An argument against neural reductionism based on the necessity of abstract ideas

Question

Reductionism in philosophy of mind is the view that mental phenomena, such as thoughts, emotions, and consciousness, can be reduced to, or explained solely in terms of, the physical or neurobiological processes which are the subjects of neuroscience. Reductionism maintains that such mental events can be explained in terms of physical reactions in the brain, such as the patterns of the firing of neurons and the release of neurotransmitters across synapses. That is contrasted with non-reductionist views that hold that mental phenomena are irreducible and cannot be explained in such terms.

While there's a lot of talk about 'qualia' in this context, I think a more robust defense of non-reductionism can be made in terms of the mechanisms of meaning. Whenever we make an argument for processes being physical in nature, or mental in nature, we constantly invoke comparisons - that such and such is this, or is not that, or is the same as this, or is different to that, or means this, or does not mean that. And I think it's those commonplace instances of the use of inference that can't be explained in terms of neuroscientific analysis, because they are dependent on the intellectual act of abstraction. And abstractions are only perceptible by a mind, as a matter of definition. They're not physical things.

The precursor for this kind of argument goes back to the Phaedo, in the dialogue on the nature of equality. The argument goes that we are only able to say that two things are the same (equal) because we already have an idea of equal - we already know what the meaning of equal is. Socrates famously argues that this is because the soul already possesses the idea of equal and is recalling it, but for our purposes, it suffices to say that the idea of equal, or the ability to recognise equal, is innate. Socrates points out that in real life, no two things are exactly equal - two lengths of wood, or two stones, are not exactly equal, but because we already know the idea of equal, we can say of those two things that they're the same. And much of our comparative and analytical thinking is based on this. In practice, the only time that equality is absolute is in respect of number or symbolic logic - the equals in 2 + 2 = 4 , or in the statement that A = A, is exact, because symbols are defined in precise terms. In real life, the objects we encounter day to day are only amenable to that level of exactitude to the degree to which they can be represented mathematically (whence the extraordinary power of mathematical physics and other quantitative sciences.)

Now I suppose a brain-mind identity theorist might respond that even though an abstraction is not physical, the act of mental representation that stands in for it is. He might say that every instance of such an idea is 'really' just a configuration of neurons or a pattern of excitation. The problem with that is, that there is not even a demonstrable relationship between the symbolic expression of such ideas as 'is not' or 'is equal to' and the meaning they embody. Any such expression can be encoded in any number of symbolic forms or in different languages, while still retaining their meaning; the meaning is separable from the symbolic form. So it seems to me that trying to show that a neural pattern could literally be the same as such an expression doesn't allow for the flexibility that is characteristic of thought and speech. It doesn't allow for abstraction. And furthermore, such an explanation must invoke the very power that it's seeking to explain; you couldn't even get to first base without the ability to make such judgements about what the neural data means. It's not as if you could stand outside the act of judging and see it objectively, without judgement, as then you will have divested it of meaning.

The purpose of this thread is not so much to debate the pros and cons of this argument, as to find out whether it is a recognised line of argument in philosophy of mind and whether there are philosophers who pursue something like this line of argument.

Answer 1

I think a more robust defense of non-reductionism can be made in terms of the mechanisms of meaning.

I'd agree with that assessment. Various computational models of vision grow more and more sophisticated, such that it's arguable a computer system that sees a face does so by recognizing colors, contours, shading, and so on. That is, an electromechanical system that converts wavelengths to integers that stand for RGB in some way begins to mimic how we see red, green, and blue; or straight lines and curves by interpreting a CCD's state much like the state of a retina is observed. For meaning, despite that we can locate difficulties in the application of semantic function to Wernicke's area when particular types of aphasia are observed, meaning in the brain as a whole is far less understood than visual processing.

The purpose of this thread is not so much to debate the pros and cons of this argument, as to find out whether it is a recognised line of argument in philosophy of mind and whether there are philosophers who pursue something like this line of argument.

