Linkposts now live!

26 Vaniver 28 September 2016 03:13PM

 

You can now submit links to LW! As the rationality community has grown up, more and more content has moved off LW to other places, and so rather than trying to generate more content here we'll instead try to collect more content here. My hope is that Less Wrong becomes something like "the Rationalist RSS," where people can discover what's new and interesting without necessarily being plugged in to the various diaspora communities.

Some general norms, subject to change:

 

  1. It's okay to link someone else's work, unless they specifically ask you not to. It's also okay to link your own work; if you want to get LW karma for things you make off-site, drop a link here as soon as you publish it.
  2. It's okay to link old stuff, but let's try to keep it to less than 5 old posts a day. The first link that I made is to Yudkowsky's Guide to Writing Intelligent Characters.
  3. It's okay to link to something that you think rationalists will be interested in, even if it's not directly related to rationality. If it's political, think long and hard before deciding to submit that link.
  4. It's not okay to post duplicates.

As before, everything will go into discussion. Tag your links, please. As we see what sort of things people are linking, we'll figure out how we need to divide things up, be it separate subreddits or using tags to promote or demote the attention level of links and posts.

(Thanks to James Lamine for doing the coding, and to Trike (and myself) for supporting the work.)

A model of arguments

-7 Elo 01 September 2016 12:41PM

Original post: http://bearlamp.com.au/a-model-of-arguments/

Why do we argue, when we could be discussing things in a productive manner? Arguments often occur because the parties involved simply don't have the tools to transmit their ideas clearly.  In this kind of situation, the whole conversation can completely break down. It’s easy to spend a lot of time saying "You're wrong", without accomplishing anything.  

Let's imagine two people having an argument, represented by a Venn diagram.  A in the black, and B in the blue.  They each see the issue slightly differently. The blue circle to represent B’s opinion and a black circle to represent A’s opinion.  The concept of “You’re wrong” falls into the area of describing the other person’s ideas.

Person A says, "You're wrong" to Person B.  A description of the state of affairs of B’s ideas.  Not one that really represents A’s own ideas.  Naturally, Person B says, “No you’re wrong” back, equally making the unfounded claim on A’s conceptual real-estate.  The thing that is hard to demonstrate is the conflict that is accidentally generated by crossing into each other’s territory to declare things.

2

To do this creates a crossing-over of ideas.  

1D

We don’t even need to know what the argument is about, but we can expect something like this to happen:

3

4edit

Now suppose instead of Person A saying, "You're wrong", where they place the burden of argument (and proof) on the opposition, they now say, "We disagree".  

5

Person B can now continue to make the same argument of "You're wrong".  But so long as Person A shrugs and replies "We disagree", there is no conflict in the argument.  

5b

For some Person Bs, Person A might get lucky, and the two could end up with a happy middle ground of  "Yes, we disagree".  This is already a step in the right direction, and will let the pair continue to sort out precisely where and how they disagree  On the other hand, a stubborn Person B will still present a problem.

6

Hey, that's the internet for you! You win some, you lose some.

a1b

Nonetheless, the shared ground offered by "We disagree" will often spur constructive discussion.

a1


As it turns out, there is another way.  When you go to understand someone else's idea, instead of starting with "You are wrong", consider starting with, "I am wrong". Right from the start, this gives you an advantage. Rather than starting off from a position of conflict, you start off in a position of equality.

8

Sometimes the other party won't accept your peace offering. They will bristle and rage and prepare for the offensive.

a2

But it's far more common to see an offer of equality met by an acceptance of that equality. Instead of things going downhill, this usually happens:

9

Or this:

10

11

And a pleasant discussion can ensue.

Why is this so great? Because what we're aiming for here - what we really want out of discussions - is this:

13

14

16

17

What we are aiming for is to trade knowledge until we can conclude the answers in the end.

This style of measured, polite and constructive conversation can only occur when parties meet each other on equal terms.

