Neural Annealing: Toward a Neural Theory of Everything (crosspost)

The fol­low­ing is QRI’s unified the­ory of mu­sic, med­i­ta­tion, psychedelics, de­pres­sion, trauma, and emo­tional pro­cess­ing. Im­pli­ca­tions for how the brain im­ple­ments Bayesian up­dat­ing, and fu­ture di­rec­tions for neu­ro­science. Cross­posted from http://​​openthe­


Con­text: fol­low-up to The Neu­ro­science of Med­i­ta­tion and A Fu­ture For Neu­ro­science; a unifi­ca­tion of (1) the En­tropic Brain & REBUS (Carhart-Har­ris et al. 2014; 2018; 2019), (2) the Free En­ergy Prin­ci­ple (Fris­ton 2010), (3) Con­nec­tome-Spe­cific Har­monic Waves (Ata­soy et al. 2016; 2017), and (4) QRI’s Sym­me­try The­ory of Valence (John­son 2016; Gomez Emils­son 2017).

0. Introduction

Why is neu­ro­science so hard?

Part of the prob­lem is that the brain is com­pli­cated. But we’ve also mostly been do­ing it wrong, try­ing to ex­plain the brain us­ing meth­ods that couldn’t pos­si­bly gen­er­ate in­sight about the things we care about.

On QRI’s lineages page, we sug­gest there’s a dis­tinc­tion be­tween ‘old’ and ‘new’ neu­ro­science:

Tra­di­tion­ally, neu­ro­science has been con­cerned with cat­a­logu­ing the brain, e.g. col­lect­ing dis­crete ob­ser­va­tions about anatomy, ob­served cyclic pat­terns (EEG fre­quen­cies), and cell types and neu­ro­trans­mit­ters, and try­ing to match these facts with func­tional sto­ries. How­ever, it’s in­creas­ingly clear that these sorts of neat sto­ries about lo­cal­ized func­tion are ar­ti­facts of the tools we’re us­ing to look at the brain, not of the brain’s un­der­ly­ing com­pu­ta­tional struc­ture.
What’s the al­ter­na­tive? In­stead of cen­ter­ing our ex­plo­ra­tion on the sorts of raw data our tools are able to gather, we can ap­proach the brain as a self-or­ga­niz­ing sys­tem, some­thing which uses a few core prin­ci­ples to both build and reg­u­late it­self. As such, if we can re­verse-en­g­ineer these core prin­ci­ples and use what tools we have to val­i­date these bot­tom-up mod­els, we can both un­der­stand the in­ter­nal logic of the brain’s al­gorithms — the how and why the brain does what it does — as well as find more el­e­gant in­ter­ven­tion points for al­ter­ing it.

That’s a big check to try to cash. What might this look like?

I. An­neal­ing metaphors for the brain

In my post about the neu­ro­science of med­i­ta­tion, I talked about simu­lated an­neal­ing, a nat­u­ral im­pli­ca­tion of Robin Carhart-Har­ris’s work on en­tropic dis­in­te­gra­tion in the brain:

An­neal­ing in­volves heat­ing a metal above its re­crys­tal­liza­tion tem­per­a­ture, keep­ing it there for long enough for the microstruc­ture of the metal to reach equil­ibrium, then slowly cool­ing it down, let­ting new pat­terns crys­tal­lize. This re­leases the in­ter­nal stresses of the ma­te­rial, and is of­ten used to re­store duc­tility (plas­tic­ity and tough­ness) on met­als that have been ‘cold-worked’ and have be­come very hard and brit­tle— in a sense, an­neal­ing is a ‘re­set switch’ which al­lows met­als to go back to a more pris­tine, nat­u­ral state af­ter be­ing bent or stressed. I sus­pect this is a use­ful metaphor for brains, in that they can be­come hard and brit­tle over time with a build-up of in­ter­nal stresses, and these stresses can be re­leased by pe­ri­od­i­cally en­ter­ing high-en­ergy states where a more nat­u­ral neu­ral microstruc­ture can reemerge.

In his work on the en­tropic brain, Carhart-Har­ris stud­ies how psychedelics like LSD and psilo­cy­bin add enough en­ergy (neu­ral ac­tivity) to the brain that ex­ist­ing neu­ral pat­terns are dis­rupted, much like how heat­ing a metal dis­rupts its ex­ist­ing molec­u­lar bonds. Re­cently, Carhart-Har­ris and Fris­ton have unified their frame­works un­der the REBUS (RElaxed Beliefs Un­der pSychedelics) model, which also im­ports the an­neal­ing metaphor for brains:

The hy­poth­e­sized flat­ten­ing of the brain’s (vari­a­tional free) en­ergy land­scape un­der psychedelics can be seen as analo­gous to the phe­nomenon of simu­lated an­neal­ing in com­puter sci­ence—which it­self is analo­gous to an­neal­ing in met­al­lurgy, whereby a sys­tem is heated (i.e., in­stan­ti­ated by in­creased neu­ral ex­cita­bil­ity), such that it at­tains a state of height­ened plas­tic­ity, in which the dis­cov­ery of new en­ergy min­ima (rel­a­tively sta­ble places/​tra­jec­to­ries for the sys­tem to visit/​reside in for a pe­riod of time) is ac­cel­er­ated (Wang and Smith, 1998). Sub­se­quently, as the drug is me­tab­o­lized and the sys­tem cools, its dy­nam­ics be­gin to sta­bi­lize—and at­trac­tor bas­ins be­gin to steepen again (Carhart-Har­ris et al., 2017). This pro­cess may re­sult in the emer­gence of a new en­ergy land­scape with re­vised prop­er­ties.

It’s a pow­er­ful metaphor since it ties to­gether and re­con­tex­tu­al­izes so many core neu­ro­science con­cepts: free en­ergy land­scapes, Bayesian mod­el­ing, the ‘hand­shake’ be­tween bot­tom-up sense-data and top-down pri­ors. For a gen­eral overview of the math, see Wikipe­dia on simu­lated an­neal­ing, Metropo­lis-Hast­ings al­gorithm, Par­allel tem­per­ing; for more on Carhart-Har­ris’s and Fris­ton’s work, see Scott Alexan­der’s and Milan Griffes’ com­men­tary. There seems to be some con­ver­gence on this metaphor: as Scott Alexan­der noted,

F&CH aren’t the first peo­ple to dis­cuss this the­ory of psychedelics. It’s been in the air for a cou­ple of years now – and props to lo­cal blog­gers at the Qualia Re­search In­sti­tute and Mad.Science.Blog for get­ting good ex­pla­na­tions up be­fore the parts had even all come to­gether in jour­nal ar­ti­cles. I’m es­pe­cially in­ter­ested in QRI’s the­ory that med­i­ta­tion has the same kind of an­neal­ing effect, which I think would ex­plain a lot.

The ba­sics: how does an­neal­ing work?

Carhart-Har­ris’s and Fris­ton’s model does many very clever things and is a sub­stan­tial ad­di­tion to the liter­a­ture; I start from a similar frame but de­scribe the pro­cess slightly differ­ently. The fol­low­ing is QRI’s model (based on my talk on the Neu­ro­science of Med­i­ta­tion in Thailand):

  • First, en­ergy (neu­ral ex­ci­ta­tion, e.g. Free En­ergy from pre­dic­tion er­rors) builds up in the brain, ei­ther grad­u­ally or sud­denly, col­lect­ing dis­pro­por­tionately in the brain’s nat­u­ral eigen­modes;

  • This build-up of en­ergy (rate of neu­ral firing) crosses a metasta­bil­ity thresh­old and the brain en­ters a high-en­ergy state, caus­ing en­tropic dis­in­te­gra­tion (weak­en­ing pre­vi­ously ‘sticky’ at­trac­tors);

  • The brain’s neu­rons self-or­ga­nize into new multi-scale equil­ibria (at­trac­tors), aka im­plicit as­sump­tions about re­al­ity’s struc­ture and value weight­ings, which given pre­sent in­for­ma­tion should gen­er­ate lower lev­els of pre­dic­tion er­ror than pre­vi­ous mod­els (this is im­plic­itly both a resyn­chro­niza­tion of in­ter­nal pre­dic­tive mod­els with the en­vi­ron­ment, and a min­i­miza­tion of dis­so­nance in con­nec­tome-spe­cific har­monic waves);

  • The brain ‘cools’ (neu­ral ac­tivity lev­els slowly re­turn to nor­mal), and parts of the new self-or­ga­nized pat­terns re­main and be­come part of the brain’s nor­mal ac­tivity land­scape;

  • The cy­cle re­peats, as the brain’s mod­els be­come out­dated and pre­dic­tion er­rors start to build up again.

Any ‘emo­tion­ally in­tense’ ex­pe­rience that you need time to pro­cess most likely in­volves this en­tropic dis­in­te­gra­tion->search->an­neal­ing mechanism— this is what emo­tional pro­cess­ing is.

And I’d sug­gest that this is the core dy­namic of how the brain up­dates its struc­ture, the mechanism the brain uses to pay down its ‘tech­ni­cal debt’. In other words, en­ter­ing high-en­ergy states (i.e., in­tense emo­tional states which take some time to ‘pro­cess’) is how the brain re­leases struc­tural stress and adapts to new de­vel­op­ments. This pro­cess needs to hap­pen on a reg­u­lar ba­sis to sup­port healthy func­tion, and if it doesn’t, psy­cholog­i­cal health de­grades— In par­tic­u­lar, men­tal flex­i­bil­ity & emo­tional vibrancy go down — analo­gous to a drop in a metal’s ‘duc­tility’. Peo­ple seem to have a strong sub­con­scious drive to­ward en­ter­ing these states and if they haven’t ex­pe­rienced a high-en­ergy brain state in some time, they ac­tively seek one out, even some­times in de­struc­tive ways.

How­ever, the brain spends most of its time in low-en­ergy states, be­cause they’re safer: sys­tems in noisy en­vi­ron­ments need to limit their rate of up­dat­ing. There are of­ten spikes of en­ergy in the brain, but these don’t tend to snow­ball into full high-en­ergy states be­cause the brain has many ‘en­ergy sinks’ (in­hibitory top-down pre­dic­tive mod­els) which soak up ex­cess en­ergy be­fore en­tropic dis­in­te­gra­tion can oc­cur.

But the brain can en­ter high-en­ergy states if these en­ergy sinks are:

(1) De-ac­ti­vated, if cer­tain evolved trig­ger con­di­tions are pre­sent- e.g., death of a loved one, fal­ling in love, good sex, so­cial re­jec­tion, get­ting bit­ten by a weird an­i­mal, failing some im­por­tant pre­dic­tion. In these cases there seems to be some sort of adap­tive gat­ing mechanism that dis­ables the typ­i­cal en­ergy sinks in or­der to al­low en­tropic dis­in­te­gra­tion->search->an­neal­ing to hap­pen.

(2) Over­whelmed, if there’s an enor­mous mag­ni­tude of en­ergy com­ing in, faster than the en­ergy sinks can mop it up- e.g., watch­ing a hor­ror movie, di­rect brain stim­u­la­tion, first day of school, be­ing sleep de­prived, mil­i­tary boot camp, cult in­doc­tri­na­tions, your wed­ding day.

(3) Avoided, if se­man­ti­cally-neu­tral en­ergy is ap­plied to the sys­tem. Essen­tially, co­her­ent en­ergy which isn’t strongly linked to any cog­ni­tive, emo­tional, or sen­sory pro­cess will be par­tially illeg­ible to most ex­ist­ing en­ergy sinks, and so it can per­sist long enough to build up ba­si­cally ‘hack­ing’ the brain’s ac­tivity nor­mal­iza­tion sys­tem. (Hold that thought this is the most in­ter­est­ing one. We’ll re­turn to it later.)

This is the ‘view from 30,000 feet’ for how simu­lated an­neal­ing in the brain works. If you stopped read­ing here, you’d walk away with a rea­son­able toy model of QRI’s “Neu­ral An­neal­ing” frame­work.

But there’s a lot more to the model! The rest of this writeup is an iter­a­tive tour us­ing Neu­ral An­neal­ing to ex­plain med­i­ta­tion, trauma, love, de­pres­sion, psychedelics, and effec­tive ther­apy, with each sec­tion adding a vari­a­tion on the core theme.

