I think the brain is a computational system and what we generally refer to as the mind and consciousness are some of its computations. But I’m also aware that the brain works very differently from how a typical digital computer works. One criticism of computationalism that I have some sympathy with is the word “computation” can lead people to think about neural processing in the wrong way.
In technological computing, we generally have a distinction between code and data, between processing and storage. We write programs, lists of instructions, that operate on that data. Of course we often forget that code is itself data. (Which causes a lot of security issues.)
But even at the hardware level, there is a distinction. Most of the computing device you’re using to read this is dedicated to storage: memory chips, SSD storage, etc. The actual doers of the system, the processors, actually make up a small (but expensive) portion. This works because the system has the ability to accurately copy data from storage into processor registers, act on them, and then copy the results back into storage, the benefits of a discrete digital system.
But that’s not how neural networks work. Instead, data and processing are together. We can insist that the strength of synaptic connections count as storage, but since they mediate signaling between neurons, they also count as a kind of code. So the brain, being a physical neural network, doesn’t seem to store information and then act on it. The storage and the acting are one and the same.
What that means is that talk of data being moved, sent, or copied between brain regions is probably a shaky metaphor at best. Instead, some regions react to incoming signals in certain ways, propagate those reactions in signals to other regions, which react in their own way. We can think of the sensory regions as the early reactions and the later higher order regions as reacting to the earlier reactions.
Some of those reactions signal back to the earlier regions, creating recurrent signaling loops. And eventually some of the signaling reactions reach motor regions, which cascade down efferent connections back to muscles, glands, etc. Of course this is what brains are for, to figure out what the best course of action is for the organism.
People often scoff at the idea of a grandmother or Jennifer Aniston neuron, but under this view, it’s more plausible than it might seem. It’s important to remember that this isn’t the only neuron that fires with these concepts, just that the convergence of many other reactions (a human, a woman, certain hair color, certain facial features, certain behaviors, etc). It should be seen as the very tip of a vast reactional iceberg.
And the convergence point is probably different today than it was years ago, as the concept of grandmother or Jennifer Aniston, and what it means for us, changes over time. These changes, called “representational drift“, concern some theorists. But to me they’re only a concern for someone caught up in thinking of a neural representation as data in the same sense as it would be in a technological computer. Once we realize it isn’t, that it’s a set of reactive dispositions, dispositions which can change over time as new things are learned, the drift makes a certain kind of inevitable sense.
This unity of data and processing is why I think understanding information as causation works better than most people realize. The most common reactions against it seem to stem from the idea that causation is action while information itself is inert. But that’s working from the distinction used by current technology. Evolution doesn’t seem to have ever used or cared about that distinction.
Consider what this means for the description of a leading theory of consciousness: global workspace theory. It’s often described in a manner that makes it sound like information in the brain needs to get into a certain location, and once in that location that it’s broadcast throughout the brain. I’ve used that language myself. It’s natural to fall into since it quickly gets the idea across. But it sets up a picture that is arguably misleading, leading some theorists to create models based on that picture rather than the messier reality.
Consider instead a system, a network of neural networks. The sensory portion of that network reacts in certain ways to stimuli. The various reactions all try to propagate. The signal forward is an attempt to create a circuit of reactions. The signals to the sides, to the other reactions happening in parallel, is inhibitory (lateral inhibition). In other words, we have a series of reactions creating multiple parallel signaling circuits, each trying to propagate while inhibiting the others.
As the signals propagate into the network of neural networks, creating ever more elaborate circuits, some of those circuits excite the same regions, which in turn signal back exciting the contributing circuits even more. This leads to coalitions of circuits, and competition between those coalitions, each trying to build themselves while inhibiting the other coalitions.
In standard global workspace theories, eventually one of the coalitions crosses a phase shift boundary and becomes omni-dominant. Its reactions have “entered the global workspace”. In closely related theories, like Daniel Dennett’s multiple draft model, the final resolution is less clear cut. No coalition ever wins completely. It’s the coalition that manages to excite the language and episodic memory centers in certain ways that are retroactively considered to be conscious.
But in both cases, it’s the coalition of circuits which end up dominating the system that become reportable and memorable, that essentially become perceived as part of the stream of consciousness. So alternate names for global workspace could be “global coalition of circuits” or “global coalition of dispositions”, while Dennett’s could be called the “multiple coalitions model”.
Understanding this unity of data and processing also makes it more clear why the idea of everything being piped to one control center is misguided (at least a control center other than the brain overall). Different mental content is different parts of the network reacting, not any one central part reacting to the data. Of course, it doesn’t feel like that, but that’s because we can’t introspect the parts.
Unless of course I’m missing something?
SelfAwarePatterns 
An excellent post, and some excellent ideas. So the following is just my take.
