EA Concepts: Share Impressions Before Credences
Hello, readers! I’m trialing for the Content position at CEA; as such, I’ve been asked to draft a couple of posts for the concept map. These are meant to be close to the current style (no links in body text, fairly concise).
I’d love to hear your feedback on this post. Specific questions:
1. What are your favorite words for “beliefs before updating on outside information” and “beliefs after updating on outside information”? We’re trying to draw that distinction with “impression” and “credence”, but those may not be the best options.
2. When you imagine this from the view of a reader who is newish to EA, and clicked on a link to read about the importance of “sharing your impressions”, does it make sense? Is it clear why this concept is useful
3. Are there any other links we should add to “further reading”? (In particular, I think that a link to the Soviet example of “everyone hates the government but is afraid to say so” might be relevant, but I couldn’t find a good article summarizing the example.)
Thanks for your help! The other concept draft is here.
Share Impressions Before Credences
When we think through a question by ourselves, we form an “impression” of the answer, based on the way we interpret our experiences. (Even if you experience something that others have also experienced, what you take away from that is unique to you.)
When we discuss a question with other people, we may update our “impression” into a “credence” after updating on their views. But this can introduce bias into a discussion. If we update before speaking, then share our updated credences rather than our impressions, our conversation partners partly hear their own views reflected back to them, making them update less than they should.
Consider two friends, Aaron and Max, who are equally good weather forecasters. Aaron has the impression that there is a 60% chance of rain tomorrow. He tells Max about this. Max had formerly had the impression that there was an 80% chance of rain tomorrow, but he updates on Aaron’s words to reach a credence of 70%.
Aaron then asks Max for his view. Max tells him he thinks there’s a 70% chance of rain, so Aaron updates to reach a credence of 65%. Both friends used the same decision algorithm (average both probabilities), but because Aaron shared his impression first, and Max shared a view that “reflected” that impression, Aaron failed to update in the same way as Max.
This dynamic explains why it can be important to share your initial impressions in group discussions, even if they no longer reflect your up-to-date credences. Doing so helps all participants obtain as much information as possible from each participant’s private experience.
Further Reading:
Kawamura, Kohei, and Vasileios Vlaseros. 31 July 2014. “Expert Information and Majority Decisions”.
- Independent impressions by (26 Sep 2021 18:43 UTC; 153 points)
- 's comment on MichaelA’s Quick takes by (14 Apr 2021 6:43 UTC; 24 points)
- 's comment on Pablo Stafforini’s Forecasting System by (17 Sep 2020 9:28 UTC; 6 points)
- 's comment on My mistakes on the path to impact by (8 Dec 2020 0:57 UTC; 4 points)
- Opinioni indipendenti by (18 Jan 2023 11:21 UTC; 1 point)
Almost all information is outside information (eg. the GDP of the US, the number of employees at CEA), so I’d prefer saying “beliefs before updating on other people’s beliefs” instead of “beliefs before updating on outside information”.
I’ve been using “impressions” and “beliefs” for these terms, but “credence” does seem better than “belief”.
Yeah. As I’ve said before, it’s good to be fully aware of what you understand, what model your inside view is using, and what credence it outputs, before/separate to any social updating of the decision-relevant credence. Or at least, this is the right thing to do if you want to have accurate models in the long run, rather than accurate decision-relevant credences in the short run.
Despite widespread agreement with the principle, I find this pretty rare to happen in practice. Do you have ideas for how this could become a more catchy social convention?
It’s definitely rare in practice. I’d imagine that we could change this by using some kind of stock phrase that works as a “pause” button in a conversation.
For example: “I hear what you’re saying. But before we get farther, can I share what I thought coming into the conversation? I’m already starting to change my mind, but I think it would be useful to clarify where we both started.”
But that’s pretty long, so hopefully we could eventually condense the idea to something like “let’s both give our impressions before we update” or “tell me what you think, I’ll tell you what I’ve been thinking, and then we can talk it out”. Or someone can come up with a catchy acronym!
It would be useful to define what you mean by impressions and credences. Especially “credences” is not a word commonly used outside of EA or fields related to decision theory.
Thanks for this feedback! I’d tried to define “credences” implicitly, as the things you have after you update, but I think that making it more clear in the final article will be really helpful (especially given that I’m using the term in an unusual, perhaps inadvisable, way).