Open in TheBrain

Chap 5 Theorizing Consc

<p>Feb 22, 2014 02:31 PM</p>

<p>

first loc in Chapter 5 Theorizing consc is &nbsp;&nbsp;<b>2707</b>...</p>

<p>

we need a theory to explain how subjective introspec relates to objective measurements.</p>

<p>

he introduces Global neuronal workspace.</p>

<p>

propsaol &nbsp;consc is brain-wide info sharing.</p>

<p>

efficient long-distance networks espec in PFC &nbsp;to select relevant info and share it.</p>

<p>

consc... evolved to attend a a piece of info and keep it active within broadcast system.

then can be flexibly routed according to current goals.</p>

<p>

we can name it, evaluate it, memorize it, or plan with it.</p>

<p>

computer sims of neural networks: &nbsp;it gens precisely the sigs that we see.</p>

<p>

*</p>

<p>

why should late neuronal firing, cortical ignition and brain-scale synchrony create a subjective state of mind?

how do they elicit experience. &nbsp;Why should firing in V4 elicit color or v5 motion?</p>

<p>

required: an overarching theory:</p>

<p>

these enigmas are not so difft from &nbsp;atomic theory: &nbsp;whence the solidity of tables made of void + atoms.

what is a liquid?</p>

<p>

needed a math model &nbsp;&nbsp;2732.

kinetic theory of gases: &nbsp;Maxwell and Boltzmann, &nbsp;pressure and temp.</p>

<p>

neurosci needs a series of bridging laws.</p>

<p>

an arrangement of Russian dolls:</p>

<p>

even a single neuron with its tens of 1000s of synapses is a unive of trafficking molecles.</p>

<p>

his colleages JP Changeux,and Lionel Naccache. &nbsp;their theory global neuronal workspace..</p>

<p>

CONSC is Global Info Sharing:</p>

<p>

info has ented a specific storage area tha make it available to the rest of the brain.</p>

<p>

millions of reps cross our brain &nbsp;unconsc, &nbsp;but ONE is selected cuz of its relevance to out present goals.

CONSC makes it golbal available to high level decision systems.</p>

<p>

Bernard Baars calls it &nbsp;global workspace:</p>

<p>

Fig 24 &nbsp;BAARS &nbsp;GWS: &nbsp;attention, Value, evaluation, &nbsp;motor,perceptual systems, &nbsp;LT mem,</p>

<p>

brain brought it into the workspace; &nbsp;can name it with language; &nbsp;use it for future plants.</p>

<p>

1870: &nbsp;loc 2763: &nbsp;Hippolyte Taine: &nbsp;theater of consc;</p>

<p>

long quote: &nbsp;single man in the footlights with a chorus in the bkgrnd.</p>

<p>

our mind: &nbsp;enormous varieyt of unconsc proc. &nbsp;&nbsp;a massive support staff for a one-man show.

a backstage ballet that remains hidden from sight.</p>

<p>

Dennett: &nbsp;ond discharges fancy homunculi by organizing armies of idiots to do the work;</p>

<p>

collective intel arises from broad exchanges of messages.

it dates back &nbsp;2777 to &nbsp;AI researchers subprograms that shared a blackboard.</p>

<p>

Taine: &nbsp;limited capacity system.</p>

<p>

2791 &nbsp;Don Broadbent: &nbsp;airplane pilots &nbsp;... can only atend one train of speech, not 2;

limited cpacity channel.</p>

<p>

discovery of attn blink in chap 2... &nbsp;&nbsp;a central bottleneck &nbsp;a VIP lounge that only

a happy few are admitted.</p>

<p>

1960s and 1970s &nbsp;consc as high level supervision system; &nbsp;central executive.</p>

<p>

CONSC &nbsp;as Wm James notes: &nbsp;an organ added to steer the CNS.</p>

<p>

PFC &nbsp;takes the lead over low level systems;</p>

<p>

Michael Posner &nbsp;Tim Shallice: &nbsp;&nbsp;proposal: &nbsp;info is consc if repr in high level system....