I think one thing you’re missing, not in the sense of missing the mark but in the sense of not yet utilizing, is Ruth Millikan’s unitrackers. It’s not so much that you have a Jennifer Anniston neuron but a Jennifer Anniston unitracker, which is a set of neurons which essentially do the things you mention: inhibit nearby competing unitrackers, nudge more distant unitrackers both upstream and downstream, and potentially contribute to (at least one) distant sub-cortical site. Seems to me that describes a cortical column pretty well.
*
[gonna add other comments separately]
LikeLiked by 1 person
I don’t think the total act of cognition of Jennifer Anniston resides in a single or even a small set of neurons. There are neurons firing all over the brain when we see a picture of Jennifer Anniston. It might be that object recognition and object recognition alone resides in a small set of neurons. In other words, the conscious quale tied to that neuron has probably little visual or factual information, In has nothing more than a recognition of a face which another part of the brain can tie to the name “Jennifer Anniston.”
So I don’t know how useful the concept of “unitrackers” is to what is happening.
LikeLike
Agreed the total act of cognition does not reside in one unitracker. The activation of a high level unitracker would involve the prior and ongoing activation of multiple other unitrackers, and would subsequently prime a host of yet other unitrackers, such as the one for Rachel, her character on Friends, as well as the other characters on Friends, as well as the other actors on Friends, etc. And as you say, the unitracker for Jennifer Anniston would not so much hold any visual information, but its activation would prime associated unitrackers in the vision areas, or may activate specific image memories, which would then activate those vision areas.
*
LikeLiked by 1 person
Maybe I don’t understand what unitrackers are but the term begins with “uni” which usually is an indication of one. I’m also not sure there are “levels” either of unitrackers or whatever we call them. They seem more like just separate units of cognition, no levels required for the simplest explanation.
LikeLike
A unitracker essentially “tracks” one concept. “Tracking” can mean becoming activated when a particular pattern of other unitrackers are activated, but there can be multiple such other patterns. Such other patterns can include words on a screen, or an image, or a sound. Also, activation of that one tracker could prime all the related trackers. So yes, they are separate units of cognition, and you don’t need levels for the simplest explanation, but you do need levels for the full explanation.
LikeLiked by 1 person
If your definition of unitracker/concept is broad enough to include sense qualia, then I could agree in part except I don’t see any idea to use a new term for qualia. They are all artifacts of consciousness or units of cognition.
If your definition isn’t that broad, then I would need to understand the boundaries of “concept” and find out, for example, where color would appear in the scheme.
When you talk about unitrackers activating related trackers, it sounds like you are really just talking about clusters of neurons. And I don’t know where the levels are in the brain. There are certainly different regions that do different things but the only “level,” if it could be called that, is the distance from sense information. Again, I don’t see any advantage to a new term for a cluster of neurons.
LikeLike
I think my (well, Millikan’s) definition of unitracker is broad enough to include sense qualia, except we may have a discrepancy in how we use “qualia”. As for the levels, I think you have it right in that they correlate with the distance from the sensory events. BTW, this is pretty much how they work in the LLM’s like ChatGPT. Simple patterns build up to complex patterns in the higher layers.
In the case of the human brain, unitrackers are “just” clusters of neurons, but they are (conjectured to be) organized in a unit (cortical column). My current hypothesis is that for each cortical column (so, unitracker) there is exactly one neuron that sends an axon to the thalamus where it can potentially become integrated w/ other unitrackers for purposes of higher functions. Attention determines which cortico-thalamic activations are effective. (40hz?)
*
LikeLiked by 1 person
Sounds like we are somewhat closer in view than I ever thought we would be. Except I still don’t like the term “unitracker” and fail to see what advantage it has over neuron clusters or maybe assemblies.
I would doubt that your current hypothesis is correct. I think the active units probably span columns in most cases. You might look at some of the research on the vortex patterns in neuron firings which some researchers believe to be associated with cognition.
“Simple patterns build up to complex patterns in the higher layers”.
If you stack a bunch of simple patterns, a more complex one might seem to appear to some outside observer but that wouldn’t make it “higher” in terms of the brain itself. It would be more like an emergent property of dynamic system.
LikeLiked by 1 person
Gotta say I don’t like “information as causation”. I understand the impetus, and it seems associated with all the talk of “downward causation”, but I think you can get where you want to be without abusing the physical understanding of “causation”.
[just looked up my comment on your linked post, so won’t go over all that, except to say …] I think my preference is just to use “information” as referring to mutual information, and causation with respect to mental activities, i.e., what neural networks are doing, as (mutual) information processing.
*
LikeLiked by 1 person
Thanks. On the unitracker, I definitely don’t rule them out. Based on everything I’ve heard about them, it seems like every neuron is basically a unitracker, it’s what they do. So I’m still leery of equating them to cortical columns, although I don’t have strong feelings about it.
On mutual information, I think we’ve talked about this before, but the question for me is how mutual information becomes mutual, and what enables us to recognize it as mutual. T Recognizing the mutual information between the word “mule” and an example of any particular hybrid offspring of a horse and donkey seem to require a causal history that, at some point, overlapped.