In­ter­lude: FEP, CSHW, and EBH/​REBUS

QRI’s “Neu­ral An­neal­ing” frame­work is es­sen­tially a unifi­ca­tion of Karl Fris­ton’s Free En­ergy Prin­ci­ple (FEP), Se­len Ata­soy’s Con­nec­tome-Spe­cific Har­monic Waves (CSHW), Robin Carhart-Har­ris’s En­tropic Brain Hy­poth­e­sis (EBH), and QRI’s own Sym­me­try The­ory of Valence (STV). Re­cently, Fris­ton and Carhart-Har­ris have unified their re­spec­tive paradigms with the Re­laxed Beliefs Un­der pSychedelics (REBUS) model. I be­lieve com­bin­ing all three is ex­po­nen­tially more pow­er­ful, not only giv­ing the com­pu­ta­tional-level story of REBUS, but also giv­ing us a model for how the brain may be phys­i­cally im­ple­ment­ing REBUS, and Bayesian up­dat­ing in gen­eral, with a cor­re­spond­ingly richer set of pre­dic­tions.

First, here’s a quick re­cap: to para­phrase what I wrote el­se­where,

Karl Fris­ton’s Free En­ergy Prin­ci­ple (FEP) is the lead­ing the­ory of self-or­ga­niz­ing sys­tem dy­nam­ics, one which has (in var­i­ous guises) pretty much taken neu­ro­science by storm. It ar­gues that any self-or­ga­niz­ing sys­tem which effec­tively re­sists di­s­or­der must (as its core or­ga­niz­ing prin­ci­ple) min­i­mize its free en­ergy, that free en­ergy is equiv­a­lent to sur­prise (in a Bayesian sense), and that this sur­prise-min­i­miza­tion drives ba­si­cally all hu­man be­hav­ior. This min­i­miza­tion of sur­prise re­volves around Bayesian-type rea­son­ing: the brain is always get­ting bot­tom-up sense data flow­ing in, more than it can han­dle. So it re­lies on top-down pre­dic­tive mod­els that at­tempt to sort through all this data so we can fo­cus on the sur­pris­ing stuff, the stuff that can’t be effortlessly pre­dicted. The core of the FEP is the de­tails of how this ‘hand­shake’ be­tween bot­tom-up and top-down hap­pens, and what can in­fluence it. See Fris­ton’s pri­mary work; Scott Alexan­der’s at­tempt to dis­till it. Re­lated to (and some­times used syn­ony­mously with) Ac­tive In­fer­ence, the Bayesian Brain, and Pre­dic­tive Pro­cess­ing /​ Pre­dic­tive Cod­ing.

Robin Carhart-Har­ris’s En­tropic Brain Hy­poth­e­sis (EBH) is es­sen­tially an at­tempt to im­port key con­cepts such as en­tropy and self-or­ga­nized crit­i­cal­ity from statis­ti­cal physics into neu­ro­science, in or­der to ex­plain psychedelic phe­nom­ena. As I noted above, it sug­gests that cer­tain con­di­tions such as psychedelics can add enough en­ergy to brain net­works that they un­dergo ‘en­tropic dis­in­te­gra­tion’, and then self-or­ga­nize into new equil­ibria. See Carhart-Har­ris 2018.

Se­len Ata­soy’s Con­nec­tome-Spe­cific Har­monic Waves (CSHW) is a method for ap­ply­ing har­monic anal­y­sis to the brain: ba­si­cally, it uses var­i­ous forms of brain imag­ing to in­fer what the brain’s nat­u­ral res­o­nant fre­quen­cies (eigen­modes) are, and how much en­ergy each of these fre­quen­cies have. The core work­flow is three steps: first com­bine MRI and DTI to ap­prox­i­mate a brain’s con­nec­tome, then with an em­piri­cally-de­rived wave prop­a­ga­tion equa­tion calcu­late what the nat­u­ral har­mon­ics are of this con­nec­tome, then es­ti­mate which power dis­tri­bu­tion be­tween these har­mon­ics would most ac­cu­rately re­con­struct the ob­served fMRI ac­tivity. This frame­work offers sev­eral no­table things: (a) these con­nec­tome-spe­cific har­monic waves (CSHWs) are nat­u­ral Schel­ling points that the brain has prob­a­bly self-or­ga­nized around (and so are worth talk­ing about); (b) a plau­si­ble mid-level bridge con­nect­ing bot­tom-up neu­ral dy­nam­ics and high-level psy­cholog­i­cal phe­nom­ena, (c) some­thing we can ac­tu­ally mea­sure. CSHW is an em­piri­cal paradigm, which is very un­com­mon in the­o­ret­i­cal neu­ro­science. Here’s a tran­script of Ata­soy’s ex­pla­na­tion; I also wrote ex­ten­sively about CSHW in A Fu­ture for Neu­ro­science.

In short: each of these three paradigms is a de­scrip­tion of how the brain self-or­ga­nizes. Fris­ton’s work un­der­stands the self-or­ga­ni­za­tion from a com­pu­ta­tional lens; Carhart-Har­ris an en­er­getic lens; Ata­soy a phys­i­cal lens.

Fi­nally, I’d offer two fur­ther pieces of back­ground con­text:

QRI’s own Sym­me­try The­ory of Valence (STV), which hy­poth­e­sizes that given a math­e­mat­i­cal rep­re­sen­ta­tion of an ex­pe­rience, the sym­me­try of this rep­re­sen­ta­tion will en­code how pleas­ant the ex­pe­rience is (John­son 2016). We fur­ther hy­poth­e­size that con­so­nance be­tween a brain’s con­nec­tome-spe­cific har­monic waves (CSHWs) will be a rea­son­able proxy for this sym­me­try (Gomez Emils­son 2017).

Marr’s Three Levels: as ex­plained on our lineages page,

David Marr is most fa­mous for Marr’s Three Levels (along with To­maso Pog­gio), which de­scribe ”the three lev­els at which any ma­chine car­ry­ing out an in­for­ma­tion-pro­cess­ing task must be un­der­stood:”
>Com­pu­ta­tional the­ory: What is the goal of the com­pu­ta­tion, why is it ap­pro­pri­ate, and what is the logic of the strat­egy by which it can be car­ried out?
>Rep­re­sen­ta­tion and al­gorithm: How can this com­pu­ta­tional the­ory be im­ple­mented? In par­tic­u­lar, what is the rep­re­sen­ta­tion for the in­put and out­put, and what is the al­gorithm for the trans­for­ma­tion?
>Hard­ware im­ple­men­ta­tion: How can the rep­re­sen­ta­tion and al­gorithm be re­al­ized phys­i­cally? [Marr (1982), p. 25]
This frame­work sounds sim­ple, but is re­mark­ably im­por­tant since ar­guably most of the con­fu­sion in neu­ro­science (and phe­nomenol­ogy re­search) comes from start­ing a sen­tence on one Marr-Pog­gio level and finish­ing it on an­other, and this frame­work lets peo­ple de­bug that con­fu­sion.

Back to an­neal­ing

As noted, Carhart-Har­ris and Fris­ton have unified their paradigms un­der REBUS by un­der­stand­ing pre­dic­tion er­rors as the ‘en­ergy’ pa­ram­e­ter which drives dis­rup­tion (en­tropic dis­in­te­gra­tion) in the brain’s net­works. Over time, this drives an evolu­tion­ary search func­tion which at­tempts to min­i­mize these pre­dic­tion er­rors. I think this is a very beau­tiful de­scrip­tion of a very clever sys­tem, and one which al­lows us an op­por­tu­nity to cross-val­i­date each model, and jump be­tween lev­els of de­scrip­tion if we get ‘stuck’. But it’s still miss­ing a story about phys­i­cal im­ple­men­ta­tion. What is this ‘en­ergy’, phys­i­cally speak­ing?

II. How med­i­ta­tion works: se­man­ti­cally-neu­tral annealing

I be­lieve that al­most all tech­niques that in­ten­tion­ally ‘hack’ the brain’s an­neal­ing pro­cess share a com­mon mechanism: a build-up of se­man­ti­cally neu­tral en­ergy. “Se­man­ti­cally neu­tral en­ergy” refers to neu­ral ac­tivity which is not strongly as­so­ci­ated with any spe­cific cog­ni­tive or emo­tional pro­cess. As I note above, usu­ally en­ergy build-up is limited: once a per­tur­ba­tion of the sys­tem neatly falls into a pat­tern rec­og­nized by the brain’s pre­dic­tive hi­er­ar­chy, the neu­ral ac­tivity prop­a­gat­ing this pat­tern is dis­si­pated. But if a pat­tern never quite matches any­thing, or takes ad­vance of im­ple­men­ta­tion-level struc­ture to per­sist, and es­pe­cially if it’s get­ting con­tinu­ally re­in­forced by some ex­ter­nal or in­ter­nal dy­namic it can per­sist long enough to build up. I think med­i­ta­tion is a perfect ex­am­ple of a pro­cess which adds se­man­ti­cally-neu­tral en­ergy to the brain: effort­ful at­ten­tion on ex­ci­ta­tory bot­tom-up sense-data and at­ten­u­a­tion of in­hibitory top-down pre­dic­tive mod­els will nat­u­rally lead to a build-up of this ‘non-se­man­tic’ en­ergy in the brain. From The Neu­ro­science of Med­i­ta­tion:

Fur­ther­more, from what I gather from ex­pe­rienced med­i­ta­tors, suc­cess­fully en­ter­ing med­i­ta­tive flow may be one of the most re­li­able ways to reach these high-en­ergy brain states. I.e., it’s very com­mon for med­i­ta­tion to pro­duce feel­ings of high in­ten­sity, at least in peo­ple able to ac­tu­ally en­ter med­i­ta­tive flow. Med­i­ta­tion also pro­duces more ‘pure’ or ‘neu­tral’ high-en­ergy states, ones that are free of the in­ten­tional con­tent usu­ally as­so­ci­ated with in­tense ex­pe­riences which may dis­tort or limit the scope of the an­neal­ing pro­cess. So we can think of in­ter­me­di­ate-to-ad­vanced (‘suc­cess­ful flow-state’) med­i­ta­tion as a re­heat­ing pro­cess, whereby the brain en­ters a more plas­tic and neu­tral state, re­leases pent-up struc­tural stresses, and re­crys­tal­lizes into a more bal­anced, neu­tral con­figu­ra­tion as it cools. Iter­ated many times, this will drive an evolu­tion­ary pro­cess and will pro­duce a very differ­ent brain, one which is more unified & anti-frag­ile, less dis­torted to­ward in­ten­tion­al­ity, and in gen­eral struc­turally op­ti­mized against stress.
An open ques­tion is how or why med­i­ta­tion pro­duces high-en­ergy brain states. There isn’t any con­sen­sus on this, but with a nod to the pre­dic­tive cod­ing frame­work, I’d offer that bot­tom-up sense-data is gen­er­ally ex­ci­ta­tory, adding en­ergy to the sys­tem, whereas top-down pre­dic­tive Bayesian mod­els are gen­er­ally in­hibitory, func­tion­ing as ‘en­ergy sinks’. And so by ‘not­ing and know­ing’ our sen­sa­tions be­fore our top-down mod­els ac­ti­vate, in a sense we’re di­vert­ing the ‘en­ergy’ of our sen­sa­tions away from its usual coun­ter­bal­anc­ing force. If we do this long enough and skil­lfully enough, this en­ergy can build up and lead to ‘en­tropic dis­in­te­gra­tion’, es­sen­tially push­ing enough en­ergy into the sys­tem that ex­ist­ing at­trac­tors are dis­rupted and an­neal­ing can oc­cur.

A nat­u­ral ques­tion here is what *is* this ‘se­man­ti­cally neu­tral en­ergy’ ex­actly? an ab­stract an­swer here is “se­man­ti­cally neu­tral en­ergy” can be thought of as an in­crease in brain ac­tivity which is (1) illeg­ible to Marr’s se­man­tic/​com­pu­ta­tional level, but (2) co­her­ent with re­gard to Marr’s al­gorith­mic or im­ple­men­ta­tional lev­els (an­other term for this might be ‘se­man­ti­cally-illeg­ible en­ergy’). But my con­crete an­swer is that se­man­ti­cally neu­tral en­ergy is a build-up of en­ergy in the brain’s nat­u­ral res­o­nances — en­ergy ac­cu­mu­lat­ing in CSHWs. And so it’s this that builds up dur­ing med­i­ta­tion, and this that starts a se­man­ti­cally-neu­tral an­neal­ing pro­cess which has a unique effect pro­file.

