🧠 The Wet Logic of Being: Why Silicon Dreams Can't Wake Up

📖 Full essay: 

Evolution spent half a billion years solving the consciousness problem under ruthless metabolic constraints. Maybe we should pay attention to the solution it found rather than assuming our 80-year-old computing paradigm can simply replicate it if we add enough layers.

The question isn't whether machines can think—they already do, in ways that matter. The question is whether they can be, in the continuous, integrated, metabolically-embedded way that biological systems are.

And right now, the answer might simply be: not yet, and not like this.

On biological and artificial consciousness: A case for biological computationalism

This is Heliox: Where Evidence Meets Empathy

Independent, moderated, timely, deep, gentle, clinical, global, and community conversations about things that matter.  Breathe Easy, we go deep and lightly surface the big ideas.

Thanks for listening today!

Four recurring narratives underlie every episode: boundary dissolution, adaptive complexity, embodied knowledge, and quantum-like uncertainty. These aren’t just philosophical musings but frameworks for understanding our modern world. 

We hope you continue exploring our other podcasts, responding to the content, and checking out our related articles on the Heliox Podcast on Substack. 

Transcript

0:25

Welcome back to the deep dive where we take the densest source material and really try to transform it into the knowledge you need, complete with some surprising facts and hopefully some clear insights. And today we are definitely deep in it. We're tackling one of the biggest questions out there right now, a question that's been turbocharged by the rise of these huge language models. Will artificial systems, you know, the ones built on silicon and code, will they ever truly become conscious? It's a debate that's moved way past just philosophy now. We have sources that are digging into the physics of it, the actual architecture of computation. For centuries, we've wondered what the line is between just a clever machine and a real inner life. And now we're looking at things like energy budgets, fluid dynamics... All suggesting the answer isn't just in what the brain computes, but how it does it. Okay, so let's just jump right into that theoretical battleground. I guess the optimism that you see everywhere, especially in AI, it's built on one big idea, computational functionalism. Right. And functionalism is so popular because it's elegant. It basically says consciousness is just about the right pattern, the right algorithm. The information processing. Exactly. The information processing. It says the physical stuff running the algorithm, whether it's neurons or silicon chips or, you know, a bunch of dominoes, it just doesn't matter. So if you can get the inputs and outputs right, consciousness should just sort of... emerge that's the theory the substrate is just scaffolding but then you have the other side of the ring biological naturalism and this view says no wait a minute that conscious experience is fundamentally maybe even intrinsically linked to the messy wet biological processes of a living system. So the stuff it's made of isn't a bug, it's the main feature. It's the whole point. And this is where our sources really lay out the mission for this deep dive. We can't just wave our hands and say the brain is special. We have to define why. We need a framework for it. A biological theory of computation. Yes. Something that spells out the real differences between how a brain, which is constrained by physics and evolution, computes, versus how a digital machine, constrained by engineering principles, computes. And our mission today is to really get into that gap and to look at computation, not as some abstract algorithm, but as a process that's tied to architecture and energy. And just to give a little preview, it really boils down to two huge differences. First is something called scale inseparability. Okay. Digital systems work because you can break everything down into neat separate modular parts. The brain does the opposite. Information is processed across all these scales at once. Molecular, cellular, whole populations of neurons, and they're all linked together. You can't separate them. And that's not just some accident of evolution. No, it's a strategy. It's an evolved metabolic optimization strategy. To save energy. To save a catastrophic amount of energy The system raises computational work across all these levels. And the second big principle is hybrid computation. Right. Digital is all discrete. Zeros and ones. Exactly. Yeah. But the brain, because it's this fluid electrochemical soup, it does both. It has discrete computations, the all-or-nothing spikes... And it has continuous ones like electric fields and voltages. And that hybrid approach is, we think, fundamental. To really get why the biological way is so different, we have to start with the digital machine.

