Consciousness: The last 50 years (and the next)
journals.sagepub.com
journals.sagepub.com
Overall, I'd say we are no closer to understanding consciousness than we were 30 years ago. As a science, we are still poking at the brain to gather data and proposing hand-wavy models of low level behavior. As philosophy, we still don't agree on what the phenomenon we need to study is.
My gut feeling is that consciousness/intelligence is an emergent phenomenon of the physical brain (a very high-dimensional non-linear system not amenable to reductionism) and that the simplest model of the brain will be too complex for our brains to understand.
Notably, this does not stop us from making new cars.
There would need to be some substantial basis for a claim that a human brain will not soon be a similar deal: no one will undertand the whole system completely, but we will have experts on different areas and aspects of consciousness which will be more than enough for progress and specifically, strong AI.
Surely, it is likely evolved and not directly designed, but if you look at it more closely - it is modular in almost every important area! Organisms oragnize into distinct non-mixable speices/genotypes, we have organs in our body that have specific functions, we have sub-organs, we have specialized distinct types of cells, distinct sub-cellular structures inside the cell, distinct proteins with different functionalities (with some groupings of course), even higher level functionality is modular, like for example sleeping: all processes that need to be done offline are refactored away into a "sleep" phase which is enforced in many ways, even though this enforcement turned out to have so many parts and being so complicated that newborns need to learn to to start sleeping. The sleep itself it split into distinct phases, different processes happening in the them; humans auto-organize their societies into specific roles, etc.
And, my personal favorite - the bulk of the information about a human is in a genome: a specified code utilizing 3-letter, 4-character words and sentences to describe almost everything needed to make a human. How did it evolve into a "code", a language??? Modern designed x86 architecture is more complex and has more ugly legacy crap than the evolved genetic code which has the perfect number of letters, perfect control structures, perfect number of aminoacids that is enough to build everything! (Ok, yes I know it is not "perfect" but it is very very sophisticated and, apparently, well-thought-through.)
So I think this modularity, which exists both in evolved and designed systems tells about the nature of the universe, and/or complexity, information and knowledge: the information and its physical representation conspires to become modular and produce more complex designs and entities that way, through abstraction, through logical levels and modular architecture. Not only through sheer number of interconnections (surely, there is enough of that too, but is it really that much different to a number of transistors in a recent x86 chip?), but through modularity.
It is easy to miss this because nature is arguably "less" modular than a man-made car. But when looking at some design choices under the hood or talking to mechanics, or looking at leaked software code for big projects - there is just enough spaghetti code there to prove that such designed entities have a lot of ugly poorly understood interconnectivity that is just made to "work good enough". I guess it is when this horizontal complexity becomes unsustainable - the designer, be it a human or a natural process of evolution - has to come up a new abstraction level, an organ, a genetic language - to allow for further progress.
We don't understand brains either but it doesn't stop us from making new brains.
The best way to make a brain is still the old way.
The fundamental insight it makes is that there is a distinction between attention and awareness: that attention corresponds to neural firing patterns, and that awareness is a schema (model) of that attention. Awareness then feeds back into attention — moving it or sharpening it.
Think of it as a high-level programming SDK for the brain.
AST also helps explain why awareness evolved: it is a prediction engine for brains, both our own and others. And once competition began, an arms race of prediction begins that ultimately ends with a non-determinative ‘free-will’ that defies prediction.
If AST is correct, then consciousness is a simulation. It also means we have a roadmap to machine consciousness. Graziano has written articles that ask for engineers to try this model, but from what I can tell, no one has really tried (by intention). However, model-based reinforcement learning is pretty close. Especially the ones that are beginning to use action-prediction.
It won’t be long before we have paranoid androids.
But pure awareness is entirely different. You can't get around qualia. Because in essence, qualia is pointing to the idea that "you are." You have an experience of being. Saying here is an explanation of how the simulation arises just pushes the question back. Now I'm wondering, "to whom is the simulation presented?"