I think se­man­ti­cally-neu­tral an­neal­ing is the best kind of an­neal­ing for psy­cholog­i­cal health, be­cause:

(1) By mostly avoid­ing en­ergy sinks, the same en­tropic dis­in­te­gra­tion->search->an­neal­ing pro­cess can hap­pen us­ing less to­tal en­ergy, which is less dis­rup­tive to the fine de­tails of the sys­tem;

(2) Since this en­ergy is se­man­ti­cally-neu­tral, it doesn’t de­pend on or trig­ger as many se­man­tic pro­cesses in the brain (which can have un­pre­dictable effects), and like­wise it doesn’t nec­es­sar­ily rely on anti-in­duc­tive ‘hacks’ to trick the pre­dic­tive pro­cess­ing sys­tem, and these fac­tors make it a more re­li­able and re­peat­able source of an­neal­ing;

(3) Very very im­por­tantly: similarly to how vibra­tory en­ergy ap­plied to a tun­ing fork quickly col­lapses to the nat­u­ral res­o­nant fre­quency of the tun­ing fork, I’m spec­u­lat­ing that co­her­ent, se­man­ti­cally-neu­tral en­ergy added to the brain will nat­u­rally cluster in the brain’s nat­u­ral con­nec­tome har­mon­ics, which will thus drive an an­neal­ing pro­cess which strength­ens a con­so­nant sub­set of the brain’s nat­u­ral har­monic res­o­nances in the long-term— es­sen­tially ‘re­tun­ing the brain’ to­ward more res­o­nant/​flow states. For more de­tails, see The Neu­ro­science of Med­i­ta­tion;

(4) Fi­nally, this pro­cess should feel re­ally re­ally good and in the long-term, re­tune the mind to be more pleas­ant to in­habit. QRI’s work on the Sym­me­try The­ory of Valence (STV) and our method of ap­ply­ing this to the brain (CDNS) sug­gests that har­mony in the brain is liter­ally syn­ony­mous with plea­sure, and so pro­cesses which ‘deepen the grooves’ of core har­monic res­o­nances will tend to boost the mind’s de­fault he­do­nic level (likely helping sig­nifi­cantly with neu­roti­cism and emo­tional re­silience).

I.e., Med­i­ta­tion is a re­mark­ably clever tech­nique which pig­gy­backs on sev­eral of the brain’s core prin­ci­ples of self-or­ga­ni­za­tion: first, effort­ful at­ten­tion on (ex­ci­ta­tory) sense-data and in­hibit­ing (in­hibitory) pre­dic­tive sto­ry­tel­ling nat­u­rally pushes the brain into a high-en­ergy state and makes it more malle­able; this ex­cess en­ergy dis­pro­por­tionately col­lects in nat­u­ral brain har­mon­ics, and as the brain ‘cools’ from its high-en­ergy state, these en­er­gized har­mon­ics be­come ‘deeper’, lead­ing to more psy­cholog­i­cal ro­bust­ness. Less neu­roti­cism and more flow. I think this is where a large por­tion of the benefits of ad­vanced med­i­ta­tion comes from.

Med­i­ta­tion isn’t the only method to in­duce build-up of se­man­ti­cally-neu­tral en­ergy; the “Big Three” are:

Med­i­ta­tion, which seems to work by both in­creas­ing ex­ci­ta­tory sense-data and de­creas­ing in­hibitory top-down pre­dic­tive mod­els (en­ergy sinks);

Psychedelics, which in­tu­itively may func­tion by dis­abling ex­ist­ing en­ergy sinks (or per­haps over­load­ing them by in­creas­ing baseline firing rates or in­creas­ing the branch­ing fac­tor of neu­ral ac­tivity).

Mu­sic, a sen­sory in­put which seems to ex­ist on the knife’s edge be­tween ex­hibit­ing highly or­dered pat­terns (some of which will hit nat­u­ral con­nec­tome har­mon­ics and so al­low ac­cu­mu­la­tion of en­ergy through res­o­nance) on one hand, and on the other hand not be­ing too pre­dictable (thus dodg­ing most in­hibitory top-down pre­dic­tive mod­els);

Hy­brid ap­proaches also ex­ist: e.g. ex­er­cise, dance, sex, tantric prac­tices, EMDR, and breath work are es­sen­tially com­bi­na­tions of the rhyth­mic por­tion of mu­sic and the sen­sory por­tion of med­i­ta­tion. The fact that psychedelics re­li­ably en­hance the po­tency of each and ev­ery one of these prac­tices is not a co­in­ci­dence, but due to shared mechanism.[1]

III. De­pres­sion as a di­s­or­der of an­neal­ing; bipo­lar de­pres­sion dou­bly so

To de­scribe de­pres­sion in one sen­tence: “De­pres­sion is a self-re­in­forc­ing per­tur­ba­tion from the nat­u­ral an­neal­ing cy­cle.” There are two re­lated as­pects to this: (1) an in­abil­ity to an­neal nor­mally, and (2) an­neal­ing ab­nor­mally (more speci­fi­cally, an­neal­ing new at­trac­tor bas­ins which are high in dis­so­nance, or an­neal­ing a patholog­i­cal change in en­ergy pa­ram­e­ter dy­nam­ics).

Most peo­ple have a sim­ple model of de­pres­sion as “be­ing sad all the time” but I think a two-fac­tor model look­ing at en­ergy pa­ram­e­ter and valence offers a lot of clar­ity and pre­dic­tive util­ity. Roughly speak­ing, this sug­gests parametriz­ing de­pres­sion into three core types:

I. De­pres­sion with no high en­ergy states, char­ac­ter­ized by a lack of an­neal­ing (emo­tional clar­ity and dy­namism) in gen­eral;

II. De­pres­sion with high-en­ergy nega­tive states, which over time an­neals minds to­ward suffer­ing and hope­less­ness;

III. Bipo­lar de­pres­sion with high-en­ergy pos­i­tive & nega­tive states, which over time an­neals minds to­ward the dra­matic.

Th­ese cat­e­gories aren’t ex­clu­sive or static; too much time in one will in­crease the prob­a­bil­ity one may also fall into the oth­ers.

Not an­neal­ing fre­quently enough may be the most im­por­tant ‘non-ob­vi­ous’ cause of de­pres­sion. Brains es­pe­cially younger ones, since they’re chang­ing so much re­ally do need to an­neal reg­u­larly to pay down their ‘tech­ni­cal debt’, and if they don’t, they grow brit­tle and neu­rotic. (Tech­ni­cal debt in the brain builds up as we twist our ex­ist­ing brain net­works to ac­com­mo­date new facts; this debt is ‘paid down’ when we en­ter high-en­ergy states and let new brain net­works which fit these con­straints self-or­ga­nize) The ‘an­neal­ing pres­sure’ also in­creases over time, and if a whole­some an­neal­ing op­por­tu­nity fails to pre­sent it­self, the brain will pro­gres­sively lower its stan­dards look­ing for any op­por­tu­nity for an­neal­ing. Espe­cially if done re­peat­edly, this can cause long-term dam­age to the brain’s at­trac­tor basin land­scape. (We see this in nega­tive cop­ing strate­gies such as cut­ting, drama-seek­ing, and so on if some­one is en­gag­ing in such, they’ve prob­a­bly an­nealed poorly, and also likely have few re­al­is­tic op­por­tu­ni­ties for healthy an­neal­ing.) Many forms of en­ter­tain­ment we think of as pal­li­a­tives in to­day’s so­ciety (e.g. movies, video games) may be weak-and-in­com­plete-but-still-nonzero drivers of an­neal­ing. Not as good as the real thing, but bet­ter than noth­ing if that’s your only op­tion.

At the high-en­ergy ex­treme, it seems likely and tragic that de­pres­sion com­pounds it­self by re­peat­edly caus­ing in­tense nega­tive emo­tion (high-en­ergy states) which an­neals the brain to­ward these pat­terns, and to­ward as­sign­ing salience on the set of prob­lems and types of thoughts (at­trac­tor land­scape) fac­ing a de­pressed per­son — many of which are their own cause and would weaken if ig­nored. Re­lat­edly, I sus­pect some CSHW- and mu­sic-the­ory-re­lated math could be found de­scribing how de­pres­sion an­neals what I would call a brain’s ‘con­nec­tome key sig­na­ture’ (CKS) to­ward a ‘minor key’, an in­ter­nal logic which feels tragic/​hope­less (has fewer har­mo­nious ar­range­ments and pro­gres­sions), which the brain then uses as build­ing blocks for its re­al­ity.

Bipo­lar de­pres­sion seems a lit­tle more strange; the ex­treme highs and lows may in ag­gre­gate pro­duce cra­zier an­neal­ing pat­terns than just one or the other — es­sen­tially there’s a ‘tug of war’ be­tween pat­terns an­nealed dur­ing each ex­treme, which pri­ori­tizes the sur­vival of the class of pat­terns that ex­ist dur­ing both ex­tremely pos­i­tive and ex­tremely nega­tive states. In prac­tice, over time this an­neals a mind’s sto­ries to­ward the dra­matic, and to­ward re­duc­ing the ac­ti­va­tion en­ergy needed to flip the brain be­tween ma­jor and minor keys (the psych liter­a­ture calls this ‘kindling’). Each of these ‘key sig­na­ture flips’ would it­self re­lease a great deal of pent-up en­ergy, fur­ther driv­ing the an­neal­ing pro­cess. As I note in A Fu­ture for Neu­ro­science:

This is not to say our key sig­na­tures are com­pletely static, how­ever: an in­ter­est­ing thread to pull here may be that some brains seem to flip be­tween a ma­jor key and a minor key, with these keys be­ing lo­cal max­i­mas of har­mony. I sus­pect each is bet­ter at cer­tain kinds of pro­cess­ing, and al­though parts of each can be com­pat­i­ble with the other, each has el­e­ments that pre­sent as defec­tion to the in­ter­nal logic of the other and so these at­trac­tors can be ‘sticky’. But there can also be a buildup of ten­sion as one gath­ers in­for­ma­tion that is in­com­pat­i­ble with one’s key sig­na­ture, which gets pro­gres­sively more difficult to main­tain, and can lead to the sort of in­ten­sity of ex­pe­rience that drives an an­neal­ing-like pro­cess when the key sig­na­ture flips. And in the case of re­peated flips, the pat­terns which are com­pat­i­ble with both key sig­na­tures will be the most strongly re­in­forced.

In some ways a bipo­lar brain may re­sult in sig­nifi­cant cog­ni­tive and cre­ative ad­van­tages: per­haps the biggest is more ac­cess to high-en­ergy states, which in the short term helps cre­ativity by al­low­ing more ex­plo­ra­tion and also steeper valence gra­di­ents to fol­low, and iter­ated over the long term al­lows sig­nifi­cantly more op­ti­miza­tion pres­sure on the sub­sys­tems that are re­peat­edly an­nealed. How­ever this has cor­re­spond­ing episte­molog­i­cal down­sides as noted above; fuel­ing cre­ative work with valence deltas is likely to ‘warp the en­g­ine’ over time, to para­phrase Shinzen. Fris­ton’s no­tion that ‘sys­tems max­i­miz­ing long-term sta­bil­ity spend most of their time in a small num­ber of states’ seems par­tic­u­larly rele­vant to mood di­s­or­ders. (My col­league An­drés sug­gests this ‘bipo­lar effect pro­file’ may be repli­cated by valence-en­hanc­ing drugs with a short du­ra­tion and hang­over, such as co­caine- this at least fits stereo­types.)

I find my­self won­der­ing if neu­roti­cism can be thought of as an­cient neu­ral tech­nol­ogy in­tended to re­duce an­neal­ing fre­quency in the an­ces­tral en­vi­ron­ment — es­sen­tially if we look into the brains of highly neu­rotic peo­ple, we might find strong en­ergy sinks lo­cated around nat­u­ral con­nec­tome har­mon­ics which pre­vent se­man­ti­cally-neu­tral en­ergy build-up. This likely con­tributes to cer­tain forms of de­pres­sion (and leads to per­ni­cious feed­back cy­cles — the less one an­neals, the more neu­rotic one gets, the less able to reach high-en­ergy states one be­comes), but might also help pre­vent seizures or in­ap­pro­pri­ate up­dat­ing/​an­neal­ing, and may have fre­quency-de­pen­dent benefits. E.g., a group with 19 care­free an­neal­ers and 1 neu­rotic guardian will act more wisely than one with 20 care­free an­neal­ers or 20 neu­rotic guardians. The ‘neu­roti­cism=en­ergy sinks’ frame seems to sug­gest how to re­duce neu­roti­cism (an­neal more of­ten, es­pe­cially se­man­ti­cally-neu­tral an­neal­ing), and also offer clues as to how neu­roti­cism is im­ple­mented in the brain: we might look into the math­e­mat­ics of An­der­son lo­cal­iza­tion in the con­nec­tome: topolog­i­cal fea­tures that can ‘eat’ waves.

