I think it is more helpful to maintain distinctions between linguistically and culturally elaborated conceptualizing capacities, and the primordial somatically-based embodied cognition we share with animals. — Janus

I don't see how they can be distinguished. When a chimpanzee rejects a previously gratefully accepted cucumber as reward on the grounds that the other chimp has been given a grape, are they not using a model of justice, or fairness? We verbalise such a concept, chimps don't, but the model (in terms of expectations and behavioural response) is still in some social animals. I'm struggling to think of an example where we, as humans, might verbalise a concept which is completely absent in its entirety in other social animals.

When a chimpanzee rejects a previously gratefully accepted cucumber as reward on the grounds that the other chimp has been given a grape, are they not using a model of justice, or fairness? — Isaac

Or perhaps they just wanted a grape...

Or perhaps they just wanted a grape... — creativesoul

Nah, Chimpanzees also favour fair (50:50 split) offers in Ultimatum Game experiments to unfair ones (80:20 split), even when the unfair split is in favour of the proposer.

Well on an ecological level, we simply wouldn't behave as efficiently as a creature which had a clear answer, on an epistemic level, how would we experience that with some means of modelling it... — Isaac

I like this conjecture a lot. Explaining the unity of consciousness in terms of our body's self modelling processes as realising a single action-sensation-internal state from a space of possible ones is pretty neat. We must act in some specified way, and that specification coincides with a collapse (through sampling) into a unique state.

Yes. The great thing about the free-energy approach is that it gives both a mechanism and an evolutionary story to the Bayesian modelling system which we already have a good idea the brain uses. — Isaac

One thing I'm generally apprehensive about with Bayesian brain approaches is the parameter optimisation mechanism. A lot of effort is put into the neural implementation of the modelling procedure, but less seems to be put into the neural implementation of the optimisation mechanism.

A rough guideline for how this works is that there is some neuroscientific model $M(\theta)$, where $M$ is the total model (the whole dependency loop in the referenced paper) and $\theta$ are the model parameters. In order to make a prediction of a specific value; say the ridge width in your example; the model has to be mapped to a specific set of parameters $\hat{\theta}$ which then allow the model to output a prediction based on the mapping; our predictions of the ridge width at a given time. This mapping is called a point estimation procedure.

One strength of the paper (again, if I read it right) is that it dodges the problem of "which point is output from a probabilistic model for our action?" by having random sampling from a posterior distribution obtained by variational inference (what sampling procedure though? How is it in the wetware?). But the optimisation; model choice steps; in constraining the M3 functional forms are conjectured to work through steepest gradient descent.

Without any math detail, steepest gradient descent is like pouring water onto a landscape defined by a function; a valley with elevations and contour lines; it travels down into the valley down the (locally) steepest slopes, the 'flow of water' is the iteration of the algorithm from step to step down the valley towards its basin. Unlike in nature, the water's movement needs to 'know' which directions are which and what they mean. In the algorithm, the "valley" is a cost function, whose "elevation" and "angle from the bottom on a contour" are parameters (directions) which are optimised over; there is also some geometry applied to the directions; how far is what from what. Steepest gradient descent finds the closest thing it can to the minimum of the cost function with some parametrisation and some geometry.

It is plausible that if we had complete model specification of the human mind, it would not really matter so long as some part of its dynamics implemented the steepest gradient descent. But whether and how the process internally optimises with steepest gradient descent is left unconsidered.

This is kinda first world problems at this point, considering it's usually easier to model something first then pick an appropriate estimation procedure, but I'm still a bit skeptical of "model first; then estimate" when both are running simultaneously in the same wetware. On a computer, you input the model and the optimisation constraints (cost function, also space geometry if you're pedantic) to find the parameter estimates; our brains seem to model and estimate at the same time.

This is alleviated in the paper by having parameters sampled from models, rather than deterministically outputted from them (if I've read it right, again) globally through some optimisation algorithm, so I can see an argument that specifying 'the general statistical structure' of our mind 'before' looking at how it estimates itself is a better approach, I'm just suspicious that something of fundamental importance is being missed.