I mean, it's an engineering marvel, but it's built from the ground up on one idea: 3:48

separation. It all goes back to the von Neumann blueprint. This design has defined computing for, what, almost 80 years? Yeah. And it formalizes computation as just manipulating discrete symbols according to a set of rules. And the architecture itself, the physical layout, enforces that separation. It does. You've got those three distinct units. There's the memory, which just holds all the data and instructions passively. Okay. Then there's the arithmetic logic unit, the ALU. which is the part that actually does the work that manipulates the symbols. Yeah. And a control unit to manage the whole show. And the key separation is that the instructions and the data are physically separate from the processor. They are. They have to be shuttled back and forth across a bus to get anything done. And this leads directly to what everyone calls the von Neumann bottleneck. Yeah. Most people think of that as just a performance problem. You know, the bus is too slow. Yeah. But you're saying it's actually part of the design, a feature. In a way, yes. It's tolerated and maybe even desired because it creates this incredibly clean separation. It makes it easier to debug, to build modules, to prove a program is correct. It isolates the algorithm, the symbols, from the messy physics of the processor. And that separation isn't just in the hardware. It goes all the way up the stack. This idea of engineered separability. Exactly. I mean, think about the instruction set architecture, the ISA. It's like a contract for the low-level code. It abstracts away the physics of the circuits. So an engineer doesn't need to know if the chip is made with old transistors or new ones. The code just works. The code just works. And then compilers add another layer, translating something like Python down into that machine code. The algorithm is kept pure, totally insulated from the physical thing it runs on. And this is kind of the holy grail of that RISC design philosophy, right? Reduced instruction set computer. Keep it simple, keep it separate. The software is treated like this transcendent thing, totally divorced from the hardware that's actually running it. So here's the kicker. Modern AI, our big LLMs and neural networks, they just inherit this entire blueprint. The whole thing. They run on von Neumann hardware, even if we call them GPUs or TPUs. When you process a tensor in a deep learning model, you are constantly shuttling that data back and forth between different kinds of memory, layer by layer. It's still the bottleneck. It's still the bottleneck. And the software frameworks we use, like PyTorch, they do the same thing. They build a static computational graph that is logically separate from the data that flows through it. And the learning rule itself, backpropagation. It's pure math. It's totally abstract. It just manipulates numbers. It doesn't care if those numbers are stored on a magnetic drive or in a flash chip. And that fungibility, that's what lets you train a model across thousands of different GPUs at once. So that complete independence of the algorithm from the thing it runs on That's a direct result of this engineered separation. And that brings us right back to computational functionalism. It does. The philosophy is just the engineering principle writ large. It says if you can make a machine that functions like a brain, then the physical details are irrelevant. But this is where the biological view has a really sharp critique. Right. The critique of the limits of functionalism. because if the implementation doesn't matter, then neither does the energy cost or time. Exactly. It completely ignores the thermodynamic reality of it. Computation isn't some abstract idea. It's a physical process. It takes energy. It generates heat. The human brain runs on 20 watts and it's doing computation through these incredibly expensive concrete physical processes. So if the brain's entire architecture is shaped by the need to manage those costs, and a digital system has basically no similar constraint, then they can't be functionally equivalent. Not in a meaningful way. In the brain, the electrochemical processes are implemented directly in the substrate. The algorithm and the implementation collapse into the same thing. The physical dynamics are the software. Whereas a digital simulation, by definition, creates that separation. It does, and in doing so, it misses the energetic and temporal reality that defines the biological system. The separation gives you modularity, sure, but it creates a world where what is computed is totally insulated from how it's physically realized, and we're arguing that consciousness emerges from that inseparable how. Okay, to really drive this point home, let's look at the formal world of math and logic. because it turns out that even in pure logic, these discrete step-by-step systems run into some very deep problems. Right. We start with the absolute foundation, which is Turing and the idea of computability. A Turing machine basically defines what any mechanical discrete step process can possibly solve. And their modern digital computers are just, you know, physical versions of that universal machine. Built on binary logic gates, zeros and ones. Yes. But the problem isn't the machine. It's the kind of math it's forced to operate on. The discrete math. This is where Gödel comes in. The problem of incompleteness. Exactly. Gödel proved that any formal system that's powerful enough to do basic arithmetic, just counting numbers, you will always inevitably find statements that are true, but they cannot be proven true or false from within the rules of that system. So the system is fundamentally incomplete. It has built-in blind spots that it can't see on its own. It does. And the real aha moment here is what that implies. To prove the truth of one of these Goodall statements, let's call it statement K, you have to jump outside the system. You have to appeal to a meta system, K plus one. So you need a higher level to judge the truth of a lower level. You absolutely do. The valuation of truth demands this kind of implicit hierarchical structure. If you're stuck on just one symbolic level, you are forever incomplete. And this idea gets formalized with Tarski hierarchies. Right. Tarski showed you can't define truth for arithmetic within arithmetic itself. You're forced into this infinite ladder of levels k, k+1, k+2, each one resolving truths that the one below it couldn't touch. It reveals this necessary multi-scale structure, even in pure logic. Now, I know the physicist Roger Penrose really ran with this idea. He did. Penrose's argument is