conductor(generic_systems_fields): Phase 1 Acquire - transcript (885 clean segments, 30KB) + 58MB mp4

2026-06-21 23:16:13 -04:00
parent 900b68009b
commit 99e955795f
4 changed files with 13217 additions and 0 deletions
@@ -0,0 +1,11 @@
 Phase 1 Acquire for generic_systems_fields: https://youtu.be/QeMajYvhEbI
 Artifacts: C:\projects\manual_slop\conductor\tracks\video_analysis_generic_systems_fields_20260621\artifacts
 Step 1: extract_transcript (yt-dlp VTT directly)
  OK: wrote C:\projects\manual_slop\conductor\tracks\video_analysis_generic_systems_fields_20260621\artifacts\transcript.json (1751 segments)
 Step 2: download_video
  OK: wrote C:\projects\manual_slop\conductor\tracks\video_analysis_generic_systems_fields_20260621\artifacts\video.mp4 (60755152 bytes)
 {
  "status": "ok",
  "video_path": "C:\\projects\\manual_slop\\conductor\\tracks\\video_analysis_generic_systems_fields_20260621\\artifacts\\video.mp4",
  "transcript_path": "C:\\projects\\manual_slop\\conductor\\tracks\\video_analysis_generic_systems_fields_20260621\\artifacts\\transcript.json"
 }
@@ -0,0 +1,885 @@
 I want to talk today about interesting
 behavior by generic systems.
 And the first part of this will be a bit
 of a review and then the second half
 will be more u a report on new work uh
 since the last time I talked which
 uh as a general presentation was over a
 year ago. I did a couple of more
 specialized presentations last year. So
 this is in the context of the diverse
 intelligence project
 which raises these questions u how
 diverse is intelligence
 and what what are the limits of
 intelligence uh if there are limits and
 how do we find out so that's what I'm
 going to talk about today
 uh to start with um
 we can look at this definition of
 intelligence that we've been using in
 the lab from William James as as kind of
 a guideline
 that intelligence is a fixed goal
 achievable with variable means of
 achieving it. So intelligence involves
 ing it. So intelligence involves
 some level of flexibility
 and this definition itself raises
 questions. Uh does any goal count? Uh
 are many any kinds of means allowed
 and does anything fall outside this
 definition? Are there any are there
 systems that don't achieve this kind of
 flexibility that James was interested
 in? And um it's not clear from from
 James the extent to which he was willing
 to extend intelligence
 uh to generic systems,
 but at at least some of his writings
 suggest that that may have been the
 case.
 Um
 some sort of methodological guidelines
 uh have appeared recently. Uh, for
 example, in this paper that Mike and
 David Resnik published last year,
 uh, emphasizing that the
 diverse intelligence project or the the
 TAM framework,
 um,
 emphasizes methodologically
 a reliance on empirical research, not on
 intuition for thinking about
 intelligence.
 and um an insistence on testable methods
 for understanding uh what is and is not
 intelligent and the extent to which
 things are intelligent. So I I want to
 follow this kind of empirically oriented
 framework as opposed to making
 philosophical assumptions upfront.
 And in thinking about this um an obvious
 place to look is the literature of the
 free energy principle.
 So here's um a recent paper from Carl
 Fristen's group
 um their most recent actual technical
 elaboration of the free energy principle
 and they describe it as describing a
 simple relationship between the dynamics
 of random dynamical systems and
 inference
 and that of course has been the theme of
 the free energy principle for almost 30
 years now that random dynamical systems
 uh can be considered inferential.
 But when you look into this paper
 um here's figure two
 already we find an assumption that
 there's some systems that are inert that
 don't act on the world at all.
 And that's a problem because a system
 that doesn't act back on the world when
 the world acts on it is violating
 Newton's third law. So it's it's
 violating a very basic physical symmetry
 principle.
 So
 this picture uh at least appears to
 involve uh intuitions about inertness.