Or maybe I should ask, can you think of examples of a shared causal history that are not, at least in some sense, mutual? Or examples of mutual information that don’t have that shared causal history? (Aside from ones that are such low probability that the universe itself ends up being that shared causality.)
LikeLike
[was trying to keep these two tracks separate, but ah well … 🙂 ]
A single neuron is a pattern recognizer, but I don’t think it would count as a sophisticated unitracker with the functionality described above (promote these, inhibit those, compete to get in way over there, become activated from way over there, etc.). [Do we know of any neurons that inhibit some neurons while exciting others, and also can be completely controlled (turned on or off) by yet other neurons besides the ones that usually activate them?]
On mutual information, the stuff you said is essentially correct. All causation, i.e. physical processes, create/adjust mutual information. And any object will have some mutual information with respect to everything in its causal history. This mutual information is an affordance. Making use of the mutual information requires coordinating the response to it. So, by evolution, the frog has coordinated the recognition of a moving spot with the tongue flicking response. That worked because under normal circumstances the moving spot has a high mutual information value relative to a fly. Of course, once the response is set, it can be fooled. So again, it’s the coordination process that determines which mutual information is being utilized.
*
LikeLike
Sorry, I’m lazy. I’ll try to remember next time you lay things out that way.
On exciting and inhibiting other neurons, yes! Although it typically requires intermediate neurons. For example, a neuron may work to excite its neighbors, but some of those neighbors may be inhibitory neurons that subsequently inhibit the its downstream neurons. So the original neuron in effect ends up exciting some and inhibiting others.
On being turned off by neurons other than the ones that usually excite it, sure. It requires a connection with that other neuron, either direct or through intermediaries. But a neuron that doesn’t usually have causal effects on a neuron can inhibit it.
On mutual information and the frog, right. Thanks. We’re on the same page, just with different language preferences.
LikeLiked by 1 person
This post is somewhat related to a discussion that I’m currently having with AJOwen’s at his site. Hopefully it will help me craft a better response. My point is that for causality to be preserved, there needs to be some sort of mechanism that brain algorithms animate to exist as consciousness. I’ll be responding to this comment: https://staggeringimplications.wordpress.com/2023/10/01/the-hard-problem-of-objectivity/#comment-655
For this post however, one thing that I would have included is a rough basic conception of how our manufactured computers and evolved brains do what they do. I’ll now provide such an account, as well as incorporate my own three step reduction of all computation. My reduction is that computers (1) accept input potential information that’s (2) algorithmically processed into new potential information that (3) goes on to inform the right sort of stuff. Here the potential information of (1) and (2) become actual information by means of the step which follows. Theoretically there will otherwise be no computation.
Computer software, or code, may roughly be though of in the form of statements like “IF (something)… THEN (do this)…. ELSE (do this)”. As I understand it such algorithms ultimately reduce back to binary code which take the form of “AND”, “OR” and “NOT” conditions.
While a constant supply direct current electricity (the voltage difference that drives a flow of charged particles) is what forces our computers to process input potential information (1) into new potential information (2) that may animate the right sort of stuff (3) for full causal computation, the power source of brain function is more complex. Enough charged particles in a neuron will cause it to fire, and neurons are electrically attached with other neurons, though heavily gated by means of synapses. So a given firing will disperse charged particles back through the system, though it’s important to note that relations exist between neurons in the form of “AND”, “OR”, and “NOT”. So for example if two neurons are firing and another has an “AND” relationship with them, then this will cause it to fire as well if it isn’t already. Thus both our computational machines and brains function as computers which cause things to happen algorithmically.
Beyond these technicalities I think it’s important to note that while our computers preform extremely fast serial operations (or process one step at a time), brains preform relatively slow though massively parallel operations. Perhaps this makes sense. We think serially and so should tend to build machines that function serially, though evolution doesn’t think and so is able to build massively parallel computational brains.
On the famous “Jennifer Aniston neuron”, obviously it didn’t evolve in someone specifically in case she were ever encountered, but rather must have algorithmically taken this role as the person progressively became familiar with her. I presume the neuron fires as images and voices get close enough to standard portrayals of Jennifer, though it’s gated such that nothing else can make it fire. Furthermore as portrayals of her image and voice change, I presume that this neuron’s firing broadens.
LikeLiked by 1 person
We’ve debated your causality and three step thing into the ground, so I’ll let AJOwens take his shot for a while.
On synapses, one thing I’d point out is that that there are both electrical and chemical synapses. Electrical ones aren’t too interesting. They just pass current from one neuron to another and seem to get used for relatively hard coded responses.
The logical processing comes from the chemical ones and the collage of neurotransmitters they use. Essentially the action potential in the source neuron gets converted into chemical messengers, which then get transmitted in vesicles. And its the receiving neuron which decides what the effects of those messengers are.
On the Jennifer Aniston neuron, definitely the relationship isn’t genetic, but learned. It actually could be possible that other things might cause it to fire, but the pattern combination with other neurons could be different and have different downstream causal effects. Neural nets are complicated beasts.