I'm struggling to think of an example where we, as humans, might verbalise a concept which is completely absent in its entirety in other social animals. — Isaac

Because this is also interesting, @StreetlightX had a thread on this at some point in the past. The gist of it was that relational concepts admit of a degree of abstraction; "analogy (is) the core of cognition" as Hofstadter puts it. Monkeys can learn relationships between tokens quite easily (this is red, that is red), but they find it much harder to learn relationships between generated types (this is coloured, that is coloured) or (this is a pair, that is a pair) (in Street's example); they also seem to need a token based learning exercise for more abstract concepts. I think it would be quite surprising if a monkey could be taught what tuple is in the general case; there probably aren't enough tokens to serve as an external memory bank for an arbitrarily relationally complex abstract object for monkeys to learn it.

But I suppose that depends on the scope of "in its entirety"; abstracting a type from tokens is something monkeys can do, abstracting higher order types into abstract objects to synthesise even higher order types is probably not on the cards arbitrarily. Humans struggle with tuples (until we learn what's in 'em doesn't matter, nor does their length...).

I doubt all of our precursors would struggle though? The boundary between human and human precursor (in terms of cognitive development) seems pretty fuzzy.

Just quickly - need to sleep - my hunch is that everything turns on negation, and the ability to treat negation as positive: to treat 'not-X' as a positive variable and not simply the absence of X. Only this allows for abstraction proper: abstraction without a ground-level, 'material' token, the double-incidence of token and type birthed at the same time - that which is not itself. Once you can do that, concepts, proper concepts, bloom full bore. No negation, no concepts.

Explaining the unity of consciousness in terms of our body's self modelling processes as realising a single action-sensation-internal state from a space of possible ones is pretty neat. We must act in some specified way, and that specification coincides with a collapse (through sampling) into a unique state. — fdrake

Yeah, I think the idea goes way back to Geoffrey Hinton in the early 90s. It also explains the self/other divide (which we know to be spurious) because one has to distinguish those actions which arose from the models from those actions which are to be input into the priors, otherwise the model becomes too self referential to be truely adaptive. M3 is 'self'.

As to the issues with parameters...

It's very complicated and I'm somewhat out of my depth with both the maths, and the neurobiology, but I'll do my best to impart my understanding (most of this is from Friston's earlier work).

Neural architecture is built on three major properties - connection strength, connection heirachy (disputed), and connection locality (specialisation and integration between regions).

Memory (which, when it comes down to it, is what we're talking about with regards to priors) is primarily about connection strength in later life, but to do with heirachy and locality immediately post-natally. So parameters are mediated differently depending on when the priming sensory input (forward neural connections) arose. Your priors about cultural effects might be mediated by connection strength, but your priors about facial recognition be mediated by locality and connection architecture.

I think this is why people often struggle with neural correlates of concepts, they try to fit them to a single model of parameter encoding in the hardware (wetware) when there is actually a range of means by which parameters are encoded with quite radically different implications.

our brains seem to model and estimate at the same time. — fdrake

Yes, unlike computers with fixed architecture, our neural hardware is responsive to use, so you have forward acting connections driving processes, backward acting connections modulating responses, and connection building/pruning in response to both (pruning connections being as important as building them, of course).

I'm just suspicious that something of fundamental importance is being missed. — fdrake

Have a look at the function of backward acting modulating neural connections, you may find there the missing piece. They modulate the subsequent forward driving connections probabilistically by restriction of signal at an asymmetric pace with the forward connections. I don't want to drown you in reading material, but the idea is explained in this paper. It may answer some of your concerns.

I don't want to drown you in reading material, but the idea is explained in this paper. It may answer some of your concerns. — Isaac

Eh, I'm still a noob in this, I'll take any free education I can get. The forward/backward propagation steps in Box 3 in the Frisk paper are probably worth me reading more closely too (with the neural implementation of gradient descent through message passing in mind).

The forward/backward propagation steps in Box 3 in the Frisk paper are probably worth me reading more closely too (with the neural implementation of gradient descent through message passing in mind). — fdrake

Yeah, I think it might be what you're looking for. Now I've got to try and get my head round what you and @StreetlightX are talking about wrt concepts.

I'm afraid I use the term rather loosely and it seems to have caused some not inconsiderable concern. For me, 'concept' rather pragmatically encodes what one can do with the data (neural connections/architecture) which the label is collecting. Its all about doing, so I struggle with concepts divorced from actions. I tend to come at this from an input-response model with, if necessary, post hoc analysis into verbally mediated concept talk. Even something abstract like maths, I consider the 'doing' of maths first and the verbal translation of that behaviour secondary.

