Softky 2014 Elastic Nanocomputation Brain
<p>Oct 28, 2015 09:49 PM
paper copy goes into top file drawer with Neurology/ consciousness folders... in front.</p>
<p>
Oct 19, 2014 02:11 PM</p>
<p>
open access on Arxiv. 2014 by Bill Softky; (also in my neuro pdf files)</p>
<p>
abstract: his theory is based on information theory, whose parent is thermodynamics.</p>
<p>
a brain is a specialized real-valued RT 3d proc detecting discontinuities in spacetime...
and reconstitutes a continu reality based on them.</p>
<p>
continuous, but how to reconcile with separate neurons;</p>
<p>
vast discrep in efficiency (btwn neurons and ? the 3d model);</p>
<p>
requires strong priors embedded in 3d edge detec algos...</p>
<p>
priors which open an unpatchable security hole to automated persuasion.</p>
<p>
3d model explains which persuasive tech alters percep of reality.
and protects against cog manip.</p>
<p>
Brain must repr Newton's laws.</p>
<p>
ideal brain requires a skull, constant temp; neurons and spikes...
interwoven with nanoscopic mesh.</p>
<p>
ideal brain is a multiscale Netonian sim and controller.</p>
<p>
ideal brain much better at repr capacity, energy effic, and speed than
circuits only composed of neurons.</p>
<p>
=> a latent reservoir of "dark computation"</p>
<p>
brains are spec. RT signal proc... not genl purpose thinking machines.</p>
<p>
end of abstract *</p>
<p>
homonids: walking... <b>an inverted compound pendulum</b>... welll enough to run!</p>
<p>
bottom line: Simulatrix + neurons is > 108 more effic at repr of 3d dynamic world
than neurons alone;</p>
<p>
page 3: KEY PARA!!!</p>
<p>
<b>Dynamic balance (espec of an unstable struc) requires simulating</b>
<b> a set of continuously evolving differential equations, </b>
<b>a task whose tight constraints and specificity cry out for specialized analog hardware. circuit stability must depend on</b>
<b>copious high-res data abt the current phys world at varying timescles and</b>
<b>contin self-correcting closed-loop control equations at arbitr levels of resolution.</b>
<b>This analog computation is ill-suited for digital hardware but the reverse is not true!</b>
<b>analog substrates can and do support quantized computation</b>.....</p>
<p>
biped's tighter spatiotemp precision in continuous domain enabled homo sapiens to grow a novel
quantized form of storage atop the contin substrate: persistent, self-reinforcing states
and narrative struc, such as osc loops, could underline our unique mental abilities.</p>
<p>
page 3: We a Hybrid system made of two interleaved OSs, contin RT and quantized... each running at many
timescales simult.</p>
<p>
our key topic: how to rep the 3d world at the local scale.
gudiing principle is "resonant computing." most effic sim of a phys system
should MATCH the target system at the finest level.</p>
<p>
key problem control a truly cont body-control problem using discreteneurons...
that is millions fold LESS efficient han phys possible.</p>
<p>
page 4: hypothesize SIMULATRIX nanoscopic matrix: interpenetrating and
interacting with neocortical neurons to help them sim the outside world.
108 advantage over neural net alone!!!</p>
<p>
page 5 SIMULATRIX: </p>
<p>
an ideal brain must compute at most granular res with the least poss energy:
its contin comp substrate: SIM = simulatrix; must be locally isomorphic to outside world.</p>
<p>
simulatrix reps outside 3d world by itself BEING a 3d medium. (elastic 3d mesh like an aerogel.
support longit elastic waves:</p>
<p>
SIM stores and repr scalr 3d probab info: positions of object contours, orientation of a skel
joint and 3d abstractions of arbit sens input.
These are physical surfaces in a 3d gel: their own momentum could repr outside momentum!!!!</p>
<p>
In SIM, P-waves are the native repr of changes across space and time.</p>
<p>
INPUT to SIM is tiny, point-like synap events at sub ms and sub micron scale of
syanp ves release. iniate small spherical P-wave
who correlation and coherence learn, repr and reconst the struc of th world outside...
by overlapping spherical ripples (as from raindrops in a 3d pond)</p>
<p>
can have accumulating interference fringes 3d priors as in holograms.</p>
<p>
page 6: how to spend its limited energy budget?</p>
<p>
nearby neurons in parallel, triggered by same P wavve could teleport the wavefront
elsewhere via coherent volley of spikes.</p>
<p>
Elastic computer need vibration protection (eg a skul));</p>
<p>
wavefront teleportation requires consistent phase, axonal cond veloctiy and brain temp
tightly regulated.;</p>
<p>
page 7 figure 1:</p>
<p>
page 8: Ideal Brain specialized for 3d computation;</p>
<p>
tempting to think of brain as genl algo: canonical micro-cricuit or
Hopfield net.</p>
<p>
</p>
<p>
page 24: Interpersonal communication:
inspect microexpressions, postural tics etc.