LikeLiked by 1 person
I wasn’t angling here for another discussion with you on the requirement (or not) for processed information to inform something appropriate to exist as such Mike. I was just adding some things that I consider important that you might have added to your post. In any case hopefully your post will help with my response to him.
Also I suppose that in my last comment I was erroneously implying that neurons connect electrically rather than chemically. Apparently you’re saying that’s not entirely false however. Are there indeed certain kinds of neurons that wire up electrically by means of conductive material of some sort? And for “hard coded” function? That sounds interesting to me.
LikeLiked by 1 person
Electrical synapses are basically points where the membranes of the two neurons are practically touching (although not quite). It allows electrical activity in one neuron to affect activity in the other. It’s fast and bi-directional, but not flexible, so it shows up in defensive reflex type circuits.
The wikipedia for it isn’t bad: https://en.wikipedia.org/wiki/Electrical_synapse
LikeLike
In the aforementioned discussion with Eric, I had to take issue with the parse that the system “has the ability to accurately copy data from storage into processor registers, act on them, and then copy the results back into storage”—not realizing it was your parse. It was necessary to point out that when the system “acts,” it isn’t doing anything in some mysterious space outside the register storage; it’s just shuffling bits within the storage. Consequently, when it arrives at a result, it doesn’t need to “copy the results back into storage.”
I think you know this, and in the context of this piece about the unity of storage and processing, it only reinforces the point. But having explained it to Eric, I may need your corroboration.
LikeLiked by 1 person
To this I must add that in a computer, the conceptual division between data and processing happens elsewhere in the architecture. For example, computers use dedicated areas of storage for static and dynamic memory, and dedicated lines of communication in the form of data and processing buses.
The conceptual difference is often reflected by default in programming languages, where data can be parked in variables for access by various functions. Not all languages are like this, but I’ll get into that separately, if I can wrangle my thoughts into presentable form.
LikeLiked by 1 person
On the registers, right. I didn’t intend my statements there to be a general one about all computation, just a quick description on the types of operations that typically happen in common commercial computers. (And I oversimplified, particularly in relation to modern processors which blur all kinds of boundaries with memory caches, microcode, and a lot of other stuff I’m not up to speed on. And I’m ignoring GPUs.) The whole point was to contrast the traditional serialized computing approach with how nervous systems do it.
LikeLiked by 1 person
Well, I prefer to put it the other way: causation is flow of information — or, as physicists prefer to think of it, exchange of physical information. But… same difference, I reckon.
An yes, there is, in principle, no data/code distinction. It is often forgotten that we invented the distinction in order to simplify debugging of our programs. In the very early days of computing, it was sometimes used by coding wizards to cram as much functionality as possible into the minuscule amount of storage available. I’ve probably already quoted the Saga of Mel (https://mipmip.org/tidbits/mel-saga.html), but please allow me to do so anyway. 🙂
LikeLiked by 2 people
Wow, Mel is probably the most extreme case I’ve heard of for that. I learned programming right on the transition from that kind of programming in microprocessors to more structured approaches. I remember typing in machine code programs from magazines that did all kinds of clever hacks for my Atari 400. Most of them broke when the new models came out a year or two later, but it would have cost memory and processing time to have made them more resilient, and when you only have a 1 MHz processor and few K of memory to work with, well.
LikeLike
I too remember writing raw machine code and mixing it with Algol — a bit earlier, though. Just before “mini” computers started eating into the mainframe business. Used to know ICL instruction codes by heart. Those were heady days. 🙂 And I do agree with the author of the Mel Saga that it was an experience to be treasured and as such should be missed by younger programmers.
But back to causation/information. I’ve been having hard time persuading people in our philosophical society that beyond causation as a flow of physical information, what is generally meant by causation is just causal story telling. We mostly conflate causation with causal explanation. A lot of philosophical puzzlement over causation is occasioned thereby.
There is also another cognitive tool relevant here, which should be used with care. We are so used to scientific analysis amounting to decomposition of complex systems to simpler subsystems, that we tend to forget its limitations. It crucially depends on the assumption of linearity (or an approximation thereof) of behaviours being decomposed. To what extent this assumption is justified when thinking about biological brains is still an open question. Evolution was not constrained by consideration maintainability or comprehensibility. Hence I very much doubt the strong eliminativist view of consciousness. There simply is no reason why the inner, 1st person discourse of our experience should be fully (or even mostly) mappable onto the 3rd person physical discourse of brain biochemistry (whether or not quantum spookiness is involved, a la Penrose).
LikeLike
On conflating causation with causal explanation, what would be examples of each? Just curious.
I remember your pessimism about understanding consciousness.
Definitely evolution wasn’t constrained in the way engineering typically is. But it seems like the same situation applies to biology overall, and we’re making steady progress. Which isn’t to say the ins and outs of DNA, genetics, epigenetics, the kreb cycle, proteomics, and all the rest aren’t horrendously complicated. But it’s been a long time since anyone was tempted to think it’s unsolvable in principle.