In a sense I think I'm agreeing with what you seem to be saying, but I think I would tend to frame it as some behaviour still - the 'talking about' is actually the thing that encodes the concepts behind the words, the 'doing the sum' is the thing that encodes the concepts behind the maths.

...but I'm still not sure if that's quite what you both mean.

Final thought on this matter before I retire for the evening. In a striking coincidence, I remember reading about some experiments with monkeys (or possibly chimpanzees), where they were trying to get them to sum quantities of different types of object (so types of types). The first experiment failed, the second had partial success. Their suggestion of a cause...the first set believed too much in the significance of the colour - their Bayesian priors let them down!

Quark.

An odd word, n'est-ce pas? I submit for your consideration, that you, immediately upon reading the word here and now, referenced your experience with it. You didn’t look at the composition of the word by letter or order of letters, you related the word itself, as an established member of a particular language, with some referent, with that which the word is intended to represent. In other words, you already understand the word as a representation, you already understand the word relates to something, because antecedently, you understand and infer from that antecedence, that no word serves any informative purpose if it doesn’t refer to something.

But what, exactly, did Gell-Mann do in 1964, as causality for the absolutely very first instantiation of this particular representational indicator “quark”**? Without regard to the inherent silliness of the word, we can reasonably suppose he wanted nothing but a way to identify this theoretically mandated physical reality, even if such reality had never yet been demonstrated, and may never have been in the case Gell-Mann made a mistake in his theorizing.

From this, two significant predications arise:
1.).....the totality of an existence need not be given in order for a representation to be assigned to it; the totality of its possible existence needs merely be thought;
2.).....simply from the silliness of the name, that even if it is merely the framework for a name that is given, the name does not require any symbolic likeness to it, which in turn is sufficient reason to permit that names can be spontaneously generated without regard to representational pertinence**;
3.)....the spontaneously generated name “quark”, and the framework the name is to represent, in order to maintain logical consistency, must be the same thing, and therefore, immediately upon being named, the concept obtains.
4.)....combining 2.) re: spontaneous generation, and 3.) re: simultaneity, we can conclude that the relation to concepts, that is, their purpose in speculative epistemological methodology, does not in any way depend on a relation in concepts with respect to their development. Thought develops concepts relative to something, but thoughts are not the constituency of them.

** I’m aware of “Finnegan’s Wake”, and the historical precedence of the word. Hopefully, no rebuttal to the philosophical point being made, ensues, for it is obvious Joyce’s and Gell-Mann’s use of the word are only related accidentally.
————————-

Try this: concepts do not begin with naming, but end with it. This way, the presupposition of names is eliminated, as well as their constituency, because the concepts are the names.
— Mww

This approach puts all concepts on equal footing as being the names. It would only follow that there are no concepts prior to naming. I could agree actually, but something tells me that you may not? My agreement to that would lead to a denial that that which exists prior to it's namesake is a concept. — creativesoul

YES!!!! Concepts and naming ARE on equal footing, there are no concepts prior to naming, and it SHOULD be denied that that which exists in its entirety prior to its namesake, is a concept. We are not thereby denying existence of namesakes, whatever its entirety, that is, that which lends itself to being nameable, but rather, we are demanding the occasion for it.

And here, I think, lay the altogether more importance of source, which is the same as occasion, as opposed to constituency, of concepts, with respect to the correlations we both acknowledge for, or by, them.

What I meant by "it's not just a matter of which came first", was that that is an gross oversimplification of the methodological approach needed in order to even be able to acquire the knowledge we're seeking to obtain here. I think you'll agree with this? — creativesoul

Yes. Even if there must be a first, here meaning a first in a methodological approach, a first by itself is meaningless.
——————

Knowing which came first requires knowing what all thought, all belief, and all concepts consist of. For when we know what each consists of, it offers us solid ground to be able to deduce which came first, by knowing what each is existentially dependent upon. — creativesoul

This would be true enough, again with respect to a methodological approach, if it were not for the fact that there are firsts in that approach that have nothing to do with it, per se, but serve as means for its inspiration. You are correct for those considerations within the approach, but we still need a reason, an occasion, to use the method to begin with. And rather than knowing what all thoughts, beliefs and concepts consist of, it is better suited for the method, to know what they do. If they do what they do without contradiction or inconsistency, their constituency may not matter. That being said, there are those things the constituency of which is quite relevant, but it remains to be seen whether the constituency is a population given to them by their use. In other words, a faculty certainly has a population of a priori objects of reason for its constituency, but each a priori object of reason that is a constituent, may not consist of anything. We must nip inevitable infinite regress at the root somehow.

