Latest / The Last Observer: Science and Beyond / Why Your Eyes Never Show You Reality

Transcript

Why Your Eyes Never Show You Reality · The Last Observer: Science and Beyond

• 0:00
  Speaker 1: You know, for our entire lives, we just sort of
• 0:02
  Speaker 1: walk around with this incredibly fundamental assumption about how we
• 0:06
  Speaker 1: interact with the physical universe.
• 0:07
  Speaker 2: Yeah, we totally do, right.
• 0:09
  Speaker 1: We believe, like almost entirely without questioning it, that our
• 0:12
  Speaker 1: visual system operates basically like a high definition video.
• 0:15
  Speaker 2: Camera, exactly like a live feed.
• 0:17
  Speaker 1: Yeah. The underlying assumption is that we are just passively
• 0:20
  Speaker 1: capturing reality as it exists, like we're streaming that raw,
• 0:24
  Speaker 1: unfiltered objective feed directly into our visual cortex and then reacting.
• 0:29
  Speaker 2: To it, which feels completely logical.
• 0:32
  Speaker 1: I mean, we assume a tree is simply green, the
• 0:34
  Speaker 1: sky is just blue, and you know, the precise three
• 0:38
  Speaker 1: dimensional shape of the coffee mug on the table is
• 0:41
  Speaker 1: an objective physical fact that our eyes are merely transmitting.
• 0:44
  Speaker 2: It is arguably the most persistent and maybe the most
• 0:47
  Speaker 2: intuitive illusion in the entire human experience.
• 0:51
  Speaker 1: An illusion.
• 0:51
  Speaker 2: Oh absolutely, it feels undeniably true because the phenomenology, the
• 0:55
  Speaker 2: actual subjective experience of seeing is just so incredibly amless.
• 1:00
  Speaker 2: There's no lag exactly, You open your eyes and the
• 1:03
  Speaker 2: world is just there. It's instantly rendered in perfect detail.
• 1:07
  Speaker 2: There are no buffering screens, no loading bars.
• 1:09
  Speaker 1: Thank goodness for that, right, But.
• 1:12
  Speaker 2: The moment we start looking at the actual computational neurobiology
• 1:17
  Speaker 2: of vision, that entire passive camera metaphor, it just completely.
• 1:21
  Speaker 1: Falls apart, completely falls apart. And that is exactly what
• 1:24
  Speaker 1: we are dismantling today.
• 1:25
  Speaker 2: It's a massive paradigm shift.
• 1:27
  Speaker 1: It really is modern neuroscience. Specifically, this whole paradigm of
• 1:32
  Speaker 1: predictive processing has revealed that your brain is absolutely not
• 1:36
  Speaker 1: a passive recording device, not even close. It's not sitting
• 1:39
  Speaker 1: in the dark theater of your skull just watching a
• 1:41
  Speaker 1: live feet of the universe. It is actively violently almost
• 1:46
  Speaker 1: constructing the reality you are experiencing right.
• 1:48
  Speaker 2: Now, generating it from scratch.
• 1:50
  Speaker 1: Yes, everything you see, everything you feel, your entire conscious
• 1:54
  Speaker 1: visual experience at this very second is actually an internal
• 1:57
  Speaker 1: simulation being generated by billions of neurons.
• 2:00
  Speaker 2: And you know, the initial cognitive dissonance you feel when
• 2:02
  Speaker 2: you first really internalize this concept is completely valid. It's terrifying.
• 2:06
  Speaker 1: Honestly, it is deeply unsettling to realize that your eyes
• 2:10
  Speaker 1: are not giving you a complete, high fidelity picture of
• 2:12
  Speaker 1: the world. Yeah, in reality, the sensory data reaching your
• 2:16
  Speaker 1: brain is incredibly noisy, it's highly compressed, highly fragmented, and
• 2:21
  Speaker 1: this is significantly delayed, which.
• 2:24
  Speaker 2: We'll get into because that delay is wild.
• 2:27
  Speaker 1: Right, your brain is performing an astonishing amount of computational
• 2:30
  Speaker 1: heavy lifting behind the scenes. It's constantly filling in massive
• 2:34
  Speaker 1: blind spots, smoothing over erratic eye movements, and building a
• 2:38
  Speaker 1: cohesive three dimensional model.
• 2:40
  Speaker 2: So it's basically doing post production in real time.
• 2:43
  Speaker 1: Exactly what you are perceiving right now isn't reality itself.
• 2:47
  Speaker 1: It is simply your brain's most statistically probable guess, it's
• 2:50
  Speaker 1: best prediction of what is out there.
• 2:52
  Speaker 2: Okay, let's unpack this, because if my brain is just
• 2:55
  Speaker 2: simulating reality, you know, if it is essentially hallucinating a
• 2:59
  Speaker 2: model of the world and guessing what is out there,
• 3:02
  Speaker 2: how is the biological hardware doing that so perfectly that
• 3:05
  Speaker 2: I never ever noticed the seams.
• 3:07
  Speaker 1: It's the ultimate magic trick, it really is. I want you,
• 3:11
  Speaker 1: the listener, to just take a second and look around
• 3:14
  Speaker 1: the room you are in right now. Notice the walls,
• 3:17
  Speaker 1: the specific texture of the furniture, the way the light
• 3:20
  Speaker 1: is coming through the window and reflecting off the floor.
• 3:22
  Speaker 2: It feels completely effortless.
• 3:24
  Speaker 1: Right, It feels entirely effortless. It feels like reality is
• 3:27
  Speaker 1: just presenting itself to you for free. But what we're
• 3:30
  Speaker 1: going to explore is how that room you are looking
• 3:33
  Speaker 1: at is actually a highly edited, delayed, and internally constructed projection.
• 3:39
  Speaker 2: It is a projection, yeah, and it's one born out
• 3:41
  Speaker 2: of absolute biological necessity.
• 3:44
  Speaker 1: Because we wouldn't survive otherwise.
• 3:45
  Speaker 2: Precisely, to understand why the brain operates this way, we
• 3:50
  Speaker 2: have to look at the immense computational bottleneck of the
• 3:53
  Speaker 2: visual system, the hardware limitation, yeah, the wetwear limitations yea.
• 3:57
  Speaker 2: We have to start viewing the brain not as an
• 3:59
  Speaker 2: information sssor but as a prediction engine.
• 4:02
  Speaker 1: A prediction engine, Okay.
• 4:03
  Speaker 2: The entire architecture of human perception exists to resolve the
• 4:06
  Speaker 2: tension between the physical environment and the internal models your
• 4:10
  Speaker 2: brain has built to survive in that environment.
• 4:12
  Speaker 1: So if we look at the evolutionary pressure here, why
• 4:17
  Speaker 1: the need for the illusion? I think the classic almost
• 4:20
  Speaker 1: naive view of evolution is that it optimizes for perfect accuracy.
• 4:24
  Speaker 2: Right, survival of the most accurate.
• 4:25
  Speaker 1: Yeah, why would natural selection just design a better organic
• 4:28
  Speaker 1: video camera? Why fake it?
• 4:30
  Speaker 2: Well? Because a perfect camera doesn't solve the fundamental physical
• 4:33
  Speaker 2: problem of biological latency, the delay. Yes, and it certainly
• 4:39
  Speaker 2: doesn't solve the problem of data compression. Let's look at
• 4:42
  Speaker 2: the actual biological hardware from moment.
• 4:44
  Speaker 1: Okay, lay it on me.
• 4:45
  Speaker 2: You have over one hundred million photoreceptors in the retina.
• 4:49
  Speaker 1: Of each eye, one hundred million per i.
• 4:52
  Speaker 2: But the optic nerve, which is the only cable transmitting
• 4:55
  Speaker 2: that data back to the brain, only has about one
• 4:58
  Speaker 2: million nerve fibers.
• 4:59
  Speaker 1: Wait. So fromundred million down to one.
• 5:01
  Speaker 2: Million, exactly right at the very first step of vision,
• 5:04
  Speaker 2: you have this massive data compression bottleneck. The eye is
• 5:07
  Speaker 2: basically forced to throw away a huge amount of raw
• 5:10
  Speaker 2: visual information before the signal even leaves the retina.
• 5:13
  Speaker 1: Wow. Okay, so we are already working with a highly degraded,
• 5:17
  Speaker 1: lossy file format just to get the data out of
• 5:20
  Speaker 1: the eyeball.
• 5:21
  Speaker 2: A very lossy format.
• 5:22
  Speaker 1: Yes, But even beyond the bandwidth issue, there's the time delay,
• 5:26
  Speaker 1: which I think is where the passive camera theory really
• 5:28
  Speaker 1: breaks down for me.
• 5:29
  Speaker 2: Oh, it shatters it, because processing sensory information physically takes time.
• 5:34
  Speaker 1: Right, It's not magic, no, it's.
• 5:35
  Speaker 2: Chemistry and electricity. Photons of light bounce off an object
• 5:39
  Speaker 2: and hit the retina. The photoreceptors have to undergo a
• 5:43
  Speaker 2: chemical conformational change to convert that light into electrical action potential.
• 5:47
  Speaker 1: Which takes a fraction of a second.
• 5:49
  Speaker 2: Right, then those electrical signals have to physically travel down
• 5:53
  Speaker 2: the optic nerve, hit the lateral geniculate nucleus and the thalamus,
• 5:57
  Speaker 2: and then project all the way to the primary visual
• 5:59
  Speaker 2: cortex at the very back of the occipital lobe.
