Instrumental Rationality 4.1: Modeling Habits

[In­stru­men­tal Ra­tion­al­ity Se­quence 4.1/​7.]

[This is part 1 of a 3-part se­quence on habits, which is it­self part of the greater In­stru­men­tal Ra­tion­al­ity Se­quence. This was ini­tially one mon­strous ar­ti­cle; in the in­ter­ests of read­abil­ity, I’ve de­cided to split it into three es­says.]

Out­line:

The Habits 101 mini-se­quence is bro­ken up into 3 sec­tions:

  1. In­tro­duc­tion, Models, and Statis­tics:
    We cover a ba­sic model of how habits work, three of their prop­er­ties (in­sen­si­tivity to re­ward changes, in­de­pen­dence of in­ten­tions, and au­to­mat­ic­ity), and some clos­ing re­marks on base rates for ha­bit­u­a­tion.

  2. Tech­niques for Creat­ing Habits:
    We cover three tech­niques for habit cre­ation: Trig­ger Ac­tion Plan­ning (TAPs), Sys­tem­atic Plan­ning (which has three sub-tech­niques), and Scal­ing Up.

  3. Tech­niques for Break­ing Habits and Con­clu­sion:
    We cover two tech­niques for break­ing habits: Go­ing Up­stream and Sub­sti­tu­tion (which has two sub-tech­niques).

(The cur­rent ev­i­dence base has many more in­ter­ven­tions for form­ing habits than break­ing them, so that’s why there’s an asym­me­try be­tween parts 2 and 3. Also, this is prob­a­bly just a good thing to keep in mind, the fact that form­ing habits is eas­ier than break­ing ex­ist­ing ones.)

++++

In­tro­duc­tion:

Peo­ple, as the say­ing goes, are crea­tures of habit. Many of our ac­tions ev­ery day are re­peated of­ten, typ­i­cally with­out much thought.

This type of thoughtless­ness, though, isn’t nec­es­sar­ily bad. Habits can help re­duce cog­ni­tive load, al­low­ing us to get through the day. Imag­ine if you had to ex­plic­itly weigh the pros and cons of be­hav­ior like flush­ing the toi­let ev­ery sin­gle time you did them.

Mak­ing in-depth, rea­soned de­ci­sions all the time can be very costly in terms of both time and at­ten­tion. Habits al­low us to com­part­men­tal­ize cer­tain be­hav­iors, so that our en­ergy and fo­cus can move to other, per­haps more im­por­tant, things.

A fre­quent part of our lives, habits make up at least roughly 40% of our ev­ery­day ac­tivi­ties, which thus makes them a strong can­di­date to tar­get when we’re think­ing about be­hav­ior change *1.

Knowl­edge about ex­actly how habits work, how we can cre­ate new ones, and how we can re­move old ones thus seems very use­ful be­cause they rep­re­sent a way of pro­vid­ing benefit con­tin­u­ously over time. It’s a lit­tle like your time spent learn­ing in school: while the ac­tual knowl­edge may soon fade, the au­to­matic re­sponse pat­terns you de­velop can you serve you well for a life­time.

My goal here is twofold: One is to give an overview of the mechanisms be­hind habits, and two is to give ev­i­dence-backed con­crete tech­niques to af­fect them.

++++

The Stan­dard Habit Model:

Roughly speak­ing, the stan­dard defi­ni­tion of a habit is that of “an au­to­matic be­hav­ior that is cued by con­text from the situ­a­tion” *4 †1.

Ba­si­cally, habits can be thought of as your de­fault re­sponses to differ­ent situ­a­tions.

This model for habits is com­posed of two parts: the con­text cue and the re­sponse.

The con­text cue, also called the “trig­ger” or “situ­a­tion”, is what first kicks off the en­tire pro­cess. Con­text cues are typ­i­cally ex­ter­nal things in the en­vi­ron­ment, from peo­ple to sen­sory de­tails to pre­ced­ing ac­tions.

Once the cue oc­curs, the re­sponse is gen­er­ated.

The re­sponse, also called the “ac­tion”, is the be­hav­ior that fol­lows the cue. Re­sponses are typ­i­cally small, atomic ac­tions, but there is some re­search sug­gest­ing that a se­ries of ac­tions can be “chun­ked” to­gether into a unit that fol­lows from the re­sponse *5.