What I'm left wondering still, is not only what exactly is it that you're claiming "always comes first", but "first" - as in prior to what else? I want to say that the primary namesake comes first, but I'm hesitant for you may be saying that the phenomenal object comes first. If it's the latter, then I would agree that some conceptions are of phenomenal objects and in those cases the object 'comes first'. — creativesoul

You got it...it is the object, the primary namesake, some thing of perception, to which the methodological approach can be directed. It appears from your terminology, you accept that the namesake has not been named, and no concept yet applies to it. If we wish to go even deeper into the particulars of the methodological approach, the namesake, while it suffices to call it a phenomenal object, because it is an object that will be named a phenomenon, technically it’s a sensation, and stands as the inspiration for the inductively quantitative evolution, which you readily grasped.
——————-

There is deductively qualitative evolution, as the procedural method itself reduces from the compendium of possible named identities for a phenomenal object to a particular named identity judged as belonging to it, which always comes last.
— Mww

This bit I cannot understand. Could you set it out with an example? — creativesoul

Keyword....judged. If it be granted the quantitative, re: numerical, evolution concerns itself with part of the methodological approach, then the deductive qualitative, re: logical authority, should concern itself with another part. In short, when we understand a particular concept belongs to....names....a phenomenon, without contradicting what the concept has previously named (what we already know), we at once deduce what was previously a general unnamed sensation, is indeed cognizable as a certain named object. We are in effect, evolving from the general to the particular. It is how we arrive at experience, the last of the cognitive chain, called empirical knowledge.

I thought we would agree there. — creativesoul

I’ll address this, and forward my agreements with your comments shortly. I don’t want you to think there’s no common ground going on here.

I don't see how they can be distinguished. When a chimpanzee rejects a previously gratefully accepted cucumber as reward on the grounds that the other chimp has been given a grape, are they not using a model of justice, or fairness? — Isaac

What would you even be talking about when you say things like this given that you don't think the world has any properties? Are you just talking about things your own mind creates?

When a chimpanzee rejects a previously gratefully accepted cucumber as reward on the grounds that the other chimp has been given a grape, are they not using a model of justice, or fairness? — Isaac

This seems to be all in the interpretation: alternatively, it could be down to a feeling of envy or a preference for grape over cucumber.

Yes, bridging would certainly be better than conflation...

I would also propose that some of our concepts are capable of describing and/or pointing towards that which existed in it's entirety prior to our reports.
— creativesoul

Of course, no argument here. Their names are in the literature, if one knows where to look. Do you have names for them of your own, or from some other literature?
— Mww

Do I have names for those concepts of my own? (...) To directly answer your question, or at least what I think you're asking me for...

Thought, belief, meaning, and truth all exist in their entirety(on the most basic level/degree of complexity) prior to our conceptualizations/names of/for them... that is... prior to common language use. — creativesoul

OK, I’ll buy that. At the very lowest level, these are concepts, and they can be empty of content. There must be truth, on order for something to be true.
——————-

, if all A's consist of B, then no A exists prior to B. If all A's consist of B, then each and every A is existentially dependent upon B. That which is existentially dependent upon something else cannot exist prior to that something else. These are the sorts of reasoning that come into play here. — creativesoul

Ok, sound logic, yes.
——————

There's no difference between our conception of games and games. — creativesoul

Yes. Reflecting back to names and concepts being equal here, in this dialogue, and the point I was making awhile ago in another where it was proposed that the criterion of the possibility of a conception (not of its object) is the definition of it. It is the name of the objects of concepts that are different than the name of the concept.
———————

Correlations drawn between different things are the building blocks of everything ever thought, believed, spoken, written, and/or otherwise uttered. — creativesoul

Agreed, without equivocation. I think we need to stipulate correlations, in order to proceed.