• 6:02
  Speaker 1: That's a long trip for a biological signal, it is.
• 6:04
  Speaker 2: And from there it still has to feed forward into
• 6:06
  Speaker 2: higher order areas for you, object recognition and motion processing.
• 6:10
  Speaker 1: And none of that neurochemical transmission is instantaneous. I mean,
• 6:13
  Speaker 1: we're talking about wetwear here, not fiber optic cables.
• 6:17
  Speaker 2: It's wetwear, and it is relatively slow. By the time
• 6:21
  Speaker 2: those visual signals have been fully processed and bubble up
• 6:24
  Speaker 2: into your conscious awareness, anywhere from fifty to maybe a
• 6:28
  Speaker 2: couple of hundred milliseconds might.
• 6:30
  Speaker 1: Have passed a couple one hundred milliseconds. That sounds tiny though.
• 6:34
  Speaker 2: Well in the grand scheme of the universe, one hundred
• 6:36
  Speaker 2: milliseconds is nothing, right, But in evolutionary terms, when you
• 6:39
  Speaker 2: are a primate trying to catch a spear or avoid
• 6:43
  Speaker 2: a striting snake. One hundred milliseconds is an absolute eternity. Oh,
• 6:48
  Speaker 2: I see, by the time you consciously perceive the snake,
• 6:51
  Speaker 2: if you are relying purely on bottom up sensory data,
• 6:55
  Speaker 2: the snake has already moved.
• 6:57
  Speaker 1: It's already biting you.
• 6:58
  Speaker 2: Basically, yeah, the reality you are perceiving is strictly in
• 7:01
  Speaker 2: the past.
• 7:02
  Speaker 1: Wow.
• 7:02
  Speaker 2: If your brain waited for the complete, accurate sensory picture
• 7:05
  Speaker 2: to finish processing before it allowed you to react, you
• 7:08
  Speaker 2: would be operating on a fatal delay.
• 7:10
  Speaker 1: So looking at it through a computational lens, the brain
• 7:13
  Speaker 1: is essentially functioning like the predictive text or like aggressive
• 7:18
  Speaker 1: autocorrect on a smartphone.
• 7:20
  Speaker 2: That is a great way to put it.
• 7:21
  Speaker 1: Like it's continuously typing out the reality of the next
• 7:24
  Speaker 1: millisecond based on statistical probabilities of what usually happens, and
• 7:28
  Speaker 1: it only pauses to correct the simulation if the incoming
• 7:32
  Speaker 1: delayed sensory data flags a major error.
• 7:35
  Speaker 2: What's fascinating here is that the autocorrect analogy maps perfectly
• 7:38
  Speaker 2: onto the mathematical models of predictive coding, particularly when we
• 7:42
  Speaker 2: discuss metabolic efficiency.
• 7:43
  Speaker 1: Metabolic efficiency, so saving energy.
• 7:46
  Speaker 2: Yes, the brain is the most energy hungry organ in
• 7:50
  Speaker 2: the body. It consumes about twenty percent of our total energy,
• 7:53
  Speaker 2: despite being a tiny fraction of our resk.
• 7:56
  Speaker 1: It's an energy, a massive one.
• 7:57
  Speaker 2: If the brain tried to reconstruct every single pixel of
• 8:00
  Speaker 2: the visual field from scratch bottom up, every single millisecond,
• 8:04
  Speaker 2: the metabolic cost would be unsurvivable. We'd starve to death
• 8:07
  Speaker 2: just looking at things that is wild. So the system
• 8:10
  Speaker 2: found a highly elegant shortcut Bayesian inference.
• 8:14
  Speaker 1: Okay, Bayesian inference right.
• 8:16
  Speaker 2: Rather than building reality from the ground up, the higher
• 8:19
  Speaker 2: cognitive areas of the brain generate an ongoing top down expectation,
• 8:23
  Speaker 2: a prior probability of what is most likely to be
• 8:26
  Speaker 2: out there.
• 8:26
  Speaker 1: But if the top down prediction is doing all the
• 8:29
  Speaker 1: heavy lifting, you know, generating the simulation of the room
• 8:32
  Speaker 1: I'm in right now based on what was there a
• 8:34
  Speaker 1: millisecond ago, what is the actual purpose of the raw
• 8:37
  Speaker 1: light hitting my retina?
• 8:39
  Speaker 2: It's a constraint.
• 8:40
  Speaker 1: Is the bottom up sensory data just functioning as an
• 8:43
  Speaker 1: error checking mechanism.
• 8:44
  Speaker 2: That is exactly its primary function. The incoming sensory data
• 8:48
  Speaker 2: is not there to build the picture. It is there
• 8:50
  Speaker 2: to constrain the internal simulation.
• 8:53
  Speaker 1: It keeps the hallucination in check.
• 8:55
  Speaker 2: Precisely, the brain compares the top down prediction against the
• 8:58
  Speaker 2: bottom up sensory of it. If reality matches the internal model,
• 9:02
  Speaker 2: which you know in a static room, it does about
• 9:04
  Speaker 2: ninety nine percent of the time. The brain spends almost
• 9:08
  Speaker 2: zero metabolic energy processing that incoming.
• 9:11
  Speaker 1: Data because it already guessed it right exactly.
• 9:13
  Speaker 2: The signals basically cancel each other out. The brain just
• 9:16
  Speaker 2: lets the simulation roll. It only fires up significant neural
• 9:19
  Speaker 2: activity when it encounters prediction errors.
• 9:22
  Speaker 1: So a prediction error is when the bottom up data
• 9:25
  Speaker 1: contradicts the top down model. Yes, like if I'm looking
• 9:29
  Speaker 1: at a static room and suddenly a big falls off
• 9:31
  Speaker 1: the shelf, my brain wasn't predicting that right.
• 9:34
  Speaker 2: The prior model assumed the book would stay on the.
• 9:36
  Speaker 1: Shelf, so that unexpected motion creates a massive prediction error,
• 9:41
  Speaker 1: which forces the top down model to immediately update.
• 9:44
  Speaker 2: You've got it. The brain is constantly trying to minimize
• 9:48
  Speaker 2: surprise or minimize prediction error. You don't consciously perceive the world.
• 9:54
  Speaker 2: You consciously perceive your brain's internal model of the world
• 9:58
  Speaker 2: which is continuously updated by error signal.
• 10:01
  Speaker 1: I have to play Devil's advocate here, though, Go for it,
• 10:03
  Speaker 1: because as much as the math of predictive coding makes
• 10:06
  Speaker 1: sense for metabolic efficiency, my subjective experience does not feel
• 10:10
  Speaker 1: like an internal model. No, of course, not sitting here,
• 10:13
  Speaker 1: my vision feels continuous, flawless, and completely externally driven. If
• 10:19
  Speaker 1: my brain is essentially hallucinating this room and only checking
• 10:22
  Speaker 1: it against occasional error signals, I need to see physical
• 10:25
  Speaker 1: proof of the simulation in action. Fair enough, Is there
• 10:28
  Speaker 1: a way we can actually prove this control hallucination is
• 10:31
  Speaker 1: happening biologically right in front of the listener's eyes.
• 10:33
  Speaker 2: There is, and it is a stark physiological reality that
• 10:38
  Speaker 2: every human being carries around with them every single day. Okay,
• 10:41
  Speaker 2: I'm listening, Yet almost no one actively notices it because
• 10:45
  Speaker 2: the brain is so incredibly good at covering its tracks.
• 10:49
  Speaker 2: We are talking about the physiological blind spot.
• 10:52
  Speaker 1: Ah, the blind spot. I know we briefly touched on
• 10:55
  Speaker 1: the optic nerve earlier, but let's dive deep into the
• 10:58
  Speaker 1: actual anatomy of why the blind spot exists. Let's do
• 11:01
  Speaker 1: it because it perfectly illustrates the brain's complete refusal to
• 11:05
  Speaker 1: accept missing data.
• 11:06
  Speaker 2: It really does. Let's look at the architecture of the retina.
• 11:09
  Speaker 2: The retina is this thin sheet of neural tissue lining
• 11:13
  Speaker 2: the back of the eye, packed with those one hundred
• 11:15
  Speaker 2: million photoreceptors, right, But the wiring of the human eye
• 11:18
  Speaker 2: is famously and honestly somewhat comically backwards backwards. How well,
• 11:22
  Speaker 2: the nerves that connect to the photoreceptors, they don't go
• 11:24
  Speaker 2: out the back of the eye. They bundle together on
• 11:27
  Speaker 2: the inside surface of the retina, wait.
• 11:29
  Speaker 1: On the inside, which means all of those millions of
• 11:32
  Speaker 1: nerve fibers have to somehow get through the retina to
• 11:36
  Speaker 1: travel back to the brain.
• 11:37
  Speaker 2: Precisely, they are basically in the way of the light.
• 11:40
  Speaker 1: That seems like a terrible design evolution.