(How­ever, we also know that ac­tions which re­quire more thought and con­scious effort don’t be­come ha­bit­ual, even when re­peated in the same con­text *6. Thus, I’ll be recom­mend­ing sim­pler re­sponses when we get to cre­at­ing our own habits.)

This [Con­text cue] → [Re­sponse] model is the core of how habits work.

While this model might seem ob­vi­ous or sim­plis­tic, I’d like to stress the use­ful­ness of this defi­ni­tion. This model al­lows for much of what we in­tu­itively la­bel as “habits” to fit this tem­plate of [Con­text cue] → [Re­sponse].

Here are some ex­am­ples of how typ­i­cal habits fit un­der this model:

  1. Ring! Your alarm shakes you awake. In re­sponse, your arm slaps the alarm, hit­ting the Snooze but­ton, and you go back to sleep.
    [Con­text cue] Shrill sound of alarm go­ing off.
    [Re­sponse] Turn it off and go back to sleep.

  2. “Hey!” Some­one asks you “How are you do­ing?” and you in­stantly re­spond with “Good, you?”
    [Con­text cue] The words “How are are you do­ing?”
    [Re­sponse] Im­me­di­ately say­ing “Good, you?”

  3. Beep beep! You open the car door and get in­side. After step­ping into the car, your hands are already look­ing for the seat­belt.
    [Con­text cue] Open­ing the car door.
    [Re­sponse] Put­ting on the seat­belt.

More­over, I think this model is im­por­tant in that it stresses how much of our be­hav­ior isn’t di­rectly un­der our con­trol. It high­lights how habits can be seen as a way of out­sourc­ing our be­hav­ior to the en­vi­ron­ment.

There’s a very real sense in which the cen­tral point of con­trol shifts from in­ter­nal to ex­ter­nal.

++++

But how ex­actly do habits form in the first place?

In sim­plified terms, the mechanism be­hind how habits form looks like this:

  1. Perform an ac­tion that isn’t too com­plex.
    EX: You floss your teeth.

  2. Keep perform­ing the ac­tion in a sta­ble con­text.
    EX: You always floss your teeth af­ter brush­ing.

  3. Your brain be­gins to make as­so­ci­a­tions be­tween the con­text when the ac­tion is performed and the ac­tion it­self.
    EX: <Insert your brain ac­tivity here.>

  4. Con­tinue perform­ing the be­hav­ior fre­quently in the con­text.
    EX: Richard keeps up his floss­ing habit af­ter he brushes his teeth.

  5. Over time, the en­tire habit loop be­comes in­ter­nal­ized and largely au­to­matic.
    EX: Richard ends up with a floss­ing habit.

Much of the ac­tual com­plex­ity is in Step 3, where our brains are able to some­how store the in­for­ma­tion about both the con­text cue and re­sponse to­gether. There’s a ques­tion here of “How are habits ac­tu­ally stored in the brain?”

There are two sug­gested mechanisms for how this ac­tu­ally hap­pens: mo­ti­vated cu­ing and di­rect cu­ing *7.

The first mechanism, mo­ti­vated cu­ing, sug­gests that cer­tain cues can cause us to act be­cause we an­ti­ci­pate a re­ward as a re­sult of our ac­tions.

Thus, the cue it­self brings to mind a sense of “de­sired­ness” which leads us to act.

A good ex­am­ple is no­tifi­ca­tions on Face­book. Many peo­ple feel a nag­ging draw to click on the red no­tifi­ca­tion but­ton as soon as they see it, like a sort of men­tal itch they need to scratch. There’s a two-step phe­nomenon here, some­thing like [See Face­book no­tifi­ca­tion] → [Click on it].

Mo­ti­vated cu­ing says that this is be­cause past ex­pe­rience with the cue (i.e. the red no­tifi­ca­tion icon) has led to re­wards (i.e. in­for­ma­tion about on­line ac­tivity that in­volves you).

This means that one way habits could come about is by trig­ger­ing a mo­ti­va­tion to act when we ex­pe­rience just the situ­a­tion it­self (even if the ac­com­pa­ny­ing re­ward hasn’t shown up yet).

