Chain Breaking

post by Mark Xu (mark-xu) · 2020-12-29T01:06:04.122Z · LW · GW · 6 comments

This is a link post for https://markxu.com/chain-breaking

Contents

  Algorithm
  Examples
    Example: messenger
    Example: distracted while exercising
    Example: sleep
    Example: Facebook
    Example: morning phone usage
    Failed example: sleep
    Failed example: being late to meetings
    Failed example: distractions
  Virtue: Specificity
  Limitations
    Some chains cannot be specified
    False narratives
    Links are replaceable
None
6 comments

This is a rationality technique I’ve been experimenting with. Thank you to Jack Ryan, Thomas Kwa, Sydney Von Arx, Noa Nabeshima, and Kyle Scott for helping me refine the method.

Algorithm

1. Pick something that has happened before that you would prefer not to happen again.
   1. Examples include: not exercising, not flossing, not sleeping on time, eating too many chocolate almonds, spending too much time on Facebook, being late to meetings, forgetting to schedule things, etc.
   2. In theory, you could pick something that hasn’t happened yet that you want to prevent. This is a more advanced version of the technique. Try it at your own peril.
2. Why do you think it might happen again? Determine the specific chain of events that will lead to this undesirable outcome.
3. A chain is only as strong as its weakest link. Locate the link in the chain that seems most fragile.
4. Intervene to break that link (and thus the entire chain). If it isn’t easy to think of an intervention, that might mean you haven’t gotten specific enough. Recurse.

Examples

Many of these examples do not involve me. I wrote them all in the first person for easier comprehensibility.

Example: messenger

1. I use messenger in a way that disrupts my work routine.
2. When doing pomodoros, when I check messenger during breaks. There are often many messages to respond to, some of which require thought. Even worse, sometimes people respond to my responses in almost real-time. Since I do not like not responding to messages, I respond to all messages, nearly always taking more than the five minutes allotted for my break.
   1. More compactly: break -> messenger -> many unread messages -> respond
3. Once I look at my messages, the compulsion is very strong to respond to all of them. In particular, if I read them, but don’t respond, I might forget to respond. I could make this easier by keeping track of which messages to respond to, but that seems too tedious. Instead, I think violating break -> messenger is easier.
4. Intervention: don’t check messenger during breaks.

Example: distracted while exercising

1. I often don’t finish my exercise routine in the time allotted.
2. When exercising, I have time in-between sets to rest. I often perform miscellaneous activities during these times, such as read articles on my phone. This causes me to lose track of time, so the next set is delayed. This both causes my exercise routine to take too long and reduces its efficacy.
   1. More compactly: start exercising -> down time -> activity -> distracted -> fail to do next set in time
3. It seems difficult to eliminate downtime during exercise. If I’m distracted, I will certainly fail to do the next set. Thus either I should stop performing activities during downtime or stop doing activities that are potentially distracting. The latter would require the activities to be capped at approximately 1 minute (the amount I rest between sets), which seems infeasible. This suggests that I should stop doing activities during downtime.
4. I notice that I’m not sure how to stop myself from doing activities during downtime. Recursing…
   1. During downtime in my exercise routine, I often perform activities that are distracting.
   2. When I’m done with a set, I’ll often pick up my phone and open apps until something interests me. These apps are most often messenger or various reading apps.
      1. More compactly: down time -> pick up phone -> activity
   3. It seems like a crucial node in the chain is my phone.
   4. Intervention: preventing my phone from having access to interesting apps while I’m exercising seems like it solves the problem. This can be done with the app Lock Me Out for android.^[1]

Example: sleep

1. Sometimes I sleep too late.
2. Normally I sleep at around 11pm, which is desirable. Sometimes, however, I will be doing something at around 10pm which will prevent me from beginning my bedtime routine. This something nearly always involves an electronic device, like my phone, laptop, or iPad.
   1. More compactly: 10 pm -> doing something -> don’t begin bedtime routine
3. Beginning my bedtime routine while doing something is not feasible. Thus the weakest link is me doing something at 10pm.
4. Intervention: use a combination of apple’s screen time and Lock Me Out to make me unable to do things past 10pm

Example: Facebook

1. I use Facebook too much.
2. I sometimes get slightly bored when using my computer, so press ctrl+t to open a new tab and start typing fa-. My browser autocompletes to facebook.com, then I press enter. My browser then takes me to Facebook. I have 12 notifications, so I click the notification button to check if any of them are interesting. None of them are, usually, so I look at my newsfeed. After reading everything on the first screen, I scroll down. Since the thing I see when I scroll down is interesting, I continue scrolling. This happens for the next 30 minutes.
   1. More compactly: bored -> ctrl+t -> fa- -> autocomplete -> facebook.com -> Facebook -> notifications -> newsfeed -> scrolling
3. Once I am on Facebook, I have already lost. Can’t really prevent myself from being bored, so preventing facebook.com -> Facebook seems feasible.
4. Intervention: use Freedom to block Facebook.

