subfunctional overlaps in attentional selection history implies momentum for decision-trajectories

post by Emrik (Emrik North) · 2024-12-22T14:12:49.027Z · LW · GW · 1 comments

Contents

  ASH-SuFuOvMo
  the countdown-trick helps when:
  prospective start-cue salience-sharpening (ProStaCuSalSharp)
  proxy-choice pre-insertion (ProChoPreIn)
  ASH-SuFuOvMo has major implications for self-steering & learning strategies
None
1 comment

ASH-SuFuOvMo (momentum from the subfunctional overlap in attentional selection-history) explains why & when the countdown-trick to get out of bed when super-groggy actually works.

This is a trick that I use for getting out of bed in the morning - quite literally:  I count down from 10 and get out of bed after the "1".
  (Eliezer, 2009 [LW · GW])

ASH-SuFuOvMo

the decision to do the countdown is motivated by a set of neuremic [LW(p) · GW(p)] subfunctions which greatly overlaps with the set of subfunctions contributing to your motivation for getting out of bed.  so when you make a top-down goal-directed decision to do the former, those subfunctions are reinforced in ASH [LW(p) · GW(p)], and that leaks over to net reinforcing the brain's total decision-weight for the latter.

the countdown-trick helps when:

  1. the subfunctions contributing to [get out of bed] are subthreshold for decision-selection, while the set for [do the countdown] is above it; and
  2. [do the countdown] has net ASH-SuFuOvMo in the direction of [get out of bed], and that overlap suffices to take the latter above threshold.

prospective start-cue salience-sharpening (ProStaCuSalSharp)

another important reason (probably weighs more than the above most of the time tbh) the countdown-trick works is that it's a type of prospective start-cue salience-sharpening, achieved via prospective motivation-scope restriction.

the problem is that [get out of bed] is a motivation with no clearly scoped context for execution, so the total salience that goes into it is normalized across all prospective contexts it bids for.  (see salience-normalization [LW · GW].)

when you just think to yourself "I really need to get out of bed sometime soon!!", the [get out of bed]-motivation is normalized over a broad range of temporal contexts, representing the brain's uncertainty about when it's scoped for.  by deciding to do it immediately after the countdown, you squish the distribution into that precise context.

(but then if you fail to wake up after the countdown, now the brain's probably going to be more uncertain about the temporal scope of the action, and rationally less sensitive to update its distribution based on the evidence from your conscious decisions.  ⚠ hence why if you do do the countdown-trick, you better not flimsy the commitment!)

proxy-choice pre-insertion (ProChoPreIn)

in my notes, the countdown-trick is the prototypical example of a ProChoPreIn, and a technique for salience-charging—relying on incoherence in the brain's utility-function (or subfunctions thereof) across time to iteratively pump more motivation into what you want to want to do.

...in other words:  Dutch-booking yourself to become more agentic.  ^^

ASH-SuFuOvMo has major implications for self-steering & learning strategies

when you make a decision D₁ which causally depends on a set of neural subfunctions {a,b,c,d}… then you reinforce all other self-consistent decisions propto their subfunctional overlap w D₁, due to Attentional Selection-History + neural modularity.  (eg, Dₓ={f,b,c,g} gets reinforced half as much as D₁, assuming there aren't specific subfunctions within D₁ that inhibit specific subsets of Dₓ)  this btw is rly important for self-steering: WHEN consider decision THEN ➀ analyze subfunctions for each decision-option, ➁ cek which other decisions will incidentally be reinforced by them, ➂ cek/predict utility of those learning-trajectories, ➃ factor (3) into object-level
(source)
(source)

1 comments

Comments sorted by top scores.

comment by CBiddulph (caleb-biddulph) · 2024-12-22T17:30:55.925Z · LW(p) · GW(p)

Strong-upvoted just because I didn't know about the countdown trick and it seems useful