You can rack up massive amounts of data quickly by asking questions to all your friends

post by Neil (neil-warren) · 2024-01-21T01:27:21.701Z · LW · GW · 2 comments

Contents

2 comments

tl;dr I asked 15 friends and acquaintances a question out of the blue because I was curious about something. I collected a massive amount of evidence, and it was fun, and I’m more agentic now than I was yesterday.  Do it too. 

Last night I randomly remembered that I had once asked a friend whether they would one-box or two-box in Newcomb’s paradox (which they had very little chance knowing about). I decided to ask another friend, and sent them a message. And then I thought "well we know what Bethany thinks of this, but now what about Steve?" And so I merrily copied and pasted my way toward contacting over 15 friends. 

A few things this lets you do: 

1. It’s an excuse to contact friends out of the blue.  

2. You can calibrate yourself by writing down your predictions for each person’s response on fatebook.io before resolving the question (very tight feedback loop). This makes the whole process more worth your while, since you’re harvesting more interesting data than you otherwise would. (How well do you know your friends? Is there a hidden correlation between what you think of them and what their response is?)

3. I got to extract further data yet, because I then asked friends who had answered what they believed other friends would answer. So,  I not only tested my own models of my friends, but my friends’ models of their friends as well! (Do one-boxers model two-boxers better than the opposite?) 

4. This was an excuse to nerd out about decision theory to those who were interested. I got to send 15 Wikipedia links about a subject I think more people should know about. 

5. The whole process was kind of distracting during the day since friends kept messaging me, but it would have been easy for me to bundle responses. Sending messages and replies in the first place takes very little time, so the opportunity cost for doing this is very low.  

In summary, I suggest you add “ask a dozen friends about something I'm curious about” to your list of “things I’m allowed to do”. Often, what stops people from being agentic is not having a long enough list of things that they can, in fact, do.  

Now if you’ll excuse me, I’m going to double down on this and create a whole bunch of formal surveys to continue cheerfully exploiting my friends in order to batter my map into the territory hard enough that it takes its shape. 

PS: In case you’re curious (compare with your own survey!): 61% of the friends who’ve answered so far (n=13) would one-box. I got a brier score of 0.28 in my predictions. The few friends I asked to model my other friends were correct in their models (since one-box/two-box is a Boolean, I lack information about what their brier score would have been had I asked for actual probabilities—but I would estimate it’s somewhere around what I got). I’m in high school, which means that my current friends aren’t as aligned with my interests as much as I expect them to be when I'm an adult. I predict the average LW user will have a higher proportion of one-boxer friends, although I have no way of testing this hypothesis. Wait a minute... yes I do! It's you! 

2 comments

Comments sorted by top scores.