If there's one lesson to take home from this post, it's that the way you deliver your argument can easily be what makes it powerful. If you come in throwing punches, ready to take your opponent down a notch or two, you might enjoy yourself - but don't expect to have a constructive discussion. Whereas if you approach your opponent as an equal from the shared ground of "We disagree", or even from the vulnerable position of "I am wrong" - well, what reasonable opponent could disagree with that?


This post took me weeks of thinking about, and only 3 hours to write down and draw the first time.  But it was rubbish.  Didn’t make sense.  The rewrite was contributed by the Captain and the slack, taking another 2 hours.  This version gets the point of the idea across.  I sent the original post to Tim@waitbutwhy but he is very busy and declined to draw pictures to go along with it.

Cross posted to lesswrong: 

Superintelligence and physical law

10 AnthonyC 04 August 2016 06:49PM

It's been a few years since I read http://lesswrong.com/lw/qj/einsteins_speed/ and the rest of the quantum physics sequence, but I recently learned about the company Nutonian, http://www.nutonian.com/. Basically it's a narrow AI system that looks at unstructured data and tries out billions of models to fit it, favoring those that use simpler math. They apply it to all sorts of fields, but that includes physics. It can't find Newton's laws from three frames of a falling apple, but it did find the Hamiltonian of a double pendulum given its motion data after a few hours of processing: http://phys.org/news/2009-12-eureqa-robot-scientist-video.html

confirmation bias, thought experiment

1 Douglas_Reay 15 July 2016 12:19PM

Why do people end up with differing conclusions, given the same data?

 

Model

The information we get from others can not always be 100% relied upon.  Some of the people telling you stuff are liars, some are stupid, and some are incorrectly or insufficiently informed.  Even in the case where the person giving you an opinion is honest, smart and well informed, they are still unlikely to be able to tell you accurately how reliable their own opinion is.

So our brains use an 'unreliability' factor.  Automatically we take what others tell us, and discount it by a certain amount, depending on how 'unreliable' we estimate the source to be.

We also compare what people tell us about 'known reference points' in order to update our estimates of their unreliability.

If Sally tells me that vaccines cause AIDS and I am very much more certain that this is not the case, than I am of Sally's reliability, then instead of modifying my opinion about what causes AIDS, I modify my opinion of how reliable Sally is.

If I'm only slightly more certain, then I might take the step of asking Sally her reason for thinking that, and looking at her data.

If I have a higher opinion of Sally than my own knowledge of science, and I don't much care or am unaware of what other people think about the relationship between vaccines and AIDS, then I might just accept what she says, provisionally, without checking her data.

If I have a very much higher opinion of Sally, then not only will I believe her, but my opinion of her reliability will actually increase as I assess her as some mould-breaking genius who knows things that others do not.

 

Importantly, once we have altered our opinion, based upon input that we originally considered to be fairly reliable, we are very bad at reversing that alteration, if the input later turns out to be less reliable than we originally thought.  This is called the "continued influence effect", and we can use it to explain a number of things...

 

Experiment

Let us consider a thought experiment where two subjects, Peter and Paul, are exposed to input about a particular topic (such as "Which clothes washing powder is it best to use?") from multiple sources.   Both will be exposed to the same sources, 100 in favour of using the Persil brand of washing powder, and 100 in favour of using the Bold brand of washing powder, but in a different order.

If they both start off with no strong opinion in either direction, would we expect them to end the experiment with roughly the same opinion as each other, or can we manipulate their opinions into differing, just by changing the order in which the sources are presented?

Suppose, with Peter, we start him off with 10 of the Persil side's most reputable and well argued sources, to raise Peter's confidence in sources that support Persil.

We can then run another 30 much weaker pro-Persil sources past him, and he is likely to just nods and accept, without bothering to examine the validity of the arguments too closely, because he's already convinced.