So it sounds plau­si­ble enough. But how do know some­thing like this is ac­tu­ally hap­pen­ing in the brain? One good piece of ev­i­dence is a clas­sic study in­volv­ing mon­keys and juice that you might have seen in a differ­ent con­text.

Here’s what hap­pened: We started with some mon­keys, some juice, and a light. We trained the mon­keys to push a lever when they saw the light flash, which would then re­ward them with the juice. When the mon­keys re­ceived juice, we saw a spike of brain ac­tivity (as we might have ex­pected).

Even­tu­ally, though, we saw the mon­keys’ brain ac­tivity shift. Rather than spik­ing when they re­ceived the juice, we be­gan to see the spike when they merely saw the light. In other words, the mon­keys seemed to re­act to the con­text cue that sig­naled the re­ward rather than the re­ward it­self *8.

Ba­si­cally, this means that one way habits could come about is by trig­ger­ing a mo­ti­va­tion to act when we ex­pe­rience cer­tain con­text cues.

How­ever, I think the mo­ti­vated cu­ing model is un­satis­fac­tory for sev­eral rea­sons. Many of our rou­tines don’t always have well-defined re­wards, like fold­ing laun­dry or dry­ing off with a towel. Ad­di­tion­ally, as we’ll see in the next sec­tion, re­wards seem to have an over­all nega­tive effect on habit for­ma­tion.

This is where the sec­ond model, di­rect cu­ing, comes in. Direct cu­ing sug­gests that when we perform the same ac­tions of­ten enough, one af­ter an­other, a link will form be­tween them.

For ex­am­ple, con­sis­tently putting on the seat­belt af­ter get­ting into the car can lead to an as­so­ci­a­tion form­ing as we chunk the two ac­tions in our brains, one af­ter an­other. Thus, once we get into the car, our next au­to­matic step is to look for the seat­belt.

Other sim­ple ac­tions are things like ty­ing our shoelaces right af­ter we put on our shoes or us­ing the same con­ver­sa­tion starters (EX: “Have you seen any­thing good lately?”) with the same friend again and again.

In di­rect cu­ing, the most im­por­tant con­sid­er­a­tion is rep­e­ti­tion.

But didn’t I say ear­lier that fre­quency wasn’t enough? And how does the brain know which ac­tions are the ones to chain to­gether to form habits any­way?

To an­swer the first one, I hope I’ve made clear that, while fre­quency or rep­e­ti­tion isn’t the whole story, it’s an im­por­tant piece. There are other im­por­tant fac­tors like a sta­ble con­text which also seem to im­por­tant to habit for­ma­tion.

As for the sec­ond one, I’m ac­tu­ally not quite sure. All I know is that in pro­ce­du­ral mem­ory (the part of mem­ory re­spon­si­ble for our ac­tions) some­thing called Heb­bian learn­ing hap­pens. And Heb­bian learn­ing is a mechanism for neu­rons to link to­gether af­ter they fire in se­quence *9.

So this is a rough idea of how linked ac­tions or thoughts could form habits in the brain us­ing the di­rect cu­ing model. It’s definitely on shak­ier ground than mo­ti­vated cu­ing.

How­ever, I think that di­rect cu­ing still seems a lit­tle more plau­si­ble be­cause it’s able to (sort of) model a greater range of habit for­ma­tion, while mo­ti­vated cu­ing is stuck to a stric­ter type of habit.

(Ul­ti­mately, I just want to stress that I am not an ex­pert in neu­ro­science. Please take both ex­pla­na­tions as my hum­ble at­tempt at a plain English trans­la­tion. In ad­di­tion to sim­plifi­ca­tions I made to aid in ex­pla­na­tion, I’m sure I also made a few straight-up er­rors along the way.)

++++

Habit Prop­er­ties:

Apart from the Stan­dard Habit Model, there’s some ad­di­tional de­tail that I think is worth a deeper look. Habits have sev­eral in­ter­est­ing prop­er­ties that set them apart from other be­hav­iors. We’ll be go­ing over how habits are in­sen­si­tive to re­ward changes, in­de­pen­dent of in­ten­tions, and au­to­matic de­faults.