famous. He basically said, well, since we humans can see that these Godel statements are true, our minds must be doing something non-algorithmic, something beyond computation. He pointed to quantum physics, I think? Yes, but the sources we're looking at offer a pretty sharp critique of that. They say Penrose kind of missed two big things. First, he ignored the Tarski hierarchies. Maybe the brain isn't non-algorithmic. Maybe it's just implementing this scale-integrated computation biologically. It's building the ladder itself instead of meeting an external one. Exactly. And second, and this is the really critical part for us, he completely overlooked the power of continuous valued systems. Which brings us to this amazing sort of counterintuitive idea, the continuous advantage. It's a total inversion. While discrete arithmetic is incomplete, Tarski also proved that arithmetic over the real numbers, the continuous numbers that describe things like voltages and gradients and ion flows, is actually complete and decidable. Wait, say that again. That's a huge point. So the math that underpins a digital computer, the discrete steps, is formally incomplete. Yes. But the math that underpins the actual physics of the brain, the continuous flow, is formally complete. That's the striking inversion. The logic of continuous processes is, in a very deep sense, more powerful and more complete than the logic of discrete steps. Digital computers are built on the discrete model, so they have to constantly simulate continuous reality. They're always approximating. But the brain just does it. The brain implements these continuous dynamics directly in its physics. The flow of ions, the change in voltage, these are real valued processes. It doesn't need symbols or approximations. It's natively operating in a mathematical domain that is formally less constrained than the one our machines are built on. So the brain's scale and separability, this coupling between levels, might actually be the physical solution to a formal mathematical problem. That's the connection. It ties the formal limits of the functionalist view to the architectural necessity of the biological view. The substrate isn't just some random medium. It physically embodies a computationally superior mathematical framework. So we've laid out the digital world, all clean and separated. Now let's dive into the biological paradigm, which is, well, it's the exact opposite. It's messy. It's not a top-down design. Brains are what we call evolved systems, built by accretion. Meaning things are just layered on top of older things. Exactly. You're layering new functions on these ancient mechanisms, and they stay deeply dependent on them. You can't just neatly separate the new wiring from the old foundation. It's all intertwined. And that creates this dynamico-structural dependency. The structure shapes the dynamics, and the dynamics, in turn, change the structure. And the thing that drives all of this, the biggest difference from a digital system, is the overriding constraint. The metabolic cost. Right. The brain uses 20% of our energy, but it's only 2% of our body mass. It's an extreme energy crisis, 2047. Evolution has been forced to optimize for energy efficiency above almost everything else. If an LLM used 20% of a city's power grid, we'd shut it down. The brain just has to make it work. Which means optimization isn't an afterthought. It happens at the lowest possible level. All the way down. Microscale optimization. I mean, generating a single action potential. a spike, is incredibly expensive. It relies on these little ion pumps that are the biggest energy hogs in the brain. So individual neurons are tweaking their own hardware to save power. They are. They'll do things like boost their potassium conductance so they can fire faster without having to build more of the metabolically expensive sodium channels. The energy calculation is baked right into the molecular design. It's not a budget. It's a design principle. And this desperate need for efficiency leads to this core strategy you mentioned. Coarse graining. This is the brain's really elegant solution. Spiking is useful for sending clear, discrete signals, but it's expensive. So wherever possible, the brain uses continuous processes to aggregate information, making a lot of that expensive spiking redundant. And the best example of this is the non-spiking neurons. You find them all over the place in sensory systems and central processing, and they communicate using what are called graded potentials. These are analog continuous changes in voltage. And they're cheaper because they don't have to do the whole all-or-nothing spike. Far cheaper. They don't have to go through the whole costly process of reloading and firing. And here's the payoff. Okay. Studies show that these continuous signals can relay up to five times more information per second than their spiking cousins in the same network. Five times? That's huge. It's massive. And it's the perfect example of spatial coarse graining. You take a bunch of expensive, high-detail discrete inputs and you aggregate them into a more reliable, information-dense and continuous signal. So this whole architecture, built by layering and driven by energy scarcity, it rejects the clean layers of a digital system. This leads to what you call heterarchy, not hierarchy. Right. In a digital hierarchy, information flows neatly up or down a stack. In a heterarchy, which some people call rainforest realism, it rejects the idea that things operate independently on their own level. It's all interconnected. So what does that look like in the brain? It means the scale integration is bidirectional. The lower scales, like your ion channels, they generate the higher scales, like membrane potentials and electric fields. But at the exact same time, those higher level fields are reaching back down and constraining the behavior of the lower levels. They're co-determining each other in real time. Yes. And that's totally different from how a digital system might use a hierarchy. In a digital oracle, a lower level asks a higher level for an answer, and the higher level just provides it. It's a one-way street. But in the brain, the higher level isn't pre-computed. It's not. It's being actively generated by the lower levels, and then it immediately acts back on them. You know, a big brainwave emerges from thousands of neurons firing, and then that very same wave synchronizes the timing of those individual neurons. It's a feedback loop born out of metabolic necessity. So the energy crisis literally forced the system to evolve this complex, multiscale, hybrid structure. It turned a constraint into a computational superpower.