At this point, when he'll consider a source to be a bit suspect, straight away, just because they don't support Persil, we introduce him to the pro-Bold side, starting with the least reliable - the ones that are obviously stupid or manipulative.   Further more, we don't let the pro-Bold side build up momentum.   For every three poor pro-Bold sources, we interrupt with a medium reliability pro-Persil source that's rehashing pro-Persil points that Peter is by now familiar with and agrees with.

After seeing the worst 30 pro-Bold sources, Peter now don't just consider them to be a bit suspect - he considers them to be down right deceptive and mentally categorises all such sources as not worth paying attention to.   Any further pro-Bold sources, even ones that seem to be impartial and well reasoned, he's going to put down as being fakes created by malicious researchers in the pay of an evil company.

We can now, safely, expose Peter to the medium-reliability pro-Bold sources and even the good ones, and will need less and less to refute them, just a reminder to Peter of 'which side he is on', because it is less about the data now, and more about identity - he doesn't see himself as the sort of person who'd support Bold.   He's not a sheep.  He's not taken in by the hoax.

Finally, after 80 pro-Persil sources and 90 pro-Bold sources, we have 10 excellent pro-Bold sources whose independence and science can't fairly be questioned.   But it is too late for them to have much effect, and there are 20 good pro-Persil sources to balance them.

For Paul we do the reverse, starting with pro-Bold sources and only later introducing the pro-Persil side once a known reference point has been established as an anchor.

 

Simulation

Obviously, things are rarely that clear cut in real life.   But people also don't often get data from both sides of an argument at a precisely equal rate.   They bump around randomly, and once one side accumulates some headway, it is unlikely to be reversed.

We could add a third subject, Mary, and consider what is likely to happen if she is exposed to a random succession of sources, each with a 50% chance of supporting one side or the other, and each with a random value on a scale of 1(poor) to 3 (good) for honesty, validity and strength of conclusion supported by the claimed data.

If we use mathematics to make some actual models of the points at which a source agreeing or disagreeing with you affects your estimate of their reliability, we can use a computer simulation of the above thought experiment to predict how different orders of presentation will affect people's final opinion, under each model.   Then we could compare that against real-world data, to see which model best matches reality.

 

Prediction

I think, if this experiment were carried out, one of the properties that would emerge naturally from it is the backfire effect:

" The backfire effect occurs when, in the face of contradictory evidence, established beliefs do not change but actually get stronger. The effect has been demonstrated experimentally in psychological tests, where subjects are given data that either reinforces or goes against their existing biases - and in most cases people can be shown to increase their confidence in their prior position regardless of the evidence they were faced with. "

 

Further Reading

https://en.wikipedia.org/wiki/Confirmation_bias
https://en.wikipedia.org/wiki/Attitude_polarization
http://www.dartmouth.edu/~nyhan/nyhan-reifler.pdf
http://www.tandfonline.com/doi/abs/10.1080/17470216008416717
http://lesswrong.com/lw/iw/positive_bias_look_into_the_dark/
http://www.tandfonline.com/doi/abs/10.1080/14640749508401422
http://rationalwiki.org/wiki/Backfire_effect

Now is the time to eliminate mosquitoes

21 James_Miller 06 August 2016 07:10PM

“In 2015, there were roughly 214 million malaria cases and an estimated 438 000 malaria deaths.”  While we don’t know how many humans malaria has killed, an estimate of half of everyone who has ever died isn’t absurd.  Because few people in rich countries get malaria, pharmaceutical companies put relatively few resources into combating it.   

 

The best way to eliminate malaria is probably to use gene drives to completely eradicate the species of mosquitoes that bite humans, but until recently rich countries haven’t been motivated to such xenocide.  The Zika virus, which is in mosquitoes in the United States, provides effective altruists with an opportunity to advocate for exterminating all species of mosquitoes that spread disease to humans because the horrifying and disgusting pictures of babies with Zika might make the American public receptive to our arguments.  A leading short-term goal of effective altruists, I propose, should be advocating for mosquito eradication in the short window before rich people get acclimated to pictures of Zika babies.   