• 11:42
  Speaker 2: Is a tinkerer, not an engineer. So all these fibers
• 11:46
  Speaker 2: converge at a specific location called the optic disc where
• 11:49
  Speaker 2: they have to plunge through the retinal tissue to form
• 11:51
  Speaker 2: the optic nerve. Okay, And because of this structural bottleneck,
• 11:55
  Speaker 2: there is a distinct circular area on your retina, roughly
• 11:58
  Speaker 2: fifteen degrees off center, where there are absolutely zero photoreceptors,
• 12:03
  Speaker 2: no rods, no cones, nothing, nothing. It is a biological
• 12:06
  Speaker 2: dead zone. Any photons of light that strike this exact
• 12:09
  Speaker 2: spot are simply not registered. You are entirely completely blind
• 12:14
  Speaker 2: in a specific region of your visual field in each eye.
• 12:17
  Speaker 1: Well wait, if that's anatomically true, and I have a
• 12:20
  Speaker 1: literal gap in my visual field where no data is
• 12:23
  Speaker 1: being collected, why doesn't my subjective experience reflect.
• 12:26
  Speaker 2: That, because your brain won't allow it.
• 12:28
  Speaker 1: If I close one eye right now and look at
• 12:30
  Speaker 1: the wall, I don't see a black, empty void. I
• 12:33
  Speaker 1: don't see a static filled hole hovering in my peripheral vision.
• 12:36
  Speaker 1: The wall looks completely continuous.
• 12:39
  Speaker 2: Because your brain absolutely abhors a vacuum in its predictive model.
• 12:43
  Speaker 1: It just refuses to show me the hole exactly.
• 12:45
  Speaker 2: The predictive engine we just discussed cannot process nothing, so
• 12:49
  Speaker 2: it performs an extraordinary computational feat called spatial filling.
• 12:53
  Speaker 1: In spatial filling, in.
• 12:55
  Speaker 2: Right, the primary visual cortex looks at the statistical probabilities
• 12:59
  Speaker 2: of the visual data surrounding the blind spot. It samples
• 13:02
  Speaker 2: the color, the luminance, and the texture immediately adjacent to
• 13:06
  Speaker 2: the biological hole, and it actively invents the missing region
• 13:09
  Speaker 2: to match.
• 13:10
  Speaker 1: Here's where it gets really interesting, because the visual cortex
• 13:13
  Speaker 1: isn't just ignoring the blind spot. It is actively generating
• 13:17
  Speaker 1: synthetic data.
• 13:18
  Speaker 2: It is fabricating reality.
• 13:20
  Speaker 1: It is literally acting like the clone stamp tool in Photoshop,
• 13:24
  Speaker 1: continuously sampling nearby pixels and pasting them over the missing
• 13:28
  Speaker 1: data in real time.
• 13:29
  Speaker 2: It is the biological equivalent of generative AI filling in
• 13:32
  Speaker 2: the background of an image.
• 13:33
  Speaker 1: Wow.
• 13:34
  Speaker 2: If you look at a highly textured surface, say a
• 13:37
  Speaker 2: complex brick wall or a patterned carpet, and your blind
• 13:40
  Speaker 2: spot happens to land precisely in the middle of that pattern.
• 13:43
  Speaker 1: Which happens all the time constantly.
• 13:45
  Speaker 2: Your brain calculates the trajectory of the mortar lines, it
• 13:48
  Speaker 2: measures the statistical distribution of the colors, and it actively
• 13:52
  Speaker 2: hallucinates the missing texture.
• 13:54
  Speaker 1: That is wild.
• 13:55
  Speaker 2: If a continuous straight line passes right through your blind spot,
• 13:59
  Speaker 2: the early visual air area is predict that the line
• 14:01
  Speaker 2: continues unbroken, and they simply fabricate the middle of the line.
• 14:05
  Speaker 1: That is a profound realization. It is when I look
• 14:08
  Speaker 1: at a patterned wallpaper and a piece of that pattern
• 14:11
  Speaker 1: falls into my blind spot, I'm literally seeing a design
• 14:15
  Speaker 1: that never entered my eye. You are the specific photons
• 14:18
  Speaker 1: from that part of the wall never registered yet I
• 14:21
  Speaker 1: experienced the distinct visual sensation of seeing the pattern. My
• 14:26
  Speaker 1: brain generated it from scratch to keep the internal simulation seamless.
• 14:30
  Speaker 2: You are perceiving a texture generated entirely by the lateral
• 14:34
  Speaker 2: connections in your visual cortex, projected seamlessly into your conscious awareness.
• 14:38
  Speaker 1: It's just astonishing.
• 14:40
  Speaker 2: And this spatial prediction, you know, it doesn't just stop
• 14:42
  Speaker 2: it covering up a biological design flaw in the retina.
• 14:45
  Speaker 1: Oh there's more.
• 14:46
  Speaker 2: Oh yeah. If the brain's predictive engine is willing and
• 14:49
  Speaker 2: able to fake missing spatial data to maintain the illusion
• 14:53
  Speaker 2: of a continuous world, it is equally capable of faking
• 14:56
  Speaker 2: missing temporal data.
• 14:58
  Speaker 1: Temporal data, so time exactly.
• 15:00
  Speaker 2: If it can fake space, it can fake time.
• 15:03
  Speaker 1: Okay, let's pivot to temporal prediction, because we established earlier
• 15:06
  Speaker 1: that the biological latency, the time it takes for a
• 15:10
  Speaker 1: signal to travel from the retina to the visual cortex,
• 15:13
  Speaker 1: creates a massive survival problem.
• 15:16
  Speaker 2: Huge problem.
• 15:17
  Speaker 1: If we are faking time, we are essentially talking about
• 15:19
  Speaker 1: the brain predicting the future. How exactly does the visual
• 15:23
  Speaker 1: system compensate for a moving object.
• 15:26
  Speaker 2: We can see this beautifully illustrated in something called the
• 15:29
  Speaker 2: flash lag effect. But to put it in real world terms,
• 15:32
  Speaker 2: let's look at high speed.
• 15:33
  Speaker 1: Athletics like baseball.
• 15:34
  Speaker 2: Perfect example, consider a professional baseball batter facing a one
• 15:38
  Speaker 2: hundred mile per hour fastball. Okay, the light bounces off
• 15:41
  Speaker 2: the baseball, hits the batter's retina, and that whole chemical
• 15:44
  Speaker 2: transduction process begins.
• 15:45
  Speaker 1: Right one hundred millisecond delay exactly.
• 15:48
  Speaker 2: By the time the visual cortex fully processes that image
• 15:51
  Speaker 2: and generates conscious awareness, roughly fifty to one hundred milliseconds have.
• 15:56
  Speaker 1: Elapsed, which is too slow, way too slow.
• 15:58
  Speaker 2: In one hundred milliseconds, a ball traveling at one hundred
• 16:01
  Speaker 2: miles per hour has moved roughly fifteen feet.
• 16:04
  Speaker 1: Oh wow, fifteen feet.
• 16:05
  Speaker 2: Yes.
• 16:06
  Speaker 1: So if the batter's motor cortex relied solely on the raw,
• 16:10
  Speaker 1: bottom up sensory data to determine where the ball actually is,
• 16:15
  Speaker 1: their conscious perception would be constantly lagging reality by fifteen feet.
• 16:19
  Speaker 2: It would be swinging at ghosts.
• 16:20
  Speaker 1: They would swing the bat at empty air because they
• 16:23
  Speaker 1: are effectively looking into the past exactly.
• 16:25
  Speaker 2: If we experience the world strictly based on the delayed
• 16:28
  Speaker 2: sensory data interacting with dynamic, fast moving environments would be
• 16:33
  Speaker 2: totally impossible.
• 16:34
  Speaker 1: It makes sense.
• 16:34
  Speaker 2: Imagine driving a car on a highway. If you had
• 16:37
  Speaker 2: to break suddenly because the car in front of you
• 16:39
  Speaker 2: slammed on its brakes and you're relying on the delayed
• 16:42
  Speaker 2: visual signal, you would hit the brakes too late every
• 16:45
  Speaker 2: single time.
• 16:45
  Speaker 1: Because you wouldn't see them stop until a fraction of
• 16:47
  Speaker 1: a second later.
• 16:48
  Speaker 2: Right, the car in front of you is physically closer
• 16:51
  Speaker 2: than it appears in your delayed conscious experience, so the brain.
• 16:54
  Speaker 1: Has to employ some kind of temporal extrapolation algorithm. It
• 16:59
  Speaker 1: does it has to calculate where the object is going
• 17:01
  Speaker 1: to be and inject that future position into our present
• 17:06
  Speaker 1: conscious awareness.
• 17:07
  Speaker 2: That is precisely what area MT or V five in
• 17:10
  Speaker 2: the visual cortex does.
• 17:11
  Speaker 1: Very MT.
• 17:12
  Speaker 2: Yeah, air MT is highly specialized for motion processing. It
• 17:15
  Speaker 2: takes the delayed positional data, calculates the object's velocity and
• 17:19
  Speaker 2: its trajectory vectors, and it actually pushes your conscious perception
• 17:23
  Speaker 2: into the future.
• 17:24
  Speaker 1: Wait, it pushes it into the future.
• 17:25
  Speaker 2: Yes, it actively overrides the raw sensory data to show
• 17:29
  Speaker 2: you the object not where it physically was when the
• 17:32
  Speaker 2: light hit your eye, but where the internal physics engine
• 17:35
  Speaker 2: calculates it will be right now.
• 17:37
  Speaker 1: That is absolutely staggering.
• 17:39
  Speaker 2: It's mind bending, isn't it.
