tired because it's finicky. I could default to my usual 900 page essays on trivial bollocks, but I'll try and make it briefer than usual.

:up:

To be sure, neural models are not representational... So that doesn't look right. — Banno

Ooh. What are they for you then?

So for me there is a veil that doesn't exist...? Again, that doesn't look right. — Banno

I don't think you think there is a veil. What I intended to convey (but failed) was the sub discussion we had about "On the Very Idea of a Conceptual Scheme" - that having one shared one is very much the same as there being none at all, so the idea of a conceptual scheme turns out to be too confused to use. Why I think of it as a veil still is because the world ends up "English shaped" due to that shared relation and how it relates to shared environments; whereas I claim we know it isn't "English shaped" and even speak as if isn't. But we can get into that entirely different sub discussion at a later date.

It's funny; I pictured you crying in your beard, but no matter, as the saying goes: "ignorance is bliss".

Why the insults?

the world ends up "English shaped" — fdrake

"language shaped" would be better, since firstly a corollary of the argument On the very idea... is that all languages are inter-translatable, hence in effect the same; and secondly whatever we choose to talk about will be, by the fact that we talk about it, shaped by the language we use.

So your claim to know that "it" isn't "English shaped" is either wrong or I've misunderstood it.

As to whether or not the hidden states are hidden from perception... — Janus

If your use here of "hidden states" is supposed to be the same as @Isaac's, then is seems you have made a category error.

Why the insults? — Banno

I find it amusing to supplement passive insult with active?

If your use here of "hidden states" is supposed to be the same as Isaac's, then is seems you have made a category error. — Banno

With no explanation of what you take Isaac's conception of "hidden states" to be...see the problem? I'm trying to draw you into examining your ideas...

Can we please postpone that discussion for now? I don't think our quibble there is too much related to the semantic content and denotation quibbles we're having. We both seem to agree "the kettle" denotes, and that we can say ""the kettle is boiling" is true" asserts the same thing as "the kettle is boiling".

So your claim to know that it isn't "English shaped" is either wrong or I've misunderstood it. — Banno

My worries for that are kinda orthogonal to the current discussion. I've got two flavours of worries; over reliance on an implicit ontology we create through how we use language. The second flavour of worry is here, (make the interpretation discussed linguistic or linguistically mediated). The first one is maybe related to the discussion; there's perhaps relevance in talking about "the kettle" as a construct of folk psychology vs "the kettle is boiling" being an accurate statement when the kettle boils (I agree that it's accurate because the appropriate court of evaluation for it is the pragmatic context it's in). The second one would take us much farther afield.

@Banno

My exegesis here for the second quibble is more detailed.

I find it amusing to supplement passive insult with active? — Janus

Fair enough; you've so little else to work with.

With no explanation of what you take Isaac's conception of "hidden states" to be... — Janus

See

One is a mathematical simplification, the other a philosophical confusion. — Banno

Hidden states are inputs in recurrent neural networks, not things-in-themselves or some other philosophical notion. Confusing the two seems to be the source of the present impasse. — Banno

...and the conversation with @Isaac back a few pages. That you did not recognise the answer is perhaps indicative of not understanding your own question.

Can we please postpone that discussion for now? — fdrake

Of course that will never happen, but by all means have your rest. I'm off to lunch soon anyway, but first have to repot a passionfruit blown over by a gale.

I've got two flavours of worries; — fdrake

Both look very interesting. But on to more mundane things for us both, for a while.

As I said to another poster a few days ago, all this says is that we determine the meaning of a proposition. It doesn't follow from this that we determine the truth of a proposition. — Michael

But I said more than that. I said that whatever the proposition means must be related to what is actually the reality of the situation, and through this comparison, it is judged for truth. That's how we determine the truth of a proposition, through judgement. How could the truth of a proposition be determined, except by a judgement?

Our language use determines the meaning of the proposition "water is H2O". John believes that this proposition is true and Jane believes that this proposition is false. The laws of excluded middle and non-contradiction entail that one of them is right and one of them is wrong, irrespective of what they or I or anyone else judges to be the case. — Michael

Actually, what you've just stated, that one must be right and the other wrong, is just a judgement itself, made by you, as Mww has already pointed out.

