Type 2 as an aggregation of Type 1 processes

This post as­sumes ba­sic knowl­edge of Type 1/​Type 2 (Sys­tem 1/​Sys­tem 2) cat­e­go­riza­tion of men­tal pro­cesses.

Back­ground (safe to skip)

After my first re­ac­tion of sur­prise (con­sum­ing per­haps a few months) to the topic of heuris­tics and bi­ases, and af­ter a few more read­ings on neu­ropsy­chol­ogy, I started re-vis­it­ing my first re­ac­tion in more de­tail. Should it re­ally be sur­pris­ing to learn that hu­mans are not ra­tio­nal? Any­one with a ba­sic con­nec­tion with hu­mans should eas­ily see that we act ir­ra­tionally in many situ­a­tions – snap de­ci­sions, im­pulses, etc. – so what was the source of my sur­prise?

My best guess (know­ing my limits of in­ter­po­la­tion) was that my sur­prise was not a re­sult of dis­cov­er­ing that we’re ir­ra­tional, but rather that there was a sci­en­tific ap­proach in ex­is­tence aiming at find­ing more about those ir­ra­tional­ities, and that re­sults of pre­dictable ir­ra­tional­ity were ap­pear­ing; that might even­tu­ally lead to unify­ing differ­ent bi­ases un­der the same the­ory or source.

The no­tion of Type 1 and Type 2 think­ing (or Sys­tem 1 and Sys­tem 2) is for me a the­ory that has the power to unify most of the bi­ases and per­haps pre­dict oth­ers. Kah­ne­man’s Think­ing Fast and Slow adopts such an ap­proach, at­tempt­ing to ex­plain many bi­ases in terms of Type 2 thought.

Now, this con­nected with a ques­tion I had back in col­lege when I first learned about Ar­tifi­cial Neu­ral Net­works (I was lucky to chose this as a topic to re­search and give a lec­ture to my col­leagues on): “if this is how the brain works, how does log­i­cal/​ra­tio­nal thought emerge?”

To my un­der­stand­ing, Con­nec­tion­ism and the self-or­ga­niz­ing pat­tern­ing sys­tem that is the brain would nat­u­rally re­sult in Type 1 thought as a di­rect con­se­quence. The ques­tion that I had per­sis­tently is how can Type 2 thought emerge from this hard­ware? Jonah Lehrer’s The De­ci­sive Point sug­gests that differ­ent brain ar­eas are (more) as­so­ci­ated with each type of thought, but es­sen­tially (un­til proven oth­er­wise), I as­sume that they all rely on essence on a pat­tern­ing pro­cess, a con­nec­tion­ist model.

Mi­gra­tion of Skills

We know that many skills start in Type 2 and mi­grate to Type 1 as we get more “ex­pe­rienced” in them. When we first learn driv­ing, we need to con­sciously think of ev­ery move, and the se­quence of steps to perform, etc. We con­sciously en­gage in ex­e­cut­ing a known se­quence be­fore chang­ing lanes (for ex­am­ple): look at the side mir­ror, look at the side to cover the blind spot, de­crease speed, etc.

As we get more driv­ing ex­pe­rience, we stop to con­sciously pro­cess those steps, they be­come au­to­matic, and we can even en­gage in other con­scious pro­cesses while driv­ing (e.g. hav­ing a con­ver­sa­tion, think­ing about a meet­ing you have later, etc).

I be­lieve this is key to un­der­stand­ing the re­la­tion be­tween both types of thought, since it pro­vides a kind of in­ter­face be­tween them, it pro­vides a way to com­pare the same pro­cess ex­e­cut­ing by both sys­tems.

Sim­ple Type 2 operations

So, hav­ing to ex­per­i­men­tal ap­para­tus at hand, I had only the weak in­stru­ment of per­sonal in­ter­po­la­tion plus child­hood mem­ory. Start­ing with a sim­ple op­er­a­tion, I de­cided to at­tempt to com­pare its ex­e­cu­tion by both sys­tems. The op­er­a­tion: sin­gle digit ad­di­tion.

As a child, 3+2 could have mul­ti­ple in­ter­pre­ta­tions de­pend­ing on pre­vi­ous ed­u­ca­tion. Two ex­am­ples might be: (1) vi­su­al­ize 3 ap­ples, vi­su­al­ize 2 ap­ples, count how many ap­ples “ap­pear” in work­ing mem­ory, and that gives you the an­swer. (2) Hold your fist in front of you, stretch out each finger, count­ing in­cre­men­tally un­til you reach 3, then start new “thread” at 0, stretch more fingers count­ing un­til you reach 2, while also in­cre­ment­ing the first thread that stopped at 3 – the re­sult then is the num­ber reached by the first thread.