Your turn. Agree, disagree, question all the above as you wish.

There's much to be said about our discussion. There are certainly several aspects I'm looking forward to getting into.

However, I think that I just realized that there's a bit of misunderstanding happening with regard to my own notion of elemental constituency and/or constituents(of our concepts). Specifically, it seems that you've taken that to include things like letters and such. Now, to be clear, that's not wrong(per se) but it is an incomplete understanding. By my lights, understanding is imperative to good, productive, and valid discourse. I'm assuming we both seek just that...

Just as I hope to better grasp the position you're arguing for/from, I also hope to be able to clearly state my own so that those who are capable can have a good grasp upon it as well. You seem perfectly capable. I mean anyone who can follow Kant, and particularly those who can use his framework from memory, which you seem to be doing, ought be able to follow the position I'm working from and developing.

In short, I would think that it would help you (or anyone else for that matter) to better understand my position if you kept in mind that I reject several different historical dichotomies. Most of the historical ones actually. This pertains directly to elemental constituency and existential dependency because correlations are neither immaterial nor material, external nor internal, mind nor body, objective nor subjective...

None of these dichotomies are capable of taking proper account of that which consists of both sides of the dichotomy. They are all inherently inadequate for taking proper account of correlations drawn between different things. They(correlations) consist of both... material and immaterial, external and internal, mind and body, objective and subjective.

The consequential scope of this could seem daunting. That feeling passes as things come into clearer view. That clearer view is practically inevitable when one consciously rejects the aforementioned historical inadequate dichotomies and the frameworks surrounding them. Witt's "bewitchment of language(use)" applies here. Rejecting inherently inadequate frameworks and dichotomies is akin to letting the fly out of the bottle, so to speak.

All this will apply to my next reply to you, as well as all the rest, and all previous ones. You may want to re-read some of them in order to sharpen your understanding given the new translation tools. The "quark" episode looks promising. Very timely. Very relevant.

Kudos.

When a chimpanzee rejects a previously gratefully accepted cucumber as reward on the grounds that the other chimp has been given a grape, are they not using a model of justice, or fairness?
— Isaac

This seems to be all in the interpretation: alternatively, it could be down to a feeling of envy or a preference for grape over cucumber. — Janus

Just wanted to mention something here. Do with it what you may. It's just dead on point relevant... on my view anyway.

Interpretation is always of something already meaningful. Our interpretations of what's going on in the chimps mind are nothing more and nothing less than reporting upon the meaningful thought and belief of the chimp at the time. Those are meaningful to the chimp... we can get them wrong if we do not work from an adequate criterion for non linguistic thought and belief.

I cannot make sense of any notion, conception, or model of fairness/justice without rather complex language use that talks in terms about what one wants as compared to what one gets(or has). Without this comparison, there is no notion of fairness or justice.

Can chimps compare what they have actually received with what they want?

Seems to me that that is impossible without the ability to compare one's own thought and belief(what they want) with what's happened and/or is happening(what they received). Thus, on these grounds(the chimp doesn't have what it takes to have a notion of fairness/justice) I can only conclude that the chimp quite simply did not want the cucumber, but rather wanted the grape.

...rather than knowing what all thoughts, beliefs and concepts consist of, it is better suited for the method, to know what they do. If they do what they do without contradiction or inconsistency, their constituency may not matter. That being said, there are those things the constituency of which is quite relevant, but it remains to be seen whether the constituency is a population given to them by their use. In other words, a faculty certainly has a population of a priori objects of reason for its constituency, but each a priori object of reason that is a constituent, may not consist of anything. We must nip inevitable infinite regress at the root somehow. — Mww

Acquiring well-grounded true belief about what all thought, belief, and concepts consist of is a methodological approach. I also do not think that these are mutually exclusive goals. We can do both, acquire knowledge of what our thought, belief, and concepts do alongside acquiring knowledge concerning what they consist of.

A faculty, which I'm taking to be the Kantian notion, could be sensibly said to have 'a population of a priori objects of reason' for it's constituency. I'm a bit confused here though. Which a priori object of reason does not consist of anything?

Infinite regress is the least of my concerns. It's not a problem on my view.

