Hello readers! It’s been a bit since my last blog post for Halfhaven, but I don’t want to fail with abandon, so I’m continuing to write! I’m currently participating in a Center For Applied Rationality (CFAR) workshop, thanks to a recommendation from my friend Ronny. Today was the first full day of the workshop, and that’s what I’ll be writing about this evening. Session-by-session, here’s what’s happened so far.

Orientation

The other participants and I, about 20 in all, arrived to the venue yesterday evening. After being greeted with a hot meal catered by a local restaurant, we settled into the cozy living room. The opening session consisted of several components, with different staff members trading off leadership and introducing themselves in the process.

First, we read two poems, line by line one person at a time, in order to “bring our voices into the space”. After reading poems, we played nameball: we passed a small ball around the room, and each time a person caught it, everyone who knew that person’s name said it together. I liked this way to play; it brought in nice elements of spaced repetition and social learning.

Having introduced ourselves, we were given a history of CFAR as an organization, and of how its conception of “rationality” has evolved over time. What started as a model of people with static goals trying to form true beliefs in order to achieve them, has evolved to treat goals as dynamic and incorporate a sense of Or "self-authorship", "pride", "ego", "spark", on the theory that if you don't have a good term for a concept, try using several.

There was a brief talk titled “TIME”, situating the workshop in the span of life, the orientation in the span of the workshop, and the current talk in the span of the orientation. After that, three different instructors each presented a tip/request for our engagement with the workshop:

  1. Be A Hobbyist: pursue our interests earnestly and deeply
  2. Respect Authorship: understand and encourage the perspectives/stories of ourselves and each other
  3. Try Things: have lots of experiences, so we can find and do more of the ones we like

A theme of the workshop is that this is the Center For Applied Rationality, so we do active practice rather than just listen to lectures. To this end, we were asked to write down “bug lists”: problems or areas for improvement in our lives. Later sessions would pull from these bug lists for examples of targets for the techniques being taught. My bugs are somewhat abstract, which works well for practicing some techniques and presents difficulties for others.

Finally, we went over some operational details for the workshop, several of which were about protecting the welfare of the two adorable kittens at the venue.

Kittens!

After a bit of freeform chatting with the other participants, I went to bed, earlier than my usual to accommodate an 8:30 wakeup for coffee and breakfast before the start of classes at 9:00 sharp!

Goal Factoring

The class starts with a hypothetical. Imagine you wake up one morning with a desperate craving to eat an orange. You make your way to the grocery store and reach to grab the last orange there, but just as you do, another hand grabs for it as well. It seems the stranger wants that orange just as desperately as you; what to do? Well, in this toy example, you can talk to the stranger and learn that they aren’t desperate to eat the orange. They’re desperate to make mulled wine, and just need the peel. You can both get what you want!

When you consider an action and wrestle with tradeoffs, goal factoring asks: how can you get all of the good? The technique of goal factoring is to break down what you want to get from an action, and then consider how each of those goods might be achieved independently with alternative actions. It may well be that if you consider why you want to do the action in more granularity, you’ll find that you don’t need to make costly tradeoffs in order to get everything you wanted from it!

Some technical details to keep in mind for goal factoring effectively:

  1. Make sure that the goals you factor out of your action are relevant, resonant, and complete.
    • Relevant: if a goal you factor doesn’t seem like it’s actually important to you, strike it from the list! You don’t need to come up with alternative actions to achieve goals you don’t care about.
    • Resonant: does the goal feel like something the proposed action was actually going to accomplish? If not, that goal may be important to you and worth noting separately, but it doesn’t need to be part of the replacements for this action.
    • Complete: if you imagine satisfying all of the goals you wrote down, and you still feel resistance to dropping the proposed action, you probably haven’t factored out all of the goals. A successful goal factoring should feel like a way to get everything you wanted from an action, not like you’re begrudgingly giving something up.
  2. Focus on positive goals: what you want, not what to avoid. You can write down an action that achieves a positive; it’s much harder to come up with an action that on its own avoids a negative. Also, this accords with the general heuristic to focus your attention on what you want to see more of.