• 17:40
  Speaker 1: It implies that my experience of the present moment, right now,
• 17:44
  Speaker 1: as I watch a bird fly across the sky or
• 17:46
  Speaker 1: a car drive past, is technically a hallucination of the
• 17:50
  Speaker 1: immediate future.
• 17:51
  Speaker 2: You got it.
• 17:51
  Speaker 1: I am living slightly ahead of the sensory data just
• 17:54
  Speaker 1: so I can interact successfully with the physical present.
• 17:57
  Speaker 2: It is the only way biological wetwear could solve the
• 18:00
  Speaker 2: physics of time delay. Your brain is constantly running these
• 18:03
  Speaker 2: highly sophisticated forward models predicting momentum, gravity trajectory.
• 18:08
  Speaker 1: It's like a built in physics engine.
• 18:09
  Speaker 2: Exactly like that. And for the most part, the prior
• 18:13
  Speaker 2: models are so highly tuned that the simulation is flawless.
• 18:17
  Speaker 2: The batter hits the ball, you hit the brakes in.
• 18:19
  Speaker 1: Time because the math is right right.
• 18:22
  Speaker 2: The top down prediction allows you to seamlessly interact with
• 18:25
  Speaker 2: a fast moving universe. But you know, this overarching reliance
• 18:29
  Speaker 2: on predictive processing doesn't just dictate how we perceive motion
• 18:33
  Speaker 2: and space.
• 18:34
  Speaker 1: What else is there?
• 18:35
  Speaker 2: It fundamentally alters how we experience static properties too, even
• 18:39
  Speaker 2: something as seemingly objective as color.
• 18:42
  Speaker 1: Okay, I can wrap my head around motion being an
• 18:44
  Speaker 1: active extrapolation to solve a latency problem. Sure but surely
• 18:48
  Speaker 1: something as fundamental as the color of an object is
• 18:51
  Speaker 1: an objective physical reality, right you think so? I mean,
• 18:54
  Speaker 1: if I look at a red apple sitting on a desk,
• 18:57
  Speaker 1: the redness is a physical property of the apple. It's
• 18:59
  Speaker 1: just fact that my eyes are accurately reporting.
• 19:02
  Speaker 2: This is where predictive processing forces us to abandon our
• 19:05
  Speaker 2: most deeply held intuitive beliefs. From a pure physics standpoint,
• 19:11
  Speaker 2: colors do not exist as objective properties in the external world.
• 19:16
  Speaker 2: A red apple is not inherently red.
• 19:18
  Speaker 1: If we are going to claim the apple isn't red,
• 19:20
  Speaker 1: we need to break down the actual physics of what
• 19:22
  Speaker 1: is happening between the light sorts the apple and the eye.
• 19:25
  Speaker 2: Let's do it so. Physical objects consist of atomic structures
• 19:28
  Speaker 2: that absorb certain wavelengths of electromagnetic radiation and reflect others.
• 19:32
  Speaker 1: Basic physics.
• 19:33
  Speaker 2: Right. The skin of the apple contains pigments that absorb
• 19:36
  Speaker 2: most of the visible spectrum but reflect longer wavelengths around
• 19:40
  Speaker 2: six hundred to seven hundred minimeter range. Okay, but those
• 19:43
  Speaker 2: wavelengths themselves bouncing off the apple do not contain the
• 19:46
  Speaker 2: property of red. They are just electromagnetic radiation.
• 19:49
  Speaker 1: So where does the red come from?
• 19:51
  Speaker 2: Redness is a purely subjective phenomenal experience, a quail, as
• 19:55
  Speaker 2: philosophers call it, constructed entirely inside your visual cortex when
• 19:59
  Speaker 2: it interprets those specific wavelengths.
• 20:01
  Speaker 1: So my brain is just arbitrarily assigning a qualitative label
• 20:05
  Speaker 1: the conscious sensation we call red, to a specific frequency
• 20:09
  Speaker 1: of radiation exactly. But if it's just a direct translation
• 20:12
  Speaker 1: of wavelength to color, where does the production engine come in.
• 20:16
  Speaker 1: Why wouldn't that just be a bottom up feed forward process.
• 20:19
  Speaker 2: Because the bottom up data is incredibly unreliable.
• 20:22
  Speaker 1: Unreliable?
• 20:23
  Speaker 2: How well, if your brain just directly translated incoming wavelengths
• 20:27
  Speaker 2: into color, the world would be a psychedelic, chaotic mess.
• 20:31
  Speaker 1: Really.
• 20:31
  Speaker 2: Oh yeah, thing about looking at a white sheet.
• 20:34
  Speaker 1: Of paper, Okay, a standard white piece of print paper,
• 20:36
  Speaker 1: got it right.
• 20:37
  Speaker 2: Imagine looking at that paper outside under the blazing midday sun.
• 20:41
  Speaker 2: The ambient light is very bright, with a heavy skew
• 20:44
  Speaker 2: toward the blue end of the spectrum. Sure, now take
• 20:47
  Speaker 2: that exact same piece of paper indoors, close the blinds,
• 20:50
  Speaker 2: and turn on a very warm, yellowish tungsten incandescent lamp.
• 20:54
  Speaker 1: Okay, entirely different lighting.
• 20:56
  Speaker 2: Exactly, the actual physical wavelengths of light bouncing off that
• 21:00
  Speaker 2: paper and hitting your retina in the second scenario are
• 21:02
  Speaker 2: drastically different. Under the tungsten lamp, the paper is reflecting
• 21:07
  Speaker 2: predominantly long, yellow and red wavelengths.
• 21:10
  Speaker 1: But my subjective experience doesn't change. I don't suddenly perceive
• 21:15
  Speaker 1: a yellow piece of paper. I still clearly see a
• 21:18
  Speaker 1: white piece of paper.
• 21:19
  Speaker 2: If we connect this to the bigger picture of predictive processing,
• 21:22
  Speaker 2: this phenomenon is called color.
• 21:24
  Speaker 1: Constancy color constancy.
• 21:26
  Speaker 2: And it is a massive, top down computational achievement. The
• 21:30
  Speaker 2: brain's goal is not to function as a scientific spectrometer
• 21:34
  Speaker 2: passively measuring exact wavelengths right.
• 21:36
  Speaker 1: It doesn't care about the exact math of the light.
• 21:38
  Speaker 2: Its goal is to reliably identify objects across vastly different
• 21:43
  Speaker 2: environments for survival. Exactly, if a poisonous barrier looked entirely
• 21:47
  Speaker 2: different depending on whether it was in direct sunlight or
• 21:49
  Speaker 2: the shade of a canopy, your ability to survive would
• 21:52
  Speaker 2: be severely compromised. You eat the wrong thing.
• 21:55
  Speaker 1: So the brain has to perform a kind of reverse engineering.
• 21:58
  Speaker 1: It has to untangle the c of the light source
• 22:01
  Speaker 1: from the actual reflective properties of the object itself.
• 22:04
  Speaker 2: Exactly, and it does this using Baesan imprints.
• 22:07
  Speaker 1: Again.
• 22:08
  Speaker 2: The brain surveys the entire visual scene globally. It compares
• 22:12
  Speaker 2: the light arriving from multiple surfaces, looks at shadows, and
• 22:15
  Speaker 2: makes a rapid unconscious statistical estimate of the ambient illumination.
• 22:20
  Speaker 1: It figures out the lighting of the room.
• 22:22
  Speaker 2: Right, It establishes a prior for the lighting conditions. It
• 22:25
  Speaker 2: basically says, the ambient light in this room is highly
• 22:28
  Speaker 2: skewed toward yellow. Okay, Then it mathematically subtracts that inferred
• 22:33
  Speaker 2: yellow illumination from the incoming sensory data of the objects
• 22:36
  Speaker 2: that is wild.
• 22:38
  Speaker 1: The brain evaluates the scene, realizes the light hitting the
• 22:41
  Speaker 1: eye from the paper is yellow, but applies the top
• 22:44
  Speaker 1: down prior that says, you know, the whole room is yellow,
• 22:47
  Speaker 1: and therefore deduces that the object itself must be white. Yes.
• 22:51
  Speaker 1: And the crucial part is that it doesn't just unconsciously
• 22:54
  Speaker 1: know this, It literally edits my conscious phenomenology. It changes
• 22:57
  Speaker 1: the picture, it alters the simulation, so I subjectively experience
• 23:02
  Speaker 1: a weight piece of paper, even though the raw data
• 23:04
  Speaker 1: hating my eye is straight up yellow.
• 23:06
  Speaker 2: You never see the raw data ever, Wow, you only
• 23:10
  Speaker 2: see the output of the Bayesian equation. You see what
• 23:13
  Speaker 2: the brain calculates the object's inherent reflectants should be based
• 23:17
  Speaker 2: on its prediction of the lighting. The brain actively edits
• 23:20
  Speaker 2: reality to force consistency onto an inconsistent physical world.
• 23:25
  Speaker 1: But if the visual system is constantly performing this massive
• 23:29
  Speaker 1: real time editorial control, subtracting inferred lighting, faking missing blind spots,
• 23:35
  Speaker 1: predicting motion vectors, what happens when the top down priors
• 23:38
  Speaker 1: get the context wrong.
• 23:39
  Speaker 2: Ah, Now we're getting into the fun stuff.