I tend to agree: I actually encountered a report of a study which addressed this very issue, and concluded that it was just a preference, but I haven't made much effort to find it again.

From memory as rewards they gave the chimp cucumber first, then grapes; and found that after having the grapes the chimp rejected the cucumber (not sure of it was grapes and cucumber in the study but I stuck to those as they were the tendered examples).

Unfortunately many purported experts in the field of the human mind do not work from a framework that is even capable of drawing and maintaining the actual distinction between linguistic and non-linguistic thought and/or belief.

The result is misattributing meaningful content that quite simply is not there. A sense of justice/fairness is a noble aim to identify prior to language. I mean, that would be great step in the right direction for a well-grounded version of universally applicable ethical considerations. However, there's inadequate evidence to suggest that the chimp is either involved in Bayesian reasoning(I know that's currently popular) or working from a modell of justice/fairness.

Those kinds of social situations(not having what one wants, and not receiving what one expects) can most certainly be a part of one's developing such 'senses' as justice and fairness, but that's only after one begins talking about situations where one's expectations had already long since run afoul. Fairness is an assessment of what's happened.

Ah well... you'll have that!

...there are those things the constituency of which is quite relevant, but it remains to be seen whether the constituency is a population given to them by their use. — Mww

Remaining to be seen presupposes the possibility of something or other actually happening. If we are looking to see if we populate the concept and we hold that there are no concepts prior to names, then we cannot admit anything else but...

We populate concepts by virtue of using names.

So...

What gives?

It doesn't remain to be seen. We cannot see such things aside from seeing the consequence of our premisses.

Either all names are concepts, or the elemental constituency of some concepts does not include naming practices and other such language use(descriptive practices) even though our knowledge and/or awareness of them does.

That which exists prior to our naming practices are not concepts if concepts are equal to names. How do you square this with the equivalency between concept and name that you drew earlier?

Sorry... the critical hat was on.

:wink:

Is anyone interested in talking about the phenomenological approach (Husserlian)?

I just don’t really see where this discussion is going? There appears to be a lot of emphasis on ‘concepts’ and ‘naming concepts’ ... why?

I just don’t really see where this discussion is going? There appears to be a lot of emphasis on ‘concepts’ and ‘naming concepts’ ... why? — I like sushi

How did Banno say it???

It's just a bunch of esoteric language use that only philosophers care about.

The focus is on non linguistic thought and belief, per the OP. Some experience is had by non linguistic creatures. All experience consists entirely of the thoughts/beliefs of the creature having the experience. Some of us, myself included, are offering argument in support of our earlier remarks regarding "what it's like to experience X" where "X" is something that happened to a thinking and believing subject.

I claim that we cannot even offer an adequate report if we do not know what all thought and belief consists of. Experience is thought and belief based. All of it. That's my synopsis anyway.

So it is a phenomenological approach then? There was some theory crafting regarding neuroscience. I don’t seem to be able to find a way in here :(

Say what you're thinking.

The neuroscience is beyond my comprehension. I've been waiting for them(the experts, specialists, and groupies in/of the field) to admit that there is no one to one mapping between brain activity and particular thought. Many different thoughts correspond to virtually the exact same brain state. Admittedly, our technologies are offering more and more knowledge. However, there's no indication that the hard sciences(or soft ones for that matter) have acquired enough knowledge about our own thought and belief to be able to assert much at all. Thought and belief(thinking about stuff) involves firing neurons, and different physiological biological structures and systems, but they most certainly do not consist entirely thereof.

So it is a phenomenological approach then? — I like sushi

I believe that Mww is arguing from such a position. I do not. It denies direct perception the actual role it plays in rudimentary level thought, belief, and experience. However, I'm currently considering another's position. Sometimes I'm pleasantly surprised.

:smile:

I believe that Mww is arguing from such a position. I do not. It denies direct perception the actual role it plays in rudimentary level thought, belief, and experience. — creativesoul

It does no such thing. Anyway, I’ll just keep at eye out and jump in if I can. Probably not though tbh

Thanks

Ugh!

Really?

Phenomenology works from the notion of indirect perception. It presupposes two worlds. The real one(which we can know nothing about aside from it's existence) and that which appears to us(the phenomenal world)...

That's indirect perception of the real world.