Is sleep a nat­u­ral an­neal­ing pro­cess? If so, this could cleanly ex­plain the con­nec­tion be­tween de­pres­sion and chronic sleep dis­tur­bances — poor sleep as both a cause and effect of in­fre­quent an­neal­ing. And it would in­di­cate a treat­ment path: a restora­tion of nor­mal an­neal­ing pat­terns may help im­prove both mood and sleep. I hold the fol­low­ing lightly, but we might model nREM as the heat­ing-up phase (un­damp­ened har­mon­ics) and REM as the neu­ral search & cool­ing pro­cess. From a re­view draw­ing par­allels be­tween sleep and jhana (in­tense med­i­ta­tive) states:

This pa­per is a pre­limi­nary re­port on the first de­tailed EEG study of jhana med­i­ta­tion, with find­ings rad­i­cally differ­ent to stud­ies of more fa­mil­iar, less fo­cused forms of med­i­ta­tion. While re­main­ing highly alert and “pre­sent” in their sub­jec­tive ex­pe­rience, a high pro­por­tion of sub­jects dis­play “spin­dle” ac­tivity in their EEG, su­perfi­cially similar to sleep spin­dles of stage 2 nREM sleep, while more-ex­pe­rienced sub­jects dis­play high voltage slow-waves rem­i­nis­cent, but sig­nifi­cantly differ­ent, to the slow waves of deeper stage 4 nREM sleep, or even high-voltage delta coma. Some oth­ers show brief pos­te­rior spike-wave bursts, again similar, but with sig­nifi­cant differ­ences, to ab­sence epilepsy. Some sub­jects also de­velop the abil­ity to con­sciously evoke clonic seizure-like ac­tivity at will, un­der full con­trol. (Den­ni­son 2019)

It seems plau­si­ble that broad rhyth­mic brain ac­tivity helps with cer­tain ‘phys­i­cal house­keep­ing’ tasks in the brain as well, and if one an­neals reg­u­larly they may need some­what less sleep (see re­cent re­search on Alzheimer’s, sleep, and rhyth­mic stim­u­la­tion helping break up brain plaques).

The ‘dead neu­ron’ model of neu­roti­cism and de­pres­sion:

Deep learn­ing mod­els can ex­hibit ‘dead neu­rons’: neu­rons whose ac­ti­va­tion func­tion gets ‘stuck’ on the on or off po­si­tion, for in­stance when a sig­moid func­tion gets too high or too low and its slope drops to al­most zero. Th­ese ‘dead’ neu­rons can be nigh-im­pos­si­ble to ‘re­vive’ within the model, since it can be the case that their gra­di­ent (im­plicit sen­si­tivity to in­put) is so shal­low that there sim­ply aren’t in­puts that will nudge it in one or an­other di­rec­tion.

Graphic: sig­moidal func­tion. This loses sen­si­tivity when val­ues get too high or too low. Differ­ent ac­ti­va­tion func­tions can lose sen­si­tivity (lead to ‘dead neu­rons’) un­der differ­ent sce­nar­ios-
ReLU is no­to­ri­ous for this.

Th­ese ‘dead’ neu­rons tend to cause lots of prob­lems, since their “always-on” or “always off” sig­nal tends to prop­a­gate through the net­work very strongly, caus­ing later neu­rons in the chain to also ex­hibit less sen­si­tivity to in­put. (Some­times this pro­cess will cas­cade, some­times not, much like ma­lig­nant vs be­nign tu­mors.)

I sus­pect this might be a strong frame for un­der­stand­ing the ‘psy­cholog­i­cal cruft’ which builds up in brains, and how and why reg­u­lar an­neal­ing is so healthy: over time, sen­si­tive neu­rons can slide into this bro­ken state, shift­ing from con­di­tional val­ues to the neu­rolog­i­cal equiv­a­lent of static 0s and 1s. In this case I would ex­pect more neu­roti­cism, less flex­ible think­ing, lower emo­tional re­silience, and worse episte­mol­ogy from peo­ple who haven’t an­nealed re­cently: lots of all-or-noth­ing think­ing. But by in­ject­ing lots of en­ergy into the sys­tem, enough of the in­ter­nal and ex­ter­nal con­text of these neu­rons is shifted such that some of them may get ‘re­set’ and re­gain their con­di­tional pro­cess­ing state. At the very least, this self-re­or­ga­ni­za­tion pro­cess can al­low these neu­rons to move to less-crit­i­cal points in pro­cess­ing net­works.

An idea re­lated to this frame is that a core func­tion of neu­ral an­neal­ing is to main­tain a smooth gra­di­ent of har­mony in the brain (and mind) – to make it pos­si­ble to “fol­low your joy” to­ward bet­ter out­comes. If this breaks down and you can’t “fol­low your joy”, con­sider putting your­self in a situ­a­tion which could plau­si­bly kick­start an an­neal­ing pro­cess (even if you don’t feel emo­tion­ally mo­ti­vated to do so).

IV. The na­ture of trauma and the im­ple­men­ta­tion of the Bayesian Brain

Trauma is one of the worst el­e­ments of the hu­man con­di­tion. It’s easy enough to ac­cu­mu­late that we all have some, and it’s hard to get rid of. But what is it?

Scott Alexan­der re­cently re­viewed a core work in the PTSD liter­a­ture, The Body Keeps The Score, and offers some con­text:

The book stressed the va­ri­ety of re­sponses to PTSD. Some peo­ple get anx­ious. Some peo­ple get an­gry. But a lot of peo­ple, what­ever their other symp­toms, also go com­pletely numb. They are prob­a­bly still “hav­ing” “emo­tions” “un­der” “the” “sur­face”, but they have no per­cep­tion of them. Some­times this men­tal deficit is ac­com­panied by equally sur­pris­ing bod­ily defic­its. Van der Kolk de­scribes a study on stereoag­nosia in PTSD pa­tients: if blind­folded and given a small ob­ject (like a key), they are un­able to rec­og­nize it by feel, even though this task is easy for healthy peo­ple. Some­times this gets even more ex­treme, like the case of a mas­sage ther­apy pa­tient who did not re­al­ize they were be­ing mas­saged un­til the ther­a­pist ver­bally ac­knowl­edged she had started.
The book is called The Body Keeps The Score, and it re­turns again and again to the idea of PTSD pa­tients as dis­con­nected from their bod­ies. The body sends a rich flow of in­for­ma­tion to the brain, which is part of what we mean when we say we “feel al­ive” or “feel like I’m in my body”. In PTSD, this flow gets in­ter­rupted. Peo­ple feel “like noth­ing”. …
There’s some dis­cus­sion of the neu­ro­biol­ogy of all this, but it never re­ally con­nects with the vivid­ness of the anec­dotes. A lot of stuff about how trauma causes the lizard brain to in­ap­pro­pri­ately ac­ti­vate in ways the ra­tio­nal brain can’t con­trol, how your “smoke de­tec­tor” can be set to over­drive, all backed up with the proper set of big words like “dor­so­lat­eral pre­frontal cor­tex” – but none of it seemed to reach the point where I felt like I was mak­ing progress to a gears-level ex­pla­na­tion. I felt like the level on which I wanted an ex­pla­na­tion of PTSD, and the level at which van der Kolk was ex­plain­ing PTSD, never re­ally con­nected; I can’t put it any bet­ter than that. …
There are a lot of al­ter­na­tive treat­ments for PTSD. Neu­ro­feed­back, where you at­tach your­self to a ma­chine that reads your brain waves and try to ex­plore the effect your thoughts have on brain wave pro­duc­tion un­til you are con­sciously able to ma­nipu­late your neu­ral states. In­ter­nal fam­ily sys­tems, where a ther­a­pist guides you through dis­cov­er­ing “parts” of your­self (think a weak ver­sion of mul­ti­ple per­son­al­ities), and you talk to them, and figure out what they want, and make bar­gains with them where they get what they want and so stop caus­ing men­tal ill­ness. Eye move­ment di­rected re­pro­cess­ing (al­ter­na­tive when the book was writ­ten, now ba­si­cally es­tab­lish­ment) where you move your eyes back and forth while talk­ing about your trauma, and this seems to some­how help you pro­cess it bet­ter. Acupunc­ture. Mas­sage. Yoga. …
Maybe the most con­sis­tent les­son from this book’s tour of suc­cess­ful al­ter­na­tive ther­a­pies – keep­ing with the theme of the ti­tle – is that it’s im­por­tant for PTSD pa­tients to get back in touch with their bod­ies. Mas­sage ther­apy, yoga, and acupunc­ture ad­dressed this di­rectly, usu­ally cre­at­ing gen­tle, com­fortable sen­sa­tions that pa­tients could take note of to grad­u­ally re­lax the ab­solute fire­wall be­tween bod­ily sen­sa­tion and con­scious pro­cess­ing.

The sim­ple Neu­ral An­neal­ing take on trauma is that sig­nifi­cant nega­tive events can push the brain into a high-en­ergy state filled with ‘trauma pat­terns’, and as the brain cools, some of these trauma pat­terns crys­tal­lize/​an­neal in a very durable form, which pre­sent as PTSD.

I think this is a more use­ful an­swer than what’s out there cur­rently, offer­ing straight­for­ward in­tu­itive an­swers for (1) what kinds of things are most likely to cause PTSD, (2) why PTSD is so ‘sticky’, and (3) an in­tu­itive solu­tion to PTSD: an­neal over the bad pat­terns with bet­ter pat­terns.

But Scott’s de­scrip­tion seems to point at some­thing fur­ther: that there’s a dis­con­nec­tion hap­pen­ing with trauma. To ad­dress this, I pro­pose the Neu­ral An­neal­ing model for how CSHW could im­ple­ment the Bayesian Brain model of cog­ni­tion. We’ll then cir­cle back and dis­cuss what might be go­ing wrong dur­ing trauma.

Last year in A Fu­ture for Neu­ro­science, I shared the frame that we could split CSHWs into high-fre­quency and low-fre­quency types, and per­haps say some­thing about how they might serve differ­ent pur­poses in the Bayesian brain:

The math­e­mat­ics of sig­nal prop­a­ga­tion and the na­ture of emo­tion­s
High fre­quency har­mon­ics will tend to stop at the bound­aries of brain re­gions, and thus will be used more for fine-grained and very lo­cal in­for­ma­tion pro­cess­ing; low fre­quency har­mon­ics will tend to travel longer dis­tances, much as low fre­quency sounds travel bet­ter through walls. This paints a pos­si­ble, and I think use­ful, pic­ture of what emo­tions fun­da­men­tally are: semi-dis­crete con­di­tional bun­dles of low(ish) fre­quency brain har­mon­ics that es­sen­tially act as Bayesian pri­ors for our lim­bic sys­tem. Change the har­mon­ics, change the pri­ors and thus the be­hav­ior. Panksepp’s seven core drives (play, panic/​grief, fear, rage, seek­ing, lust, care) might be a de­cent first-pass ap­prox­i­ma­tion for the at­trac­tors in this sys­tem.

I would now add this roughly im­plies a con­tinuum of CSHWs, with scale-free func­tional roles:

  • Re­gion-spe­cific har­monic waves (RSHWs) high fre­quency res­o­nances that im­ple­ment the pro­cess­ing of cog­ni­tive par­tic­u­lars, and are lo­cal­ized to a spe­cific brain re­gion (much like how high-fre­quen­cies don’t travel through walls) in the­ory quan­tifi­able through sim­ply ap­ply­ing Ata­soy’s CSHW method to in­di­vi­d­ual brain re­gions;

  • Con­nec­tome-spe­cific har­monic waves (CSHWs) low-fre­quency con­nec­tome-wide res­o­nances that act as Bayesian pri­ors, car­ry­ing rel­a­tively sim­ple ‘emo­tional-type’ in­for­ma­tion across the brain;

  • Sen­so­rium-spe­cific har­monic waves (SSHWs) very-low-fre­quency waves that span not just the con­nec­tome, but the larger ner­vous sys­tem and parts of the body. Th­ese en­code so­matic in­for­ma­tion – in the­ory, we could in­fer sen­so­rium eigen­modes by ap­ply­ing Ata­soy’s method to not only the con­nec­tome, but the ner­vous sys­tem, ad­just­ing for vari­able nerve-lengths, and val­i­date against some­thing like body-emo­tion maps.[2][3]

Th­ese waves shade into each other – a ‘low-fre­quency thought’ shades into a ‘high-fre­quency emo­tion’, a ‘low-fre­quency emo­tion’ shades into so­matic in­for­ma­tion. As we go fur­ther up in fre­quen­cies, these waves be­come more lo­cal­ized.