 Uh and the the trademark inert system is
 a rock uh in much of this literature
 that may not stand up to scrutiny from
 the point of view of fundamental
 physics.
 So uh in that case I think we need to
 take an approach where we look at
 physics first and worry about intuitions
 later.
 So if we have a system which I'll call a
 and it's interacting with something else
 which I'll call a bar or the complement
 of a for reasons that will become clear.
 How do we guarantee that this
 interaction respects all of the physical
 symmetries that it needs to respect to
 comply with known physical theory or
 empirically supported physical theory.
 And what we want is a generic case that
 describes any interacting system
 regardless of its scale or its
 structure.
 And um I'll add here since
 the whole talk will respect this uh
 regardless of its embedding in spacetime
 what it looks like as a as a thing in
 spacetime.
 So how do we do that? And the answer is
 we start with the generic case.
 So let's assume the simplest thing we
 can assume
 which is a system that doesn't interact
 with anything. So a system that doesn't
 have an environment.
 Um so a system that's isolated.
 And if we start with that, um, we know
 that we're going to respect the various
 symmetries that have to do with not
 having singularities because there's no
 there's no place for information to flow
 from into this system and no place for
 information to flow to because there's
 no environment. So there aren't going to
 be sources or syncs.
 So whatever dynamics this system has, it
 has to conserve momentum and energy and
 information.
 U so
 the technical term for conserving
 information is unitarity
 and it turns out that this is a very
 productive assumption.
 If we have conservation of information
 in other words unitarity
 then we can represent the dynamics as a
 linear operator
 um because nonlinearities
 uh don't conserve information.
 So this dynamics that I've called P of U
 the propagator of U is a linear operator
 on some state space. And in fact we can
 make the state space a Hilbert space
 which is a particularly simple kind of
 vector space. It just takes every
 possible value of every possible degree
 of freedom and makes it a basis vector.
 So if we assume some background time t
 which is just a symbol that lets us talk
 about change.
 Uh we can write this propagator
 uh as a periodic function of another
 operator which represents the energy of
 the system. That operator is the
 Hamiltonian
 which plays essentially the same role
 here that it plays in classical physics.
 It's a measure of energy
 and when we write this equation um it
 requires introducing a constant which is
 finite and which has the units of action
 and a finite value of this constant
 um corresponds to there being no no
 singularities. Clearly if that symbol h
 bar was zero
 there would be a singularity and if it
 was infinity there would be a
 singularity. So it needs to be finite.
 Now this theory is quantum theory
 of an isolated system.
 So in a sense isolation is all you need
 to get quantum theory. Uh it's a it's a
 good way to start because it gets you
 someplace that we understand
 and that we have empirical reasons to
 think is a good description of generic
 systems.
 So if everything is linear, the dynamics
 is linear, we can do a linear
 decomposition of the dynamics.
 So we can split this isolated system U
 into the components that we're
 interested in some system that we want
 to talk about uh A and and its
 environment now A bar which is just
 everything else in U that isn't A and
 that introduces the idea of a boundary
 between the two. Um and the linear
 decomposition is a decomposition of this
 operator h the Hamiltonian
 and we can decompose it into the sum of
 three terms. A term for the left side, a
 term for the right side and a term for
 the interaction.
 So a nice thing about quantum theory, it
 has this simple representation of
 interaction and it's linear.
 So this is a generic symmetry preserving
 representation of interaction between a
 system and and everything else.
 Now we can continue decomposing this
 interaction term again additively
 um as a sum
 of operators that act on single quantum
 bits. And some some of you have seen
 this kind of picture before in uh
 previous talks, but these operators are
 very simple. They're just operators that
 measure the spin of a quantum bit. And
 each one of them is equipped with a
 reference frame that says what direction
 counts as Z since this is a Z spin
 operator. And Z can be chosen to be
 anything. and it can be chosen uh in
 different ways for different cubits.
 But this turns this boundary B into a
 holographic screen. So there's a lot of
 physics that can now be imported to talk
 about this boundary. And I'm not going
 to really talk about that physics today.
