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:

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.
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.
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:

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.
“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?”
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:

Perform an action that isn’t too complex.
EX: You floss your teeth.
Keep performing the action in a stable context.
EX: You always floss your teeth after brushing.
Your brain begins to make associations between the context when the action is performed and the action itself.
EX: <Insert your brain activity here.>
Continue performing the behavior frequently in the context.
EX: Richard keeps up his flossing habit after he brushes his teeth.
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.

++++

Instrumental Rationality 4.1: Modeling Habits

Outline:

Introduction:

The Standard Habit Model:

Habit Properties:

Insensitivity to Reward Changes:

Independence of Intentions:

Automatic Defaults:

Habit Statistics:

Next Essay