The continuous dynamics are the key to the whole thing: 16:18

the efficiency, the formal power, and the glue that holds all the discrete bits together. Okay, let's get into the weeds now. If this theory is right, we should see this hybrid of continuous and discrete processing at every single level of the brain. And we do. Starting right down at the subcellular level, you see this pattern of continuous accumulation driving discrete events. A decision isn't just to switch being flipped. Okay, give us an example, the pKa1. Right, protein kinase A. Right. It's a perfect molecular example of an evidence accumulator. It sits there inside the cell, monitoring and integrating signals over time, and its internal chemical state just builds up continuously. It's not just on or off. Not at all. And only when that continuous state hits a specific internal decision point, a threshold, does it trigger a big, sharp, discrete event, like a massive synchronized burst of calcium that changes synaptic strength across a whole area. And we see something similar with the support cells too, right? The astrocytes. Absolutely. Astrocytes use these continuous waves of calcium that spread through their branches, and they act like a system-level gate. A local calcium wave will only go system-wide if a spatial threshold is met. And there's a specific number for that, right? There is. It has to activate about 23% of the astrocytes' physical territory. If it doesn't hit that spatial mark, the global event doesn't happen. So the geometry, the physical space, is literally part of the computation. It is. The shape and spread of the continuous signal determine the discrete outcome. It's physical computation. Okay, now let's move up to the neuron itself. We have to get past that old simple model, McCulloch-Pitts neuron. Yeah, the idea that the neuron is just a simple summer, you know, it adds up inputs and fires if it hits a threshold. We now know that's way too simple. Real neurons are dominated by these huge branching dendritic trees. And those dendrites are not passive wires? Not at all. They are active, nonlinear processors, full of their own voltage-gated ion channels. They make the neuron itself a hybrid computer. And you can see this with specific molecules like the NMDA receptors. Yes, NMDA receptors are really special. They generate these local dendritic spikes. which are like little discrete events happening out in the branches embedded in the overall continuous voltage of the neuron. This allows them to perform what are called non-Markovian computations. Okay, let's break that term down. What does non-Markovian mean here? Well, a simple unit in an artificial neural network is usually Markovian. Its next state only depends on its current input. But non-Markovia means the system's behavior depends on the sequence and timing history of past inputs. So it's not just what you're hearing now, but the order you hid things in before. Exactly. The neuron is remembering the context, the chronology events. And the computational power you get from this is just enormous. This brings us to the XOR analogy. Right. The XOR problem is a classic. It's a nonlinear problem that famously requires multiple layers in a standard simple neural network to solve. But a real neuron can do it differently. Studies on human cortical neurons, specifically from layer 23, found that the complexity of a single neuron's dendritic tree is can solve XOR-like problems all by itself. So one neuron is doing the work of, what, an eight-layer deep network? It can match or even exceed the functional capacity of an eight-layer network. That is a massive gain in efficiency. It shows the biological solution isn't always to stack more simple layers, it's to build more complexity into the foundational unit itself. Now let's go up another level to whole populations of neurons, and here we find continuous communication happening even without synapses. the electric fields. Right, this is a faptic coupling. The collective activity of all these neurons generates these subtle, local electric fields. And these fields can then influence other nearby neurons, changing their excitability without any physical connection. So it's a kind of wireless communication. A very efficient, continuous, wireless communication. And these fields have that dual definition we talked about. They're a communication channel, yes, but they're also a dynamical control parameter. They're the output, but they also become a top-down constraint. Exactly. The field emerges from the activity, and then it enslaves or synchronizes the very activity that created it. It seems like these fields would be too weak to do much, but the sources say that even very weak fields matter. They absolutely do. Experiments show that the brain's own endogenous fields, which are only a few millivolts per millimeter, are strong enough to significantly change how neurons behave, to amplify oscillations, to synchronize their firing. It's a real causal influence. So in terms of information transfer, the continuous field is doing a lot of the heavy lifting. It is. Models show that these fields can guide all the noisy, high-dimensional activity of individual neurons onto more stable, global states. They act like a low-cause broadcast system, consolidating information. And these fields aren't random. They are structured as brainwaves, as oscillatory modes. We usually think of these as just a side effect, but you're saying they are the computational synthesis. They're not epiphenomenal. They are constitutive informational carriers. These waves are the brain's own coarse graining mechanism. They provide a neural syntax. They structure time, telling the system when a local spike is important and how it should be interpreted in the bigger picture. So the continuous wave is the context, and the discrete spike is the context. And you can't have one without the other. And it gets even wilder. We have to look beyond just the neurons. There are new findings showing that even the cerebrospinal fluid, the CSF, participates in this. The fluid in the brain's ventricles. Yes. That fluid has coherent dynamical modes that actually couple to the cortical networks. And specific patterns of this brain ventricle coupling can predict cognitive ability. That just makes the problem of simulating this on a dry, rigid silicon chip so much harder. It really emphasizes that brain computation is a volumetric, chemical, fluid-based process.