You can say, "to no one—it's all a trick." But who is being tricked. Ultimately I am. The idea that my perceptions arise mechanistically presents no problem to me. The idea that the original background intelligence itself is a simulation seems obviously false. So obvious it precedes a logical argument because it is the intelligence from which logics derive their authority. Crudely speaking, it makes 1+1=2 true.
I think a lot of really smart people look right past the most basic reality that they must exist prior to their perceptions for perceptions, logic, or a simulation to exist at all. If you can say that awareness is secondary and arises from these mechanistic explanations, who is left to evaluate the validity of this proposition?
It's way simpler than we're all making it. People just have to sit alone with the knowledge that they are.
A "clever" turn of phrase, nothing more. Illusions don't require a subject. An illusion is a perception that entails a false conclusion if taken at face value, the false conclusion in this case being that subjective experience exists.
> If you can say that awareness is secondary and arises from these mechanistic explanations, who is left to evaluate the validity of this proposition?
Don't you see how this question assumes the conclusion? A mechanistic process can validate propositions, obviously.
The poorly defined word of consciousness may mean we’re asking the wrong question, but I find it pretty unlikely it’s too complex to understand.
https://www.lesswrong.com/posts/8QzZKw9WHRxjR4948/the-futili...
Why? We already have models of the brain and cognition, and they're improving over time. Just because we can't gain a sufficient understanding within a time scale suitable for you doesn't allow you to call it "too complex for our brains".
We'd have a much better understanding if we conducted live vivisection of prisoners, but we're not cruel. We have to glean what we can from shadows of knowledge. One day maybe our children will see the shape of reality. Even the things we think we understand now will be greatly expounded upon.
I'd be very happy to listen to arguments against that pessimistic viewpoint because then the hundreds of hours I've spent reading and thinking about the topic won't have been wasted.
The last book I read on the subject was Hofstadter's "I am a strange loop". I found myself agreeing with almost everything he wrote, so the book wasn't very exciting for me, though I should look up some of the bits of music he mentioned.
The author (Peter Hankins) does an excellent job of summarizing and critiquing current research papers in philosophy and neuroscience, and the comments on the articles are often very informative.
Can we the mystery of human consciousness in the 21st century? Ok, ok, maybe we should start with a definitions...
Or, perhaps, the implications of that.
I think one of the underlying problems is that people want a cold, objective quantification of a word that is inherently emotional and subjective. So when the objective definition does not align with the emotional experience, they dip into mysticism or philosophy, neither of which is testable science.
Really, the word "consciousness" is not a scientific word, it's kind of like trying to quantify God.
Can we test a being's self-awareness, decision-making ability, etc.? Yes. Is there something "consciousness" that is not testable? Then that part is not science. We can really really want it to be science, and keep looking for ways to test it, but until we do, it's just conjecture. (Personally, I think that once we're able to model brains that we agree "are conscious", all of this feeling of mystery will fall away.)
We don't push back those models because they are "uncomfortable", make us feel bad, or inferior, or anything like that - but because they are simply non-functional! There are many aspects and phenomena in the world, in human interactions, in my own consciousness - which are simply not explained by the reductionist view! "Based on some complicated network of neurons" is not good enough! I clearly have these phenomena and experience, and this "complicated network" explanation just gives no explanation at all really! While other less scientific models do.
I'm like "sure, I can accept whatever science you can give me and become a fully logical being, but don't ask me to ignore obvious phenomena that I know I have and experience, that is just stupid!".
Sometimes they mean that the phenomenon under consideration is always reducible to a common set of basic physical underpinnings, which is something for which we have never conclusively observed a contradiction.
Other times, they mean to say that the phenomenon in question can be well understood using a lower level set of abstractions than we initially considered and that there is no need for further abstraction. This is observably not true, i.e. the workings of an internal combustion engine are not practically modelled by combining a model for each of its individual atoms.