• 23:41
  Speaker 1: Because the system making this many complex statistical guesses has
• 23:44
  Speaker 1: to trip up occasionally. I imagine this is exactly what
• 23:47
  Speaker 1: optical illusions exploit.
• 23:49
  Speaker 2: It is exactly what they exploit. When people experience optical illusions,
• 23:53
  Speaker 2: the intuitive reaction is to think that their eyes are
• 23:55
  Speaker 2: failing them, or that their brain is broken or glutching right.
• 23:58
  Speaker 1: The feeling is always my eyes are playing tricks on me.
• 24:01
  Speaker 1: We assume the hardware is malfunctioning.
• 24:03
  Speaker 2: But optical illusions are not malfunctions. They are the exact.
• 24:07
  Speaker 1: Opposite, the opposite.
• 24:09
  Speaker 2: They are the rare, highly engineered edge cases where the normal,
• 24:13
  Speaker 2: everyday predictive algorithms your brain relies on to survive are exposed.
• 24:18
  Speaker 2: They're functioning perfectly, they're just being fed adversarial data.
• 24:22
  Speaker 1: We brought up the autocorrect analogy earlier, and it feels
• 24:25
  Speaker 1: incredibly relevant here. Definitely, most of the time, predictive text
• 24:29
  Speaker 1: is brilliant. You type incredibly fast, your thumbs miss the
• 24:33
  Speaker 1: right keys, and the top down software silently fixes the
• 24:37
  Speaker 1: typos in the background based on the statistical probability of
• 24:40
  Speaker 1: the words you use.
• 24:41
  Speaker 2: It saves so much time.
• 24:42
  Speaker 1: Speed and efficiency are prioritized over absolute perfection. But occasionally
• 24:47
  Speaker 1: you type a slightly ambiguous word in an unusual context
• 24:51
  Speaker 1: and the autocorrect aggressively forces it into a completely wrong.
• 24:54
  Speaker 2: Word, and you stare at your screen wondering why I
• 24:56
  Speaker 2: did that right.
• 24:57
  Speaker 1: The software didn't break, it faithfully executed it's core logic,
• 25:01
  Speaker 1: but this specific context trick the prior.
• 25:03
  Speaker 2: That is a phenomenal analogy. Optical illusions are visual autocorrect errors.
• 25:08
  Speaker 1: Visual autocorrect errors.
• 25:09
  Speaker 2: I love that evolution designed the human visual system to
• 25:12
  Speaker 2: interpret incredibly ambiguous, noisy information almost instantaneously. Imagine our ancestors
• 25:20
  Speaker 2: walking through a dense, dappled forest.
• 25:22
  Speaker 1: Environment lots of shifting light.
• 25:24
  Speaker 2: Right, the interplay of light and shadow is chaotic. A
• 25:27
  Speaker 2: predator hiding in the brush is obscured by shifting contrast,
• 25:31
  Speaker 2: the brain cannot wait to perfectly resolve the image.
• 25:34
  Speaker 1: Oh, you're dead right.
• 25:35
  Speaker 2: It relies on strict, hard coded priors about how light, shadow,
• 25:39
  Speaker 2: and perspective operate in the natural world to instantly construct
• 25:43
  Speaker 2: a coherent model.
• 25:44
  Speaker 1: And an optical illusion is essentially a visual stimulus deliberately
• 25:48
  Speaker 1: engineered by researchers or artists to hack those specific hard
• 25:52
  Speaker 1: coded priors precisely.
• 25:54
  Speaker 2: Take Edward Adelson's famous checkerboard shadow illusion. Have you seen
• 25:58
  Speaker 2: that one?
• 25:59
  Speaker 1: I think so, but describe us.
• 26:00
  Speaker 2: You have an image of a checkerboard with light and
• 26:02
  Speaker 2: dark squares and a cylinder casting a shadow diagonally across
• 26:06
  Speaker 2: the board. Okay, In the illusion, a dark colored square
• 26:09
  Speaker 2: that is sitting directly in the bright light is actually
• 26:11
  Speaker 2: the exact same shade of gray, the exact same pixel
• 26:14
  Speaker 2: value as a light colored square sitting inside.
• 26:17
  Speaker 1: The shadow, the same exact color identical.
• 26:19
  Speaker 2: If you isolate them on a computer screen, they are
• 26:21
  Speaker 2: the same gray, but.
• 26:23
  Speaker 1: When you look at the full image, your brain absolutely
• 26:26
  Speaker 1: refuses to let you see them as the same color.
• 26:30
  Speaker 1: The one in the shadow looks drastically.
• 26:31
  Speaker 2: Lighter because your visual cortex is aggressively applying the rule
• 26:35
  Speaker 2: of color constancy and shadow compensation we just discussed.
• 26:38
  Speaker 1: It's due in the math.
• 26:39
  Speaker 2: Exactly, the brain's top down prior understands that a shadow
• 26:43
  Speaker 2: reduces the luminance of a surface. When it looks at
• 26:46
  Speaker 2: the light square sitting inside the shadow, it automatically calculates
• 26:51
  Speaker 2: this surface is in a shadow, which means less light
• 26:54
  Speaker 2: is bouncing off it.
• 26:55
  Speaker 1: Right.
• 26:55
  Speaker 2: Therefore, to reflect this much light while in a shadow,
• 26:59
  Speaker 2: the physical surface itself must be highly reflective. It must
• 27:02
  Speaker 2: be a light colored square.
• 27:04
  Speaker 1: So it mathematically brightens the square in your conscious perception
• 27:08
  Speaker 1: to communicate the meaning of the object, completely disregarding the
• 27:12
  Speaker 1: physical reality of the photons hitting the retina.
• 27:14
  Speaker 2: Yes, it prioritizes the object's inherent property over the raw
• 27:18
  Speaker 2: sensory data.
• 27:19
  Speaker 1: That is incredible in the natural world.
• 27:21
  Speaker 2: This aggressive autocorrect keeps you alive. It allows you to
• 27:25
  Speaker 2: recognize objects rapidly under terrible fluctuating lighting.
• 27:29
  Speaker 1: Conditions, but the illusion breaks it.
• 27:31
  Speaker 2: The illusion only works because it intentionally isolates and exploits
• 27:34
  Speaker 2: the very Baesian priors that typically ensure our survival, which.
• 27:38
  Speaker 1: Brings up a fascinating structural problem.
• 27:40
  Speaker 2: What's that.
• 27:40
  Speaker 1: Well, all of these computations shadow compensation, color constancy, motion prediction,
• 27:46
  Speaker 1: they all begin with light hitting the retina. But the
• 27:49
  Speaker 1: retina is literally a two dimensional sheet of tissue at
• 27:51
  Speaker 1: the back of a sphere. It is a flat canvas,
• 27:55
  Speaker 1: totally flat. So how does the predictive engine reconstruct a rich,
• 28:00
  Speaker 1: uh complex, three dimensional spatial reality if the only sensory
• 28:04
  Speaker 1: data it has to work with is entirely two D?
• 28:07
  Speaker 2: This represents one of the most profound computational feats of
• 28:10
  Speaker 2: the central nervous system. As you said, the image on
• 28:12
  Speaker 2: the retina has zero depth. It is a completely flat projection.
• 28:16
  Speaker 1: It's like a movie screen exactly.
• 28:18
  Speaker 2: The three dimensional world you experience is not a direct perception.
• 28:21
  Speaker 2: It is a total ground up reconstruction. It is an
• 28:25
  Speaker 2: unconscious inference.
• 28:26
  Speaker 1: But inference requires data. Yes, if depth isn't inherently contained
• 28:31
  Speaker 1: in the two D retinal image, how does the brain
• 28:34
  Speaker 1: infer a spatial dimension that it literally cannot see.
• 28:38
  Speaker 2: By integrating a multitude of independent, highly ambiguous sensory cues
• 28:43
  Speaker 2: and running them through a probabilistic calculation.
• 28:46
  Speaker 1: So it's looking for clues tons of them.
• 28:48
  Speaker 2: The brain utilizes both binocular and monocular cues. Binocular disparity
• 28:53
  Speaker 2: is the one most people are familiar with. Because human
• 28:56
  Speaker 2: eyes are set slightly apart on the face, each retina
• 28:59
  Speaker 2: receives a slightly different two D projection of.
• 29:01
  Speaker 1: The world right the classic viewmaster or VR headset mechanism exactly.
• 29:06
  Speaker 2: The visual cortex, specifically areas like V one and V two,
• 29:10
  Speaker 2: contains neurons that are highly tuned binocular disparity. They compare
• 29:14
  Speaker 2: the two flat images, calculate the minute difference in viewing
• 29:18
  Speaker 2: angles between corresponding points, and use that geometrical discrepancy as
• 29:22
  Speaker 2: a strong data point to infer depth.
• 29:24
  Speaker 1: I have to challenge the primacy of binocular disparity.
• 29:27
  Speaker 2: Though, oh why is that?
• 29:29
  Speaker 1: Because if I cover one eye right now, the world
• 29:31
  Speaker 1: doesn't suddenly collapse into a flat two D painting. I
• 29:34
  Speaker 1: still possess a highly functional, highly accurate sense of depth.
• 29:39
  Speaker 1: I can still reach out and grab my coffee mug
• 29:42
  Speaker 1: without knocking it over, so the brain must be relying
• 29:44
  Speaker 1: heavily on cues that only require a single two D image.