An in­ter­est­ing im­pli­ca­tion here is we may es­sen­tially get Bayesian up­dat­ing to nat­u­rally emerge from this ty­pol­ogy, through in­ter­ac­tions be­tween these var­i­ous waves: es­sen­tially, I think it’s ‘in­jec­tion-lock­ing all the way down’. (In­jec­tion-lock­ing is when har­monic os­cilla­tors (like CSHWs) es­sen­tially ‘sync up’ their pe­ri­ods and phases.) Speci­fi­cally:

Low-fre­quency CSHWs carry pri­ors, higher fre­quency RSHWs deal with par­tic­u­lars. Lower fre­quen­cies span the brain; higher fre­quen­cies res­onate within more lo­cal re­gions of the brain — the higher the fre­quency of the wave, the smaller the re­gion it tends to res­onate in. The RSHWs in differ­ent re­gions can’t talk to each other di­rectly, since (defi­ni­tion­ally) these waves can’t travel across re­gional bound­aries. But they can talk to each other in­di­rectly, through in­ter­act­ing with low-fre­quency CSHWs. More speci­fi­cally, I spec­u­late that re­gions and CSHW-en­coded pri­ors in­ter­act through a power-weighted av­er­ag­ing be­tween CSHWs and RSHWs, as me­di­ated by the math of in­jec­tion-lock­ing and in­jec­tion-pul­ling. This al­lows both func­tional par­ti­tion­ing and also global up­dat­ing: re­gions get some iso­la­tion in or­der to perform their spe­cial­ized com­pu­ta­tions, but they also get ex­po­sure to data about the over­all Bayesian prior situ­a­tion, aka what we call ‘emo­tional in­for­ma­tion’. I.e. Re­gion A syncs up with CSHWs, which carry the in­for­ma­tion to Re­gion B and sync up with the RSHWs there, and so on. Of note, there’s a del­i­cate, power-weighted hand­shake be­tween CSHWs and RSHWs: low-fre­quency har­mon­ics (emo­tions /​ Bayesian pri­ors) carry more power per har­monic (lower due to fre­quency, much higher due to am­pli­tude) but there are many more high-fre­quency har­mon­ics (sen­sory+cog­ni­tive par­tic­u­lars). Strong emo­tions like anger likely pump huge amounts of en­ergy into CSHWs and up­end this bal­ance, forc­ing RSHWs to syn­chro­nize with CSHWs. We can think of this as sac­ri­fic­ing the del­i­cate episte­mol­ogy-har­mo­niza­tion hand­shake in fa­vor of unity of pro­cess­ing and clar­ity of ac­tion — or put sim­ply, forc­ing per­cep­tion to match top-down ex­pec­ta­tions.

On en­tropic dis­in­te­gra­tion, search, and an­neal­ing in evolved har­monic sys­tems:

The noisy, stochas­tic na­ture of brain ac­tivity, along with prac­ti­cal re­quire­ments for home­osta­sis, will lead to a strong op­ti­miza­tion of the CSHW+RSHW con­figu­ra­tion for lo­cal min­ima which are re­sis­tant to change. How­ever, a large enough per­tur­ba­tion will push the sys­tem out of this basin (en­tropic dis­in­te­gra­tion step). The neu­ral search step is es­sen­tially the sys­tem stochas­ti­cally test­ing differ­ent har­monic con­figu­ra­tions; the neu­ral an­neal­ing step is the sys­tem ‘set­tling into’ a con­figu­ra­tion as its top-down pre­dic­tive mod­els get suffi­ciently good at sop­ping up the ex­cess en­ergy in the sys­tem, es­sen­tially form­ing a new basin it will again take a large amount of per­tur­ba­tion to get out of. The strength of an­neal­ing can be thought of as the steep­ness of this basin, and also the Heb­bian re­in­force­ment of sys­tem at­trac­tors (“neu­rons that fire to­gether, wire to­gether”). In­so­far as par­ti­tion­ing is pos­si­ble in a broadly-cou­pled har­monic sys­tem, these per­tur­ba­tions will tend to be ‘lo­cal’ as the brain has strong in­cen­tives to pre­serve struc­ture that doesn’t need up­dat­ing.

Toward a gen­er­al­ized defi­ni­tion of trauma: a break­down of in­for­ma­tion-prop­a­ga­tion-via-in­jec­tion-locking

I pro­pose that some­times the brain needs to rapidly halt in­for­ma­tion prop­a­ga­tion across re­gions to pre­vent cas­cad­ing sys­tem failure (a metaphor that comes to mind is an un­con­trol­led prion-like change in the lo­cal key sig­na­ture that rip­ples out from a trau­ma­tized re­gion, pro­gres­sively break­ing cy­ber­netic cal­ibra­tions). I be­lieve the brain uses two in­ter­linked mechanisms to do this: (1) weak­en­ing CSHWs, thus weak­en­ing in­for­ma­tion prop­a­ga­tion through­out the brain, and (2) ar­rang­ing differ­ent brain re­gions into fre­quency regimes which make in­for­ma­tion trans­fer difficult be­tween them (the golden mean is the math­e­mat­i­cally-op­ti­mal ra­tio for non-in­ter­ac­tion). Once this hap­pens, it can be very hard to re­verse, since it forms a self-sus­tain­ing cy­cle: (1) causes (2) and (2) causes (1). We call this ‘trauma’.

Some pre­dic­tions from this I’d ex­pect to see sub­stan­tially less en­ergy in low-fre­quency CSHWs af­ter trauma, and sub­stan­tially more en­ergy in low-fre­quency CSHWs dur­ing both ther­a­peu­tic psychedelic use (e.g. MDMA ther­apy) and dur­ing psy­cholog­i­cal in­te­gra­tion work. Stretch­ing a lit­tle, per­haps we could also ap­ply Ata­soy’s CSHW al­gorithm to in­di­vi­d­ual brain re­gions and com­pare their spec­trums (and those of CSHWs), to quan­tify the ex­pected fre­quency-cou­pling be­tween each re­gion.[4] Pos­si­bly these two mea­sures could be de­vel­oped into a causal quan­ti­ta­tive met­ric for trauma.

This gen­er­al­ized ‘break­down of com­mu­ni­ca­tion’ defi­ni­tion of trauma neatly fits with the story Scott tells about PTSD, where peo­ple

[A]re prob­a­bly still “hav­ing” “emo­tions” “un­der” “the” “sur­face”, but they have no per­cep­tion of them … PTSD pa­tients as dis­con­nected from their bod­ies. The body sends a rich flow of in­for­ma­tion to the brain, which is part of what we mean when we say we “feel al­ive” or “feel like I’m in my body”. In PTSD, this flow gets in­ter­rupted. Peo­ple feel “like noth­ing”.

It also fits with the ther­a­pies that seem to work: EMDR, neu­ro­feed­back, In­ter­nal Fam­ily Sys­tems (IFS), yoga, mas­sage — the con­sis­tent thread that con­nects these is they all plau­si­bly help restart and strengthen com­mu­ni­ca­tion within the brain (which I hold is strongly me­di­ated by CSHWs). Scott doesn’t men­tion mu­sic, but I’d ex­pect it to be sur­pris­ingly effec­tive at boost­ing emo­tional in­te­gra­tion — and I’d ex­pect the most effec­tive mu­sic will have strong low-fre­quency rhythms.

This shades into novel types of ther­a­peu­tic ap­proaches: per­haps we could sim­ply pump en­ergy into lower-fre­quency bands (per­haps har­monic stim­u­la­tion cen­tered at ~3-6hz) to kick­start emo­tional in­te­gra­tion.[5]

Si­de­note on mu­sic: The sim­ple de­scrip­tion I gave of mu­sic was

[A] sen­sory in­put which seems to ex­ist on the knife’s edge be­tween ex­hibit­ing highly or­dered pat­terns (some of which will hit nat­u­ral con­nec­tome har­mon­ics and so al­low ac­cu­mu­la­tion of en­ergy through res­o­nance) on one hand, and on the other hand not be­ing toopre­dictable (thus dodg­ing most in­hibitory top-down pre­dic­tive mod­els).

Armed with the CSHW/​RSHW dis­tinc­tion, we can give this a sec­ond pass. In short, I ex­pect the above story to be true, but in a frac­tal way: mu­sic will be hit­ting both CSHWs and RSHWs. Nat­u­rally, differ­ent re­gions will have differ­ent sets of har­mon­ics, which means sim­ple tones are un­likely to pro­duce much cross-re­gional res­o­nance. In­stead, the mu­sic which is the most effec­tive at in­creas­ing the brain’s en­ergy pa­ram­e­ter will tie to­gether and layer a di­verse set of mo­tifs, with two goals: (1) hit­ting as many con­nec­tome-spe­cific *and* re­gion-spe­cific res­o­nances as pos­si­ble, and (2) en­train­ing dis­parate re­gions and pul­ling them into sync es­sen­tially us­ing in­jec­tion-lock­ing to pull RKSs (Re­gional Key Sig­na­tures) into sync with each other and the CKS (Con­nec­tome Key Sig­na­ture).

Could we quan­tify what the ‘perfect song’ would be, for a given con­nec­tome? Not ex­actly, since so much of mu­sic’s effects rely on get­ting through the brain’s pre­dic­tive pro­cess­ing gaunt­let and the state of this gaunt­let isn’t well-cap­tured by a static con­nec­tome, but we could pos­si­bly use this frame­work to de­sign (po­ten­tially much) more evoca­tive songs.

It’s also worth not­ing that bet­ter mu­sic and bet­ter ways to listen to mu­sic shade quickly into po­ten­tial ther­a­pies for trauma un­der this model.

V. On psychedelics:

As noted above, Neu­ral An­neal­ing sug­gests a very sim­ple model for un­der­stand­ing the effects of psychedelics: as sub­stances which “may func­tion by dis­abling ex­ist­ing en­ergy sinks (or per­haps over­load­ing them by in­creas­ing baseline firing rates or in­creas­ing the branch­ing fac­tor of neu­ral ac­tivity),” dra­mat­i­cally in­creas­ing se­man­ti­cally-neu­tral en­ergy. Psychedelics share a ‘char­ac­ter­is­tic feel­ing’ (and char­ac­ter­is­tic emo­tional af­tere­ffects) with each other and with ac­tivi­ties such as med­i­ta­tion, listen­ing to mu­sic, EMDR, breath work, and so on, be­cause all of these things in­crease the en­ergy pa­ram­e­ter of the brain. Psychedelics are par­tic­u­larly in­ter­est­ing be­cause they do this so pow­er­fully, effortlessly, and nois­ily (with the effects bleed­ing over into sen­sory modal­ities, not just ac­cu­mu­lat­ing in har­mon­ics).