I think consciousness might eventually be reducible in the first, but not in the second sense. Of course, crude initial models don't explain a lot, but I don't think that has much bearing on a thing being reducible.
Less scientific models can be simpler and functional, but that doesn't mean that they’re more correct.
(This would work for a purely mechanical computer made out of rocks, too.)
Personally, with enough knowledge about how modern AI works, and a reasonable understanding of the neuroscience literature, I feel there is no fundamental mystery to consciousness, only awe at the level of complexity harnessed.
In the same sense one could say that people on your side of the argument are simply "uncomfortable" with the notion that not everything can be understood logically and predicted, including big important things in their own life, but also the larger truths of the universe. And so instead of looking at the other experiences and data, which don't fit into an understandable paradigm - they tend to ignore or deny them.
I agree that many people tend to “over-science” and ignore inconvenient data, but this is not a core component of the basic scientific method. In fact, many of the most useful predictions are probabilistic in nature, and so you would expect the data to be fuzzy and the models to be imprecise in the early stages of formation.
Scientific method is a great tool - for the tasks that it can handle. I can argue that there are many tasks which it cannot handle. Areas of knowledge which cannot be studied with scientifid method exactly.
Another limitation of it is obviously that it needs a hypothesis - which means it requires a preconceived notion of the world, before any testing can actually take place. Which means that the tester is responsible for the basic paradigm and worldview, from which they ultimately produce the hypothesis to test, to begin with. Obviously, the scientific method is less proficient in handling these preconceived prejudices and worldviews, because it comes in after they have been already applied. (It can help indirectly by verifying the hypothesis and perhaps sort out some worldviews, but it can naturally not create a new worldview by itself.)
So what happens when there are multiple hypothesis from completely different paradigms that show some evidence? Or what happens when there is a hypothesis which you just cannot disprove directly? This is where other paradigms ultimately come in.
I suspect that the question of consciousness is ultimately one of these questions, where we will need help from many other areas, before the scientific method can ever be applied here. Before a testable model can even be conceived.
If you're talking about the neural correlates of consciousness, you're right that we're removing a lot of the mystery. But if you're talking about the mystery that is your own subjective being, then we may never solve that problem. As Susan Greenfield states, "we can't turn the water into wine"
Sam Harris does a great 5m summary on the hard problem of consciousness that's worth a watch:
There should be nothing mysterious in 2018 about a biological being that has an internal self-model, memory, and processes sensory stimuli and internal neural firings over time, both of which have a direct relation to the physical basis of that self-model. That is what self-awareness is.
"Experience can't be explained by science because science only arises in experience" sounds profound but contributes nothing -- if you invalidate the content of your experience, then you no longer have any reliable axioms on which to build any argument whatsoever, and we might as well all go home.
The Sam Harris video you linked makes a good point -- that we want an explanation of consciousness that intuitively makes sense. I'm saying that if you study enough AI and neuroscience, it does intuitively make sense.
His counterexample of explaining the "brittleness of glass" is intuitively OK with us even if we don't actually personally understand all of the details of the molecular and atomic interactions, because it does not fundamentally challenge our beliefs of who we are. If there's some explanation of consciousness that does challenge our beliefs of who we are, you better believe you're going to need a more comprehensive and complex explanation before you're going to intuitively believe it.
The OP article makes the point that we haven't got near to cracking the problem.
Personally, I think we have a category issue. We simply don't know what category to put consciousness in, and all attempts to solve it thus far have been category mistakes.
For your consideration:
https://www.frontiersin.org/articles/10.3389/fpsyg.2015.0050...
> Personally, I think we have a category issue. We simply don't know what category to put consciousness in, and all attempts to solve it thus far have been category mistakes.
Consider the possibility that the category you've implicitly created for consciousness doesn't actually exist.
It'd be nice to think that consciousness wasn't real, because then science could explain mostly everything. Very comforting. But it can't.