That and I think what we call 1st person experience is hopelessly clouded by the limitations of introspection. Those limitations tempt us to inflate practical difficulties into impossibilities in principle. But like biology, eventually most of us will focus on the solvable problems.
LikeLike
Ah. I was rather hoping that you would be familiar with the causality/causal explanation distinction. As all such philosophical slogans, it attempts to cover a lot of ground and volumes can be (and have been) written on the subject. Still, having incautiously opened my mouth, I ought to try my inadequate best to elucidate, though it will take me a bit of time to organise my thoughts into something sufficiently concise without (I hope) being misleading.
In the meantime, let me deal first with another, closely related, matter. Am I pessimistic about understanding consciousness? Well, yes and no. It really depends on what you mean by “understand”. Case in point: do we “understand” generative AI systems? We know how to construct them and to train them. We are fully cognisant with their low level working. We could, in principle, trace each individual decision they make to its low-level causes. In that sense we clearly do understand them. And yet their emergent capabilities took even their creators by surprise. Even in these pretty simple (compared to brains) systems, it is hard to argue that we really understand their behaviour.
This should come as no surprise. I keep citing Game of Life (and cellular automata more generally) as a paradigmatic example of weakly emergent behaviour being exhibited even by quite trivial computational systems. This would seem to do with such systems being Turing-complete — i.e. being capable of implementing Turing’s model of universal computation. Human brains clearly are capable of doing so, so why would we expect to understand their high level behaviour any better?
That is not to say that emergent behaviour is necessarily “inexplicable”. It means that in explaining it we are not going to get far by relying on reductively causative explanations. Yet their behaviour may well make sense in terms of emergent ontologies, appropriate to the level of behaviour being explained. This is where the distinction between causation and causal explanation raises its head (beautiful or ugly, according to taste :-)).
More on that later (may take me a day or two). But as to my pessimism… I have no doubt that we’ll eventually arrive at some understanding of consciousness in the weaker sense. As for any reliable mapping of it onto the underlying level of biochemistry (let alone physics) — forget it.
LikeLike
On the causality distinction, if there’s a SEP or Wikipedia article, or sub-section of an article you could link to, I’d be fine with that. Or if not, please don’t feel obligated. No worries at all. I was just curious.
On strong vs weak understanding, I guess the standard I’m hopeful for is the weaker one. I can’t even say I have an understanding in the strong sense of complex business systems I helped design and develop. So the fact that minds, either evolved or engineered, may never be fully predictable seems very probable. But to me, the strong version seems to be expecting too much from an explanation.
LikeLike
OK, so we don’t differ that much after all. But I am not talking just
about predictability. Unpredictability is cheap. 🙂 To quote a well
known philosopher 🙂 : “Prediction is very difficult, particularly
about the future.”
As highly iterative systems brains are likely to exhibit some chaotic
behaviour and since unpredictability can be an evolutionary advantage,
my bet is that evolution did not neglect this possibility. But I am
suggesting something stronger: inability to account for behaviour even
after it is observed. Just think of AlphaGo’s famous winning move, which
caused sucha stir in Go circles. Why was it chosen? Even
retrospectively,while we could, in principle, account for the how of it,
the why is a different kettle of fish altogether. And that’s a
relatively simple, very vaguely brain-like system.
As for causal explanations… I can’t think of a simple summary. SEP has
a nice piece on the war between neo-Russelians and “causal imperialists”
(https://plato.stanford.edu/entries/causation-physics/#CausExpl) but
that’s a very particular argument. So let me try to sketch the more
general picture, as I understand it. The below is based partly on a
philosophy of causation course given by Marianne Talbot in oxford some
years ago and partly on my unsystematic reading and thinking before and
after that. So just bear in mind that this is a personal account of the
matter. 🙂 And pardon any typos. I am unlikely to have the time to
proof-read properly (I am a lousy proof-reader anyway).
Firstly, our account of causal connections mostly amounts to selective
causal stories. As a trivial example, we may explain a fire by a
short-circuit in some electrical equipment without any reference to
presence of oxygen. Such stories select factors amenable to control and
wave the rest aside under “all other things being equal”, without
specifying what is meant by “all other things”. This avoids the
difficulty posed by the frame problem, but at the cost of not giving a
proper causal account.
Secondly, our explanations tend to be too generic — e.g. a car skidded
because the road was slippery, without noting that a road can be
slippery for a number of unrelated reasons and its slipperiness in any
case depends on the state of car tyres too. This precludes further
causal reduction.
Thirdly, our explanations can feature entities with no causal powers —
a car skidding because the corner was to sharp for the car’s speed. The
sharpness of the corner has an explanatory function but not a causative one.