Insen­si­tivity to Re­ward Changes:

One prop­erty of habits is that they are largely in­sen­si­tive to re­ward changes, mean­ing the habit per­sists even when the re­wards are al­tered or re­moved *10.

As ev­i­dence, there was one study where we trained peo­ple to press a but­ton for a food re­ward. Yet, even when the re­ward was re­moved, we saw that they con­tinued to re­spond by press­ing the but­ton *11.

This gives some in­sight into why us­ing only re­wards and in­cen­tives usu­ally isn’t enough to change your habits. Once you’ve in­ter­nal­ized the habit loop, changes to the out­come don’t have much effect on al­ter­ing your be­hav­ior.

Hold on, though. Are all in­cen­tives re­ally worth­less? Surely peo­ple are more likely to act on cer­tain be­hav­iors if you pay them, right?

Well, we do see that fi­nan­cial in­cen­tives, as one ex­am­ple of a re­ward, ac­tu­ally are of­ten good at en­courag­ing short-term ac­tivi­ties. Com­pared to a con­trol group, we see that peo­ple who re­ceive in­cen­tives are more likely to perform the tar­get be­hav­ior.

But once the re­wards stop, the be­hav­ior of­ten does not ha­bit­u­ate *12. For ex­am­ple, a re­cent study with over 1,000 sub­jects ex­am­ined whether pay­ing peo­ple to go to the gym would lead to in­creased gym habits. Alas, we found that about two months af­ter the pay­ment stopped, the in­crease in be­hav­ioral fre­quency went back down to roughly pre-in­cen­tive lev­els *13.

As a fi­nal piece of ev­i­dence, we of­ten see that drug ad­dicts con­tinue to use sub­stances, even when such be­hav­ior turns self-de­struc­tive †2. This has also been repli­cated in an­i­mal stud­ies, where rats con­tinue to re­spond ha­bit­u­ally, even when the in­cen­tive is changed to a dis­in­cen­tive (EX: a poi­son) *14.

The take­away here is that sim­ply chang­ing re­wards isn’t enough to cre­ate or break habits.

++++

In­de­pen­dence of In­ten­tions:

A com­mon di­chotomy for be­hav­ior is that be­tween “goal-di­rected ac­tions” and “ha­bit­ual ac­tions”.

Our in­ten­tions, i.e. our de­sires and thoughts, do a good job of pre­dict­ing which goal-di­rected be­hav­iors we’ll carry out. Goal-di­rected be­hav­ior refers to the cat­e­gory of ac­tions where we act con­sciously on our prefer­ences.

On the flip side, ha­bit­ual be­hav­ior is quicker and largely un­af­fected by re­wards, as we cov­ered above *15. They are largely in­de­pen­dent of your in­ten­tions, mean­ing that they per­sist even if you de­sire for the habit to stop.

(Such a view has many par­allels to other dual-pro­cess the­o­ries like Sys­tem 1 and Sys­tem 2, pop­u­larized by Daniel Kah­ne­man in his phe­nom­e­nal book Think­ing Fast and Slow.)

In the ear­lier ex­am­ple with drug use, for ex­am­ple, it’s quite plau­si­ble to as­sume that the ad­dict would have liked to stop due to the nega­tive effects of con­tinued use, but still found it hard to quit.

We see that one’s in­ten­tions only af­fect be­hav­ior when the habit is weak. For ex­am­ple, peo­ple’s in­ten­tions to pur­chase fast food only af­fected their ac­tual pur­chases in the ab­sence of a strong habit *16. When habits be­come well-de­vel­oped, in­ten­tions mat­ter lit­tle.

On the flip side, sev­eral stud­ies have shown that in new, un­fa­mil­iar con­texts, our habits be­come dis­rupted, and our in­ten­tions once again be­come a stronger guide to be­hav­ior *17. We’ll ex­plore this dy­namic later in the form of a tech­nique called Cue Dis­rup­tion in to help with cre­at­ing and break­ing habits.

Thus, there’s a sort of in­verse re­la­tion­ship here be­tween one’s in­ten­tions and one’s habits.