• 29:48
  Speaker 2: You are absolutely right, and in many scenarios monocular cues
• 29:52
  Speaker 2: actually override binocular cues. Really, oh absolutely. Motion parallax is
• 29:57
  Speaker 2: a massive one.
• 29:57
  Speaker 1: What's mosher parallax?
• 29:59
  Speaker 2: When you move your head e even slightly, objects that
• 30:01
  Speaker 2: are physically closer to you shift rapidly across your retinal field,
• 30:05
  Speaker 2: while objects far away on the horizon barely.
• 30:07
  Speaker 1: Move at all, like looking out the window of a
• 30:09
  Speaker 1: moving train.
• 30:10
  Speaker 2: Perfect example. The visual cortex instantly translates those differential motion
• 30:15
  Speaker 2: vectors into a highly accurate spatial map.
• 30:17
  Speaker 1: So it's extracting three D structure from two D motion.
• 30:20
  Speaker 2: Yes, But it goes far beyond motion. The brain analyzes
• 30:24
  Speaker 2: texture gradients, like the way the grain of a wooden
• 30:26
  Speaker 2: table becomes denser and less distinct the further away it is.
• 30:30
  Speaker 2: It utilizes linear perspective, where parallel lines converge at a
• 30:33
  Speaker 2: vanishing point. It analyzes acclusion, where one object blocking another
• 30:37
  Speaker 2: definitively proves which one is closer. That makes sense, and
• 30:40
  Speaker 2: it relies incredibly heavily on shape from shading. The brain
• 30:44
  Speaker 2: has a deeply embedded prior that light generally comes from
• 30:48
  Speaker 2: above because the sun, and it uses the way shadows
• 30:51
  Speaker 2: fall across a two D circle to instantly pop it
• 30:54
  Speaker 2: out into a three D sphere in your perception.
• 30:57
  Speaker 1: So, if we synthesize all of this, the brain is
• 31:00
  Speaker 1: acting less like a passive observer and more like a
• 31:04
  Speaker 1: forensic detective piecing together circumstantial evidence. Yes, because none of
• 31:09
  Speaker 1: those monocular cues a shadow, a blurry texture, a converging line,
• 31:14
  Speaker 1: none of them individually guarantees distance.
• 31:16
  Speaker 2: That is the perfect way to frame it. Vision is
• 31:18
  Speaker 2: not photography. Vision is active inference.
• 31:21
  Speaker 1: Active inference.
• 31:22
  Speaker 2: The great nineteenth century physicist Hermann von Helmholtz coined the
• 31:26
  Speaker 2: term unconscious inference for this exact process.
• 31:29
  Speaker 1: Oh wow, so they've been thinking about this for a
• 31:30
  Speaker 1: long time.
• 31:31
  Speaker 2: Over a century. The brain takes all of this ambiguous,
• 31:34
  Speaker 2: circumstantial two D evidence, weighs it against its internal priors,
• 31:37
  Speaker 2: and generates a probabilistic hypothesis.
• 31:40
  Speaker 1: It's just playing the odds.
• 31:41
  Speaker 2: Basically, It effectively asks the Bayesian equation. You know, given
• 31:45
  Speaker 2: the current shading, the motion, parallax, the occlusion, and my
• 31:48
  Speaker 2: past experience with these objects, what is the most statistically
• 31:52
  Speaker 2: probable arrangement of physical matter in three D space that
• 31:56
  Speaker 2: would cause this specific two D pattern.
• 31:58
  Speaker 1: On my retina, and then it just builds.
• 32:00
  Speaker 2: The three D world you consciously experience is simply the
• 32:03
  Speaker 2: brain's final winning hypothesis presented to you as objective reality.
• 32:08
  Speaker 1: Since you mentioned that the brain weighs this evidence against
• 32:11
  Speaker 1: past experience with these objects, I want to steer the
• 32:15
  Speaker 1: conversation into the role of memory.
• 32:17
  Speaker 2: Okay, memory is huge here, because.
• 32:19
  Speaker 1: If the brain's predictive engine is relying on prior knowledge
• 32:23
  Speaker 1: to resolve the ambiguity of the present moment, then our
• 32:26
  Speaker 1: past memories and experiences must be literally shaping the fundamental
• 32:30
  Speaker 1: structure of our current visual perception.
• 32:33
  Speaker 2: They absolutely are. In the predictive processing framework. We cannot
• 32:37
  Speaker 2: cleanly separate perception from memory.
• 32:39
  Speaker 1: They're intertwined.
• 32:40
  Speaker 2: They are part of the same hierarchical generative model. Let's
• 32:43
  Speaker 2: think back to how the brain fills in the spatial
• 32:45
  Speaker 2: void of the blind spot. It performs the exact same
• 32:48
  Speaker 2: autocomplete function for object recognition in the real world. In
• 32:52
  Speaker 2: any natural environment, objects are constantly occluding each other. Your
• 32:56
  Speaker 2: laptop partially hides your coffee cup. A parked car obscures
• 33:01
  Speaker 2: part of a fire hydrant.
• 33:02
  Speaker 1: Right, We rarely see the full one hundred percent outline
• 33:06
  Speaker 1: of anything exactly.
• 33:07
  Speaker 2: Your retinas are constantly receiving fragmented, disconnected patches of visual data.
• 33:12
  Speaker 1: But phenomenologically, I don't perceive a chaotic collage of floating
• 33:17
  Speaker 1: handles and disconnected rims. When I look at the coffee
• 33:20
  Speaker 1: cup behind the laptop, I subjectively perceive the presence of
• 33:24
  Speaker 1: a complete, solid coffee cup.
• 33:26
  Speaker 2: Right, because your higher cortical areas, specifically the inferior temporal cortex,
• 33:31
  Speaker 2: which handles complex object recognition, are feeding massive pryors down
• 33:35
  Speaker 2: into your early visual area.
• 33:37
  Speaker 1: Downloaded the file for coffee cup from.
• 33:38
  Speaker 2: My memory, That's a great way to put it. The
• 33:40
  Speaker 2: brain searches its vast database of statistical models, retrieves the
• 33:43
  Speaker 2: structural template of what a coffee cup usually looks like,
• 33:46
  Speaker 2: and actively projects the accluded edges and hidden surfaces into
• 33:50
  Speaker 2: your internal simulation.
• 33:51
  Speaker 1: So I perceive the complete object not because I see
• 33:54
  Speaker 1: the whole thing, but because my prior expectation completes the geometry. Yes,
• 34:00
  Speaker 1: what I expect to see is quite literally dictating what
• 34:03
  Speaker 1: I actually see. The prior is heavily biasing the likelihood.
• 34:07
  Speaker 2: Yes, your lifetime of previous visual experiences has built a
• 34:11
  Speaker 2: highly refined statistical model of the environment.
• 34:15
  Speaker 1: This is why familiar spaces feel so cognitively effortless to navigate.
• 34:19
  Speaker 2: Exactly when you walk into your own kitchen, your brain
• 34:22
  Speaker 2: doesn't have to work hard to process the raw sensory data.
• 34:25
  Speaker 2: It already possesses a rigid predictive model of the spatial layout.
• 34:29
  Speaker 1: It knows where the fridge is, where the island is.
• 34:32
  Speaker 2: Right, the bottom up sensory data is mostly just confirming
• 34:35
  Speaker 2: what the top down model already expects, generating almost no
• 34:38
  Speaker 2: prediction errors.
• 34:39
  Speaker 1: So what does this all mean for the concept of expertise.
• 34:42
  Speaker 1: Expertise does the acquisition of highly specialized knowledge actually physically
• 34:47
  Speaker 1: alter the biological operation of the visual system.
• 34:49
  Speaker 2: It undeniably alters the perceptual hardware. Let's consider expert radiologists
• 34:54
  Speaker 2: analyzing a complex MRI or X ray scan. When a
• 34:57
  Speaker 2: layperson looks at an MRI, they perceive a chaotic, ambiguous
• 35:02
  Speaker 2: mess of gray blobs and static.
• 35:05
  Speaker 1: It just looks like TV static to me.
• 35:07
  Speaker 2: Right, But an experienced radiologist looks at the exact same
• 35:10
  Speaker 2: image and immediately effortlessly spots a minuscule tumor or anomaly.
• 35:15
  Speaker 1: Because they've seen it a million times.
• 35:17
  Speaker 2: Exactly, it is not that the radiologist's retinal photo receptors
• 35:21
  Speaker 2: are capturing higher resolution photons than the laypersons.
• 35:24
  Speaker 1: It's that their priors are completely different.
• 35:27
  Speaker 2: Precisely, years of intense medical training and thousands of hours
• 35:31
  Speaker 2: of analyzing scams have fundamentally rewired the generative models in
• 35:35
  Speaker 2: their visual cortex. Wow, their high level priors are specifically
• 35:38
  Speaker 2: tuned to recognize the statistical signature of a tumor. Their
• 35:42
  Speaker 2: brain uses that top down expectation to literally pull the
• 35:46
  Speaker 2: anomaly out of the visual noise, making it pop into
• 35:49
  Speaker 2: their conscious awareness effortlessly.
• 35:51
  Speaker 1: While the layperson's brain, lacking that specific prior, simply categorizes
• 35:55
  Speaker 1: it as background.
• 35:56
  Speaker 2: Noise and just smooths it over.