A full Neu­ral An­neal­ing model of psychedelics will have to wait a few more months as in­ter­nal QRI dis­cus­sion set­tles on a unified story. But a few pre­limi­nary notes:

First, we could define ‘psychedelics’ in a prin­ci­pled way, as any sub­stance, pat­tern, or pro­cess that pro­duces se­man­ti­cally-neu­tral en­ergy ac­cu­mu­la­tion – any­thing that dis­ables, over­loads, or avoids the brain’s en­ergy nor­mal­iza­tion sys­tem. The im­pli­ca­tion here is in­ter­est­ing, that any­thing that adds se­man­ti­cally neu­tral en­ergy into the brain should pro­duce psychedelic effects, re­gard­less of how this is done. E.g., even things like mod­ern art may be clas­sifi­able as a psychedelic, in­so­far as it gen­er­ates se­man­ti­cally-neu­tral en­ergy (see Gomez Emils­son 2019). But we should also note that cur­rent psychedelics are not nec­es­sar­ily perfect sources of ‘clean se­man­ti­cally-neu­tral en­ergy’; they’re sub­stances that hap­pen to mas­sively in­crease the en­ergy pa­ram­e­ter of the brain, with no guaran­tees about how ‘bal­anced’ this boost is. There may be bet­ter and more tar­geted meth­ods to do this in the fu­ture. In the mean­time, I would recom­mend mod­est cau­tion with sub­stances which in­volve a hang­over af­ter use, as nega­tive valence or af­fec­tive blunt­ing dur­ing a crit­i­cal win­dow could ‘sour’ the an­neal­ing pro­cess with sub­tle long-term mood effects.[6]

As men­tioned above, I’ve been think­ing more and more that the core psy­cholog­i­cal changes driven by psychedelics are best un­der­stood in terms of the amount and ‘statis­ti­cal fla­vor’ of the se­man­ti­cally-neu­tral en­ergy they add to the sys­tem. Or, as an al­ter­nate fram­ing, psychedelics may be best un­der­stood as tem­po­rary dis­rupters of the brain’s nat­u­ral en­ergy sinks, each with a spe­cific tar­get or ‘fla­vor’ of dis­rup­tion (or psychedelics may add to neu­ral ac­tivity’s ‘branch­ing fac­tor’, which in turn will add a spe­cific fla­vor to the en­ergy). I also find my­self won­der­ing, all else be­ing equal, whether psychedelic vi­su­als ac­tu­ally are in­versely cor­re­lated with an­neal­ing effects, since by di­vert­ing en­ergy into the vi­sual sys­tem (which plau­si­bly has very effec­tive en­ergy sinks), there is less en­ergy available to drive en­tropic dis­in­te­gra­tion.[7]

As I noted in A Fu­ture for Neu­ro­science, an­other start­ing point for sort­ing through psy­choac­tive drugs would be

[T]o parametrize the effects (and ‘phe­nomenolog­i­cal tex­ture’) of all psy­choac­tive drugs in terms of their effects on the con­so­nance, dis­so­nance, and noise of a brain, both in over­all terms and within differ­ent fre­quency bands (Gomez Emils­son 2017).
In the long term, we’ll want to move up­stream and pre­dict con­nec­tome-spe­cific effects of drugs treat­ing psy­choac­tive sub­stances as op­er­a­tors on neu­roa­cous­tic prop­er­ties, which pro­duce re­gion-by-re­gion changes in how waves prop­a­gate in the brain (and thus differ­ent peo­ple will re­spond differ­ently to a drug, be­cause these sorts of changes will gen­er­ate differ­ent types of re­sults across differ­ent con­nec­tomes). Essen­tially, this would in­volve eval­u­at­ing how var­i­ous drugs change the in­ter­nal pa­ram­e­ters of the CSHW model, in­stead of just the out­puts. Mov­ing up­stream like this might be nec­es­sary to pre­dict why e.g. some peo­ple re­spond well to a given SSRI, while oth­ers don’t (no­body has a clue how this works right now).

Pos­si­bly this would al­low us to gen­er­ate a prin­ci­pled ty­pol­ogy of psy­choac­tives, and also check for miss­ing quad­rants: psy­choac­tives and psychedelics we haven’t dis­cov­ered or cre­ated yet. (See also An­drés’s no­tion of parametriz­ing the ‘in­for­ma­tion vs en­ergy tra­jec­tory’ of a trip.) We can also think of anti-psy­chotic drugs as anti-psychedelics: sub­stances that rapidly de­crease the en­ergy pa­ram­e­ter of the brain (Gomez Emils­son 2019). We at QRI strongly be­lieve this makes anti-psy­chotics more dan­ger­ous than com­monly re­al­ized: the neu­ral search pro­cess is com­plex and del­i­cate, and an ex­ter­nally-forced, un­even rapid cool­ing pro­cess may warp the in­ter­nal land­scape of the brain in sub­tle but dele­te­ri­ous ways. In the­ory, we could test this in­di­rectly by eval­u­at­ing the effects of anti-psy­chotics on sen­sory in­te­gra­tion tasks in healthy con­trols – but as noted above, this may be an un­eth­i­cal ex­per­i­ment.

Another frame would be ‘psychedelics as full-spec­trum res­o­nance agents’ CSHWs are meant to sub­stan­tially res­onate dur­ing nor­mal hu­man op­er­a­tion (fal­ling in love, or­gasm, etc) RSHWs are not. The per­cep­tual and episte­molog­i­cal changes we some­times see dur­ing psychedelics could be due to the fine log­i­cal ma­chin­ery that usu­ally deals with high-con­text sen­sory par­tic­u­lars (facts and log­i­cal in­fer­ences) start­ing to malfunc­tion as its nat­u­ral eigen­modes are ac­ti­vated. Like link­ing and rhyth­mi­cally flip­ping all the bits in a mem­ory reg­ister, ig­nor­ing what that reg­ister is “sup­posed to” com­pute. If psychedelic vi­su­als are an ex­am­ple of RSHW res­o­nance, HPPD may be an ex­am­ple of this RSHW res­o­nance an­neal­ing into durable pat­terns.

On MDMA’s strangely pow­er­ful ther­a­peu­tic effects, I’d sug­gest MDMA shares the ‘ba­sic psychedelic pack­age’ with sub­stances like LSD and psilo­cy­bin (albeit a lit­tle weaker at com­mon doses). Any­thing with this ‘baseline’ pack­age sig­nifi­cantly in­creases the en­ergy pa­ram­e­ter of the brain, which both al­lows es­cape from bad lo­cal min­ima and canal­izes the brain’s core CSHWs, which both should be highly ther­a­peu­tic. My in­tu­ition is MDMA may also have a par­tic­u­lar effect on stochas­tic firing fre­quen­cies of neu­rons, and that this effect es­sen­tially acts as an emer­gent metronome – and this metronome will drive syn­chron­ic­ity be­tween di­verse brain re­gions. Given the pres­ence of such a re­gion-span­ning ‘clean’ metro­nomic sig­nal, brain re­gions that have par­tially ‘stopped talk­ing to each other’ will re-es­tab­lish in­te­gra­tion, and some of this in­te­gra­tion will per­sist while sober (or rather, some of the rea­sons for the lack of in­te­gra­tion will have been ne­go­ti­ated away dur­ing the MDMA-driven in­te­gra­tion). Plau­si­bly this ‘emer­gent metronome’ effect may also un­der­lie the par­tic­u­lar phe­nomenolog­i­cal effects of 5-MeO-DMT as well, par­tic­u­larly in terms of sense of unity, high valence, and ther­a­peu­tic po­ten­tial.[8]

Some­what po­etic side­note: on tak­ing psychedelics:

In the ab­stract I think psychedelics are more pow­er­ful, more dan­ger­ous, and more heal­ing than com­monly as­sumed.

But we don’t live in the ab­stract. The nat­u­ral ques­tion for any given per­son is thus: should I take them?

There’s no one-size-fits-all an­swer, and I recom­mend check­ing with lo­cal laws. But I can share a sim­ple heuris­tic for who shouldn’t worry too much about the down­sides of psychedelics and who should be very care­ful: do you trust your own aes­thetic?

Psychedelics mas­sively in­crease the ‘en­ergy pa­ram­e­ter’ of the brain, so nat­u­rally there’s a large amount of very-high-di­men­sional ex­plo­ra­tion go­ing on. There are countless ‘micro-choices’ your brain makes as to how to an­neal af­ter this ex­plo­ra­tion: we can think of a per­son’s ‘aes­thetic’ as in­di­vi­d­ual var­i­ance in these an­neal­ing choices. What the self-or­ga­niz­ing sys­tem which is the brain’s sub­con­scious finds beau­tiful in the mo­ment and im­plic­itly strives to save.

Some­times, and in some peo­ple, we want the right things, we find the right things beau­tiful. Things that have a deep el­e­gance and fit with ev­ery­thing about us and fit with how re­al­ity works. We just need enough en­ergy pa­ram­e­ter to get there. Psychedelics are a great way to get there.

Other times, we might not want the right things. Evolu­tion is kind of a jerk, episte­molog­i­cally speak­ing: it cares much more about ge­netic re­pro­duc­tion than it does about deep co­her­ence and cal­ibra­tion with re­al­ity and such. Some­times we’re at a func­tional lo­cal max­ima, but we’re not pointed in the right di­rec­tion globally, and frankly speak­ing our lack of a high en­ergy pa­ram­e­ter is our sav­ing grace – our in­abil­ity to di­rectly muck up our emo­tional land­scape. In­so­far as this is true – and it will be more true at cer­tain times than oth­ers, and in cer­tain peo­ple than oth­ers, and per­haps in cer­tain com­bi­na­tions of peo­ple than oth­ers – us­ing psychedelics to crank the en­ergy pa­ram­e­ter is not good for a per­son. Our ‘Psychedelic Ex­trap­o­lated Vo­li­tion’ (PEV) is not a healthy vec­tor.[9]

The nat­u­ral fol­low-up is, how do you know whether your PEV is pos­i­tive or not?

Hard ques­tion, but prob­a­bly good to ask your friends – group episte­mol­ogy seems healthy in these cases. And in gen­eral it seems strongly prefer­able to err on the side of cau­tion. You can always take that LSD to­mor­row, or next week, or next year.

(But, don’t be too para­noid about one trip per­ma­nently break­ing your brain, ei­ther. My guess is the an­neal­ing that tends to ‘stick’ is that which ac­tu­ally finds bet­ter lo­cal min­ima (thank­fully) – if it’s an un­suc­cess­ful ex­plo­ra­tion I sus­pect the sys­tem can usu­ally climb back to where it was (with some caveats).)

A sep­a­rate fac­tor is your cur­rent en­ergy pa­ram­e­ter and how psychedelics may in­crease this baseline: if you’re drag­ging on the bot­tom of your en­er­getic at­trac­tor bas­ins, maybe a lit­tle kick could be healthy. But if you’re already ‘high on life’ – con­sider skip­ping the LSD and MDMA. In­creas­ing a high baseline can redline the sys­tem into exquisitely un­bear­able in­ten­sity.

VI. Love and other types of Neu­ral Annealing

It’s im­por­tant to note that most an­neal­ing doesn’t hap­pen in a vac­uum: just as “set and set­ting” mat­ter quite a lot for psychedelics, and for emo­tional up­dat­ing in gen­eral, the im­por­tance of con­text in the an­neal­ing model is hard to over­state. Much as hold­ing a mag­net close to iron as it cools can mag­ne­tize the metal, the in­ten­tional con­tent pre­sent when en­tropic dis­in­te­gra­tion->an­neal­ing hap­pens pro­vides im­por­tant con­straints for which new pat­terns form. I pro­pose there are four gen­eral types of neu­ral an­neal­ing:

A. An­neal­ing to an ob­ject or event. An­neal­ing which is ‘pointed at’ some­thing is by far the most com­mon type. Some ob­ject, or event, or new in­sight makes it­self known in a sur­pris­ing or oth­er­wise in­tensely salient way, and this pushes the brain into a high-en­ergy state, kick­start­ing a self-or­ga­ni­za­tion pro­cess for ac­com­mo­dat­ing the pres­ence and sig­nifi­cance of this new thing. This can in­volve in­tense pos­i­tive emo­tion — a new ro­man­tic part­ner, the birth of your child, your wed­ding day. This sort of an­neal­ing can also be caused by trauma— get­ting bit­ten by a weird an­i­mal, so­cial re­jec­tion, los­ing a close one. As I sug­gested in The Neu­ro­science of Med­i­ta­tion, neu­ral an­neal­ing may offer a rather pithy de­scrip­tion of love:

Fi­nally, to spec­u­late a lit­tle about one of the deep mys­ter­ies of life, per­haps we can de­scribe love as the re­sult of a strong an­neal­ing pro­cess while un­der the in­fluence of some pat­tern. I.e., evolu­tion has primed us such that cer­tain in­ten­tional ob­jects (e.g. ro­man­tic part­ners) can trig­ger high-en­ergy states where the brain smooths out its dis­con­ti­nu­ities/​dis­so­nances, such that given the pres­ence of that pat­tern our brains are in har­mony. This is ob­vi­ously a two-edged sword: on one hand it heals and re­news our ‘cold-worked’ brain cir­cuits and unifies our minds, but also makes us de­pen­dent: the felt-sense of this in­ten­tional ob­ject be­comes the key which un­locks this state. (I be­lieve we can also an­neal to archetypes in­stead of spe­cific peo­ple.)
An­neal­ing can pro­duce durable pat­terns, but isn’t per­ma­nent; over time, dis­con­ti­nu­ities creep back in as the sys­tem gets ‘cold-worked’. To stay in love over the long-term, a cou­ple will need to re-an­neal in the felt-pres­ence of each other on a reg­u­lar ba­sis. From my ex­pe­rience, some peo­ple have a nat­u­ral psy­cholog­i­cal drive to­ward re­flex­ive sta­bil­ity here: they see their part­ner as the source of good­ness in their lives, so nat­u­rally they work hard to keep their mind al­igned on valu­ing them. (It’s cir­cu­lar, but it works.) Whereas oth­ers are more self-re­li­ant, ex­plo­ra­tory, and restless, less prone to­ward these self-sta­ble loops or an­neal­ing around ex­ter­nal in­ten­tional ob­jects in gen­eral. Whether or not, and within which pre­cise con­texts, some­one’s an­neal­ing habits fall into this ‘re­flex­ive sta­bil­ity at­trac­tor’ might ex­plain much about e.g. at­tach­ment style, he­do­nic strat­egy, and aes­thetic tra­jec­tory.