 

Personally, I have (unsuccessfully) pitched articles on mosquito eradication to two magazines and (with a bit more success) emailed someone who knows someone who knows someone in the Trump campaign to attempt to get the candidate to come out in favor of mosquito eradication.  What have you done?   Given the enormous harm mosquitoes inflict on mankind, doing just a little (such as writing a blog post) could have a high expected payoff.

 

Purposeful Anti-Rush

4 Elo 08 March 2016 07:34AM

Why do we rush?

Things happen; Life gets in the way, and suddenly we find ourselves trying to get to somewhere with less time than it's possible to actually get there in.  So in the intention to get there sooner; to somehow compensate ourselves for not being on time; we rush.  We run; we get clumsy, we drop things; we forget things; we make mistakes; we scribble instead of writing, we scramble and we slip up.

I am today telling you to stop that.  Don't do that.  It's literally the opposite of what you want to do.  This is a bug I have.

Rushing has a tendency to do the opposite of what I want it to do.  I rush with the key in the lock; I rush on slippery surfaces and I fall over, I rush with coins and I drop them.  NO!  BAD!  Stop that.  This is one of my bugs.

What you (or I) really want when we are rushing is to get there sooner, to get things done faster.  

Instrumental experiment: Next time you are rushing I want you to experiment and pay attention; try to figure out what you end up doing that takes longer than it otherwise would if you weren't rushing.

The time after that when you are rushing; instead try slowing down, and this time observe to see if you get there faster.

Run as many experiments as you like.

Experimenter’s note: Maybe you are really good at rushing and really bad at slowing down.  Maybe you don't need to try this.  Maybe slowing down and being nervous about being late together are entirely unhelpful for you.  Report back.

When you are rushing, purposefully slow down. (or at least try it)


Meta: Time to write 20mins

My Table of contents contains other things I have written.

Feedback welcome.

Availability Heuristic Considered Ambiguous

9 Gram_Stone 10 June 2016 10:40PM

(Content note: The experimental results on the availability bias, one of the biases described in Tversky and Kahneman's original work, have been overdetermined, which has led to at least two separate interpretations of the heuristic in the cognitive science literature. These interpretations also result in different experimental predictions. The audience probably wants to know about this. This post is also intended to measure audience interest in a tradition of cognitive scientific research that I've been considering describing here for a while. Finally, I steal from Scott Alexander the section numbering technique that he stole from someone else: I expect it to be helpful because there are several inferential steps to take in this particular article, and it makes it look less monolithic.)

Related to: Availability

I.

The availability heuristic is judging the frequency or probability of an event, by the ease with which examples of the event come to mind.

This statement is actually slightly ambiguous. I notice at least two possible interpretations with regards to what the cognitive scientists infer is happening inside of the human mind:

  1. Humans think things like, “I found a lot of examples, thus the frequency or probability of the event is high,” or, “I didn’t find many examples, thus the frequency or probability of the event is low.”
  2. Humans think things like, “Looking for examples felt easy, thus the frequency or probability of the event is high,” or, “Looking for examples felt hard, thus the frequency or probability of the event is low.”

I think the second interpretation is the one more similar to Kahneman and Tversky’s original description, as quoted above.

And it doesn’t seem that I would be building up a strawman by claiming that some adhere to the first interpretation, intentionally or not. From Medin and Ross (1996, p. 522):

The availability heuristic refers to a tendency to form a judgment on the basis of what is readily brought to mind. For example, a person who is asked whether there are more English words that begin with the letter ‘t’ or the letter ‘k’ might try to think of words that begin with each of these letters. Since a person can probably think of more words beginning with ‘t’, he or she would (correctly) conclude that ‘t’ is more frequent than ‘k’ as the first letter of English words.

And even that sounds at least slightly ambiguous to me, although it falls on the other side of the continuum between pure mental-content-ism and pure phenomenal-experience-ism that includes the original description.