• 35:58
  Speaker 1: And this framework must apply broadly across all sensory modalities,
• 36:02
  Speaker 1: not just vision. Think about an expert musician, like a
• 36:06
  Speaker 1: jazz pianist or a classical conductor listening to a complex symphony. Absolutely,
• 36:10
  Speaker 1: a novice just hears a wall of sound, but the
• 36:14
  Speaker 1: expert's auditory cortex is heavily constrained by years of music theory.
• 36:19
  Speaker 1: They are likely anticipating the chord progressions and the harmonic resolutions.
• 36:22
  Speaker 1: Before the orchestra even plays the notes.
• 36:25
  Speaker 2: They are hearing the future of.
• 36:26
  Speaker 1: The song right their top down model is predicting the melody.
• 36:31
  Speaker 1: It proves that experience isn't just a passive archives of
• 36:33
  Speaker 1: memories we occasionally access. Experience literally fundamentally changes our real
• 36:39
  Speaker 1: time conscious perception of reality.
• 36:41
  Speaker 2: It does, and you know, relying this heavily on a
• 36:43
  Speaker 2: predictive framework where perception is essentially guided by what we
• 36:47
  Speaker 2: expect to encounter, has massive and often dangerous consequences for
• 36:51
  Speaker 2: conscious awareness. Dangerous how well, if our brain is metabolically
• 36:54
  Speaker 2: incentivized to only process what it expects to see or
• 36:59
  Speaker 2: what its model are specifically primed for. What happens to
• 37:02
  Speaker 2: the staggering volume of high fidelity sensory information hitting our
• 37:07
  Speaker 2: retinas that the top down model doesn't expect or doesn't
• 37:10
  Speaker 2: currently value.
• 37:11
  Speaker 1: I would assume, based on the energy conservation principle, that
• 37:15
  Speaker 1: it just gets brutally ignored.
• 37:17
  Speaker 2: Brutally is the right word, because.
• 37:19
  Speaker 1: If the brain is constantly trying to minimize prediction errors
• 37:22
  Speaker 1: and save calories, processing useless unexpected background data would be
• 37:27
  Speaker 1: a massive metablock waste.
• 37:29
  Speaker 2: It gets completely ruthlessly filtered out. Your retinas are capturing
• 37:32
  Speaker 2: megabytes of raw visual data every single second, right, but
• 37:36
  Speaker 2: only a microscopic fraction of that data ever breaches the
• 37:39
  Speaker 2: threshold of conscious awareness. In the predictive processing model, attention
• 37:43
  Speaker 2: is not a spotlight that illuminates reality.
• 37:45
  Speaker 1: Would is it? Then?
• 37:46
  Speaker 2: Attention acts as the ultimate synaptic gain control. It is
• 37:49
  Speaker 2: a prediction filter.
• 37:50
  Speaker 1: How does the brain calculate what is allowed to pass
• 37:52
  Speaker 1: through the filter and what gets permanently discarded before we
• 37:56
  Speaker 1: ever consciously perceive it.
• 37:57
  Speaker 2: The brain dynamically adjusts the precision weighting of sensory.
• 38:01
  Speaker 1: Signals precision waiting. Right.
• 38:03
  Speaker 2: It prioritizes information based on a few key criteria. First,
• 38:08
  Speaker 2: as we discussed, it prioritizes massive prediction errors data that
• 38:13
  Speaker 2: violently contradicts a strong prior.
• 38:15
  Speaker 1: Like a loud bang in a silent room.
• 38:17
  Speaker 2: Or a sudden erratic movement in a static visual field.
• 38:20
  Speaker 1: Right the tiger jumping out of the.
• 38:21
  Speaker 2: Bushes exactly second. And crucially, it applies high precision weighting
• 38:27
  Speaker 2: to information that is behaviorally relevant to your current goals. Goals, Yeah,
• 38:32
  Speaker 2: if you are walking down a crowded street and you
• 38:34
  Speaker 2: are extremely hungry, your top down model will suddenly crank
• 38:38
  Speaker 2: up the gain on the specific visual frequencies associated with
• 38:41
  Speaker 2: a brightly colored restaurant sign Oh wow. You will consciously
• 38:45
  Speaker 2: perceive the sign while simultaneously filtering out the faces of
• 38:48
  Speaker 2: hundreds of strangers, the texture of the pavement, and the
• 38:51
  Speaker 2: hum of traffic. Everything irrelevant is suppressed to keep the
• 38:54
  Speaker 2: internal simulation efficient.
• 38:56
  Speaker 1: This sounds exactly like the mechanism behind inattentional blindness.
• 39:00
  Speaker 2: Is the exact mechanism I'm.
• 39:01
  Speaker 1: Thinking of the famous psychological experiments, specifically the invisible Gorilla test,
• 39:06
  Speaker 1: where subjects are asked to watch a video of people
• 39:09
  Speaker 1: passing a basketball and are given a highly specific, intense
• 39:15
  Speaker 1: cognitive task count the passes right, count the exact number
• 39:18
  Speaker 1: of passes made by the team in white shirts. And
• 39:21
  Speaker 1: while their top down models are intently focused on tracking
• 39:24
  Speaker 1: the white shirts, something completely absurd happens in the middle
• 39:27
  Speaker 1: of the frame.
• 39:28
  Speaker 2: Yes, a person in a full gorilla suit walks slowly
• 39:31
  Speaker 2: into the center of the frame, stops, beats their chest,
• 39:34
  Speaker 2: and slowly walks off screen.
• 39:36
  Speaker 1: And they're in plain sight for nearly ten seconds.
• 39:39
  Speaker 2: Ten full seconds, and roughly.
• 39:41
  Speaker 1: Fifty percent of the people watching the video completely genuinely
• 39:45
  Speaker 1: fail to see the gorilla. It's astounding, and they are
• 39:48
  Speaker 1: usually furious when told they missed it, swearing the video
• 39:51
  Speaker 1: was altered. But the raw physics of this situation is undeniable.
• 39:55
  Speaker 1: The photons bouncing off the gorilla suit entered their corneas.
• 39:58
  Speaker 1: Yes the light hit their retinas, the photoreceptors fired action
• 40:02
  Speaker 1: potentials down the optic nerve. The sensory data physically entered
• 40:05
  Speaker 1: the hardware of their brain.
• 40:07
  Speaker 2: It was processed by the early visual cortex.
• 40:10
  Speaker 1: But because their attentional filter, their precision weighting was so
• 40:14
  Speaker 1: narrowly tuned to predicting the spatial trajectory of the white
• 40:18
  Speaker 1: basketball players. The brain's higher cortical areas deemed the guerrilla
• 40:23
  Speaker 1: data entirely irrelevant to the current predictive model.
• 40:26
  Speaker 2: And completely erased it from their conscious awareness.
• 40:29
  Speaker 1: Erased it. It remains one of the most humbling and
• 40:32
  Speaker 1: vital demonstrations of the limits of human cognition.
• 40:35
  Speaker 2: It physically proves that while our visual world feels seamlessly complete,
• 40:41
  Speaker 2: our conscious awareness is astonishingly selective and sparse. We are
• 40:45
  Speaker 2: routinely biologically blind to monumental changes in our environment. If
• 40:49
  Speaker 2: we do not expect them and are not actively attending
• 40:52
  Speaker 2: to them.
• 40:52
  Speaker 1: So we've talked a lot about top down pryors, bottom
• 40:54
  Speaker 1: up prediction errors, and attention filters. Yes, but how does
• 40:57
  Speaker 1: all of this actually map out of the wetwear on
• 41:00
  Speaker 1: a strictly anatomical level. How does a high level cognitive
• 41:03
  Speaker 1: expectation actually physically alter the electrical signals arriving from the retina.
• 41:08
  Speaker 2: To understand the physical implementation of the predictive engine, we
• 41:11
  Speaker 2: have to look at modern neuronata and functional brain imaging,
• 41:15
  Speaker 2: which has completely inverted our classical assumptions about how the
• 41:19
  Speaker 2: brain wires itself. How so well, for decades, the dominant
• 41:23
  Speaker 2: paradigm in neuroscience was a strict bottom up feed forward process.
• 41:28
  Speaker 2: Light hits the retina, travels up the optic nerve to
• 41:31
  Speaker 2: the lateral geniculate nucleus in the thalamus, projects up to
• 41:35
  Speaker 2: the primary visual cortex or V one for basic edge detection.
• 41:39
  Speaker 1: Right just a one way street exactly.
• 41:41
  Speaker 2: Then it moves up to V two and V four
• 41:43
  Speaker 2: for color and shape, and finally up to the inferior
• 41:46
  Speaker 2: temporal cortex where a grandmother sell fires and you recognize
• 41:49
  Speaker 2: an object.
• 41:50
  Speaker 1: That's how it's taught in basic biology, it was viewed.
• 41:52
  Speaker 2: As a one way street from raw sensation up to
• 41:55
  Speaker 2: complex cognition.
• 41:57
  Speaker 1: But if the predictive processing model is accurate, that classical
• 42:01
  Speaker 1: one way street is entirely backwards. Entirely the higher cognitive areas,
• 42:05
  Speaker 1: the francal lobes, the temporal lobes holding our memories and
• 42:08
  Speaker 1: statistical priors. They must have a physical way to communicate
• 42:12
  Speaker 1: back down the hierarchy to the early visual areas to
• 42:14
  Speaker 1: project the simulation.