More generally, when one considers all possible ways in which we think
and talk about A causing B, it is hard to avoid conclusion that
causation is not a unitary concept. While physical causation, in the
sense of exchange of physical information, underlies and constrains all
forms of causation (and that means setting aside purely logical
“causation”), various specialised discourses (e.g. those of special
sciences) operate on their own, domain-specific (perhaps weakly
emergent) ontologies, without reference to deeper reductive layers.
What is more, causal explanations may differ, depending on one’s stance.
Depending on your point of view, a program designed to halt with an
output 1 if a particular number is a prime and zero otherwise, halt with
a particular output either because its physical bits got flipped in a
particular way or because the number in question is or isn’t a prime.
The difficulty here is that in a modern computer, the connection between
software and its specific location and operation in hardware, is
contingent — heavily dependent on what has gone before. Hence
translating between the two stances’ causal view is only possible in
every specific case, but not generally.
Which brings us to the issue of singular causation. Causation operates
in each individual instance, but to express it we need a causal account
which relies on types ofconditions, rather than unique singular
conditions. Hence philosophical argument over singular causation, with
Humeans claiming that there can be no such thing and others pointing out
that an event may instantiate a causal law even if it is so rare that it
only happens once in the life-time of a universe. The main point,
though, is that in such case we can posit causation but ar unable to
provide a causal explanation, due to our inability to offer a general
type-based description of the event. (A different manifestation of the
frame problem, I suppose.)
Anyway, this is already too long, but it is an interesting nook of
philosophical disagreements. 🙂 Hope you’ll find it interesting too.
LikeLiked by 1 person
Thanks Mike. As soon as you said “selective causal stories” a light bulb went off and I caught the distinction you were talking about. Although I appreciate the other points. Maybe the right way to put this is that all causal explanation exist for particular purposes (engineering, legal culpability, etc) while actual causation is always broader.
LikeLike
Maybe the right way to put this is that all causal explanation exist for particular purposes
(engineering, legal culpability, etc) while actual causation is always broader.
Well, that’s a relatively uncontroversial aspect of the matter. But I feel that it glosses over what now looks to me like a more fundamental issue. Between them the issue of singular causation and the frame problem, suggest that a proper causal account (as distinct from a causal explanation) is in principle only possible if there is a way of fully characterising relevant ontology of what is being accounted for, to its reductive substrate, all the way down to exchange of physical information at the level of (currently) fundamental physics. And that’s a very tall order. Specifically, it is hard to imagine it being possible in the case of causal explanations of mental activity.
Take something as fundamental to our experience of the world as perception of colour. It used to be thought that the tri-colour theory of vision was sufficient in stepping down from personal experience to biology. But then Land blew this notion apart with his “retinex” experiments. And even his account of colour perception is now deemed too simplistic. Add to that the individual uniqueness of specifics of neural wiring and the task of reliably linking a person’s perception of a particular colour with a particulars pattern of neural activity starts looking distinctly impossible.
The realistic alternative, I think, is to fully embrace the notion of causal plurality, with each discourse domain (cognition, biology, chemistry, physics, including their distinct sub-domains) having their causal explanations based on their own (possibly emergent) ontologies. This works as long as we have good experimental reasons to believe that in specific (singular) cases we can in principle always drill all the way down to fundamental physics. (That, BTW, is how I understand Ladyman’s demand for primacy of fundamental physics.)
LikeLike
I’m more optimistic on colors. This is one of those things that I think philosophers psych themselves out on. It helps to remember that a color perception is a conclusion of our nervous system in terms of evolutionary affordances. It’s horrendously complex because it’s both innate and learned associations. I don’t want to downplay the difficulty at all, but it seems scientifically tractable.
Something like what you describe in your final paragraph is how I envision a reductive explanation working. We can, in principle, reduce the dynamics of a hurricane down to quantum physics, but I don’t know anyone who thinks we’ll ever model hurricanes using the Schrödinger or Heisenberg picture. Of course, as we discussed above, evolution is not a tidy engineer. It doesn’t stick to clean levels of abstraction. So knowing that the proteins in a synapse are molecular chemistry can be important at times.* But it doesn’t mean we have to or should try to model the entire network at the molecular level.
* I’m reminded of that time a few years ago when physicists were getting results that seemed to indicate faster than light particles. They kept checking and rechecking their data and calculations. It turned out to be a loose communication cable somewhere. Sometimes the lower abstraction layer intrudes!
LikeLike
I have no doubt there will be heuristic ways of reading off things like colour — useful, but of varying reliability. So the difference between us appears to be simply that you think it is a matter of what is possible in practice, whereas I see limitations in principle — which in practical terms amounts to the same thing. But I am fascinated by purely philosophical (epistemological) aspects of it, which appear to be implicit in any system of conceptualising (of which language is one manifestation).
LikeLiked by 1 person
A+.
Code is data, data is code. Neural weights, feedbacks, reinforcement, cognitive cohorts competing — all good stuff. All jiving well with my own interpretations.
LikeLiked by 1 person
Mostly agree with this, although I’m not sure how much “information as causation” helps with explaining. Much of what you describe fits directly into the fragmented consciousness view.