not quite right &nbsp;&nbsp;<b>2808</b>: &nbsp;Exec attn is just one of many systems that get consc info.</p>

<p>

Language, &nbsp;LT mem, attn, intention all part of the inner circle of intercommunic devices.</p>

<p>

like BAARS.. I believe consc reduces to what the workspace DOES:</p>

<p>

<b>nothing prevents thre reproduction of these functions to silicon-based hardware.</b>..</p>

<p>

2821: &nbsp;computer hold block of text but cannot decide relevance to goals.</p>

<p>

Our cortex embraces a simultaqneous, modular set of processors and flexible routing.</p>

<p>

CONSC may have evolved to mitigate this modularity.</p>

<p>

global avaialability is what ex perience as consc.</p>

<p>

decisions must be based onpooling multiple sources of know.</p>

<p>

Thirsty elepant 2821: &nbsp;alone in savannah. &nbsp;how to find a waterhold.</p>

<p>

its decision to walk toward a distant and invisible location must be based on most efficient use of info

including a mental map of space, &nbsp;visual recog of landmarks, trees, apths,

and recall of past successes and failures.</p>

<p>

Consc evolved aeons ago to flexibly tap all source sof know.</p>

<p>

2835 &nbsp;Evolved Communication Networks;</p>

<p>

consc implives connectivity; &nbsp;&nbsp;flex info sharing &nbsp;to link distant specialized regions in a coherent role.</p>

<p>

axons... up to meters in size! &nbsp;unique among cells;</p>

<p>

neurons with long distance axons most abundant in PFC.

connects to infer parietal lob, &nbsp;mid and anterior temproal lob and &nbsp;ant and post cingulate.</p>

<p>

major hubs of reciprocal connex.</p>

<p>

2861 &nbsp;at the top of cortical hierarchy &nbsp;an elitist board of executives distrib in distant territories

styas in syc by exchanging msgs.</p>

<p>

primar ivlovles &nbsp;PFC and pareital lobs &nbsp;coincides with what I described in chap 4:

our first signat of consc.</p>

<p>

we can now undersatnd ignition.</p>

<p>

2873: &nbsp;pyrams in cortex... giant cell bodies.</p>

<p>

large nerve <b>cells concentrate in layers II and III... respons for CALLOSAL</b> connex;</p>

<p>

fig 26: &nbsp;large pyrams: &nbsp;10,000 spines in PFC pyrams.</p>

<p>

layers 2 and 3: &nbsp;loc 2886 &nbsp;much thicker in PFC than in sensory areas.</p>

<p>

the thickness of layers 2 and 3 rougly delineates max active regions during consc perception.

muc larger dendr trees... &nbsp;more prom in huamns than other primates.</p>

<p>

1920s &nbsp;von Economo: &nbsp;much thicker in PFC and cingulate and assoc areas in parietal &nbsp;and temp.</p>

<p>

Guy Elston Australia and Javier DeFelipe in Spain: these giant workspace neurons

have immense dendrites... &nbsp;immense receiving antenas.</p>

<p>

Dendr SPINES...</p>

<p>

Elston and DeFelipe &nbsp;dendrites ar emch larger and spines more numerous in PFC...</p>

<p>

2902 &nbsp;dnese jung of dendrites: &nbsp;controlled by &nbsp;gmaily of genes unique mutations in humans:

FoxP2 &nbsp;modualtes language;</p>

<p>

FoxP2 family several gens resp for building neurons, dendrites axons synapses.</p>

<p>

mutant mice with 2 human FoxP2 mutations.. they grew pyrams with much larger

human like dendrites and facility to learn.</p>

<p>

FoxP2 => &nbsp;human PFC neuron may host &nbsp;15,000 spines.

perfect for collecting nfo anywhere in the brain. &nbsp;and bdcast it back to thousand sof sites.</p>

<p>

trace connex as we recog a face: &nbsp;v short connex in retina to clean up image;</p>