Example: morning phone usage

1. I use my phone immediately when I wake up.
2. I’m usually on the phone lying in bed. Then the phone shuts off automatically, so I put my phone on my desk and go to bed. When I wake up, my phone is right there and I have notifications so I look at them.
   1. More compactly: on phone in bed -> phone shuts off -> put on desk -> got to bed -> wake up -> notifications -> use phone
3. It seems like my phone just ends up wherever I happen to be when it shuts off.
4. Interventions: close to bedtime, use my phone on the couch instead of in bed.

Failed example: sleep

1. I frequently sleep after 2am.
2. I am frequently caught up doing a distracting thing, so I don’t take a shower until my computer automatically shuts itself off at 11pm. After I shower, it’s 11:40pm and my hair has to dry afterward. I’m also often hungry because it’s late, so I made some food. At this point, I’m often more bored than sleepy, so I engage in some other activity until I become more sleepy, which happens after 2am.
   1. More compactly: distracting thing -> computer turns off at 11pm -> take a shower -> hair has to dry -> hungry -> make food -> more bored than sleepy -> do activity -> after 2am
3. I notice there are many parts of this chain that can be violated without fixing the problem. For instance, if I wasn’t hungry at night, the problem would likely remain. Also, being more bored than sleepy at night is caused by not having enough stimulation during the day. It’s not clear which parts of this chain can be violated easily and also solve the problem.
   1. In particular, “bored” and “sleepy” are mental states that are already tangled in huge webs of cause and effect. It’s not clear how to intervene to make myself more or less bored at night.
4. Intervention: ???

The problem with this example is that the causal chain involved very complicated processes, so even though one could try and construct a causal chain leading to the undesirable event, there was still opacity that made it difficult to locate an intervention

Lesson: sometimes causal chains involve nodes like mental states, which are extremely difficult to get more specific about.

Failed example: being late to meetings

Warning: slightly fictional, but am confident this general pattern exists.

1. I frequently am late to meetings.
2. It’s pretty important to me to not be late for meetings, yet I’m often late for them because I don’t really try to be on time. In particular, I never feel a sense of urgency when going to meetings, so I’m about five minutes late every time.
   1. More compactly: meeting coming up -> fail to try -> am late
3. Can’t really just not have meetings, so the only place to break the chain is when I fail to try.
4. Intervention: resolve to Try Harder [LW · GW]!

This example might not be a failed example if trying harder actually ends up working. In my experience, however, resolving to try harder almost never results in changing the state of the world. It might work initially but will likely end up lapsing. In this case, the example fails because the causal chain does not reflect the true dynamics at play. A better causal chain might look like: meeting coming up -> don’t realize until one minute before -> am late, which might suggest an intervention of increasing the amount of forewarning one’s calendar gives.

Lesson: inaccurate causal chains make interventions ineffective.

Failed example: distractions

1. I often get distracted during the time I’ve set aside to do work.
2. During a pomodoro, I sometimes get bored and start aimlessly reading Wikipedia. This causes me to get distracted from work, wasting many minutes of time.
   1. More concretely: pomodoro -> bored -> wikipedia -> distracted
3. Only ever doing non-boring work would be nice, but doesn’t seem feasible. The easiest implication to violate is bored -> Wikipedia. However, getting distracted doesn’t depend on Wikipedia. If Wikipedia didn’t exist, I would just find some other website to browse. If I didn’t have internet access, I would just find some other activity in the real world.
4. Intervention: ???

This example fails because there are many many paths to getting distracted. Writing down a single causal chain will almost certainly fail to capture the complete picture. Thus, even if it seems the chain-as-written suggests an easy intervention, in practice said intervention will only end up shifting reality to a new chain. This is akin to a programmer adding if/then statements to their code so they can pass specific test cases. Focusing [? · GW] on the feeling of boredom might be a productive next step, but it is more difficult.

Lesson: sometimes the easiest link to break is also the most redundant.

Virtue: Specificity

Many problems disappear when you get specific enough [LW · GW]. For example, Murphyjitsu [? · GW] can point out specific failure modes for your plan; pre-hindsight might let me know that my plan to “book a plane ticket” might fail because I have an aversion to using the clunky interface and will constantly delay the task. Most of the time, however, it is very difficult to be specific.

However, a problem with many rationality techniques is that they benefit from specificity, but do not make it easy to be specific. This is because they ask you to consider hypotheticals. Murphyjitsu asks you to imagine your plan failed. Goal factoring asks you to speculate on possible goals an action might help you achieve. TAPs work better the more specific they are, but still ask you to come up with a specific TAP on your own. Noticing asks you to notice… what exactly?

Chain breaking has the benefit of having the bulk of the work be writing down a causal chain for something that has actually happened. It is much easier to be specific about things that actually happen. Imagine a keyboard. How many letters are between Y and P? Now look at your keyboard. How many? It’s much easier with your keyboard right in front of you; you can just count!

Similarly, while it might be difficult to imagine a plan that you haven’t done yet, it should be easy to imagine something that has happened to you many times. For instance, I find it very easy to imagine myself eating too many chocolate almonds; I just have to recall what happened 25 minutes ago. I was using my computer and I got slightly hungry. Since there is a jar of chocolate almonds conveniently within reach, I mindlessly grabbed it. I then proceeded to eat multiple handfuls.