 Uh we're just going to go on.
 Um this Hamiltonian now tells us how a
 and a bar act on each other. they act on
 each other by changing the values of
 cubits.
 Uh what we wanted though is to
 understand how a and a bar influence
 each other. How information flows from
 one to another.
 And these are the same uh if and only if
 one condition is met that a and a bar
 have conditionally independent states.
 So that we could talk about the state of
 A and the state of A bar.
 And that's a simple requirement. We have
 to require that the joint state the
 state of the two systems factors into
 the state of the one system and the
 state of the other system.
 And this is in quantum theory called
 separability. And it's the same as the
 absence of entanglement.
 So systems
 have their own states. if they're not
 entangled.
 And entanglement just means factoriz or
 non-entanglement just means
 factorizability.
 And of course that maps over into
 classical systems. If classical dynamic
 system has a markoff blanket then the
 then it factors
 and we know what separability requires.
 It requires sparse coupling.
 So formally this interaction has to have
 a small dimension and since it's an
 operator its dimension is well defined
 and by small it means its dimension is
 much less than the dimension of either
 of the other operators in the picture
 the internal operators for a and a bar.
 So intuitively the evolutions of these
 systems have to be almost independent
 for their states to be separable and for
 the interaction to actually capture all
 of the influence of one system on the
 other.
 So what we've done here is reconstruct
 the free energy principle from minimal
 physics. And again u at least some of
 you have seen this before.
 If uh these two systems A and its
 environment are separable, so they meet
 this dimensionality constraint on the
 Hamiltonians,
 then the Hamiltonian fully describes
 information exchange. The boundary
 functions as a markoff blanket. Uh the
 variational free energy is a measure of
 the interaction strength.
 uh minimizing VF VFE which is what the
 free energy principle is about is
 keeping the interaction weak while
 allowing thermodynamic exchange.
 So it's it's keeping the interaction
 between the system and its environment
 fairly weak while enviring allowing the
 system to eat some of its environment to
 to provide enough energy to live.
 And in this case predictability which is
 what the system is after is constrained
 interaction. So it's constraining the
 interaction in a way that's consistent
 in a way that's consistent
 with staying alive.
 So ANA bar maintain their identities as
 distinct systems only while their
 boundary um remains a markoff blanket.
 So there can't be any rips, there can't
 be any explosions, other sort of huge
 excesses in interaction strength.
 So that's actually what the FE is about.
 It's about keeping interactions small
 enough to maintain the integrity of a
 boundary and that's what's required for
 persistence over time as a well-defined
 system
 and this follows uh just from very
 minimal physical assumptions.
 So now if we go back to Jane's Jane's
 definition of intelligence as a fixed
 goal with various means of achieving it,
 we can ask um is there always a goal?
 Does any goal count? And the FEP always
 gives us one goal continuing to exist.
 uh any system that
 persists through time acts inferentially
 uh as if it's trying to continue to
 exist. This is what Jacob Howey calls
 self-evidencing
 as an interpretation of the FEP.
 Are any means allowed? Well, yeah.
 Whatever the internal dynamics are
 capable of is an allowed means for
 keeping the boundary intact, for keeping
 the interaction uh weak but still open.
 And does anything fall outside of this
 definition? No. It's completely generic.
 At least it's completely generic for
 systems that are physically realizable.
 um so systems that actually respect the
 sorts of symmetries that our physics
 requires.
 So from this point of view, intelligence
 looks like something that could be
 generic.
 The question then becomes not what's
 intelligent, but is its behavior
 interesting?
 Um maybe everything is intelligent but
 some things are completely boring. They
 don't behave in ways that are
 interesting.
 So our question about intelligence
 really turns into a question about
 interestingness. Are there limits on
 what kinds of systems can exhibit
 interesting behavior? And if there are,
 how do we find out? How do we find out
 what the limits are? And how do we find
 out what sorts of systems actually
 display interesting behavior?