For the applied portion, I did a dual goal factoring on “get a job”/“stay self-employed”. I didn’t get very far into it before the session ended, with an incomplete factoring and very few ideas for alternative actions, but it was an interesting exercise and I may continue it later.

Trigger Action Plans

Trigger Action Or Trigger Action Patterns, when constructed less deliberately. (TAPs) were presented as “assembly code for the mind”. TAPs are the low-level habitual nudges that make up our day-to-day decision-making. Much as an amoeba follows chemical gradients towards food and away from toxins, we follow our TAPs.

As the name suggests, a TAP consists of a trigger and an action, each consisting of a single mental chunk of complexity. For instance, “a person sits next to me” -> “I move my things to make space” or “I get out of bed” -> “I brush my teeth”. Both the trigger and the action are concrete and immediate, happening at the level of habitual action, not reasoned planning. Note that this implies that by chunking more complex concepts, we can form more effective TAPs!

By understanding TAPs and being more deliberate about which TAPs we install in our minds, we can harness their power in several ways:

  • to form a routine, set up a consistent trigger for the action we want
  • upon seeing someone do a For example, intuitively choose just the right tool for a carpentry task. figure out the TAP that led them to it
  • to uninstall a harmful TAP, install a new TAP with the same trigger to replace it, rather than just try to not do the action
  • when something goes poorly, come up with a TAP to prevent the failure modes in similar situations
  • install TAPs to coordinate with yourself across different cognitive contexts

I really like this concept and I’ll add it to my collection of useful jargon going forward, but I struggled to use it effectively during the applied portion of the lesson.

Inner Simulator

Each of us carries powerful mental machinery for simulating the world. We have a sense of what would surprise us, and we can form detailed “mental movies” when we imagine hypotheticals. By imagining the future, we can use “pre-hindsight” to get a detailed story of how things might go. This inner simulator can be very useful for making effective plans!

The technique of planning with the inner simulator, cheekily named “Murphy-jitsu”, runs as follows:

  1. Select a goal you want to achieve.
  2. Write out a plan. This should be a list of concrete steps that you can visualize in detail.
  3. Imagine that in the future, you learn that your plan failed. Are you utterly shocked to learn this? If so, you’re done!
  4. Given that your plan failed, use pre-hindsight to figure out what went wrong. This is what the inner simulator immediately visualizes having gone wrong when you imagine learning that the plan failed; your job at this step is not to brainstorm a list of possible failures.
  5. Come up with a way to bulletproof your plan against the failure you imagined, so that it no longer feels plausible to imagine.
  6. Return to step 3.

I really like this technique! In the practical portion, we paired off to plan together, and with support from my friend Ricki, I used Murphy-jitsu on my goal of “publish a blog post today”.

The Plan
  1. Open my laptop.
  2. Open the Cursor window with my blog and make a new file.
  3. Name the post “CFAR Workshop: Day One”.
  4. Look at the schedule printout and my notes to remind myself of what happened.
  5. Write section headers corresponding to class titles.
  6. Under each header, describe what I learned in that class.
  7. Embellish the post with context about why I’m here and any feelings I have about the lessons.
  8. Read my draft and have Ricki read it as well.
  9. If Ricki suggests edits, consider whether I agree, and make the edits if so.
  10. Have ChatGPT check spelling/grammar, and fix any issues.
  11. Commit and push the post to GitHub.

I found that much of the value of the technique was found in listing out these concrete steps! Breaking down my vague intention to blog into visualizable actions made me feel much better about my chances of succeeding, before I even started bulletproofing the plan. As it happened, the session ended before I could finish the iterative bulletproofing process, but I did simulate a few points of possible failure.