Here’s a vi­sual:

When habits are not well-es­tab­lished, in­ten­tions are a strong de­ter­mi­nant of be­hav­ior. When habits are well-es­tab­lished, in­ten­tions be­come ir­rele­vant *18.

But the im­por­tant take­away here is that sim­ply “in­tend­ing” to change your habits isn’t enough.

++++

Au­to­matic De­faults:

The last im­por­tant prop­erty of habits is their au­to­mat­ic­ity. As al­luded to ear­lier, habits op­er­ate with­out much con­scious con­trol. Though they might be de­liber­ately over­rid­den, they are what we de­fault to in the ab­sence of cog­ni­tive effort.

In­deed, we see that when peo­ple be­come dis­tracted, and their willpower de­pleted, that their habits take over, even in situ­a­tions where a more rea­soned de­ci­sion might have been op­ti­mal *19 †3.

This also helps ex­plain why re­wards don’t do much to in­fluence ha­bit­ual be­hav­ior: In­cen­tives of­ten work best when they are de­liber­ately con­sid­ered, but the au­to­matic na­ture of habits means that they oc­cur with­out much de­liber­a­tion.

Thus habits (due to their au­to­mat­ic­ity) can be thought to by­pass the ex­plicit con­sid­er­a­tion pro­cess (which is also where re­wards hold the great­est weight).

The au­to­mat­ic­ity of habits is a dou­ble-edged sword:

If we’re not care­ful, habits can show up when we don’t want them to. Th­ese are referred to as “ac­tion slips”, and com­mon ex­am­ples are situ­a­tions like driv­ing along the same route to one’s work­place on a Sun­day or call­ing peo­ple “Dad” (even when they aren’t your father) *18.

On the other hand, habits can of­ten free our time and at­ten­tion to fo­cus on other things. It’s what al­lows our thoughts to drift even when we’re driv­ing on the road. Some­times, the best perfor­mance in an art or sport comes from this au­to­mat­ic­ity. In a bas­ket­ball game, for ex­am­ple, there’s lit­tle time to think over ev­ery shot; good form and ac­cu­racy must be au­to­matic.

++++

Habit Statis­tics:

Be­fore we dive into some ev­i­dence-based in­ter­ven­tions to cre­ate and break habits, I think it’s use­ful to quickly look at some base-rate statis­tics. To avoid get­ting any un­re­al­is­tic ex­pec­ta­tions on how long it’ll take for some of this stuff to work, I think some Refer­ence Class Fore­cast­ing is in or­der.

How long does it re­ally take to form a habit?

Well, it varies. One study of around 100 peo­ple go­ing to the gym found that about half of the peo­ple de­vel­oped habits in about 40-50 days *21. An ear­lier study us­ing a differ­ent method­ol­ogy (and a smaller pool of peo­ple) found that the in­di­vi­d­ual times varied very greatly, but the av­er­age time worked to be about 66 days *22.

(Stats for habit cre­ation were ac­tu­ally fairly hard to mind, so par­don that there’s only two sources for this sec­tion. Definitely happy to up­date if some­one finds ad­di­tional info)

Thus, for a con­ser­va­tive es­ti­mate, you want to scale your ex­pec­ta­tions to some­thing in the bal­l­park of about two months. That’s a long-ish time. Prob­a­bly longer than what most peo­ple naively es­ti­mate, I sus­pect.

Also, in the sec­ond study, we found that only about half of the peo­ple even formed habits at all. This means the 66 day figure only came from the peo­ple who even suc­ceeded in the first place.

This also means that if you tried to start a habit right now, your chances of stick­ing with a habit would only be about 50%.

But fear not! Start­ing with the next sec­tion, we’ll be go­ing over ob­vi­ous-but-still-use­ful-ad­vice clev­erly dis­guised as spe­cial tech­niques, to im­prove your chances. My goal is that, by the end of this Habits 101 primer, you’ll be far bet­ter in­formed and in a much bet­ter po­si­tion to han­dle things.

I can’t make any con­crete promises, but it does seem plau­si­ble to me that mak­ing a habit could per­haps be in­grained in about 30 days, if you take a fo­cused, in­ten­tional stance to­wards the en­deavor.

++++

Next Essay