I think the basic, intuitive argument implicit in any philosophical position that embraces mind-body duality as fundamentally irreconcilable in the form:

Mind is mental, body is physical, they're exclusive, and all meaning is mental, therefore nothing in meaning can be reduced to the physical. These sorts of philosophical arguments might be considered folk philosophical arguments against reductionism that are rooted in psychological intuitive dualism. Take for instance the comment:

Um... Mind cannot be matter because mind does mind stuff that isn't matter. OK. Maple syrup cannot be matter because maple syrup tastes good and taste isn't matter. – Boba Fit

This is essentially the intuitional approach for resisting mind can be reduced to matter, and there's scientific evidence this is the norm in conceptual development. Consider from "Developmental and Cross-Cultural Evidence for Intuitive Dualism" by Chudek et al.:

Adult humans behave in peculiarly "dualist" ways. Adults everywhere profess beliefs in souls and the afterlife (Atran & Norenzayan, 2004; Boyer, 2001), mindless bodies (zombies), bodiless minds (ghosts, spirits) and minds entering new bodies (Cohen, 2007; Cohen & Barrett, 2008) and place unusual value on "genuine" artifacts (Bloom, 2005) as though they'd become infused with their owners' non-physical essence. Even medical professionals show reasoning biases when thinking about 'psychological' versus 'physical' disorders(Ahn et al., 2009). Cross-cultural research suggests that adults' intuitions about disembodied minds are strikingly similar across societies (Cohen et al., 2011). Young children sometimes expect minds to persist after their body dies (Astuti & Harris, 2008; Bering & Bjorklund, 2004). Both children (Notaro et al., 2001; Schulz et al., 2007) and adults (Ahn et al., 2009) struggle to draw causal connections between mind- and body-related phenomena. Recent cognitive-historical work even indicates dualist thinking in ancient Chinese texts (Slingerland & Chudek, 2011).

But they also say:

Alternatively, dualism may be learned gradually by participation in cultures with a Cartesian intellectual tradition.

So, the reason that Descartes was the target of Ryle's label as categorical error maker is that there are plenty of anti-reductionist arguments starting with and since Cartesianism.

Currently, the debate over the possibility of AGI can be seen as an extension of this divide between irreconcilable dualism and the belief that meaning is inherently a physical process of association of some sort. Hoffstadter and Sander explore the topic in Surfaces and Essences: Analogy as the Fire and Fuel of Thinking. While some objections to mapping mind to gray matter take the form of multiple realizability (SEP) as an objection which attacks reduction by looking at the relationship between how mental states might reduce to physical systems, others make an argument from the other direction: the physical systems we build fail to create mental states or possess meaning. Thus, narrow-AI is held up as an example that our most sophisticated attempts to create minds fall spectacularly short of modeling mind and meaning, and that semantic hallucinations by LLMs such as ChatGPT show that there is something quintessentially missing. Of course, the most famous philosophical argument of this sort is Searle's Chinese Room argument:

The Chinese Room Argument holds that a digital computer executing a program cannot have a "mind", "understanding", or "consciousness",[a] regardless of how intelligently or human-like the program may make the computer behave.

Thus, the question of how meaning and physical matter relate is very much a real-time argument undergoing testing and argumentation in experimental philosophy (SEP). The two sides essentially are trying from the bottom-up to reason to the best conclusion with people settling in two camps, the orthodox made famous by Hubert Dreyfus who became famous by point out what computers can't do largely because they don't understand, and the heterodox which is a rag-tag band of academic dreamers like the original optimistic AI folks, and current self-proclaimed AGI folks who believe that meaning can be built into systems one learning algorithm at a time.

Currently, one group of thinkers working on this problem are 2nd generation cogntive scientists with a philosophical bent like Nicholas Shea who are trying to show connections between physical systems and representations. If the representations have a physical basis, then by the representational theory of consciousness (SEP), the problem offered by intentionality is resolved by reducing intentionality to representations and showing how representations give rise to consciousness. To wit:

The idea of representation has been central in discussions of intentionality for many years. But only more recently has it begun playing a wider role in the philosophy of mind, particularly in theories of consciousness. Indeed, there are now multiple representational theories of consciousness, corresponding to different uses of the term “conscious,” each attempting to explain the corresponding phenomenon in terms of representation. More cautiously, each theory attempts to explain its target phenomenon in terms of intentionality, and assumes that intentionality is representation.

Where does meaning fit into this? Well, if meaning is primarily an association between representations (for instance as posited by cognitive semantics and conceptual metaphor which views meaning as being derived from source-target domains of neural computations), then there are physical computational mechanisms that can be shown to underlie the mental faculty of reasoning. This is why 2nd generation cognitive scientists offer mathematical models of neural computation and look to the neuroscience of consciousness (SEP) for inspiration. Unlike LLMs, the goal is to produce connectionist systems (like those proposed years ago by Rumelhart, McClelland, et al. in PDP vols. 1&2) that manifest behavior that might be considered purposeful and meaningful.

