Gathering thoughts on Distillation
I think one of the biggest outstanding issues facing LessWrong is good infrastructure for distilling lengthy conversations.
Wei_Dai notes in the comments of Forum Participation as Research Strategy:
There’s nothing that explicitly prevents people from distilling such discussions into subsequent posts or papers. If people aren’t doing that, or are doing that less than they should, that could potentially be solved as a problem that’s separate from “should more people be doing FP or traditional research?”
Also, it’s not clear to me that traditional research produces more clear distillations of how disagreements get resolved. It seems like most such discussions don’t make for publishable papers and therefore most disagreements between “traditional researchers” just don’t get resolved in a way that leaves a public record (or at all).
The LessWrong team has discussed this on and off for about 9 months. While we haven’t done any development work on it yet (mostly working on Open Questions instead), habryka and I at least had some sense that this might be one of the most important things LessWrong is missing.
I have some vague thoughts on what would be needed here, but am interested in others opinions.
The rough thing that seems missing to me is a pipeline that goes something like:
1. Somebody notices that there was either a post that was way longer than it needed to be (or a bit confusing, or something). Or a meandering comment conversation that was could use to be distilled down into a post. They flag that collection of posts and comments into a databse object that says “hey, this cluster of posts and/or comments could use distillation”
2. Someone (sometimes that same person, sometimes a totally different person), writes up a distillation of that cluster of posts and/or comments
3. It might be that different people write up different distillations of the same content, because they had different takeaways. Or maybe two people both agreed that a post was long, but they both found different central metaphors useful.
4. The people who wrote the original post or comments can mark distillations as “good representation of my views” or not.
In theory there’s nothing preventing this from happening with the current site technology. But, a year ago, there was nothing stopping people from making question posts (and getting answers). And yet introducing the question feature made it much more salient that “ask questions” is a thing you’re encouraged to do, and then people did it more.
I think having some kind of official distillation pipeline could be useful in a similar way.
Interested in people’s thoughts on this.
- 's comment on List of resolved confusions about IDA by (1 Oct 2019 6:39 UTC; 23 points)
- 's comment on On Internal Family Systems and multi-agent minds: a reply to PJ Eby by (31 Oct 2019 8:43 UTC; 9 points)
- 's comment on On Internal Family Systems and multi-agent minds: a reply to PJ Eby by (30 Oct 2019 21:40 UTC; 7 points)
I think a higher priority is to have some incentives or software/institutional support for actually indicating how a disagreement got resolved, or not get resolved, as opposed to a conversation just ending without anyone knowing why, except the person who stopped responding. I know I keep harping on this, and you already have a possible solution in the works, but how am I supposed to produce “clear distillations of how disagreements get resolved” if I have no idea if one of the participants actually changed their mind and is tacitly agreeing with the other person, or just got too frustrated and stopped wanting to talk with the other person, or something else?
(See here and here for two recent somewhat frustrating examples. Note that I’m not blaming the participants, but instead the current culture/norms/software/institutions.)
Thanks.
I’m not 100% sure which distinction you’re drawing. My current frame is something like “if we have official software support for distillation, that will hopefully shift the culture to acknowledge that distillation is important, and therefore, it’s more important for conversations to reach some state of summarizeability.”
This doesn’t mean all conversations need to be resolved, or subsequently distilled. Looking at your two examples, I have some sense that the first example (from “The Real Rules” post) didn’t cry out to me for resolution, largely because the people involved aren’t really collaborating on longer term projects. It’s not necessary for any given two lesswrongers be on the same page about everything. (But, maybe, this is a wrong frame/characterization on my part, either of that particular conversation or LW as a whole?).
Your conversation with Michael Cohen seemed more likely to be worth reaching some kind of resolution on (again, just based on some rough surface level heuristics).
My little habryka-simulation that runs in my head is saying “Don’t try to force people to spell out what they changed their mind about, especially not too early. It’s okay if a conversation sort of meanders and then people move on and reflect on it in the background for awhile.” Which I think I agree with.
But it still seems like it might be good, a few days or weeks after a conversation seems to have wrapped up, if you replied to the comment and said “Hey, I’d be interested in writing up my summary of this conversation. I’m not 100% sure where everyone in the convo had ended up. Would anyone be up for sharing some of their more up-to-date thoughts before I start?”.
One could start doing that unilaterally, and hope that that starts to shift conversations towards having an after-the-fact step where people crystallize their takeaways.
[note: I currently think it’s important for people to be more legible about what they’ve changed their mind about, so that people can model the overall epistemic health and progress of the community. I think habryka disagrees with some of that framing but not 100% sure]
Ok, I hadn’t thought of this causal pathway, which I admit does make sense (although I’m not sure if it’s enough). My thinking was like, “In order for me to distill discussions, I need to know what each participant thinks at the end, and that information is often unavailable, so we need to fix that first before trying to support distillation more.”
I suspect the disagreement over “act consequentialism” might be a crux in the recent debates over site culture/norms, which is definitely a long-term project that we’re involved in.