<p>

V1 etc to FFA &nbsp;2902 to face clusters:</p>

<p>

Stephanie Clark, Swiss: &nbsp;then to any corner of the brain !!!!

from R infer temp lobe, massive direct connex in a single step to distant assoc cortex and oppos hemisphere.</p>

<p>

concetnrate in infer frontal &nbsp;(broacs) and in tempr assoc &nbsp;(Wernicks);</p>

<p>

and to entire inner circle of high level extecs... in a reverberating assembly.</p>

<p>

according to my theory, &nbsp;access to this dense network is all that is need for info to become consc.</p>

<p>

SCULTPING a consc thot... &nbsp;2918</p>

<p>

brutish anger to suubtle schadenfreude, &nbsp;geograph trivia, &nbsp;hx info, math know

endless list...</p>

<p>

&nbsp;

2931 &nbsp;how does this suff gain access to neuronal workspace.</p>

<p>

Tononi: &nbsp;&nbsp;differentiation... bird, plane or Superman...</p>

<p>

1949 Hebb: &nbsp;Organiz of Behavior:</p>

<p>

slow development of a cell assembly: &nbsp;&nbsp;capable of acting as a close system.</p>

<p>

many such assemblies can activate indpendelty in difft place;</p>

<p>

cominbatorial code can repr billions of states.</p>

<p>

eg fire extingusiher: &nbsp;handle, body, hose...</p>

<p>

1959 &nbsp;John Selfridge: &nbsp;pnademonium &nbsp;AI.</p>

<p>

specialized daemons.... &nbsp;&nbsp;specialize neurons tuned to Hollywood stars and US presidents.

Selfridge: &nbsp;deomons &nbsp;yelled in proportion to match.</p>

<p>

waves of shouting propagated thur hierarchy of incr abstract units</p>

<p>

eg 3 demons shouting for eyes nose hair...

gotten spectacular support &nbsp;loc 2945;</p>

<p>

ws orig strict feed fwd; &nbsp;but &nbsp;it is NOT striclty hierarchical...

demons can yel at lower ranking demon too.</p>

<p>

cortex: full of loops and bidriec projecitons;</p>

<p>

2959 &nbsp;Attractor states: &nbsp;groups of neurons form reproduc patterns of activity of long duraiton.

stable assemply.</p>

<p>

<b>recurrent netwrks converge to a consensus</b>. &nbsp;come to agreement...

<b>a unified interp of the scene.</b></p>

<p>

<b>neurons mutally support one another and suppress alternatives</b>.</p>

<p>

<< Dennett: &nbsp;fame in the brain!>>></p>

<p>

<b>can restore missing parts and remove noise</b>.</p>

<p>

Circk and Koch: &nbsp;a winning &nbsp;neural coalition.....</p>

<p>

2979: &nbsp;must be tightly itnegrated...

eg Mona &nbsp;Lisa: &nbsp;&nbsp;not a disemboweled Picasso with detached hands and Cheshire cat smile.</p>

<p>

but how do those distantly separ groups get attached?</p>

<p>

<b>Damasio: &nbsp;convergence zones</b>. &nbsp;in PFC, &nbsp;anterior temp lobe, infer parietal lobe and

precuneus.</p>

<p>

single stage &nbsp;at once differentiated and integrated.</p>

<p>

called Re-Entry by Edelman &nbsp;2974...</p>

<p>

best poss statist interp of scene. &nbsp;each group , an expt staitican.</p>

<p>

a &nbsp;shadow expert decides that it can acct for the dark zone of image only if light comes from the top;

lighting expert agrees... &nbsp;these exchanges cotninue til every bit is tentively interpreted.</p>

<p>

the SHAPE of an Idea &nbsp;2989:</p>

<p>

Pnademonium, &nbsp;competing coalitions.</p>

<p>

My own theory .. global neuronal workspace &nbsp;... draws heavily from them.

consc state is encoded by stabel activation for a few tnes of sec of a subset of active workspace neurons.</p>