• 42:16
  Speaker 2: And that is exactly what the modern neuroanatomical data reveals. Really,
• 42:21
  Speaker 2: the brain possesses massive, extensive feedback pathways. In fact, if
• 42:26
  Speaker 2: we look at the connections between the primary visual cortex
• 42:29
  Speaker 2: V one and the thalamus, the neural connections traveling downward
• 42:33
  Speaker 2: from the cortex back to the thalamus vastly outnumber the
• 42:37
  Speaker 2: feed forward connections coming up from the eyes.
• 42:40
  Speaker 1: Wait. Really, the brain is sending more signals away from
• 42:44
  Speaker 1: the central processing areas back down toward the early visual
• 42:47
  Speaker 1: relay stations than it is actually receiving from the outside world.
• 42:51
  Speaker 2: Yes, by an order of magnitude in some pathways.
• 42:54
  Speaker 1: That is insane.
• 42:55
  Speaker 2: Higher cortical areas are continuously blanketing the lower visual areas
• 42:59
  Speaker 2: within massive streams of descending predictions.
• 43:02
  Speaker 1: And constantly shouting down at them exactly.
• 43:04
  Speaker 2: They are constantly priming the synaptic weights of the early
• 43:07
  Speaker 2: visual cortex, effectively saying, based on the prior, here's the
• 43:10
  Speaker 2: specific pattern of data you should expect to process in
• 43:13
  Speaker 2: the next one hundred milliseconds. Because of this descending architecture,
• 43:17
  Speaker 2: incoming bottom up sensory signals are being influenced, intercepted, and
• 43:22
  Speaker 2: interpreted at the earliest possible stages of visual processing.
• 43:25
  Speaker 1: So perception is not a passive, delayed reception at the
• 43:28
  Speaker 1: end of a chain.
• 43:29
  Speaker 2: No, it is an incredibly dynamic, real time negotiation between
• 43:33
  Speaker 2: descending top down expectations and ascending bottom up sensory evidence.
• 43:38
  Speaker 1: This raises an important question, Okay, well what is it
• 43:40
  Speaker 1: What happens when that negotiation fails? Ah? If our entire
• 43:44
  Speaker 1: grasp on reality, our entire conscious phenomenology depends on this
• 43:49
  Speaker 1: incredibly delicate computational balance between what our high level priors
• 43:53
  Speaker 1: expect and what are low level sensory prediction eras report.
• 43:58
  Speaker 1: What happens to human perception those synaptic weightings get thrown off.
• 44:02
  Speaker 2: This is where the predictive processing framework provides deeply profound,
• 44:05
  Speaker 2: unifying insights into neurology, psychiatry, and psychopathology. When the delicate
• 44:11
  Speaker 2: mathematical balance of precision weighting is disrupted, perception itself fundamentally
• 44:16
  Speaker 2: shatters or mutates. Yes, If the brain's neurochemistry alters, perhaps
• 44:21
  Speaker 2: due to say, dopamine diusregulation, and it begins giving massive,
• 44:25
  Speaker 2: inappropriate precision weighting to its own internal top down predictions,
• 44:30
  Speaker 2: while simultaneously suppressing or ignoring the bottom up prediction errors
• 44:35
  Speaker 2: that normally constrain.
• 44:35
  Speaker 1: Them, the simulation breaks free.
• 44:37
  Speaker 2: Exactly, the internal simulation loses its tether to the physical.
• 44:40
  Speaker 1: World, and a top down internal simulation that has lost
• 44:43
  Speaker 1: its constraint to physical sensory evidence is the exact clinical
• 44:47
  Speaker 1: definition of a hallucination.
• 44:49
  Speaker 2: Exactly under the predictive framework. Hallucinations are not random misfirings
• 44:54
  Speaker 2: of visual cortex noise.
• 44:56
  Speaker 1: They're not just static no.
• 44:58
  Speaker 2: They are overly rigid, overly dicise top down priors that
• 45:02
  Speaker 2: have become so incredibly strong they effectively override the sensory data.
• 45:05
  Speaker 2: The brain strongly predicts a reality, say a voice or
• 45:08
  Speaker 2: a figure, and refuses to accept the ascending sensory evidence
• 45:12
  Speaker 2: that dictates the room is actually empty.
• 45:14
  Speaker 1: That makes terrifying sense.
• 45:15
  Speaker 2: Many prominent neuroscientists like Carl Fristen now argue that the
• 45:20
  Speaker 2: positive symptoms of conditions like schizophrenia, the delusions and visual hallucinations,
• 45:26
  Speaker 2: can be elegantly explained as a failure of this predictive
• 45:29
  Speaker 2: precision weighting.
• 45:30
  Speaker 1: The brain generates an intense prior and the error checking
• 45:33
  Speaker 1: mechanism just fails to update it precisely.
• 45:36
  Speaker 2: Conversely, if you look at conditions like autism, some predictive
• 45:39
  Speaker 2: processing models suggest the opposite waiting issue the opposite the
• 45:42
  Speaker 2: priors might be too weak, or the sensory prediction errors
• 45:45
  Speaker 2: are weighted far too heavily, resulting in a world that
• 45:48
  Speaker 2: feels overwhelmingly intense, chaotic, and relentlessly surprising. Because the brain
• 45:53
  Speaker 2: cannot efficiently predict and suppress the sensory noise.
• 45:56
  Speaker 1: It really emphasizes how breathtakingly freshed our grip on reality
• 46:01
  Speaker 1: actually is.
• 46:02
  Speaker 2: It's hanging by a thread.
• 46:03
  Speaker 1: The reality we see, navigate and just completely take for
• 46:07
  Speaker 1: granted every single day depends entirely on billions of microscopic
• 46:11
  Speaker 1: synaptic calculations. Maintaining an absolutely perfect, delicate balance between our
• 46:16
  Speaker 1: inner statistical expectations and outer physical physics.
• 46:19
  Speaker 2: It is arguably the most remarkable biological and computational achievement
• 46:24
  Speaker 2: in the known universe. Truly, vision is not a passive window.
• 46:27
  Speaker 2: It is an active, demanding construction. Every single hue of
• 46:32
  Speaker 2: color you perceive, every distinct object you instantly categorize, every
• 46:37
  Speaker 2: visceral sense of depth, stability, and trajectory that's all built.
• 46:41
  Speaker 2: It all emerges from an endless cascade of unconscious Baesian
• 46:44
  Speaker 2: predictions operating deep beneath the threshold of your awareness.
• 46:48
  Speaker 1: The brain takes this highly compressed, massively delayed, incredibly noisy
• 46:53
  Speaker 1: stream of sensory data and continuously effortlessly transforms it into
• 46:58
  Speaker 1: a cohesive, beautiful, immersive three D simulation of reality, lawless lea,
• 47:02
  Speaker 1: And it does it so flawlessly that the simulation is
• 47:05
  Speaker 1: accurate enough to guide our behavior, keep us alive, let
• 47:08
  Speaker 1: us hit at one hundred mile per hour fastball, and
• 47:10
  Speaker 1: navigate complex social environments, all while remaining plastic and flexible
• 47:14
  Speaker 1: enough to adapt to a continuously shifting world.
• 47:17
  Speaker 2: The ultimate takeaway from peeling back these layers of neuroscience
• 47:20
  Speaker 2: is that the world you experience every second of your
• 47:22
  Speaker 2: waking life is not objective reality exactly as it exists
• 47:25
  Speaker 2: in the physical universe.
• 47:26
  Speaker 1: It's a guess.
• 47:27
  Speaker 2: It is merely the most probable, metabolically efficient version of
• 47:31
  Speaker 2: reality your brain can construct from limited, noisy evidence. In
• 47:36
  Speaker 2: a very profound literal sense, every conscious moment you experience
• 47:41
  Speaker 2: is a carefully curated, constantly updated hypothesis, refined billions of
• 47:46
  Speaker 2: times a second by nature's greatest prediction engine.
• 47:49
  Speaker 1: And unpacking all of that leads us with a truly lingering,
• 47:53
  Speaker 1: provocative philosophical thought. To end on, I love a good
• 47:56
  Speaker 1: philosophical thought. We started this whole discussion by shattering the
• 47:59
  Speaker 1: illusion of the objective video camera. We establish that our
• 48:02
  Speaker 1: previous experiences are memories, and our highly specific training create
• 48:06
  Speaker 1: the complex statistical models that literally dictate our future perception.
• 48:11
  Speaker 1: We perceive the world not as it is, but through
• 48:13
  Speaker 1: the highly customized lens of our own historical priors. So
• 48:16
  Speaker 1: what does this all mean? If every single human brain
• 48:20
  Speaker 1: is building its own internal simulation based on its own
• 48:23
  Speaker 1: entirely unique history of experiences, biases, and biological waitings, does
• 48:29
  Speaker 1: that mean two people standing in the exact same room,
• 48:31
  Speaker 1: looking at the exact same physical scene are biologically constructing
• 48:35
  Speaker 1: and consciously experiencing two completely fundamentally different physical realities. It's
• 48:40
  Speaker 1: a haunting question. The video camera's a myth. The reality
• 48:44
  Speaker 1: you see is uniquely, exclusively yours. Something to think about
• 48:47
  Speaker 1: the next time you open your eyes.