[FRIAM] model, metaphor, and things

[⛧ glen]

Very cool. It'll take me a minute to parse that 2nd paper. But I wanted to also mention that within animal learning may not be all that different from between animal learning (e.g. science). I'm always criticized for suggesting that brain function isn't very well modeled by point (artificial) neurons, because I'm too ignorant to make the argument. But things like glial cells, the enteric cluster, octopuses,  neuron types, etc. just make me feel that networks of animals may seem very different, but may not actually be different. Toss in skepticism about the separation of cultural from biological evolution, and others' confidence in their rhetoric seems pretty _sus_.

On May 1, 2022 11:40:18 AM PDT, Marcus Daniels <marcus at snoutfarm.com> wrote:
>I'm reminded of the Zen quote "In the beginner's mind there are many possibilities, but in the expert's, there are a few."
>This seems to fall out naturally from sparse coding of information.  Sparse coding in animals has been studied going back to the 1970s. [1]   In particular, the observation that relatively few neurons are active at any given time, and that there are mechanisms for lateral inhibition.  If the coding were dense (metaphors being reused and modified), then one would expect that old metaphors would be sanded down to generalities as more experience was gained, and then augmented with particulars to be useful.   If that occurred, one would see the neurons involved in the core pattern often activated and then specialization neurons activated for refinement.   This approach would have the disadvantage that unlearning would be as essential as learning to avoid getting locked in local minima.   Alternatively, if there is sparse coding using a growing and overcomplete basis set, then one would expect to see less recurring use of neurons as the inventory of patterns to compose would be larger, but there would also be a requirement that not too many of them be used at once -- more of a filtering down rather than building up would be occurring to separate patterns. [2]
>
>[1] https://scholar.google.com/scholar_lookup?title=A%20theory%20of%20cerebellar%20cortex&journal=J.%20Physiol.&volume=202&pages=437-470&publication_year=1969&author=Marr%2CD
>[2] https://www.nature.com/articles/s41467-017-01109-y
>
>-----Original Message-----
>From: Friam <friam-bounces at redfish.com> On Behalf Of glen?C
>Sent: Sunday, May 1, 2022 8:16 AM
>To: friam at redfish.com
>Subject: Re: [FRIAM] model, metaphor, and things
>
>While I technically agree, I can't help but ask: What if it's not true? How would we go about demonstrating that it's not true? What if animals actually "relax into" their envinronments in a computationally universal way? I.e. although it seems like we learn by analogy, maybe we're actually universal algorithms that are *fitted* to reality?
>
>As "models" of reality, then, we would be over-fitted to past (or elsewhere) contexts and ill-fitted to new contexts. The steepness of the learning curve would explain well enough the registration error McGilchrist describes as model choice error. So the mere ill-fittedness wouldn't be enough to test the hypothesis. We'd need something more, perhaps a *type* of ill-fitedness? What type of ill-fittedness comes from bad analogies versus the types that result from over-fitting to prior (or other) contexts?
>
>Of course, the hypothetical dichotomy (between analogical fallacy versus in-context training) is probably a false one. We most likely do *some* analogical learning and some in-context learning. But using the (false) dichotomy to test conditions/contexts will help us anyway, in categorizing which situations exhibit analogical learning versus in-context learning. And, of course, someone convicted by their confidence in "metaphors everywhere" will claim that in-context learning reduces to analogical learning. Maybe. But mere assertion is sophomoric. How can we test the assertion? That's where the work (and credibility) lives.
>
>It's just a bit galling to assert one's stance without also providing a way to think clearly about the assertion and methods to falsify/test it. So the kids get it right, yet again, by asking "Is it, though?"
>
>On 4/30/22 16:43, Prof David West wrote:
>>  From Iain McGilchrist, /The Matter of Things, vol I, Our Brains, Our Delusions, and the Unmaking of the World/. (Top 5 most important books I have read.)
>> 
>> "Explanation, science's forte, is a subset — an explicit, rigorous, disciplined subset, but still a subset — of understanding. All understanding depends on metaphor. What we mean when we say we understand something is that we see it is like something else of which we are already prepared to say 'I understand that'. That, in turn, we will have understood because we have likened it to something else we had previously understood, and so on. It's metaphors all the way down.
>> 
>> In science this inescapable role of metaphor is manifest in the model the science uses in order to seek an explanation of the phenomenon it is investigating. Models are simply extended metaphors. The choice of model is crucial here because the problem for seekers after truth is that that choice governs what we find. We find more or less according to what we put there. Since a model always highlights those aspects of what it is modelling that fit the model, any model soon begins to seem like an uncannily good fit, which means we espouse it with still greater confidence.
>> 
>> Even the sense data that go into selecting the model are not innocent. Perceptions are laden with theory. We never just see something without seeing it *as* a something. We may think that our theories are shaped by observations, but it is as true that our observations are shaped by theories. This means we can be blind to some very obvious things in our immediate environment. We don't look where we don't expect to see so that our expecations come to govern what we *can see*. This is why the model is crucial. In the past such a model was often something in the natural world — a tree, a river, a family. Nowadays, unless otherwise specified, it is the machine."
>> 
>> Quotation of Evelyn Underhill:
>> 
>> "It is notorious that the operations of the average human consciousness unite the self, not with things that really are, but with images, notions, and aspects of things. The verb 'to be', which he uses so lightly, does not truly apply to any of the objects amongst which the practical man supposes himself to dwell. For him, t/he hare of Reality is always ready-jugged/: he conceives not the living, lovely, wild, swift-moving creature which has been sacrificed in order that he may be fed on the deplorable dish which he calls 'things as they really are'."
>> 
>> Quotation from /The Function of Reason/, by Alfred North Whitehead:
>> 
>> "The man with a method good for purposes of his dominant interests is a pathological case in respect to his wider judgement on the coordination of this method with a more complete experience. Priests, scientists, statesmen and men of business, philosophers and mathematicians, are all alike in this respect. We all start by being empiricists. But our empircism is confined within our immediate interests. The more clearly we grasp the intellectual analysis of a way [of?] regulating procedure for the sake of those interests, the more decidedly we reject the inclusion of evidence which refuses to be immediately harmonized with the method before us. /Some of the major disasters of mankind have been produced by the narrowness of men with a good methodology./"
>> 
>> BTW— jugged hare is an English country dish: hare marinated in red wine and juniper berries, then slow cooked with some of the hare's blood. AKA /civet de lievre/.
>> 
>> I find McGilchrist's books exhilarating—2,000 plus pages of rich empirical evidence and densely reasoned argument that supports almost all of the essays, books, and ideas I have been espousing for 25 years. Including, a path for incorporation of all the hallucinogenic experiences.
>> 
>> 
-- 
glen ⛧