 And if we're going to ask this question,
 we have to know what interesting means.
 So here's some ways of characterizing
 interesting behavior. Um
 so if it's interesting, we want it to be
 surprising.
 uh maybe it's unpredictable in practice
 or only predictable approximate
 approximately with some amount of coarse
 graining. So if we don't measure what
 the systems doing with too much
 precision
 um maybe we can predict what's going on
 but if we really look closely it's it's
 going to be uh less predictable.
 Maybe the system is unpredictable in
 um systems seem to be interesting if
 they can learn from experience, if their
 um behavior depends on their memory, if
 it depends on the context they're in.
 Uh and we can get really radical here.
 some systems may have distributions of
 outcome values that actually violate the
 Kaggoro vacs. So the outcome probability
 distributions are undefined. That's
 really interesting.
 But operationally,
 what this all comes down to is that
 if we measure a bunch of state
 transition probabilities for some finite
 amount of time, they don't converge to
 predictive adequacy.
 So induction uh doesn't work or doesn't
 work very well or only works if we
 severely coarse grain things.
 So the sort of mechanical expectations
 that we've inherited from 19th century
 science are violated
 and we know that life does this. Um,
 if you observe a baby for a while,
 you're not going to be able to predict
 what happens as an adult uh to that
 person or how that person behaves as an
 adult. In fact, if you observe an adult
 for a while, you can't perfectly predict
 what they're going to do next. So, life
 life doesn't satisfy these sorts of
 mechanical expectations.
 But what else violates them? Uh that's
 what we're going to be interested in
 finding out.
 And we have some hints.
 So here's a old hint from the 1950s from
 cybernetics.
 Edward Moore showed that finite input
 output experiments can't uniquely
 determine the the machine table which
 just means the in internal state
 transition probabilities
 of a generic classical black box.
 And um the canonical example of that is
 a a box with a clock. So something like
 a time bomb where the behavior can
 change abruptly after some amount of
 time. And if you don't observe it
 through that change of behavior, then
 you can't predict what's going to happen
 next.
 Here's another more recent hint from um
 Conway and Koken. The free will theorem.
 Here Conway is John Conway from the game
 of life.
 They published what they called the free
 will theorem
 which shows that special relativity and
 quantum theory together rule out local
 determinism
 and rules out local determinism in a
 very strong form. Uh the whole pass like
 cone of a system can't determine its
 behavior
 if special relativity and quantum theory
 are combined. So their slogan in their
 first paper in 2006 was if experimenters
 make choices electrons do too.
 So that's a pretty strong hint that even
 electrons can exhibit interesting
 behavior.
 And then a third hint uh is this paper
 from Frank Tipler in 2014
 where he showed that the simplest way to
 remove the singularities
 from classical physics
 uh actually reproduced the quantum
 potential postulated by David Bone in
 his formulative formulation of quantum
 theory. And basically what the quantum
 potential does is make the motion of any
 given particle dependantly
 on the motion of every other particle in
 the universe.
 And interestingly uh Nicholas Jesus
 pointed out that back in the day of
 Newton and lelass
 uh before the 19th century physics
 wasn't singular because it was basically
 about gravity
 and gravity wasn't local. Every particle
 did actually behave on depend on what
 every other particle in the universe was
 doing instantaneously.
 And it was Einstein who introduced
 strict locality by requiring information
 to only flow at the speed of light.
 So these hints sort of suggest that
 generic systems can display interesting
 behavior. So the question is how do we
 make this precise? How do we understand
 it? How do we use it? So that's what I'm
 going to try to talk about in the
 remaining time.
 So here's the setting again. Uh we have
 a system. It's interacting with another
 system which is its environment. There's
 a boundary between them that the
 interaction flows through. So
 observations and actions live there on
 the boundary. U actions of A on A bar
 and actions of A bar on A.
 And u A's measurements and action
 choices are computed by its internal
 dynamics. That's true for AA bar also.
