A Concrete Multi-Step Variant of Double Crux I Have Used Successfully

post by deluks917 · 2018-03-15T01:26:16.028Z · score: 46 (11 votes) · LW · GW · 4 comments

=== Intro

Raymond Arnold has written about the failure of many discussions to result in clear progress recognizable by both sides. In general, the ideas around double crux are useful but the technique can be under-specified (especially for people who have not been to CFAR). Additionally, progress can be difficult to achieve until participants have a good understanding of the other person’s relevant mental models.

=== Step by Step Instructions

1- Both people independently write down their positions and which arguments and pieces of evidence they find the most convincing. Split this into claims/cruxes. Each claim should have a bullet point list of reasoning. If possible, give an estimate of how convinced you are by this claim. These claims should be sufficiently connected with your overall model, such that if you felt the claim was actually false you would update your views a substantial amount. Admittedly lists of claims is not a great model of how people think, but it’s helpful to try to make your views ‘cruxy’.

2 - Both people exchange write ups.

3 - After reading the other person’s write up, the participants agree on a list of ~5 topics to discuss in more detail. You want to choose topics which can be fruitfully discussed in ten to twenty minutes. This can be done via a short email exchange or via discord/irc/etc.

4 - The two people discuss the above list of topics for 1-2 hours. The main goals are to understand the confusing parts of your opponent’s argument and resolve contentious points. However, once it becomes clear a contentious point is hard to resolve, move on for now. You only have 1-2 hours; don’t waste them arguing what ‘true’ means, find fertile ground. This needs to be done in person or via real time chat; email will not work.

5 - Try to find explicit points of agreement. Find a list of non-trivial points that both people can explicitly endorse. For example myself and Jacob ‘putanumonit’ Falkovich double cruxed about which people would benefit from a CFAR workshop. In the end we agreed the following list of factors were predictive:

-- You find the CFAR teachers (and to a lesser degree alumni) impressive and would like to be more like them in some important ways

-- You would have a good culture fit among CFAR attendees

-- You find rationality techniques relatively cool even if you have trouble doing them consistently

-- Some sense that you might benefit from an attitude that 'problems are solvable'

The fourth factor relies on the first three since these make it much more likely CFAR will successfully convince you of the benefits of this mindset

6 - Make a list of hard to resolve cruxes/disagreements that came up in step 4. For each crux, discuss what plausible evidence would cause you to update, even if the update would not be huge.

=== Conditions for the Technique To Work:

1 - You need to respect each other intellectually and trust the other person to ‘actually change their mind’ .

2 - Both people need to feel safe to express their actual reasons and beliefs.

3 - Both people need to have thought about the issue in some depth. If one or both participants are new to the topic than a different technique is likely to be more productive.

4 - It’s good if both people feel somewhat surprised the other person doesn’t already agree with them (at least in outline). A more extreme level of surprise might suggest too much inferential distance.

5 - It’s important to find fruitful territory to explore with your partner, but it’s easy to get stuck on a prickly crux for 2 hours and make no progress.


Comments sorted by top scores.

comment by Jacobian · 2018-03-15T16:50:37.216Z · score: 22 (5 votes) · LW · GW

As the second crux to deluks917 (are you using your real name on LW?) in the example he described I want to add a couple of observations to this very good write-up:

1. Exchanging emails ahead of time is really important both for having a list of topics to cover and also a list of topics to ignore. For example, in the email deluks brought up the topic of how CFAR spends its budget. In the middle of a discussion, this may have felt to me like an unfair attempt to attack my pro-CFAR position with flanking action. Over email, it was much easier to calmly realize that I know very little about the subject and it has little implication for the actual question. We both just agreed to ignore it, and it didn't come up in the discussion.

2. Throughout the discussion I felt like it was really a debate with some double crux forced on the end. In retrospect I think that this is the way it should feel. In a debate I am challenging the other person for their best arguments, which I probably won't find immediately convincing. However, I can then ask myself "what should be added to their argument that would convince me?" I can start with answering all of those with "if Metatron stepped down from heaven to confirm it" and then adjust over time to something more reasonable.

Basically, having to "defend" against an argument is the way to find out where your defense is weakest and which "crux" would bring it down crashing.

3. Both #1 and #2 require a high level of trust and good faith. The fact that we did the discussion publicly as a specific "double crux demonstration" probably helped keep us both at our most charitable. Perhaps having observers who expect a double crux is important.

comment by Said Achmiz (SaidAchmiz) · 2018-03-15T01:45:25.780Z · score: 22 (5 votes) · LW · GW

Thanks for writing this up!

I do have some questions. First:

You want to choose topics which can be fruitfully discussed in ten to twenty minutes.

It hasn’t been my experience that any topic of any real import or depth can be fruitfully discussed in ten to twenty minutes. Can you give some real-world examples of topics that fit this criterion?

Second: how do you deal with the fact that parts of a person’s view on any given topic are often hard to disentangle from one another? And, separately (but relatedly to the first part): suppose we pick a topic to discuss for ten minutes, but find that our disagreement about it stems from deeper differences, and lead to other topics, etc. How to deal with such cases (which seem to me to be quite common)?

comment by deluks917 · 2018-03-15T03:40:04.554Z · score: 18 (4 votes) · LW · GW

Its perhaps best to give concrete examples. This is roughly the set of topics we discussed in the CFAR Double Crux. It seemed possible to discuss these frutifully in a small length of time. (I didn't put this in the article because I really want to avoid people making this thread into a discussion of CFAR's value)

-- How much learning in person being qualitatively different from learning via blogs

-- Whether its an isolated demand for rigor to judge CFAR relative to the best self improvement techniques you could use (since few people use those). Is CFAR competitive relative to things like 'many hours of private tutoring'.

-- How much of the value of CFAR comes form the curriculum and how much comes from the Alumni Network and Events

-- What percentage of one's money and effort should people generally be spending on self improvement

-- How skeptical should we be about the CFAR studies

You should ask to discuss topics where either agreement seems relatively easy to reach or you want to get more details from your partner. You might want details because they have info you don't, because you want to ask some key questions or because you want to hear their model explained at length. If a disagreement about a crux seems especially deep you should add it to the list of 'prickly cruxes' and handle it in step6.

In general the goal is to make progress and find non-obvious statements you can both agree to. This requires focsing the discussion on areas where progress is likely. In the case of the CFAR doucle crux I felt I learned alot and got a much better model of which people will benefit from CFAR. The goal is to make genuine mutually agreed upon headway not nescessarily to resolve all disagreement.

comment by Raemon · 2018-03-18T07:51:22.531Z · score: 6 (1 votes) · LW · GW

Just wanted to add another "thanks for writing this up" – I think both having more variations on systems-for-Doublecrux will probably make it more useful in a wider variety of contexts, and having more examples of Doublecrux will help illustrate the underlying dance of ideas that it's trying to get at.