I feel reluctant to do this, and I think it’s because (1) I don’t want people to think that I’m trying to push them to admit that I’m right and they’re wrong and (2) if they ended the conversation because it became too frustrating or something like that, it would increase their aversive feelings towards me if I explicitly asked them to continue. I think it would be better if it was a third party or some kind of impersonal norm or incentive that pushed people to give reasons for ending discussions.
I agree it’s a bit weirder for you to push to summarize discussions (esp. disagreements) that you were a part of.
Curious if there exist things-that-need-summarizing where you _weren’t_ an active participant?
A possible experiment to try is something like “have a Distillation Open Thread”, where people either propose threads that they’d like distilled (which they may or may not have been involved with), and other people do a first pass attempt to summarize the conversation. (Maybe doing the distillation in the Open Thread, but replying to the original conversation with a link to it?)
In the past, I have found distillations posts extremely helpful. I wonder if part of that is because when someone currently attempts it, they’re doing it because they’re correctly judging that: 1) it’s important, and 2) they’ll do it well. Probably as it becomes more common, one or both of these will stop being true as often.
I don’t write that frequently, but I think I would do 1-2 distillments per year if it was more socially encouraged / rewarded.
I do expect there to be some signal/noise tradeoffs. But one hope is that karma can address this pretty straightforwardly – good distillations will get upvoted and endorsed by the authors, mediocre ones less so.
Feels to me like this is a subcase of the more general problem that it’s hard to find related content, especially more recent related content. It’s easy to include a link in a new post pointing back at whatever’s being distilled, but someone who stumbles on the old post/thread has no way to know that there’s a summary, less technical explanation, more technical explanation, better central metaphor, etc available in some later post.
For instance, a sidebar of “posts which link to this one” would make distillation-type posts more visible and useful, while also solving other problems too.
It seems to me like working on visibility of related content in general would be higher-impact than working on specific use-cases, at least right now. If there were already a related-visibility system in place, then it would make more sense to add special use-cases like distillation on top of it.
I agree that adding a list of “things that linked to this post” at the bottom of a post would be a good related step (and maybe a better “next action” than any more complicated distillation pipeline stuff)
It almost[1] sounds like auto-generated sequences—that is, linked posts being collected together into a sequence, automatically—would be a good idea.
[1] This might make people more hesitant to link posts together, or create a mess—it might be hard to come up with algorithms that can tell when to make 1 sequence, and when to make 2.
The closest alternative seems like a general graph navigation system. (With post titles as nodes [2], and links between linked posts, and possibly colors indicating direction (A linked to B is red if you have A selected, but Blue if B is selected.))
[2] And a link for going to a post if you’ve selected that node.
With regards to distillation, I’d note that if a post is really long this might not be ideal, especially for users who don’t know their way around this site—those who do know their way around, and about the feature, can just click on the Comments link in the title, then scroll up to see the bottom of the post.
I would probably be a frequent beneficiary of having my content distilled.
I do not think that distilling can easily be disentangeld from the primary content generation. The step 4 saves it’s feasiblity a lot and is the essential thing such thing would revolve around.
I would even think that a discussion style where you are not allowed to state what you believe but only guess what others believe and only confirm when guesses hit home to you could be an interesting extreme way of talking. This would force the active listening part which often is a nice side bonus or only employed in special circumnstances.
I would also guess that there would be a lto of “when you put it that way....” where the confirmation steps makes you hold differnt positions than if you just left at the undistilled conversation.
I have also seen that there is language tech where a concept or symbol is powerful when it communicates a lot without it being that spesific what it communicates about. “Shrödinghers cat”. It evokes a lot of multiverse theories and measurement problem etc. But it seems to have the property that somebody that doesn’t understand the multiverse bits still correct tells the cat part or remembers that there are multiple states etc.
Shrödinger probably was important in providing some key formulas but I think the thought frameworks worked on is a part that can not be overlooked.
in order to make effective memetic talk you need catchphrases and distillation. As used in media as “sound bites” they usually make discussion problmeatic (or does the media conductive transformation on it). There is certain purity about imageboard memes. There are some fringe positions which I think have clear summations floating around that makes discussion abuot them easier. And a lot of the times it makes them easier to detect and resists.
For example “holocaust denial” is somewhat dry thing that is technically welldefined. But commenting on the spotlessness of a certain german leaders actions makes the position is so vivid that its easier to detect even in altered forms.
I also know about incels and chads, trans and huns in a way that is probaly way more agile than various physical theories which I would probably go at length to find about. But those I know about becuase of the good information design that I get them even if they are not personally so relevant to me. There is something cool about einstein saying “god does not play dice”. I rememeber tht bit, I know how that stance would apply to various situations. Yet if I had read a transcript of that conversation and that particular turn of phare was not used it would not have sticked out that well at all. It makes me presume that there is a spesific communicaiton skill that one could learn to have more of ones communition be that way. And I think the spartans deliberately sought for a kind of laconic expression. Adn the spartans knew that oneliners are not easy but require lots of education.
I think imageboards are closer in reaping the benefit that spartans saw in being laconic and I think that benefit might be relevant here as well.