<p>

During consc, &nbsp;workspace neurons alll exchange reciproc messaes im a massively parallel attempt

to a achieve a coherent and sync interpretaiton.</p>

<p>

cons cpercep is competle then they converge. &nbsp;kept in sync by long distance axons.</p>

<p>

Neural sync is a key ingredient.</p>

<p>

syc spikes with oscilations.</p>

<p>

a swarm of fireflies that harmonize.</p>

<p>

but PFC does not gain access to unconsc local groups...</p>

<p>

at moment of consc &nbsp;all relevant neurons activate in partial sync under the aegis of PFC.</p>

<p>

3016 &nbsp;Explanation of P3 wave &nbsp;large + voltage;

worksapce => &nbsp;lands on inhib neurons that are silcencers.

Please remain silent, your features are irrelevant.

small partches of active and sync neurons &nbsp;with a massive corwn of inhbite dneurons.</p>

<p>

current heading away from scale; &nbsp;But &nbsp;inhib neurons dominate the picture;

and form positive potential. &nbsp;many more inhibited than activated

these form the P3. &nbsp;what the current thot is NOT aobut.</p>

<p>

<b>Edward Vogel Univ of Oregon &nbsp;beautif demo</b> of negative voltages over parietal cotex.

that track WM for space patterns. &nbsp;volts incr when we add objects, collapse when we forget.

GET THIS ref!</p>

<p>

<b>3031 SIMULATING a CONSC IGNITION</b>...</p>

<p>

3031 &nbsp;sculpting a pattern of active and inactive neurons into our GWS.</p>

<p>

3044 &nbsp;with JP Changeux: &nbsp;computer sims of neurla nets. se fig 27;

they get IGNITION in simulation;</p>

<p>

we put in columns, respecitng development;</p>

<p>

small thalamus: &nbsp;ccoarse model of the thalamocortrical column.</p>

<p>

virtual neutons fire spontan &nbsp;and gen EEG.</p>

<p>

neurons either code for light or sound &nbsp;on or off. &nbsp;includes feedback &nbsp;3074;</p>

<p>

simple GWS.</p>

<p>

inject ucrrent into visual thalamic neurons</p>

<p>

it has global ignition. &nbsp;</p>

<p>

pulse of stim: &nbsp;slowly climbs up hierarchy; &nbsp;&nbsp;then entire set of columns began to ignite;</p>

<p>

and alternate percpet inhibited.</p>

<p>

most sim neurons showed &nbsp;late amplification.... &nbsp;in the sim &nbsp;ignition led to reverberation acorss nested loops.</p>

<p>

net effect: &nbsp;an incr in osc fluctuations with prom peak at 30 Hz... &nbsp;spikes become strongly coupled

and synced... &nbsp;the sim mimicks 4 empir sigs of consc.</p>

<p>

3087: &nbsp;consc access: &nbsp;=> &nbsp;a phase transition from one state to another.

jump from low spontan activyt to high sync spiking.</p>

<p>

150 yr old concept: &nbsp;consc has a threshold &nbsp;separaating subliminal from supraliminal.</p>

<p>

Each pyram 3114 &nbsp;received a vigilence signal... from

ACh, NE, and 5HT ascneding from bs, basal forebrain and hypothal.</p>

<p>

must incr vigilance parameter to get ignition.</p>

<p>

3114 &nbsp;spontan activity;</p>

<p>

3169: &nbsp;DARWIN in the Brain:

spontaneous activity &nbsp;is impt &nbsp;(not simply initiated from outsie)

Autonomy is primary property of CNS.</p>

<p>

relevan tpatterns are preserved in childhood, while inapprop ones are weeded out.</p>

<p>

central pattern generators &nbsp;3183...