Per­haps we can go fur­ther now, and hy­poth­e­size that ‘fal­ling in love’ is a spe­cific al­gorithm the brain runs, which is trig­gered by when the ‘felt sense’ of an­other per­son (a pat­tern dis­tributed across RSHWs, CSHWs, and SSHWs) pro­duces sub­stan­tial sys­temic res­o­nance. When this hap­pens, and in the ab­sence of warn­ing signs (dis­so­nance), a per­son will ac­tively seek to fill their sen­so­rium with this sig­nal, which am­plifies the sys­temic res­o­nance (po­ten­tially to ex­treme lev­els) and fur­ther syn­chro­nizes pri­ors and other re­gions into har­mony with the origi­nal pat­tern. As you fall in love, you liter­ally an­neal to your felt-sense of that per­son – you take their rhythm as yours, be­cause your body judged it to be so. A key which fit your con­nec­tome’s lock. This will nat­u­rally do two things: (1) fuzz bound­aries be­tween lovers, as pat­terns pro­gres­sively syn­chro­nize, and (2) add a har­monic echo, or ‘warm con­so­nant glow’ to all thoughts about the per­son. This lat­ter phe­nomenon will feel nice, but also keep it­self sta­ble: the pres­ence of this bun­dle of syn­chro­nized fre­quen­cies will sta­bi­lize (via in­jec­tion-lock­ing) many forms of drift – effec­tively pre­vent­ing cer­tain thoughts/​per­cep­tions. This may fade over time if not re­freshed, but per­haps to com­pletely ‘fall out of love’ the brain has to build a com­pet­ing key sig­na­ture el­se­where, e.g. in a golden mean ra­tio to this har­monic echo, and these ri­valrous key sig­na­tures (im­plic­itly Bayesian pri­ors about what is real and what is good) bat­tle it out. (Thanks to An­drés for dis­cus­sion on com­pet­ing key sig­na­tures.) This ‘de-an­neal­ing’ pro­cess liter­ally eras­ing some­one’s pat­terns and rhythms from your body can fol­low sev­eral tra­jec­to­ries, few of them pleas­ant, as the sys­tem rene­go­ti­ates new (or old) equil­ibria.

B. An­neal­ing to an on­tol­ogy. A much more gen­eral type of an­neal­ing is when the en­tropic dis­in­te­gra­tion->an­neal­ing pro­cess is pointed to­ward an on­tol­ogy, and the brain re­or­ga­nizes its in­ter­nal struc­ture (‘on­tolog­i­cal con­tours’) to ac­com­mo­date this new ty­pol­ogy. This can hap­pen im­plic­itly and weakly, over the course of en­tropic dis­in­te­gra­tion->an­neal­ing to mul­ti­ple sep­a­rate ideas, or ex­plic­itly and strongly, for in­stance read­ing some book in col­lege which com­pletely re­shapes one’s view of re­al­ity.

Any crafts­man, any in­tel­lec­tual, any philoso­pher worth their salt is strongly an­nealed to­ward at least one nu­anced on­tol­ogy, and in fact much of the in­fluence of the Great Philoso­phers can be found in how they’ve laid out their thoughts in a way that oth­ers can use as a co­her­ent an­neal­ing tar­get. What makes some­thing a good an­neal­ing tar­get? I’d offer it’s the pres­ence of clear archetypes ar­ranged in both a novel but ul­ti­mately cog­ni­tively effi­cient way. Th­ese archetypes can be thought of as a com­bi­na­tion of na­ture (in­nate Jun­gian-type lim­bic res­o­nances) and nur­ture (prior an­nealed pat­terns & cul­tural reifi­ca­tions).

An im­por­tant point here is that peo­ples’ con­cep­tion of where good­ness comes from is de­pen­dent upon their on­tol­ogy; change the on­tol­ogy, change the per­ceived na­ture of good­ness it­self! See e.g. John Lily’s dis­cus­sion of the supra-self-metapro­gram­mer (SSMP). This frame-shift can also man­i­fest at the ex­treme end of fal­ling in love, where all the world’s good­ness seems to come from your spe­cial per­son (a dan­ger­ous thing).

C. So­cial an­neal­ing. A spe­cial hy­brid of an­neal­ing to an on­tol­ogy and to other peo­ple is so­cial an­neal­ing, wherein a group of peo­ple un­der­goes the ‘en­tropic dis­in­te­gra­tion → neu­ral search → an­neal­ing’ pro­cess to­gether, within some shared con­text- a re­li­gious ser­vice, a sport­ing event, a re­treat. This seems like the nat­u­ral mechanism by which tribes are formed (loosely speak­ing, group syn­chro­niza­tion of con­nec­tome-spe­cific har­monic wave dy­nam­ics) and un­der­lies many of our most sa­cred ex­pe­riences. The power of so­cial an­neal­ing is such that a re­li­gious ex­pe­rience that lacks it no longer feels like a re­li­gious ex­pe­rience- merely the mouthing of dogma. On the other hand, any group ex­pe­rience that does in­crease the group’s en­ergy pa­ram­e­ter and trig­ger an­neal­ing starts to take on a pseudo-re­li­gious frame- e.g. ec­static dance, fes­ti­vals, protest marches, even con­certs.

D. Se­man­ti­cally-neu­tral an­neal­ing. Al­most all neu­ral an­neal­ing is se­man­tic an­neal­ing, or an­neal­ing to­ward some in­ten­tional ob­ject. This pro­cess is pointed at some­thing, of­ten the thing that caused the en­tropic dis­in­te­gra­tion pro­cess in the first place, be it a per­son, an event, an idea, an on­tol­ogy. But there’s noth­ing in the laws of neu­ro­science that im­plies an­neal­ing has to have an in­ten­tional ob­ject as a fo­cus. As per Sec­tion II, I be­lieve this is a par­tic­u­larly healthy form of an­neal­ing.

Toward a new psy­chol­ogy & so­ciol­ogy?

Spec­u­la­tively, we may be able to re-de­rive much of psy­chol­ogy and so­ciol­ogy from just the en­ergy-pa­ram­e­ter view of the brain: e.g.,

Gop­nik 2017 sug­gests that differ­ent de­vel­op­men­tal win­dows may in­volve differ­ent im­plicit ‘heat pa­ram­e­ters’ for simu­lated an­neal­ing, with young peo­ple hav­ing higher pa­ram­e­ters. Spec­u­la­tively, this may cor­re­spond to differ­ent ‘lived in­ten­sity of ex­pe­rience’ at differ­ent ages- young brains (and lifelong learn­ers) might not only be more plas­tic than av­er­age, but ac­tu­ally hav­ing ex­pe­rience that is ob­jec­tively more visceral. One way to frame this is that be­ing young is like micro­dos­ing on LSD all the time. This could have in­ter­est­ing im­pli­ca­tions for ethics.

Most likely, there’s been sig­nifi­cant re­cent sex­ual se­lec­tion for a higher en­ergy pa­ram­e­ter, for sev­eral rea­sons:

  • Select­ing for neoteny plau­si­bly also im­plic­itly se­lects for an en­ergy pa­ram­e­ter that starts higher and/​or de­cays less with age;

  • A high en­ergy pa­ram­e­ter would be a good proxy for cog­ni­tive-emo­tional-be­hav­ioral dy­nam­i­cism, per­haps the most strongly sex­u­ally-se­lected-for trait;

  • A high en­ergy pa­ram­e­ter would be an hon­est sig­nal of not be­ing in a bad ‘iter­ated aes­thet­ics’ at­trac­tor (oth­er­wise they would have self-de­struc­ted pre­vi­ously).

Psy­chol­ogy has var­i­ous per­son­al­ity met­rics, with the most widely used be­ing the Big 5, also known as OCEAN (Open­ness, Con­scien­tious­ness, Ex­tro­ver­sion, Agree­able­ness, Neu­roti­cism). One of the most in­ter­est­ing sub­find­ings here is that we can still get rea­son­able pre­dic­tive util­ity if we col­lapse these into a one-vari­able model: the ‘Big One’ per­son­al­ity fac­tor. Scor­ing high in this fac­tor is ”as­so­ci­ated with so­cial de­sir­a­bil­ity, emo­tion­al­ity, mo­ti­va­tion, well-be­ing, satis­fac­tion with life, and self-es­teem.” Scor­ing low is as­so­ci­ated with de­pres­sion, frailty, lack of emo­tion­al­ity, and so on. I wouldn’t be sur­prised if the ‘Big One’ sim­ply tracks how fre­quently and deeply some­one an­neals.

Con­tin­u­ing the thread on So­cial An­neal­ing, I think we can push into so­ciol­ogy with the Neu­ral An­neal­ing model too; to un­der­stand a so­ciety, we need to un­der­stand how and when an­neal­ing hap­pens in that so­ciety. To gauge the wis­dom of a so­ciety, look at how its de­ci­sion-mak­ers an­neal; to gauge the cul­tural di­rec­tion of a so­ciety, look at how its young peo­ple an­neal. To un­der­stand the strongest so­cial bonds of a so­ciety, look at the con­texts in which group an­neal­ing hap­pens.

This also sug­gests why drugs like al­co­hol and cer­tain psychedelics are rit­u­al­is­ti­cally cel­e­brated in so many cul­tures: they al­low so­cial-an­neal­ing-on-de­mand, a key tech­nol­ogy in build­ing and main­tain­ing so­cial co­he­sion and co­or­di­na­tion.

Like­wise, we could en­vi­sion a field of ‘so­cial arche­ol­ogy’ eval­u­at­ing an­neal­ing pat­terns in the past: how of­ten did peas­ants and no­bles in the Mid­dle Ages an­neal? In what con­texts did the an­neal­ing hap­pen, and which in­sti­tu­tions con­trol­led them? Per­haps most poli­ti­cal con­flicts could be rein­ter­preted as con­flicts over an­neal­ing.[10] And so on. My col­league An­drés has sug­gested that a good rule of thumb for iden­ti­fy­ing an­neal­ing (and mak­ing good movies) is that in­tu­itively, an­neal­ing defines where you should ac­tu­ally point the cam­era if you were mak­ing a movie of a his­tor­i­cal pe­riod, since where an­neal­ing is hap­pen­ing is where changes that ‘mat­ter’ are tak­ing place: cog­ni­tive up­dates, de­ci­sions about how to feel, and so on.

On the effect of pro­fes­sion on emo­tional vibrancy: It would be some­what sur­pris­ing if cer­tain re­peated com­pu­ta­tional tasks didn’t tend to push re­gions’ key sig­na­tures into be­ing highly cou­pled (=in­tense emo­tions), whereas other classes of tasks push re­gions’ key sig­na­tures into fairly or­thog­o­nal con­figu­ra­tions (=‘white noise’ as emo­tional state). A life­time of dance or po­etry might liter­ally make you feel emo­tions more strongly; a life­time of do­ing ac­count­ing might liter­ally pro­duce a seg­mented brain and af­fec­tive blunt­ing. From Dar­win’s au­to­bi­og­ra­phy:

I have said that in one re­spect my mind has changed dur­ing the last twenty or thirty years. Up to the age of thirty, or be­yond it, po­etry of many kinds, such as the works of Mil­ton, Gray, By­ron, Wordsworth, Col­eridge, and Shel­ley, gave me great plea­sure, and even as a schoolboy I took in­tense delight in Shake­speare, es­pe­cially in the his­tor­i­cal plays. I have also said that formerly pic­tures gave me con­sid­er­able, and mu­sic very great delight. But now for many years I can­not en­dure to read a line of po­etry: I have tried lately to read Shake­speare, and found it so in­tol­er­ably dull that it nau­se­ated me. I have also al­most lost my taste for pic­tures or mu­sic. … I re­tain some taste for fine scenery, but it does not cause me the exquisite delight which it formerly did. …
This cu­ri­ous and lamentable loss of the higher aes­thetic tastes is all the odder, as books on his­tory, bi­ogra­phies, and trav­els (in­de­pen­dently of any sci­en­tific facts which they may con­tain), and es­says on all sorts of sub­jects in­ter­est me as much as ever they did. My mind seems to have be­come a kind of ma­chine for grind­ing gen­eral laws out of large col­lec­tions of facts, but why this should have caused the at­ro­phy of that part of the brain alone, on which the higher tastes de­pend, I can­not con­ceive.