And consciousness does not have the long history you describe.
And "unable to model" is a huge stretch. You can't measure something that has no definition or understood boundaries. When the rest if our perceptual and cognitive faculties are explained, you'll find there's nothing left for consciousness. It's the new god of the gaps argument.
“Researchers have learned a great deal about the neural mechanisms underlying global states of consciousness, distinctions between conscious and unconscious perception, and self-consciousness. Further progress will depend on specifying closer explanatory mappings between (first-person subjective) phenomenological descriptions and (third-person objective) descriptions of (embodied and embedded) neuronal mechanisms.”
I wouldn’t call that “haven’t got near to cracking”, especially when the technologies needed to specify those closer mappings are clearly defined and already under development.
"We cannot yet know whether today’s students will find the solution to the ‘problem of consciousness’ or whether the problem as currently set forth is simply misconceived."
And all the while we're making that same category error: Mapping subjective experience to neural correlates. This is like someone claiming that they know how internal combustion works because when they push the pedal the car goes faster.
This is a case where more science could probably help philosophy.
I'd take that bet. Physics already tells us that humans are merely finite state automatons (due to the Bekenstein bound), so the logic needed to describe us needn't even be that powerful. And given our behaviours were largely shaped to be fairly energy efficient due to evolutionary pressures, consciousness will turn out to be one or more cognitive tricks that convey some adaptive advantage.
The hurdle people typically have to accepting this inevitability is because of the internal perception of cognitive continuity, but this property is clearly false. There's are many obvious cases where continuity fails, like sleep, anaesthetic, etc.
You're simply assuming there must be a separate category either because you want one to exist, or because you've been fooled into thinking consciousness has some properties that you consider irreducible to more basic components.
There are many intuition pumps discussing surgical or other interventions that tweak qualia in various seemingly reasonable ways, to the extent that the concept of qualia itself becomes increasingly unreasonable. Dennett has covered a lot of this ground.
You say conscious beings are obviously just finite state automata. That’s curious to me because it seems strange that I would essentially be a certain mathematical structure. I agree that my structure can in principle be explained as an automaton (except that my lack of a boundary means that I have to be described with a universe-sized map that can’t fit in the universe) but why is it like something to be an automaton?
Maybe you’re saying something like this. I have an illusion that there is some subjective realm of experience, exemplified by the visual field, a kind of mirror universe that is mysteriously connected to the “outer” world, and this illusion is what leads to mistakes like body-mind dualism. In fact there is no such subjective realm, and my perception is not different in kind from any other causal system’s different reactions, it’s just so complex and finely tuned that this “illusion” appears especially because of the illusion of continuity.
I can’t steelman the argument properly because I really don’t understand, and unfortunately I haven’t read Dennett.
It seems to me that why there is perception is as mysterious as why there is energy and these do seem like separate questions so far. Some kind of monism seems reasonable but I don’t see how experience can be assumed to be downstream from the mathematical laws of fundamental physics—that seems like an unproven assumption that’s philosophically tricky and extremely contentious.
Happy to hear any thoughts or pointers.
That's a complex topic, touching on metaphysics, epistemology and the philosophy of mathematics. To summarize, explaining the unreasonable effectiveness of mathematics without simply adopting a Platonist-type ontology requires making more assumptions than simply accepting the existence of mathematical objects. At that point, it's also more parsimonious to then conclude that our world is simply one such object, rather than positing that both mathematical objects and matter exist.
> but why is it like something to be an automaton?
I think your steelman is pretty close, and you can see exactly this approach in mechanistic theories of subjective experience:
https://www.frontiersin.org/articles/10.3389/fpsyg.2015.0050...
Since this is Hacker News, an apt analogy might be how single CPU computers simulate concurrency via task switching. Our brains undergo similar signal switching between perceptual signals coming from our internal model of the world, and our direct perceptions of the external world, and this creates the illusion.