<b>GWS also uses them to freely gen. new patterns, try them out and change them.</b>

Darwinian &nbsp;process ov variation.</p>

<p>

by spontan generating activty,the brain generates diversity

... &nbsp;sculpted by brain eval of future rewards..</p>

<p>

JP Changeux: &nbsp;sims: &nbsp;resolve &nbsp;Tower of London;

learning by slection

yields robust architecture...</p>

<p>

3197 &nbsp;Gener. of Diversity = GOD

sponatn activity due to excitability and random rel of vesicles of neuroxmtr.</p>

<p>

thermal noise. &nbsp;neurons amplify noise;</p>

<p>

MonteCarlo Markov chain and simulated annealing &nbsp;require noise.</p>

<p>

local noise ends up as global activity. &nbsp;3197;

ends up as covert thots and goals;</p>

<p>

stream of consc &nbsp;finds its ultimate origin in random spikes sculpted by random spikes &nbsp;3197 &nbsp;(ref??)</p>

<p>

<b>CATALOG of UNCONSC: &nbsp;3211: &nbsp;a Linnaeus taxonoy of the unconsc</b>...</p>

<p>

we are UNaware of most of what we do and know... &nbsp;a wild profusion of unconsc proc.</p>

<p>

GWS some order out of the jungle.</p>

<p>

<b>Inattn blindness</b>: &nbsp;visual stim way above threshold but we fail to notice it.</p>

<p>

I write this in my wife's birth home: &nbsp;a 17th cent. farmhouse...&nbsp;

with grandfather lcock. &nbsp;pendulum... &nbsp;vanishes .

3211</p>

<p>

fig 28: &nbsp;taxonomy of the unconsc: &nbsp;CONSC &nbsp;hi str and top down amplification</p>

<p>

<b>PRECONSCIOUS.</b>.. &nbsp;hi strength but no amplif &nbsp;&nbsp;(like the grandfather clock)

on the periphery, not attended to; &nbsp;will decay out unless amplified;</p>

<p>

mechanism: &nbsp;FRINGE of Inhibition surroundinging the active percept!

(that prevents the 2nd stim fro igniting. &nbsp;&nbsp;central competition; winner take all;</p>

<p>

some workspace neurons must be actively silenced to delimit the current consc content.</p>

<p>

prevent us from seeing two thing at once or from perfomring two effortful tasks at once.</p>

<p>

sublim stim: &nbsp;weak str.</p>

<p>

disconected proc; &nbsp;&nbsp;</p>

<p>

know encoded in connex or in micropatterns.</p>

<p>

Preconsc: &nbsp;consc in waiting: &nbsp;Freud invented term: &nbsp;capable of entering consc.</p>

<p>

3230: &nbsp;Sims of &nbsp;GWS: &nbsp;see his sims ref &nbsp;64.</p>

<p>

ignited stim in GWS creates fringe of inhibition and prevents 2nd stim from etnring.</p>

<p>

some workspace neurons may be actively silenced.

diffuse inhib creates a bottleneck.... prevents proc 2 things at once.

But early sensory areas can light up.</p>

<p>

Preconsc is temp buffered.&nbsp;</p>

<p>

<b>SUBLIMINAL STATE</b>: &nbsp;&nbsp;</p>

<p>

3245 &nbsp;see ref 66: &nbsp;My sims shows &nbsp;a brief pulse of acitv can fail to trigger global ignition cuz

the the itme the top down signal returns the orig activ is gone and replaced by the mask.</p>

<p>

Subliminal: &nbsp;incoming sensory wave dies out before creating a tsumanmi on the shores of GWS.</p>

<p>

3245 &nbsp;BREATHING. &nbsp;&nbsp;deep in brain stem &nbsp;ventilate.</p>

<p>

<b>tot. UNCONSC. &nbsp;why? &nbsp;&nbsp;It is DISCONNECTED</b>?

discon from GWS in PFC and parietal cortex.</p>

<p>

3271 &nbsp;DISCONNECTION syndrome: disconnex: &nbsp;eg &nbsp;corpus callosum syndrome:

pt hosts two impaired workspaces...</p>

<p>

4th way to be uncsonc: &nbsp;3273: &nbsp;diluted into complex firingpattern:

eg &nbsp;fine visula grating that flickers fast....

must be re-encoded in anter cortex.</p>

<p>

any face we see &nbsp;beings in this unconsc manner: &nbsp;absurdly contorted spatiotemp train of spike;

3287 &nbsp;infin amts of info abt speaker, msg, emotion... we cannot decode it.