Bulletproofing
  1. I might fail to open my laptop in the first place. To prevent this, I set an alarm in the evening telling me to commence my plan, and Ricki set a slightly later alarm to remind me.
  2. I might get stuck on filling out the sections. In that case, instead of writing what I learned, I’d just write down literal events that happened in the classes. If I still felt stuck, I’d record myself talking to someone about the classes, and then turn the transcript into a first draft.
  3. I imagined failing to come up with any embellishment, but didn’t have time to bulletproof against this.
  4. Having run out of time, I also didn’t consider a hitch that my plan has in fact run into: this post is taking a while to write, and Ricki has gone to bed, so I can’t run the draft by her! However, this step isn’t actually critical for succeeding at the goal of “publish a blog post today”, so I continue undaunted!

Reference Class Forecasting

There’s a well-established cognitive bias known as the planning fallacy; people reliably underestimate the duration and budget a project will take. To counter this, we can use a technique called “reference class forecasting”. In reference class forecasting, rather than imagine the details of a project, we look for a “reference class” of similar projects and use their outcomes to guide our estimates. Reference class forecasting involves a few steps:

  1. Find a reference class. Look for similar projects that have been done before. The more similar to your project, the better! The gold standard here is a routine task you have personally done many times.
  2. Get data from the reference class. This might involve looking up statistics, or just asking people who have done similar things how long they took.
  3. Assume your project will have similar outcomes. For large-scale, important projects, you can do serious statistics and get information like mean, median, and N-percentile time to completion, but usually this is overkill. If you have a very good reason to think you are a special case, you can apply a corrective factor, but remember that others in your reference class may have also thought they were built different and fallen prey to the planning fallacy.

While reference class forecasting was presented in the context of project planning, it was noted that it can also be used for other types of predictions.

Reference class forecasting can fail when:

  • an event is a one-off; similar things haven’t happened before
  • it’s unclear what the proper reference class is for an event
  • data about outcomes in the reference class is secret or otherwise unavailable
  • your reference class is fictional; stories are biased by narrative demands and don’t present an accurate picture of reality

I found it interesting that reference class forecasting is, in some sense, the opposite of the inner simulator technique. One says to ignore your specific details, the other says to drill into them. Each is useful for its purpose; inner simulator for making solid plans for action, and reference class forecasting for predicting outcomes even when you don’t know much about the details.

On top of the well-established technique of reference class forecasting, CFAR has developed a more experimental technique called reference class hopping. If you use reference class forecasting and find that your predicted outcomes aren’t great, what can you do? Take action to shift your reference class! For instance, move from the reference class of “people who take a test” to “people who prepare extensively for a test”. This feels pretty common-sensical to me; stepping away from the jargon, this sounds like “do things that make you more likely to succeed”.

Multiple Hypothesis Testing

We went for a walk on the beach and were asked to pair up, notice something, and form multiple hypotheses about it. My partner and I noticed that some of the logs on the beach had flat ends. Some hypotheses:

  1. Some natural process flattens the ends of driftwood.
  2. Some industrial process releases cut logs into a river.
  3. People bring logs to the beach to use as benches.
  4. Lumber falls off of ships transporting it.

I don’t know that I learned much from this session, but it was pleasant to get out in nature and walk around and observe things!

Who-Friending

This session was pulled straight from How To Win Friends And Influence People, with the twist that the techniques of being kind, appreciating and pointing out people’s contributions, and validating people’s pride, were applied internally as well as interpersonally. More could perhaps be said about this session, but frankly I’m very tired and want to wrap up writing this post. Be nice to people!

Dinner and Pair Projects

Dinner was provided by an “artistic chef” from Chicago, and quite tasty. I chatted with him a bit and I’m looking forward to getting to know him better, given my own interest in cooking! After dinner, we paired up to work on projects. Most worked on physical projects, including disassembling a trove of old electronics that CFAR provided for educational purposes, but Ricki and I both wanted to work on the digital projects that we had planned earlier: she on some web development for Arbor Trading Bootcamp, and I on this very blog post. Writing this has kept me occupied since then, so that’s all I have to say for now. Time to get some rest, as I’m sure the second day will be just as packed with learning!