Regarding the Jennifer Anniston neuron, one of the mysteries for me for a while was the apparent sparse coding of much cognition. This nagged at me when I was promoting McFadden’s views since he is promoting the unified consciousness view which would seem to require many neurons firing for consciousness to work. With fragmented consciousness, the fact that there may be a small set of neurons corresponding to cognition is perfectly consistent.
LikeLiked by 2 people
My only point about the information-causation link in this post was that the unity removes one of the objections I’ve heard, that it’s conflating something passive with something active.
On the sparseness, yeah. It’s not surprising if we only focus on what is needed to produce behavior, including the behavior of reporting on mental states. We’re only tempted to think it must be otherwise due to introspection, essentially the story the brain tells itself about its own activity.
LikeLiked by 1 person
Could information written to a disk drive cause anything by itself?
To me, causation only happens in the context of a dynamic system that can use the information. So, information can’t be abstracted out of the system that uses it without losing its causative powers.
LikeLiked by 1 person
Information on a disk, by itself, can’t cause much. But neither can only one on input to an AND logic gate. In both cases, they need another causal factor to have an effect. In the case of the input, it needs the other one to be on. In the case of the disk info, it needs a lot of other stuff (working drive head, running program requesting data, etc) to have a meaningful effect.
Consider the counterfactuals. If the other input to the AND gate is on, but the one we considered above is off, then nothing happens (except maybe waste heat). If the info on the disk isn’t there, then whatever effect it might have had won’t happen, even with all the other factors in place.
I have considered before just considering information to be a snapshot of causal processing, to recognize most people’s intuition of it being passive. In that sense, it would be information processing which would be causation. But really it seems like the information and the processing are two inputs to a complex AND gate, and we’re back to it being causal in and of itself.
Unless I’m missing something?
LikeLiked by 2 people
It seems to me a case of both-and rather than either-or.
And that seems particularly the case in the brain with your own unified view of “data and processing are together,” which I agree with very much.
LikeLiked by 1 person
Further to what I was saying in a thread started here by Philosopher Eric, there are certain so-called “functional” programming languages that deliberately blur the distinction between data and programs. I know personally of LISP (mostly AutoLISP, which was a scripting language for AutoCAD) and XSLT. It’s been decades since I used AutoLISP, but the language consisted entirely of parenthesized lists, each containing a series of arguments, the first of which could be a function—for example, (add 2 3). The result of each list, in this case 5, was not stored, but simply returned to the processor for the next step. By nesting lists, you could build up an entire program. Since the program itself was nothing but a long list, and the data on which it operated was a list of the same form, a LISP program could operate on its own content and structure, which made it interesting to AI researchers, or so I’m told. More to the present point, it did not store data. But it was an awkward language. Because of the extensive (one could say mind-blowing) nesting, the acronym (originally for LISt Processing) earned the joke expansion Lots of Infuriating and Silly Parentheses.
I have more recent experience with XSLT. This language, used for processing XML documents, consists of nodes called “templates” that are invoked using an established order of operations to process certain nodes in an XML tree (the ones “matched” by the template). The XSLT program is itself expressed in XML, so there’s no structural difference between data and programs; and as with LISP, the result of each node is returned directly for processing. Separate data storage is not required, indeed not allowed; at times this required me to resort to recursion, calling a template from itself so that it operated on its own result, and hoping it would eventually arrive at a result that would send the processor somewhere else. Miraculously, from my perspective, this happened reliably.
Both languages assume a processor that operates sequentially in a fixed way. LISP works steadily outward from the innermost nestings; XSLT matches nodes in a nominally predictable order according to a complex set of rules, although how it actually gets through the XML tree can be a bit of a black box for the programmer. I suspect this serial simplicity at their core differentiates them from neural networks, which I understand are massively parallel. It’s not only that the “data” exist in a weighted distribution of nodes that in their interaction and distribution also amount to a “program,” but that immense complications of negative and positive feedback present, by way of a result or “return value,” a distributed output best approached using chaos theory—for chaos theory is all about feedback. There seems to be a trend afoot to analyze the distributions in terms of probability, but to me such tools seem better suited to probabilistic events, for example quantum events or random vagaries of signal propagation, than to complex feedback loops.
On this score I didn’t quite follow your mention of coalitions crossing a “phase shift boundary” to become dominant. Up to now we hadn’t considered phases in the feedback; and feedback does not necessarily involve phases. You could give me feedback on my ideas without phases being involved; a neuron could get feedback on a signal sent out, without any phases. Phases imply an oscillation, at least in the context of “phase shifts.” You mentioned “recurrent signaling loops,” but even these don’t necessarily have phases.
In the type of feedback used by Jimi Hendrix, dominant frequencies emerge through phase reinforcement, not phase shifts. Phase shifts can certainly affect the reinforcement, as when Jimi moves his guitar into another part of the standing waves that have formed in the acoustic space, causing different frequencies to excite different string segments. But when it comes to neural networks, I’m not sure what phases and phase shifts necessarily have to do with anything. But with next to no knowledge of global workspace theory, I’m probably missing something.