> Some kind of monism seems reasonable but I don’t see how experience can be assumed to be downstream from the mathematical laws of fundamental physics—that seems like an unproven assumption that’s philosophically tricky and extremely contentious
Absolutely. That's why Dennett has spent 50 years researching and writing on the topic. It's a tangled web that requires what he calls perceptual inversions to make it intuitive. But of all philosophers, he's probably spent the most time with neuroscientists, physicists and other philosophers. A lot of people don't like what he says, but he's definitely worth reading if this a question that interests you.
- Why do we have subjective experience?
vs
- What causes us to have/creates subjective experience?
- For what reason/purpose do we have subjective experience?
- How is subjective experience implemented?
The last three seem to be reasonably well fleshed out in the literature. Do you interpret “why” to have an additional meaning?
Tractatus 5.6.3 presents the metaphysical problem quite clearly. The perceiving subject is a limit of science, kind of like the big bang which is also fundamentally and inescapably mysterious. You cannot prove or disprove the simulation hypothesis. Of course we exist within a deep and unfathomable mystery, no matter how accurately we can make predictions with the differential equations of physical science. It’s all fundamentally mysterious and not admitting that is ridiculous.
Math can prove things unprovable from a specific set of axioms, but in the physical sciences it's a matter of imagination. The question of what stars were made of was once unknowable; now it is a robust field of study. What happened before the Big Bang is unknowable... by current scientific understanding and techniques. Maybe once we unify physics or build colliders the size of solar systems it will become understandable; maybe not.
Humans have an instinct to make things mysterious and more meaningful than they are because it drives their curiosity and need to explain, which drives their ability to predict, which fulfills the mandate of life to win competition for resources and grow.
Introspecting and understanding the perceiver thus answers the question easily. Wittgenstein did not have fMRI machines.
For your consideration:
https://www.frontiersin.org/articles/10.3389/fpsyg.2015.0050...
From the wikipedia article (https://en.m.wikipedia.org/wiki/Attention_schema_theory): “AST is a theory of how a machine insists it is more than a machine, even though it is not.”
That would reflect a lot of the voices here! ;-)
The role of consciousness is to keep the body alive: fend dangers, find food and reproduce. It does that by taking into account the state of the environment and the internal state of the body and acting in such a way as to maximise its own rewards, learning to avoid bad situations along the way. Evolution has shaped the body and rewards in order to maximise survivability. It's interesting to see how surviving bootstraps meaning and consciousness out of itself (and its environment).
How are laptops not reproducing? If you looked at the population of laptops in the world when they were first introduced - and then tracked them through the years, they have clearly started reproducing. They even evolved! The basic shape stays the same, but they are clear evolutionary trends, like the taxonomy, outwinning the poor designs, branches of evolutionary traits etc!
Of course you can choose to look at the people that are building them, but how do you actually know that laptops are not "parasites" for example? Or a symbiotic life form like dogs? Yes, individual humans might take specific actions for building specific laptops, but so what, we take actions to raise and walk dogs also. Do we as humans really have a choice of stopping producing laptops? If not, couldn't it be just as easily seen that laptops actually use us to reproduce? Like flowers use bees to reproduce? You don't see flowers as being less alive because bees are ultimately required for them to reproduce?
Sure, one might need to talk more about the design of a laptop, the basic idea of a computer in a portable interface with certain features - not any individual physical laptop as being alive. But why not? People are surely not only alive in a physical sense. Many of the physical differences and transformations actually don't make much difference. It's the inherent internal something (entity? information? idea? soul?) that is more alive than the physical body itself.
The point is, "property of self replicators in their competition for life and reproduction" - is not a very good metric for consciousness. Many people don't see it as coming from the data itself, and more being a predetermined paradigm or outlook on the world, which someone already comes in with, when they look at the data. When someone with another outlook (for example that consciousness actually uses the physical world, including the reproduction cycle) to its own goals, not the other way around - then the data starts supportnig that too. It's all about how to look at it.