The above is an at­tempt at an­a­lyz­ing how a child, us­ing Type 2 pro­cesses, would find the an­swer to 3+2; while a grown up will sim­ply look at “3+2” and “5” would “mag­i­cally” pop up in her brain.

Now, the ques­tion is: can we in­ter­pret the child’s pro­cesses as a se­quence of Type 1 op­er­a­tions? The key op­er­a­tion here is count­ing, ev­ery­thing else can be eas­ily un­der­stood as Type 1 op­er­a­tions (for ex­am­ple, a con­nec­tion be­tween the writ­ten num­ber “3” and a pic­ture of three ap­ples can be un­der­stood as Type 1). What hap­pens in the child’s brain as he counts? As chil­dren we had to learn to count, prob­a­bly by just re­peat­ing the num­bers in or­der over and over again, to form a con­nec­tion be­tween them. After some prac­tice, the num­ber 1 form a con­nec­tion to 2, which is con­nected to 3, etc. in a linked list that ex­tends as we learn more num­bers. So, com­bin­ing this con­nec­tion, with a con­nec­tion be­tween a writ­ten num­ber and its lo­ca­tion in this list (3 is one el­e­ment higher than 2), a child can use Type 1 to count.

So, roughly and ab­stractly, a child’s brain adding 3+2 might go in a se­quence like this: the vi­sions of “3” would fire a pic­ture of 3 ap­ples (a younger child might need to perform a count­ing pat­tern to reach that step, which would also later mi­grate to Type 1), “2” would fire two ap­ples, a child then starts count­ing (each num­ber con­nected to the next, and the con­text of count­ing en­forces this con­nec­tion), cross­ing out each ap­ple with each fired num­ber, un­til all ap­ples are crossed out.

Now this in­tro­duces the fol­low­ing men­tal op­er­a­tion: vi­su­al­iz­ing ap­ples and perform­ing op­er­a­tions on this vi­sual image while count­ing (like cross­ing out or mark­ing each counted ap­ple). My wild guess here is that this, again, is re­ducible to Type 1 op­er­a­tions re­sult­ing from ba­sic teacher in­struc­tions on ad­di­tions, in­clud­ing vi­sual demon­stra­tions.

Levels of Type 1 to 2 Migration

Now, as pointed above, a younger child might need to ap­ply count­ing to con­vert “3” to an image of 3 ap­ples. As the child grows, she might have formed (by prac­tice) the di­rect grown-up pat­tern that trans­lates the image of “3+2” di­rectly to “5”. She will then use this to add a num­ber like 13+12 – uti­liz­ing “3+2”, “1+1+1”, and the carry 1 vi­sual pat­terns. So the child would ap­ply Type 2 ad­di­tion uti­liz­ing sev­eral skills re­cently mi­grated to Type 1. As the child grows up, more lay­ers of pro­cesses would mi­grate to Type 1, and the cur­rent Type 2 op­er­a­tions would be­come more effi­cient as they rely on those mi­grated skills.

So, what I am say­ing here, my guess, is that there is no clear dis­tinc­tion be­tween the two Types. That Type 2 op­er­a­tions are sim­ple those that use a large num­ber of Type 1 steps, and hence is slower, non-au­to­matic (as they are slow, there is more time for other pro­cesses to stop them from com­plet­ing, and hence they seem to be con­trol­led), and effort­ful.

Which con­nec­tion­ism pat­tern will be used

Now prob­a­bly a grown up still has all those ac­cu­mu­lated skills in place. See­ing “3+2”, I still have the abil­ity to ap­ply the ap­ple tech­nique, and also to ap­ply the di­rect con­nec­tion be­tween “3+2” and “5”. Which one I use, I sug­gest, is based on two prob­a­bly al­gorithms:

  1. Size: I use what I call the “Largest Available Rec­og­niz­able Pat­tern” (LARP). This means, how many pat­terns I need to in­voke to come to a re­sult. The brain then keeps in­vok­ing pat­terns from largest (less to­tal num­ber of pat­terns) to smaller, un­til a rea­son­able re­sult is reached

  2. Time: this is based on the quick­est pat­tern, which would usu­ally be equiv­a­lent to the largest.

And?

I to­tally con­fess that this is a wild guess, and an idea that is not at all fully de­vel­oped. I am not aware if this idea had been sug­gested in a more ma­ture way or not, so this is an at­tempt to mainly get feed­back and re­sources from you, and per­haps to build it up into bet­ter struc­ture.

The value of de­vel­op­ing such a the­ory is that at some point it can be testable, and per­haps bring a bet­ter un­der­stand­ing of how we learn new skills, and more effi­cient ways to ac­quire and de­velop our skills.