Jerry Cleaver said: “What does you in is not failure to apply some high-level, intricate, complicated technique. It’s overlooking the basics. Not keeping your eye on the ball.” It’s much easier to keep your eye on the ball when it’s right in front of you.

Limitations

As demonstrated by the failed examples, this technique has limitations.

Some chains cannot be specified

When attempting to determine that chain of events that leads to an outcome, sometimes the chain involves forces that you do not have a white box understanding of. For example, sometimes I might get distracted “because I’m tired” or fail to exercise because “I didn’t have enough energy.” It is unclear what is meant by “tiredness” or “energy”, nor is it clear how to decrease tiredness and increase energy.

In these situations, it is very difficult to get more specific. One way around this is to attempt to specify other parts of the causal chain and intervene in those places. For example, tired -> distraction could be further specific into tired -> go on Reddit -> distracted, suggesting an intervention involving blocking Reddit. If you can’t identify an intervention, further specifying any part of your causal chain is likely going to be useful.

However, there are going to be some situations where the true causal chain is almost entirely controlled by vague concepts. In these situations, chain breaking might not have any insights to offer. One possibility is to combine this with post-hoc noticing [? · GW], where you realize you’ve been on Reddit for 30 minutes and tries say/write/think all of the things you can remember about the previous 30 minutes, like “I felt a vague interest in what was going on in the world” and “there was no clear stopping point so I never stopped.” The hope is that diligent practice will shorten the lag time until you’re able to notice the relevant phenomenon in real-time.

False narratives

One benefit of chain breaking is the ease of specificity; one flaw is the existence of hindsight bias [? · GW]. When attempting to describe why something happened to you in the past, it’s easy to generate fake explanations [LW · GW]. For instance, I might be late for a meeting and conclude that it was because I didn’t remember and that my memory was poor because I drink too much alcohol. ^[2] A more realistic example might be the adoption of bird augury [LW · GW], which might have been because of false causal narratives of various bird sightings leading to increased rice yield.

In general, it’s much easier to convince yourself that you have understood than developing an actual understanding. The truth is likely that most phenomena are not even predictable in retrospect. Asking yourself “what happened?” and getting the correct answer is a non-trivial feat. For example, yesterday my body failed before I was able to complete my exercise routine. What happened? I am not sure. Maybe I ate poorly or slept poorly. One particularly tempting narrative is that I overexerted myself on Monday and hadn’t recovered yet, but I suspect that contains less truth than falsehood.

For this reason, I recommend using chain breaking on events that have happened many times in the past. If something has happened to you many times, you will have a much better sense of the various dynamics at play. For example, I had gotten distracted while exercising ~7 times before I pinned down my phone as the actual cause. Other plausible narratives included just needing longer recovery times or exercising on too low energy.

Links are replaceable

When thinking about various goals, I find that a useful question to ask oneself is “how far is this goal from the default outcome?” If my goal is to not get into a car crash, since I don’t make a habit of getting into car crashes, this goal is pretty close to the default. However, if my goal is to write 10,000 words in a day, since I usually write much less than 10,000 words in a day, this goal is pretty far from the default. One way of thinking about systemization [LW · GW] is as making achieving various goals closer to the default outcome.

Chain breaking is far more useful when it’s trying to answer the question of “why did the default outcome not happen?” as opposed to “why did I fail to prevent the default?” For instance, if I usually don’t exercise, the question “why didn’t I exercise?” is likely to have many more answers than if I do usually exercise. If I usually turn in assignments on time, me turning in an assignment late is much more likely to contain a specific causal reason for the lateness than if lateness was the default.

Default outcomes usually happen because of many redundant causal links. Outcomes that different from the default usually happen because of specific causal reasons. Trying to break a redundant causal chain is much harder than breaking a unique one. Trying to prevent yourself from pulling all-nighters is much easier than trying to prevent yourself from sleeping.

There is some tension between wanting to focus on non-default outcomes and also wanting to focus on things that have happened many times. Balancing this tension carefully is choosing a problem that you understand well enough to solve, happens frequently enough to make a solution useful, and is far enough from being a robust default that it is tractable. For instance, about once a year I get really unproductive for a week or two. Solving this problem would be great, but I don’t understand it well enough to ensure a robust solution. On the other hand, I lost about 30 minutes every day due to small transition costs between tasks or work/break cycles. It would be great if I could reclaim this time, but it happens so robustly that it is not very tractable.

---

1. I advise caution about solving problems via acting to prevent yourself from getting what you want. These sorts of strategies work for me because I actually do want to exercise, but sometimes just forget. Blocking my phone helps me remember. ↩︎

2. This is fictional. Coming up with an example of this proved too difficult. Wittgenstein: “If there were a verb meaning ‘to believe falsely,’ it would not have any significant first person, present indicative.” ↩︎

6 comments

Comments sorted by top scores.