The role of consciousness is debatable. People who died by suicide have some significance against the theory of consciousness keeping the body alive. My observation of consciousness makes me consider it an abstract layer of observation & emotions and nothing more. I don't even think emotions have to be there. It's hard to describe and I understand why some people don't even think consciousness is a real thing.
On the other hand, what are emotions? They are predictions of future cumulative rewards/bad situations. The role of emotion is to guide actions, they are not just an 'abstract layer'. And rewards are signals based on circuits designed by evolution, to maximise life.
Chaos theory is a very challenging and interesting concept which you should check out, but one of the most mundane outcomes of it is that there is no way to predict the physical world. No way to see what is going to happen. Now, obviously one could predict a simple neutonian mechanical system for a 10 second-period, or predict enough of electrical impulses in a piece of designed silicon to make a functional processor - but in general, in the widest sense, there is no way to predict the universe. Even with the assumption that there is some completely determined process at the microscopic core of every interaction - those processes, when combined into big systems, wired up with feedback loops and put in a system so tiny that one has to had a machine 2 times bigger than the original system in the first place: and we arrive at total unpredictability of the universe.
Note, this is not just practical unpredictability: the question is not about us not having enough technical prowess to predict the universe. The issue is that it is physically impossible to build any real prediction machine that would predic the whole thing: you would need a bigger universe than the universe it is trying to predict. It is even a logical contradiction!
Thus determinism becomes more of a philosophical outlook than an actual physical property that could be calculated, predicted etc.
Even the word itself - "determinism". Assumnig that the fate, the outcome of something is determined. In what way can it be determined if we know that it is physically impossible to determine it? Not because we lack machines or havent built a strong enough microscope. Our very own scientific paradigm, the scientific method, the way of thinking about the universe - the thing that lies at the core of our worldview - says that it is impossible to predict the whole system. Absolutely impossible. So it is unpredictable - ergo undetermined.
The fallacy is thinking that we must predict an exact state vs. a probabilistic prediction of an information-theoretically meaningful set of states. Reality is metastable, so there are certain attractor states, even if randomness is injected throughout.
I have no idea the exact second my plane is going to land tomorrow, but I’m reasonably confident it will land, and I’m very confident that the United States will still exist tomorrow.
I have no idea if the United States will exist in 10,000 years, but I’m pretty sure Earth will still be a rock orbiting the Sun at that time.
And so on.
Using your analogy, whether United States will exist in 10000 years might actually be undetermined. Not just unknown, unpredicted, but undetermined in the chaos theory- sense. Not only you don't know it - the universe itself does not know it. And the universe is built in such a way that it is impossible to know. Not just "very hard and requires a lot of computation". Impossible, because we can show that the computations you would need would require a computer bigger than the universe itself.
Then there is of course an extra note required here is that there is no way for me as a human to know if the question about United States actually is universally undetermined - or just unpredicted; if there is some entity that can predict that. But the chaos theory says that many aspects of the system, and the system as a whole is undetermined. (While having exceptions where simple things like your plane landing can be predicted with some accuracy, yes.)
Also please note, and I repeat myself that I argue there is a great difference between "really hard to predict" and "physically impossible to predict for any being, based on our own notion of the universe".
If it "means nothing in practice", and can be true/not true regardless of whether it is possible (even in theory) to test it - then I assume that it is unrelated to science and you are using it in a philosophical sense?
Turing machines are deterministic. Enumerating all Turing machines is deterministic. Whether any given Turing machine will terminate is unpredictable (the Halting problem).
Unpredictability does not entail nondeterminism, although distinguishing the two is not necessarily always possible.
Paraphrasing, the theoretical image of an atom, as well as the set of atoms and other particles - is perfectly deterministic. But the chaos theory talks about the real world, not the theoretical framework.