True or not true can be nothing other than a judgement. The question of the thread, I believe, is what exactly constitutes a true judgement. But we cannot remove the judgement aspect without leaving "true' as completely meaningless. That would be like asking what is "red" while insisting it's not a colour.

Yes, that is how "knowledge", as the subject of epistemology, is normally defined. But we were not talking about "knowledge", the epistemological subject, we were talking about normal use of "know" as an attitude. — Metaphysician Undercover

That's an artificial distinction. Knowledge, and thus epistemology, is grounded in ordinary life and we use ordinary (and, if need be, specialized) language to talk about it.

And the fact is that people often claim to know things, which turn out to be not the case. — Metaphysician Undercover

That's correct, and no-one disagrees. A flat-earther can claim to know that the world is flat. He nonetheless doesn't know that. So there's a distinction between knowledge and knowledge claims.

So the definitions which epistemologists prescribe as to what "knowledge" ought to mean, do not accurately reflect how "know" is truly used. — Metaphysician Undercover

Actually, it does. For example, people once said that they knew that Hilary Clinton was going to win the 2016 election. But since she didn't win, they didn't know that at all, they only thought they did. The term knew is retracted because of the implied truth condition.

This is somewhat analogous to Alice saying that "Trump won the 2020 US election". Is Alice misusing the word "won"? Does she need to consult a dictionary so she can correct her mistake? Presumably, the problem is not with her use of the word "won", it's that her perfectly well understood claim is false.

But if we conceive of "true" as I proposed earlier in the thread, to be a representation of one's honest belief, then knowing entails truth, as commonly said by epistemologists, but truth does not necessarily mean what is the case. — Metaphysician Undercover

I see that Oxford Languages lists that as an archaic usage, as in "we appeal to all good men and true to rally to us". But that context isn't propositional truth.

Just to play devil's advocate: The Myth of Factive Verbs.

The SEP article on knowledge summarises Hazlett's view as:

Hazlett takes this to motivate divorcing semantic considerations about the verb “to know” from knowledge, the state of traditional epistemic interest. Even though “knows” is, according to Hazlett, not a factive verb, even Hazlett accepts that knowledge itself is a state that can only obtain if its content is true.

This is almost exactly what Metaphysician Undercover is saying — Michael

Yes. I will have to take a look at the paper. Note that SEP put knew in scare quotes with the above Clinton example, which I think also says something about the ordinary use of the term.

I’ll have to leave that alone; I don’t see how classical can be derived from nominal, but that’s ok. Also....once again.....translator’s preference. The SEP quote is right, but mine on pg 45 herein, is also right, and different. In addition, the SEP quote, after “is assumed as granted”, leaves out “...and is presupposed”, which offers a clue as to what exactly definitions are supposed to do. — Mww

I take nominal to mean that the definition can't be employed to establish which statements are true (see Kant's comments here). That's the case with Aristotle's (classical) definition as with Kant's.

Nevertheless, there is rather apparently an intended difference between Kant and Aristotle, insofar as the former’s definition contains cognition, while the latter’s does not. They would have been much less different if Aristotle had said, “to think that what is is......”. — Mww

As I mentioned earlier, for Aristotle, to say something presupposes cognition. Keeping in mind, of course, that people sometimes do speak without thinking...

A flat-earther can claim to know that the world is flat. He nonetheless doesn't know that. — Andrew M

That's what you say. He says he knows it, you say he does not know it. It's your word against his. We can move to analyze the justification, and show that your belief is better justified than his, but this still doesn't tell us whether one or the other is true. And if you argue that his is not knowledge, it's not because his belief is not true that it's not knowledge, it's because it's not justified. So we cannot establish the relationship between knowledge and true, in this way.

Actually, it does. For example, people once said that they knew that Hilary Clinton was going to win the 2016 election. But since she didn't win, they didn't know that at all, they only thought they did. The term knew is retracted because of the implied truth condition. — Andrew M

If anything which may turn out to be false in the future cannot be correctly called knowledge, then there is no such thing as knowledge, because we cannot exclude the possibility of mistake. This is what Plato demonstrated in The Theaetetus. So, it's much better to allow that what people claim to know right now, is "knowledge", regardless of the fact that it may later turn out to be wrong. It was still knowledge, at that time, before it was proven wrong.. So, if at a later time they decide against it, it is no longer knowledge, but it still was knowledge back when it formed the principles upon which they based their decisions.