Philosophically speaking, there are still many open questions in the philosophy of language and semantics itself and a bewildering array of models and viewpoints. WP has half a dozen theories of semantics listed, and many of these aren't known in the wider philosophical communities, but are of interest to philosophers of language with a particularly strong knowledge of linguistics. Szabo and Thomason in their Philosophy of Language (GB) represent semantics as a field rife with philosophical questions and pose a number of questions and assertions their own from the origins and role of compositionality in language function to how indexicals relate to theories of modality.

As such, there might not be a single theory like multiple realizability which creates a strong objection to be overcome, but rather myriad open questions about semantics itself that prevents reductionists from articulating a strong theory. RTM, consciousness, and the study of semantics is still a bit of a frontier, and mastering its intricacies means focusing not only on historical arguments such as those Descartes makes, but on an increasingly divergent philosophies-of and sciences to find meaning in the discussion. One can say that as computer technologies increasingly become sophisticated potentially drawing from bigger distributed systems, quantum computing, and broader and stronger narrow-AI, it will certainly continue to be a lively debate at what point if ever AI can be said to possess intentionality, meaning, self-awareness, and maybe a theory of mind. This after all, may be the proof in the pudding one day to show that meaning and human-level intelligence reduce to sophisticated systems of physical computation (SEP).

Answer 2

And I think it's those commonplace instances of the use of inference that can't be explained in terms of neuroscientific analysis, because they are dependent on the intellectual act of abstraction. And abstractions are only perceptible by a mind, as a matter of definition. They're not physical things.

Here you are assuming what you wish to prove. You wish to prove that the mind is not reducible to physical things. But here you assume that something that is "only perceptible by a mind" must not be a physical thing (because it is perceived by the mind, which you assume to be non-physical).

The problem with that is, that there is not even a demonstrable relationship between the symbolic expression of such ideas as 'is not' or 'is equal to' and the meaning they embody. Any such expression can be encoded in any number of symbolic forms or in different languages, while still retaining their meaning; the meaning is separable from the symbolic form. So it seems to me that trying to show that a neural pattern could literally be the same as such an expression doesn't allow for the flexibility that is characteristic of thought and speech.

Yes, the meaning is separable from the symbolic form; I grant you that. But now answer me this: is the meaning of a symbolic form separable from the way the symbolic form is used?

To give an example, is the meaning of the word "cat" separable from the way we use the word "cat"? If everyone used the word "cat" in sentences where they presently use the word "sofa," wouldn't the meaning of the word have changed?

And this manner-of-usage is an entirely empirical matter. A scientist can study the sentences in which "cat" appears and where it appears in those sentences; that is measurable.

So it seems that a great deal of meaning, perhaps all of meaning, is reducible to something empirical and measurable. When we look not at the symbol in isolation, but when we look at how the symbol interacts with the system that it is part of, there we will find the meaning of the symbol. The meaning of the symbol is the structure of interaction between the symbol and its system. Certainly such structures of interaction can be found in the brain; neural correlates of ideas are not inert symbols, they are causally active, influencing and being influenced by the rest of the brain-system.

Answer 3

And abstractions are only perceptible by a mind, as a matter of definition. They're not physical things.

What should we make of the case of numbers? They don't seem to be physical things, so maybe we should see them as abstractions of some kind. The number "2" would be an abstraction for all sets that contain two things, maybe.

But if that is granted, don't we now have something abstract that can live outside a "mind"? For example, our computers are full of numbers, and they infer (calculate) with those numbers all the time.

Is that useful in this discussion? Maybe some mental acts, such as calculating with integer numbers, can be reduced to algorithms executed outside of "minds" (on computers)?

Answer 4

And abstractions are only perceptible by a mind, as a matter of definition.

There you go. Our definitions that arose from a time humanity spend in religiosity and dualist superstitions make it hard to reason about the apparently material reality.

When definitions become the obstacle to understanding reality, they need to be changed.

So it seems to me that trying to show that a neural pattern could literally be the same as such an expression doesn't allow for the flexibility that is characteristic of thought and speech.

Nowhere in reductionism is it claimed that a given neural pattern would be the same as an expression. That is a strawman argument.

So it seems to me that trying to show that a neural pattern could literally be the same as such an expression doesn't allow for the flexibility that is characteristic of thought and speech.

That is the argument from incredulity fallacy. "I cannot imagine neural networks to have flexibility. Even though I don't know enough about them but still I'll use my lack of imagination as proof of infeasibility. Instead I'll believe in some form of miracle instead, as that is totally possible. Miracles happen all the time around us. That's easy to believe. Flexibility if neural networks? What a crazy idea!"