Which brings us to the big idea here: 22:19

consciousness as interscale integration. We know consciousness needs integration to feel unified, but why specifically across scales? Well, a lot of theories focus on integration within a scale, like binding sites and sounds together. But we're arguing that the subjective experience, its unity, its coherence over time, requires this vertical interscale coupling from the molecules to the dendrites all the way up to the global fields. And there has to be a metabolic reason for this complexity. There is. It's the metabolic payoff. By reusing computational work across scales, the brain avoids expensive redundancy. The continuous field constrains the firing of individual neurons, so you need fewer expensive spites to send a message. The flow is the hack. And finally, that sense of time, the stream of consciousness. Yeah, that subjective feeling of flow. We suspect that this phenomenology might depend on the physical and dynamical continuity of the substrate. Digital systems move in discrete steps no matter how fast. Biological systems evolve continuously in real physical time. that might be something you can't just approximate. A series of snapshots, no matter how quick, may never actually become a river. So if this biological computationalism theory is on the right track, then just making our current AI models bigger and bigger is, well, it's not going to get us to consciousness. I mean, it suggests that path is fundamentally insufficient. The critique of the current AI trajectory is that it's still completely stuck in that architecture of separability. Our systems are time decoupled. Right. The speed of the computation has nothing to do with real physical time. They're totally unconstrained by energy and they lack that dynamico-structural coupling. The hardware is just a passive interchangeable part. So even when people try to model theories like the global neuronal workspace... They're capturing the functional flowchart, but they're running it on the wrong kind of machine. Exactly. The model might do the right thing. But if consciousness depends on the underlying physics of these hybrid, scale-integrated dynamics, then a functional simulation just isn't enough. The separation between the code and the silicon is the point of failure. So what would a consciousness supporting system actually need? You've boiled it down to these tripartite criteria. Right. First, it has to use hybrid computation. It needs to natively support continuous dynamics like fields or chemical gradients that interact with discrete events. like spikes in real physical time. Second, it needs scale inseparability and metabolic embedding. The processes can't be neatly separated into modules. They have to be constrained by a serious energy budget, which forces the evolution of these coupled heterarchical scales. And third, it needs dynamic co-structural co-determination. The physical substrate itself has to be adaptive. It has to be able to modify itself just like our neural circuits do. The hardware and the software have to be the same thing. And current digital computers fail on all three counts. But are there any promising signs, any edge cases that suggest we might be moving in the right direction? There are few. I think the most fascinating biological hint comes from these neural cultures or dish brains. These are literal brain cells in a dish. Lab-grown biological neurons. And when you put them in a closed-loop environment, they show incredible sample efficiency. They learn tasks way faster than our best deep reinforcement learning models. Which suggests the biological matter itself has some kind of built-in computational advantage. It has what we call inductive biases. The continuous substrate, the plasticity. It gives these living systems a leg up that we just haven't figured out how to engineer yet. And what about on the pure software side? We're seeing some inspiration from biology. There are these new multi-scale architectures like the Hierarchical Reasoning Model or HRM. It basically uses two coupled networks running at different timescales, a slow planner and a fast worker. So it's importing that biological idea of different interacting timescales. And it gets really strong results on tasks that need planning, like Sudoku. It shows that even if you're running on conventional hardware, just adopting the organizational principles from biology gives you new functional capacity. Okay, let's talk about the hardware. The term neuromorphic gets thrown around a lot. We need to be