Must be put into meanigful bins.</p>

<p>

Fascin consequence: &nbsp;ref 69. &nbsp;US military: &nbsp;flahsed ten per sec potos form satellite to trained observerL

monitors brain ptentials.

future: &nbsp;may grant us computer-assisted brain decoding. &nbsp;a kind of ESP.</p>

<p>

3287 &nbsp;5th categ of unconsc: &nbsp;&nbsp;latent connex...

quiescen storage in synap connex.</p>

<p>

Each synapse stores minisc bit of statist wisdom:

how likely is presynap neuron to fire just before postsynap one?</p>

<p>

years of paino practice induce change in gray matter density... due to changes in synap densi, sizes,

white mattre stuc and supporitng glial cells.

and in HC... where, when and with whom even t happened.</p>

<p>

3302: &nbsp;how to convert form dormant to active?</p>

<p>

during retr synap promote reencactment of precise pattern of firing.</p>

<p>

<b>a consc memory is just the approx reconstruc of a&nbsp;</b>

<b>preconscious pattern of activ that once existed...</b></p>

<p>

to be consc patterns must invade PFC and cingulate ;</p>

<p>

John believes he is clever; &nbsp;&nbsp;language networks wired to proc

words and phrases &nbsp;but per. &nbsp;NOT accessible to awareness.

know is is in wrong format.</p>

<p>

why are we consc aware of &nbsp;34 times 31 but not of grammar proc. &nbsp;why the diff?</p>

<p>

the knowlede is in the wrong format for consc access.</p>

<p>

<b>complex arith... is done step by step &nbsp;under direct control of PFC, cingulate, and parietal</b>...</p>

<p>

complex seq explciityly coded in PFC.</p>

<p>

Indiv cells encode intentions, plans, steps, and even errors and corrections.</p>

<p>

Grammar, by contrast: &nbsp;3316: &nbsp;L temp lobe and infer frontal gyrus.

it spares networks in DL PFC for consc effort.</p>

<p>

during anesthesia: &nbsp;speech continues to be produced.</p>

<p>

<<<< above is key and interesting distinction: &nbsp;between effortful plans that

are used in arithmetic &nbsp;13 * 31 eg &nbsp;which are under control of DL PFC &nbsp;VERSUS

grammar parsing which involves &nbsp;Wernickes and Brocas area but NOT DL PFC.

so we do NOT know the rules of grammar since we did not evoke them at recording time.>>></p>

<p>

3330: &nbsp;Subjective states of Matter....</p>

<p>

<b>info must be encoded as organiz pattern in higher cort regions</b>

<b> and pattern must IGNITE an inner circle of tightly interconne areas forming GWS</b>.</p>

<p>

I do NOT doubt that computer could capture details of consc state.</p>

<p>

my sim is rudimentary; &nbsp;more approp would have B of diff states and perform

statist infers.</p>

<p>

see refs 76, 77... <b>&nbsp;consc ignition, subset of paterns collapse into a unifed internal deciison</b>.</p>

<p>

Selfridge pandemonium struggle for coherence. &nbsp;constantly struggling for a single unified interpretation.</p>

<p>

confront piecemeal evidence and accumulate it,, &nbsp;until get coherent answer that satisfies orgs goals.</p>

<p>

Autonomy is its motto.</p>

<p>

3341 &nbsp;has the bottomline: &nbsp;as above: &nbsp;reread this entire page:</p>

<p>

3344 &nbsp;HBP &nbsp;= human brain project... want to sim human size cortical netowkrs.</p>

<p>

ref 78 &nbsp;dedicated neuromorphic chips.</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;

.</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;

.</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>