It's more accurate to define "knowledge" as the principles that one holds and believes, which they apply in making decisions. That is a person's knowledge, regardless of the fact that it may later turn out to be wrong. This way, we don't have to decide at a later date that the knowledge we held before wasn't really knowledge. And the knowledge we hold now will later turn out to be not knowledge, onward and onward so that there is no such thing as something we can truly call "knowledge" because we can never exclude the possibility of mistake..

I see that Oxford Languages lists that as an archaic usage, as in "we appeal to all good men and true to rally to us". — Andrew M

It isn't archaic usage. It is the principal meaning of "truth", employed in courts of law etc., and any time or place where people are asked to "tell the truth".

Since the world is all that is the case, it is also a collective story. — Banno

'According to the Bible' or 'Fred says that' are not restricting modifiers; they do not pass through the truth-functional connectives. 'Fred says that not P' and 'Not: Fred says that P' are independent: both, either, or neither might be true. If worlds were like stories or story-tellers, there would indeed be room for worlds according to which contradictions are true. The sad truth about the prevarications of these worlds would not itself be contradictory. But worlds, as I understand them, are not like stories or story-tellers. They are like this world; and this world is no story, not even a true story. — David Lewis, On the Plurality of Worlds

...and...?

What are we to conclude here? Modal Realism?

Because we could never be surprised to find that Aragorn was not king of Gondor, or that "Aragorn was king of Gondor" is false. Surely we know our collective fiction (which is the model, which is the world) in exactly the same way, and with the same level of surety, that we know Aragorn was king of Gondor. So, whence surprise? — Luke

From hidden states.

Then the model is not equivalent to the world; there is a distinction between them. The world is not the model or a collective fiction, because the world can surprise us. — Luke

Who said anything about the world surprising us?

My point was that I'm not assuming, either. How am I begging the question — Luke

You said...

Since it is possible that our model could be false in at least some respects, and that we could be surprised, it follows that there is more to truth than a mere "collective fiction". — Luke

This is only true if the terms are interchangeable (that truth is about the model being surprising), otherwise your conclusion doesn't follow, hence you begged the question by assuming that relation in your argument for it.

I take the position of redundancy to be that there are no matters outside of language, and that the model is equivalent to the world, whether that is your personal view or not. — Luke

I really don't know where you're getting that idea from.

If we have modeled imperfectly some detail of the hidden states, but we never encounter evidence that would encourage us to update our model, were we wrong? — Srap Tasmaner

There's a muddle of temporal terms here that I can't make sense of, You say that we "never encounter evidence...", but then ask "were we...". 'Were' from what temporal vantage point? Your first 'never' seems to disallow any point of reflection from which we can look back and ask the question.

can our model be properly said to supervene upon the hidden states? That is, can there be a change in our model without a change in the ("underlying") hidden states?

If the answer is "no," if our model is not so tightly coupled to the hidden states as that, what is the source of that relative freedom? And if our model is then, to some undetermined degree, independent of the hidden states, what entitles us to describe changes to our model as updates rather than just changes, which could, for all we know, be arbitrary, or, if not arbitrary, free? — Srap Tasmaner

Interesting question. Change and chaos. Or, less prosaically - (1) our models are based on priors and priors are necessarily based on historical inferences which will be constrained by previous states of the environment, not the current one, hence a source of de-coupling, our environment changes, our models run behind that change; and (2) there's a lot of noise in the system, neuron firing can be random, so a lot of what neural modelling does, one of the main reasons for backward acting suppression, is to make sure that noise is not mistaken for data, but this system isn't perfect, so sometimes it is. Obviously, there are then magnifying effects of both of these since we do not passively receive data from the external states, but rather we actively harvest data (and even manipulate those states) to match our models, so we're going to act in such ways as to re-affirm the model predictions insofar as that it possible, even if those model predictions have been affected by nothing but noise.

There is nothing, it seems, that we can point to as "evidence" that is outside the model, not even surprise; surprise is not a fact, but part of our model of ourselves. — Srap Tasmaner

Yes, that's right. We could model our 'surprise' as errors in our models, or as part of the narrative, as evil demons, as the tweaking of the matrix by our cruel AI overlords... The whole of active inference is a scientific model, an activity we undertake to help explain phenomena, just like philosophy. Some explanations seem better than others, but the reason why they do so can only ever itself be just another such explanation.