specific. We do. And we should be very skeptical of what gets called neuromorphic. There are really three types. What's type 1, the original vision? Type 1 is the silicon neurons, Carver-Meade's original idea. These were analog chips that used the actual physics of the transistor to mimic neural processes. The continuous physical operation of the chip was the algorithm. And then there are Memristors. Type 2, yeah. Yeah. They focus on implementing plasticity, on having a memory of their history. And then there's Type 3, which is what we see a lot of now, simulated spiking neural networks like the Spinnaker machine. Which is often hailed as the future, but you're saying it doesn't meet the criteria. It doesn't. Spinnaker is an incredible engineering feat, but it's a massive parallel digital simulation that, It uses standard processors running simplified neuron models in discrete time steps. It completely decouples computational time from physical time. So scaling it up hasn't really yielded any new computational paradigm. No, because it fails to implement the core biological principles. It's still trapped in the digital separable world. So the real path forward has to be with new physical media. where the substrate's physics is the computation. We need to embrace fluids and chemistry. We have to look at fluidic neuromorphic devices. Right. These are systems that use the actual movement of ions, diffusion, things operating in real physical time. Like that example from Xiang and colleagues. The fluidic memister. The computation isn't a switch in a solid state material. It's the history-dependent reorganization of ions in a fluid. And this is so important because it brings chemistry and structured randomness into the device, things that are essential for biology but that solid state devices are designed to eliminate. It's the ultimate embodiment of that idea that you can't separate the computation from the medium. And this brings us to the last point. If we change our idea of computation, we have to change how we measure it. Our current biomarkers for consciousness, things like PCI or integrated information, they usually look at just one level of description. They often take this rich biological signal and have to simplify it way down to analyze it. And in doing so, you lose all the critical inner scale dynamics. So we need new multi-scale metrics. We need a way to model systems where discrete events are being generated and shaped by continuous dynamics. And the mathematical framework for that is point process theory. Right. It's a type of math that's designed to describe systems where events, like spikes, are embedded within continuous underlying processes like electric fields. It's the natural language for these hybrid systems. So instead of just measuring information transfer on one level. The goal is to track how information is transformed, compressed, or reused across scales from a continuous voltage change to a discrete firing pattern and back again. It would let us actually measure something like scale inseparability. So we'd be looking for the informational signature of a biological system, not just its functional output. We'd be looking for the emergent properties of an architecture where computation, energy, scale, and time are all completely inseparable. So to wrap this all up, this deep dive has, I think, fundamentally reframed the search for artificial consciousness. The core thesis here is really that the differences between the digital machine and the biological brain, they're not just details. They're foundational ontological divides. Digital systems get their power from engineered separation and abstraction. Biological systems get their power from evolved integration, hybrid dynamics, and the constant pressure of metabolic embedding. So if we're serious about this, the path has to pivot away from abstract algorithms and toward physically grounded computation. We have to embrace true scale and separability. It's a call to move toward a genuine biological computationalism, one that demands our synthetic systems start to mirror the actual physics and chemistry of life. And that leaves us with this final thought, something for you to mull on. If consciousness really does depend on the continuous unbroken flow of physical processes unfolding in real fluid time. Can a machine that is defined by discrete time steps and mathematical approximations ever truly capture the subjective flow of conscious experience? Or will it forever be limited to just being the most brilliant, fastest sequence of stills?