Read­ing this ac­count, I find it plau­si­ble that Dar­win re­peat­edly pushed (and an­nealed) his mind to­ward RSHW-driven ‘clock­work piece­meal in­te­gra­tion’ in­ter­ac­tions rather than CSHW/​SSHW-driven global sym­me­try gra­di­ents, al­though Dar­win’s age, sick­ness, and de­pres­sion may have also con­tributed. A warn­ing sign for us the­o­rists and sys­tem­atiz­ers.


Neu­ral An­neal­ing is a neu­ro­science paradigm which aims to find the op­ti­mal trade­off be­tween el­e­gance and de­tail. It does this by iden­ti­fy­ing a level of ab­strac­tion which sup­ports par­allel de­scrip­tion un­der three core prin­ci­ples of self-or­ga­ni­za­tion: phys­i­cal self-or­ga­ni­za­tion (around con­nec­tome res­o­nances), com­pu­ta­tional self-or­ga­ni­za­tion (around min­i­miza­tion of sur­prise), and en­er­getic self-or­ga­ni­za­tion (around con­di­tional en­tropic dis­in­te­gra­tion).

There is yet much work to be done: in par­tic­u­lar, there are huge bod­ies of liter­a­ture around re­cep­tor af­fini­ties, net­work topolo­gies, re­gional anatomies and cell types, and so on. The promise of Neu­ral An­neal­ing is it’s not only a pre­dic­tive and gen­er­a­tive the­ory in its own right, but it pro­vides a level of de­scrip­tion by which to con­nect these dis­parate maps, and an ex­ten­si­ble con­text to build on as we add more and more de­tail to the model.

Fi­nally, we can ask: why does good neu­ro­science mat­ter? I would offer the fol­low­ing.

The fu­ture could be much bet­ter than the pre­sent. Much bet­ter.

Ma­te­rial con­di­tions are only very loosely cou­pled with well-be­ing. If life is to be rad­i­cally bet­ter in the fu­ture, it will be due to bet­ter neu­ro­science point­ing out how we can be kinder to our­selves and oth­ers, and fu­ture neu­rotech­nol­ogy chang­ing the he­do­nic calcu­lus of the hu­man con­di­tion.

A unified the­ory of emo­tional up­dat­ing, de­pres­sion, trauma, med­i­ta­tion, and psychedelics may give us the tools to build a fu­ture that’s sub­stan­tially bet­ter than the pre­sent. This has been my hope while writ­ing this.

–Michael Ed­ward John­son, Ex­ec­u­tive Direc­tor, Qualia Re­search Institute


[1] The ‘se­man­ti­cally neu­tral en­ergy’ model also sug­gests why tran­scra­nial mag­netic stim­u­la­tion (TMS) seems to help treat de­pres­sion – es­sen­tially, TMS in­jects a large amount of en­ergy into the brain, and this en­ergy (1) trig­gers some en­tropic dis­in­te­gra­tion, al­low­ing es­cape from bad lo­cal min­ima, and (2) may slightly col­lect in the brain’s nat­u­ral har­mon­ics, which may help pull the brain out of dis­so­nant equil­ibria. Note that this could be done much more effec­tively: in­stead of the pre­sent strat­egy of us­ing a quick flash of un­pat­terned, pulsed TMS (e.g., 5 sec­onds @ 100hz) which over­pow­ers the brain but quickly dis­si­pates and likely doesn’t lead to a sig­nifi­cant build-up in har­mon­ics, we could in­stead try an en­train­ment ap­proach via lower-power, rhyth­mic, con­tin­u­ous TMS, ap­plied for longer du­ra­tions (keep­ing the brain above its ‘re­crys­tal­liza­tion tem­per­a­ture’ for longer, al­low­ing a ful­ler self-or­ga­ni­za­tion pro­cess), per­haps paired with mu­sic.

[2] Thanks to An­drés for the idea about so­matic in­for­ma­tion, and the sug­ges­tion of sen­so­rium as the la­bel.

[3] I sus­pect that mus­cle ten­sion could be a core mechanism for reg­u­lat­ing SSHWs and per­haps CSHWs. Tens­ing mus­cles will strongly in­fluence body res­o­nance, and one’s body res­o­nance con­figu­ra­tion will likely have rip­ple effects on what sorts of fre­quen­cies per­sist in the brain. This sug­gests that tra­di­tions such as yoga are ba­si­cally right when they posit a link be­tween prob­lems in mus­cles and prob­lems in the mind: we may hold ten­sion in one sys­tem in or­der to com­pen­sate for a prob­lem in the other. Spec­u­la­tively, this com­pen­satory reg­u­la­tion may also be found across hu­mans, es­pe­cially in pair bonds: that ten­sion in your back might in some literal way be an at­tempt to help your part­ner with their emo­tional reg­u­la­tion. This would sug­gest mus­cle ten­sion should change sig­nifi­cantly af­ter a break-up. (Thanks to Emily Crot­teau, Lena Se­lesneva, and Ivanna Ev­tukhova for pieces of the puz­zle here.)

[4] My col­league An­drés sug­gests that “[A] more di­rect method, though per­haps more difficult, would be to look di­rectly for the spec­tral sig­na­tures of in­jec­tion lock­ing — we’d pre­dict you will see a se­ri­ously diminished de­gree of in­jec­tion lock­ing sig­na­tures on peo­ple who are heav­ily trau­ma­tized, and see it come back af­ter MDMA ther­apy.”

[5] Per­haps we could model Per­sis­tent Non-Sym­bolic Ex­pe­rience (PNSE) as per­sis­tent par­tial in­jec­tion lock­ing of key re­gions by low fre­quency CSHWs: es­sen­tially this would in­volve en­train­ing (and effec­tively par­tially dis­abling) the ma­chin­ery that usu­ally han­dles in­ter­pre­ta­tion of cer­tain par­tic­u­lars /​ cog­ni­tive in­ter­pre­ta­tions. Per­haps highly neu­rotic or trau­ma­tized in­di­vi­d­u­als with strong top-down con­trol ex­hibit the op­po­site: es­sen­tially try­ing to en­train CSHWs to a spe­cific re­gion (with pre­dictably poor re­sults).

[6] My col­league An­dres also recom­mends against “psychedelic sub­stances that have as part of their ac­tivity pro­file a high level of body-load, such as nau­sea and cramps as these pat­terns might them­selves be­come an­neal­ing tar­gets (cf. com­pounds no­to­ri­ous for this, ac­cord­ing to Psy­cho­nautWiki such as 2C-E, 2C-T-2, and 2C-P, are prob­a­bly best avoided as ther­a­peu­tic aids).”

[7] On psychedelic tol­er­ance: if the se­man­ti­cally-neu­tral en­ergy model of psychedelics proves out, we should also be open to sub­tle corol­laries: e.g., to what ex­tent is the tem­po­rary tol­er­ance effect of psychedelics bio­chem­i­cal (de­ple­tion of some neu­ro­trans­mit­ters, per the cur­rent story) and to what ex­tent is it in­for­ma­tion-the­o­retic —as­so­ci­ated with the re­lease and de­ple­tion of sys­temic sources of Free En­ergy? I.e., there is po­ten­tial en­ergy of a sort liber­ated when the sys­tem finds a bet­ter lo­cal min­ima, and if the sys­tem has un­der­gone strong an­neal­ing re­cently, there are fewer such ‘en­er­getic free lunches’ around to help power the psychedelic effects. (Hy­poth­e­sis held weakly, as my col­league An­drés points out there are psychedelics which do not trig­ger tol­er­ance, such as N,N-DMT and 5-MeO-DMT.)

[8] HT to Steve Le­har for point­ing at this ’nys­tag­mus’ phe­nomenon as be­ing some­how linked to MDMA’s mood-lift­ing effect, and to An­drés for call­ing my at­ten­tion to Le­har’s work and sug­gest­ing 5-MeO-DMT may also share this mechanism.

[9] This is a refer­ence to Eliezer Yud­kowsky’s “Co­her­ent Ex­trap­o­lated Vo­li­tion” (CEV) con­cept, which is an at­tempt to sketch a heuris­tic for how to use a rad­i­cally-pow­er­ful op­ti­miza­tion pro­cess (such as an AGI) safely. Essen­tially, CEV sug­gests we could ag­gre­gate all hu­man prefer­ences (vo­li­tions), find some way to merge them (make them ‘co­here’), then re­peat (ex­trap­o­late), un­til we get to a self-sta­ble loop. A ‘psychedelic ex­trap­o­lated vo­li­tion’ is a vari­a­tion of this: if it be­comes eas­ier to change your­self on psychedelics, and then that per­son you turn into can change them­selves into some­one else, and so on, where do you end up? What gen­er­ates a ‘pos­i­tive vec­tor’ here?

[10] This nat­u­rally and un­for­tu­nately makes the ac­cess to and con­texts of so­cial an­neal­ing an axis of cul­tural con­flict: those who con­trol these events con­trol the emo­tional tone and con­tours of co­or­di­na­tion of a so­ciety. Tak­ing away healthy an­neal­ing con­texts from your op­po­nents and giv­ing more so­cial an­neal­ing op­por­tu­ni­ties to your peo­ple is a key (but also very dirty) way to ‘win’ the cul­ture war. (Per­haps the opi­oid crisis, and the crack-co­caine crisis be­fore it, could in some sense be ex­ac­er­bated by a lack of healthier an­neal­ing op­por­tu­ni­ties.)


For at­tri­bu­tion in aca­demic con­texts, please cite this work as:

Michael Ed­ward John­son, “Neu­ral An­neal­ing: Toward a Neu­ral The­ory of Every­thing”, https://​​openthe­​​2019/​​11/​​neu­ral-an­neal­ing-to­ward-a-neu­ral-the­ory-of-ev­ery­thing/​​ , San Fran­cisco (2019).


I’d like to thank An­drés Gómez Emils­son for many great con­ver­sa­tions on an­neal­ing (and first call­ing my at­ten­tion to the term), en­ergy sinks, and countless other top­ics, and offer­ing care­ful feed­back on a draft of this work; Robin Carhart-Har­ris and Karl Fris­ton for a beau­tiful de­scrip­tion of simu­lated an­neal­ing; Romeo Stevens for wide dis­cus­sion about an­neal­ing & on­tolo­gies; Adam Safron for in­tro­duc­ing me to the depth of ex­pla­na­tion af­forded by pre­dic­tive cod­ing, point­ing me to­ward in­jec­tion lock­ing, and many great con­ver­sa­tions in gen­eral; Quintin Frerichs for his hard work to­ward mak­ing ther­a­peu­tic ap­pli­ca­tions of this the­ory real, and the rest of the QRI team for sup­port and in­spira­tion; Milan Griffes for care­ful feed­back on a draft of this work; Alex Alek­seyenko and James Dama for dis­cus­sions about simu­lated an­neal­ing; An­thony Mark­well for shar­ing the Bud­dhist Dhamma with me in such a thought­ful and gen­er­ous way; Justin Mares for his con­stant cu­ri­os­ity and en­courage­ment; my par­ents, for their end­less love and pa­tience; Lena Zait­seva and Lena Se­lesneva for their warmth and sup­port; and es­pe­cially Ivanna Ev­tukhova, who has made my life rad­i­cally bet­ter and whose love, en­ergy, and ob­ses­sion with Bud­dhist en­light­en­ment was why this work hap­pened.

To grat­i­tude.

Timeline: most of this doc­u­ment writ­ten ~Feb-April 2019, as a con­tinu­a­tion of The Neu­ro­science of Med­i­ta­tion and this talk, and shared in­ter­nally and with se­lect re­view­ers; sec­tion deal­ing with trauma writ­ten July 2019. Doc­u­ment re­ordered for flow and pol­ished in Oct-Nov and posted Thanks­giv­ing 2019.