 In FE language, that computation is done
 by A's generative model. Same applies to
 Abar.
 The system satisfies this dimensionality
 constraint. So we can also think of it
 as the dimensionality of the boundary is
 much smaller than the dimensionality of
 either system.
 But the thing to keep in mind is that
 inputs and outputs, so behavior is much
 less complex than the computations that
 generate inputs and outputs. So behavior
 is less complex than the computations
 that generate it generically.
 So this immediately tells us something
 important
 that recurrence of states on B.
 So seeing the same behavior does not
 imply recurrence of the internal
 dynamics of the system that's exhibiting
 the behavior.
 And in particular, the probability of a
 behavior given the internal dynamics is
 not the same as the probability of the
 internal dynamics given the behavior.
 So behavior generically depends on
 hidden states um which we can think of
 as memory or internal context or
 something like that.
 And formally we can always represent
 that as an internal geometric phase.
 what's also called a Barry phase after
 um can't remember his first name Barry
 who first characterized it almost in
 general for for adiabatic systems
 and as as Chris Fuches the uh physicist
 who in came up with the cubist
 interpretation of quantum theory put it
 all physical systems have interiority
 and here he's also very influenced by
 William James I think interiority is
 actually a Jamesian term.
 So what is this geometric phase?
 It's it's just an apparent phase change
 that results from transporting a vector
 or some collection of vectors around a
 path in an internal state space.
 So here's an example um the right hand
 side of the screen. If you have a sphere
 and you start with a planer a
 two-dimensional coordinate system up at
 the north pole and you transport that
 coordinate system smoothly down to the
 equator and then you transport it a bit
 to the east on the equator and then
 transport it back up north. It's going
 to look like you've introduced a 90deree
 phase change in the coordinate system.
 But nowhere in this process have you
 done any rotation.
 Uh you've just transported
 uh this system in a in a parallel way
 without changing the system at all
 around in the state space but the state
 space happened to be curved. So these
 are called holonomy operations.
 Uh and purists call them anholony
 operations.
 um but physicists just call them
 holottomy operations and they're not
 much talked about in biology
 but there is this paper from 2026 from
 Marcel Blatner
 that applies this sort of thinking to
 plenaria
 and he does it in a framework he calls
 tangential action spaces but if you go
 back and look what he means by that he's
 really just talking about uh holomy
 transfer formations.
 So this is this is work that's starting
 to be applied biologically
 but to see why it's important
 um it's important because non-trivial
 honomy is actually a provably sufficient
 resource for universal quantum
 computation.
 So if you want to build a universal
 quantum touring machine, you actually
 only need one ingredient and that's
 non-trivial holomy in the search space.
 And it works because it allows you to
 construct a map from an observable
 boundary state, an input and an internal
 state to some other observable boundary
 state, an output and some other internal
 , an output and some other internal
 state. for arbitrary input and output.
 . for arbitrary input and output.
 So you can you can implement any
 computation just with holonomy and this
 has been known for 25 years. Uh but it's
 only really been known in the quantum
 information community.
 So um
 computing by holom is actually stayed
 basically within the quantum computing
 community.
 So to see why this is important um or to
 see its larger implelications,
 I think it's important to think about
 what a physically implemented
 computation is.
 Uh physically implemented computation is
 just a mapping
 from the behavior of some physical
 system. The observable behavior. So the
 observable of the behavior of the
 boundary of that system
 to an abstract representation of a
 computable function.
 And
 saying that
 um some device implements a computation
 is just saying that these kinds of u
 diagrams commute
 where you're either mapping the behavior
 into the symbolic representation of the
 computation
 or you're mapping the symbolic
 representation of the computation into
 the behavior by either a projection or
 embedding.
 And obviously this only works for
 computable functions
 because if you can't compute the
 function f, you can't determine whether
 these diagrams commute. So this is a
 definition of implemented computation
 for computable functions.
 And this interpretation
 is a projection uh into the behavior. So
 that's important.