What becomes questionable is the claim that the "map" is not the territory but only a map, and the positing of a "genuine" territory out there, somewhere, that the "map" we wander around in is a copy of. That will surely strike most residents of the "map" as an article of faith. Anything can count as evidence for it, and nothing can count as evidence for it. — Srap Tasmaner

You're focussing on material construction, which is not that the model of active inference is about. The model is about data. It simply says that given a self-organising network of data nodes that is greater than a mere ring of nodes, there will inevitably be nodes which are inside a Markov blanket relative to the nodes which don't constitute the system. That has nothing to do with the material constituents of the nodes, only their informational relationship with one another. Thus, as a self-organising system, we must, by definition, have internal states, and boundary states (and there must exist external states). Without these three states we cannot say that there is a system at all, we cannot define it from 'not-system' without defining a boundary and (as far as data is concerned) that boundary must be Markov boundary if the internal network is any more complex than a single ring of nodes.

So the only way we could be informationally connected to 'the world' without a Markov boundary is if we say that we are the world, one integrated system. I don't think anyone is going there...

I don't think it will quite do to answer that "data underdetermines theory." What "data" there is, is not just theory-laden; it is crushed under the weight of the theory it's carrying on its back. It could, for all we know, be 100% theory. — Srap Tasmaner

That data underdetermines theory is implied by defining a system, as is raw data. As above, the mere definition of an informational system implies a boundary and external states. Complexity beyond a mere ring implies boundary nodes. Once you have those two elements, it is a necessary fact of the model that external states must be inferred by internal states from the states of the boundary nodes. If you introduce any variable whatsoever (active harvesting, ergodic feedback mechanisms, noise...) then it is a necessary part of the model that this inference will be prone to error - underdetermined by the external states.

So...

(1) nothing entitles us to make any claim that there is such a true state, or to make any claim about how close our conception is to it, — Srap Tasmaner

I don't think that's true. I've outline above how the very act of defining a system leads us to the conclusion that there are external, internal and boundary states and that external states can only be inferred by internal states from boundary states.

We have a model that is, for all we know, 100% mistaken, and at the same time, for all we know, all there is and no model at all. — Srap Tasmaner

Yep. I'm not sure what improvement in certainty you're looking for beyond that which we can rationally argue for.

Before you post "pragmatism" and count that as a job well done, plan on explaining exactly how pragmatism answers any of the questions I asked, or shows the questions to be ill-conceived. — Srap Tasmaner

Hopefully avoided pragmatism in my answer... but do check.

Since the world is all that is the case, it is also a collective story. That does not meant hat just anything goes. You will still burn your hand if you touch the boiling kettle. — Banno

Yes, this is what I've been trying elucidate, but perhaps you're right and my use of 'fiction' here has only confused matters. There's perhaps a better way to explain.

The idea here (for me) is that despite having to posit hidden states as part of our informational meta-thoery (see my post to Srap above), these states can still be proper objects of reference. "the kettle" doesn't refer to my model of a kettle, it refers to (in the informational model) the hidden state itself. It's like us all speculating what's in the room next door. the subject of our speculations isn't our speculations, the subject of our speculations is what's actually in the room next door. As such, the best way I can find of 'translating' an active inference model to talk of "kettles" is to say that "kettle" refers (when it refers at all, that is - not all uses are referenential, of course) to the hidden state we're modelling, the contents of the room we're speculating about.

The difference is that we have data which more or less coheres with certain speculations (still on the 'room' analogy here) such that the speculation "nothing is in the room" would be difficult to cohere with a lot of noise and shouting coming from it - ie you still burn your hand when you touch the kettle.

So, perhaps more to @Luke than yourself (I may otherwise be preaching to the choir...)...

Kettles and water and boiling might be describable as collective fictions, but the part that needs emphasising (that I ought perhaps to have emphasised) is that they're purposeful fictions. They're not fantasies like the Lord of the Rings, where anything goes, they've a job to do - that of unifying, to an extent, our individual (neural) models so that we can co-operate, and not constantly surprise one another. This hooks them in to the actual external states we exist within in a way that actual fiction need not worry itself with.

(if that's cleared anything up @Luke then you can thank Banno)