LikeLiked by 1 person
One thing I didn’t really get into in the post was that there are often software frameworks that mix data and execution. You listed LISP and XSLT. (I haven’t thought about LISP in a long time.) But of course there are software neural networks, which also conceptually mix data and execution.
But typically these software frameworks are executed with the traditional serial processing hardware, although these days there may be multiple cores involved, and as I noted above, GPUs. If you look at things at the assembly or machine code level, there’s still a processor retrieving things from memory (albeit through performance caches today), manipulating them, and then saving the result somewhere. When a programming language passes a value from a function, even if that value never goes into a language variable, it still typically goes onto the stack, a LIFO (last in first out) area of memory reserved for transient information. (At least that’s the traditional architecture. I’m not up on all the latest optimizations.)
The phase transition thing just refers to a point, posited by various global workspace theories, where a circuit coalition becomes so dominant, it manages to suppress all the others and become the only thing happening, or at least the only thing significant. For more info: https://en.wikipedia.org/wiki/Dehaene%E2%80%93Changeux_model#Self-organized_criticality
I’m not sure I buy the absolute phase transition, but I do think there are points where a circuit dramatically increases its reach, typically by recruiting a highly interconnected region, which might pragmatically look like a phase transition from a distance.
LikeLiked by 3 people
I realize you effectively just made the same point I did.
LikeLiked by 1 person
The question I would have is whether those languages are really doing anything different under the covers. It is still just data shuffling between registers and memory with operations performed on the data. The data/processing divide is really built into the architecture of the computer, isn’t it?
LikeLike
Functional languages attempt to simulate a model that blurs the distinction between data and programs, but of course the simulation is run on standard digital computer architecture, where the divide is built in, as you say. In the standard architecture (circa 1990, anyway), the processor has its own collection of dynamic registers, and it’s connected to a separate static data storage area by a dedicated bus for carrying data, and an address bus for specifying which data to access. There are also some specialized connections for direct signalling to peripheral devices and other components, collected into a processing bus.
The brain looks nothing like this, of course. But I imagine the so-called “neural networks” of advanced AI computing are also simulations built on standard digital computer hardware. How this is done exactly, I don’t know. Possibly each neuron is implemented in a virtual machine, which shares the standard architecture of modern computers (a certain allocation of CPUs, RAM, and so on), and the neural interactions are simulated at the edges of these VMs.. More probably GPUs are deployed to facilitate parallel processing, although I know nothing abut GPU architecture and tis is sheer speculation.
The point is that neural networks, as far as I know, are simulations of brain architecture on physical architecture that looks nothing like it. If functional languages take the same approach, it’s hard to fault them for it.
LikeLiked by 1 person
All of which is a good argument that whatever computing the brain is doing is nothing like what happens in a regular computer or maybe even Turing machines of any form.
Of course, the brain might be simulated on a regular computer, Eventually if the numbers after the decimal point are taken out far enough, the result becomes indistinguishable from the result from an actual brain
LikeLike
Your blog post delves into the intricate workings of the brain, particularly discussing the unity of data and processing within the neural network. Your perspective on how the brain functions as a system where data and processing are intertwined offers a valuable insight into the limitations of metaphors borrowed from traditional computing. The idea that neural processing doesn’t neatly segregate data from action, as is the case in standard computer systems, is an essential point. It challenges the common belief that data is moved or copied between brain regions.
One commendable aspect is your emphasis on a more complex and interconnected neural network where reactions, signaling, and the competition between circuits all contribute to the conscious experience. The suggestion that consciousness arises from dominant coalitions of circuits offers a unique perspective. It’s intriguing to think about consciousness as a dynamic interplay of circuits rather than a singular control center.
However, there are a few logical flaws in your argument. While your insights challenge the conventional descriptions of the brain’s functioning, it’s crucial to acknowledge that these insights don’t necessarily negate the value of other theories. The use of metaphors, like the global workspace, serves as a simplification to help us conceptualize complex processes. These models are useful for communication and understanding, even if they don’t perfectly mirror the brain’s intricacies. Rather than dismissing other theories, your perspective could be seen as a valuable addition that enriches our understanding of how the brain operates.
Overall, your blog post provides a thought-provoking exploration of the brain’s inner workings, offering a unique perspective on consciousness and information processing.
LikeLiked by 1 person
I’m grateful for the compliments. But I’m not really challenging the conventional understanding of how brains work. I think everything I say here matches the understanding of mainstream neuroscience, although some psychologists and philosophers may need to be reminded. And I’m not dismissing any theories based on this observation, at least not any in wide circulation in the cognitive fields.
But I do appreciate your thoughts!
LikeLike
Fascinating post on nervous system, Mike. Our body is really complex and we’re still learning a lot of things.
LikeLiked by 1 person