 Because embeddings are injective.
 Embeddings are one to many.
 Um you can embed lots of different
 computations in any given behavior. And
 that tells us something that poly
 computation is actually generic.
 Um and indeed managing thermodynamic
 flow. So keeping your system alive uh
 requires that the informative behavior
 that you're you're interested in
 describing computationally is just a
 proper sample of the total behavior.
 U and that's true even for your laptop.
 You know we're not looking at everything
 the computer is doing when we're looking
 at the screen particular. We're not
 looking at all the thermodynamic
 exchange that keeps the thing running.
 And the same is true for organisms.
 Uh we're not necessarily looking at the
 caloric content of what they're eating
 when we're trying to do their
 psychology.
 So um we can take this a little bit
 farther. Turns out that we can always
 think of computation as scattering in
 some sort of space of data structures.
 And some colleagues and I have a paper
 on this that's been in review for months
 now that goes fairly far and in fact
 shows that the formal representation of
 scattering can always be applied to
 computation. But we can just use this in
 a in a fairly intuitive way. If we think
 of an ideal classical computer
 implementing an algorithm for f then
 we're thinking about something like
 this. We have an input for f and we have
 some ready state of this classical
 computer. And if we combine the input
 and the ready state, something happens
 that is some sort of implementation of f
 on the input.
 And at the end of that something
 happening, we get an output of f on the
 input and the computer returns to its
 ready state. So we can do it again.
 Now this is a useful way to think in a
 coarse grained way but the observed
 ready state which is a projection of B
 of the boundary of the device that we're
 talking about does not actually pick out
 a unique machine state.
 Um,
 that's what we just saw with this
 business about holomy, but it's also
 just Morris theorem from back in the 50s
 updated.
 Uh, we don't actually know what's going
 on inside the machine. It's a black box.
 Uh, it's got a boundary and we have to
 look at what's going on on its boundary
 and that doesn't determine what's going
 on inside.
 And we can see this in practice. Um even
 a classical operating system
 uh accumulates internal state changes as
 we execute it over and over and over
 again with different inputs. Here the
 input is some program together with some
 data
 and
 uh we have the machine implement some
 program with some data and it produces
 an output. Then we give it another
 program and some more data and it
 produces another output. We're not
 keeping track of everything that's going
 on inside. We just depend on the
 operating system to to stay in some
 usable state
 uh without saying exactly what that
 state is. And after a day's use of your
 computer, the final state may be very
 different from the initial state. And
 eventually the computer has to be
 rebooted to get it back to something
 like the initial state that you started
 with today.
 So we can think of that in in this
 language of side projects. Uh the
 operating system is constantly doing
 side projects. It's it's rearranging its
 internal memory. It's moving things
 around in its long-term what used to be
 a disk memory.
 um it's um changing its internal state
 in various other ways. So this is
 inevitable in generic systems uh because
 generic systems
 u exhibit non-trivial honomy because
 they have big state spaces and more
 complicated dynamics than what they
 display on their boundaries.
 So we can also represent these sort of
 phase changes geometric phase changes as
 reference frame changes.
 So for example u go back to that picture
 of a generic interaction that shows
 operations on cubits by spin operators
 each of which has a local reference
 frame.
 If the system undergoes some honomy
 operation
 uh that introduces a phase into a vector
 that's transported to some new part of
 the state space
 and we think of what that does to these
 reference frame vectors
 um
 that tell the spin operators what
 they're doing. Then a general holomy
 transform can modify a reference frame
 and in that case the output of acting
 with that operator is different. So if
 the reference frame is up and you act on
 something with spin up stays spin up.
 But if I've tilted the reference frame
 then what I do is rotate
 um that cubit. So it when that cubit
 gets measured by another system you're
 going to get a different answer
 and in fact this representation
 of holom as reference frame change turns
 out to be generic
 and um I'm involved in another ongoing
 paper that's trying to sort out um how
 this works in general
 and Again, it involves many different
 literatures that don't communicate with
 each other very much to put together a a
 real understanding of how holomy change
 relates to reference frame change
 and it all turns out to couple to the
 theory of error correcting codes and
 hence the theory of emergent spacetime.