our bodies act as 3d display devices in the megabyte/sec range...
driven by 1000s of parallel channels.</p>
<p>
info vibr is abt 106 bits/sec
high speed, low latency bidirec channel...</p>
<p>
Info in english < 25 bits per second;</p>
<p>
page 27 Ideal Computations Must be CLOAKED..</p>
<p>
<b>Optimal Computation Resists Observation</b> .....</p>
<p>
great observation!!! computing elements which use little energy and waste none are
effectively invisible!!!! measuring what could be measured...
most curcially detectable SPIKES rather than (say) undetectable P-waves.</p>
<p>
</p>
<p>
page 26 Figure 10: David statues: bandwidth advantages for vibratory communication:
<b>body divided by words is about 105</b> !</p>
<p>
page 28 <b>QUANTIZATION</b> ... essential as in DNA; action potentials.</p>
<p>
self-reinforcing loops and kinks; </p>
<p>
<b>Adhesion</b> is Bill's term for an<b> atomic self-reinforcing + feedback loop (like an attractor state)</b>...</p>
<p>
Quantized adhesions must be the atoms of recorded,episodic memory.</p>
<p>
<b>Quantizations is a DEVIL's BARGAIN</b>!!!
some info is destroyed to save the remainder...
quantiz artifacts, threshold bias, truncation error.</p>
<p>
<b>The principle is simple: adhesion: the equilibrium of a + fb loof is held in place by LOCAL pleasure.</b>.. one must push away against
the pleasure's local gradient deliberately toward discomfort.</p>
<p>
page 30: figure 13: Reality is ground truth zero; but that truth is
accessible only the sensory spike trains and their correlation with prev muscle firings.</p>
<p>
first level of indirection: with P-waves</p>
<p>
simulatrix phonons and wave propag medium...</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
page 32 <b>Path to Physical Happiness</b>:
<b>a skeleton covered in string.</b>..
determine shape from cable vibrations.</p>
<p>
a FORCE GEODESIC a movable stripe of indiv cables;
upright stance... learning to walk effciently take stime...</p>
<p>
Neuromechanical enlightenment: a consistent body map with no more blocks in space or time;
no artif divisions btwn cognit, emotional and physical domains... an undiff circuit.</p>
<p>
ADHESIONS: local motor maps that are unfused.</p>
<p>
</p>
<p>
</p>
<p>
idealized body: muscles are millions of thin fibers:</p>
<p>
page 35: Idealized Sociability:
physio trust is immediate:"I know I am holding a glass"
it is deep and requires vastly more data.</p>
<p>
trust is informational: it ought to scale with bandwidth (amount of info);
interactivity (low latency,, faster round trip confirmation of your output signal.
Intimacy is the physical closeness to the core of the spine and viscera.</p>
<p>
Incr trust => bringing people as close as possible visually, physicall and audit with as little
noise and distraction as possible.</p>
<p>
<b>Conclusion: page 35</b>: brain's canonical problem is continuous time motor control that can be
solved by a contin time microcircuit btwn 108 and 1064 more efficient than neural circuits.</p>
<p>
It posits a a subneural substance (simulatrix) which reps a moving world with
moving waves. It is undiscovered cuz it computes almost invisibly. This is a secondary hypothesis.</p>
<p>
<b>Primary hypothesis is continuous control</b>.... </p>
<p>
Threats to human happiness:
1) sensory deprivation of NATURAL SIGNALS.
2) Muscular mis-coordiatnion and
3) social alienation: lack of interpersonal trust.</p>
<p>
page 36 and 37: humans possess untapped reservoirs of vibratory sense....
every categoroical "digital" computtion like episodic memory, logic must
be executred by smoothly chaning 3d circuits specialized for muscles.</p>
<p>
<b>any circuit optimized for body control must be sub-optimal at unrelated tasks</b>....</p>
<p>
we are actually approx unconsc 3d circuits we really are.</p>
<p>
page 36: TRUST ... a lived experience like balance, not (as with digital computers) a separate authentication channel. proximity communication outstrips verbal channels 1000X fold.</p>
<p>
Electronic mediation: our richest primal socio-sensory channels have gone dark.
We need intense proximity-practice!!!</p>
<p>
</p>
<p>
</p>
<p>
Notes at end:</p>
<p>
note 30: Simulatrix non -linear sharpening amp... to infer exact hard edges.
the dynamical eqns of sim. are difft from outside world...
it is simplified in crucial entropy-reducing ways... a Platonic Ideal.</p>
<p>
relentless quantiz. of its most atomic rep.
trans-generational waves affect xmiss prob of intermixed cultural memes,</p>
<p>
Imagine a little girl imprisoned in a cell... she loves to move wonderfully and gracefully...
she yells in vain... the cage's bars are words.</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>