 So this this issue of geometric phase
 and reference frame change is is
 actually very deep. But all we need to
 know for for now is that if the system a
 bar exhibits non-trivial honomy then how
 it acts on on the boundary so that the
 output that it produces its behavior
 it's going to change um as honomy
 operations move vectors around in its
 state space.
 So this is important um because when you
 move reference frames around
 you end up with uh behaviors that don't
 commute.
 So in particular uh this simple case of
 spin operators
 uh spin operators with different
 reference frames don't commute. Acting
 with up and then sideways is not the
 same as acting with sideways and then
 up.
 And this has a an overall global
 consequence. Whenever you have
 non-commuting reference frames,
 uh they generate non-causal context
 dependence in the behavior of the
 system. Uh and if you're you're making
 measurements with non-commuting
 reference frames, they generate
 non-causal context dependence in your
 measurement outcomes.
 And in either case, the consequence of
 that is that joint probability
 distributions on the outcomes are
 actually undefined. They violate the
 Kagoro axioms. And this is this is what
 is observed in systems that violate for
 example Bell's inequality that exhibit
 entanglement
 uh systems that violate the leot
 inequalities and so um exhibit what
 looks like entanglement in time.
 uh systems that violate the Coke and
 Specker
 uh theorem
 uh which is essentially a restatement of
 the conditions for entanglement.
 Uh so exhibit context dependence
 depending on different orders of of
 observational
 actions yield results that can't be
 given an overall joint probability
 distribution. So there's tons now of
 experimental data that it demonstrates
 this sort of
 kagorov axiom violation by physical
 systems
 and I've put down here a reference to a
 paper that talks about uh this
 non-communing QRFs generating non-causal
 context dependence in general.
 So what does this mean? It means that
 all of these interesting kinds of
 behaviors are in fact generic.
 And they're all generic for the same
 reason. That they're all the result of
 requiring that the behaving system is
 separate from the system that's exhibit
 that's observing its behavior.
 Uh and that separability requirement as
 we saw
 uh induces this dimensionality
 requirement that we have big systems
 with small boundaries and so behavior
 has much less dimensionality than the
 computation that generates it. Whenever
 that condition holds
 uh then we see these sorts of
 interesting behaviors
 and that condition holds whenever we can
 separate ourselves as observers from
 whatever system it is we're observing.
 So this tells us in fact
 that intelligence is very diverse and in
 fact interesting behavior is very
 diverse. Um it also provides us with a
 very general and strongly empirically
 validated foundation for talking about
 diverse intelligence i.e. quantum theory
 and it tells us that intelligence and
 persistent observability so separability
 between the observer and the observed
 uh go hand in hand.
 uh if if a system is different from us,
 we're observing it, then its behavior is
 going to be interesting.
 And we can make that interestingness go
 away by coarse graining or averaging or
 the common practice of throwing out
 anything that looks anomalous,
 but it's actually there. And if we look
 carefully enough, we'll see it.
 So that's it. Thank you.
@@ -0,0 +1,16 @@
 # yt-dlp log
 # url: https://youtu.be/QeMajYvhEbI
 # output: conductor/tracks/video_analysis_generic_systems_fields_20260621/artifacts/video.mp4
 # returncode: 0
 stdout:
 [youtube] Extracting URL: https://youtu.be/QeMajYvhEbI
 [youtube] QeMajYvhEbI: Downloading webpage
 [youtube] QeMajYvhEbI: Downloading android vr player API JSON
 [info] QeMajYvhEbI: Downloading 1 format(s): 400+251
 [download] video.mp4.f400.mp4
 [download] video.mp4.f251.webm
 [Merger] Merging formats into video.mp4
 stderr:
 WARNING: yt-dlp EJS not enabled; some formats may be missing.