II.

You can’t really tease out this ambiguity with the older studies on availability, because these two interpretations generate the same prediction. There is a strong correlation between the number of examples recalled and the ease with which those examples come to mind.

For example, consider a piece of the setup in Experiment 3 from the original paper on the availability heuristic. The subjects in this experiment were asked to estimate the frequency of two types of words in the English language: words with ‘k’ as their first letter, and words with ‘k’ as their third letter. There are twice as many words with ‘k’ as their third letter, but there was bias towards estimating that there are more words with ‘k’ as their first letter.

How, in experiments like these, are you supposed to figure out whether the subjects are relying on mental content or phenomenal experience? Both mechanisms predict the outcome, "Humans will be biased towards estimating that there are more words with 'k' as their first letter." And a lot of the later studies just replicate this result in other domains, and thus suffer from the same ambiguity.

III.

If you wanted to design a better experiment, where would you begin?

Well, if we think of feelings as sources of information in the way that we regard thoughts as sources of information, then we should find that we have some (perhaps low, perhaps high) confidence in the informational value of those feelings, as we have some level of confidence in the informational value of our thoughts.

This is useful because it suggests a method for detecting the use of feelings as sources of information: if we are led to believe that a source of information has low value, then its relevance will be discounted; and if we are led to believe that it has high value, then its relevance will be augmented. Detecting this phenomenon in the first place is probably a good place to start before trying to determine whether the classic availability studies demonstrate a reliance on phenomenal experience, mental content, or both. 

Fortunately, Wänke et al. (1995) conducted a modified replication of the experiment described above with exactly the properties that we’re looking for! Let’s start with the control condition.

In the control condition, subjects were given a blank sheet of paper and asked to write down 10 words that have ‘t’ as the third letter, and then to write down 10 words that begin with the letter ‘t’. After this listing task, they rated the extent to which words beginning with a ‘t’ are more or less frequent than words that have ‘t’ as the third letter. As in the original availability experiments, subjects estimated that words that begin with a ‘t’ are much more frequent than words with a ‘t’ in the third position.

Like before, this isn’t enough to answer the questions that we want to answer, but it can’t hurt to replicate the original result. It doesn’t really get interesting until you do things that affect the perceived value of the subjects’ feelings.

Wänke et al. got creative and, instead of blank paper, they gave subjects in two experimental conditions sheets of paper imprinted with pale, blue rows of ‘t’s, and told them to write 10 words beginning with a ‘t’. One condition was told that the paper would make it easier for them to recall words beginning with a ‘t’, and the other was told that the paper would make it harder for them to recall words beginning with a ‘t’.

Subjects made to think that the magic paper made it easier to think of examples gave lower estimates of the frequency of words beginning with a ‘t’ in the English language. It felt easy to think of examples, but the experimenter made them expect that by means of the magic paper, so they discounted the value of the feeling of ease. Their estimates of the frequency of words beginning with 't' went down relative to the control condition.

Subjects made to think that the magic paper made it harder to think of examples gave higher estimates of the frequency of words beginning with a ‘t’ in the English language. It felt easy to recall examples, but the experimenter made them think it would feel hard, so they augmented the value of the feeling of ease. Their estimates of the frequency of words beginning with 't' went up relative to the control condition.

(Also, here's a second explanation by Nate Soares if you want one.)

So, at least in this sort of experiment, it looks like the subjects weren’t counting the number of examples they came up with; it looks like they really were using their phenomenal experiences of ease and difficulty to estimate the frequency of certain classes of words. This is some evidence for the validity of the second interpretation mentioned at the beginning.

IV.

So we know that there is at least one circumstance in which the second interpretation seems valid. This was a step towards figuring out whether the availability heuristic first described by Kahneman and Tversky is an inference from amount of mental content, or an inference from the phenomenal experience of ease of recall, or something else, or some combination thereof.

As I said before, the two interpretations have identical predictions in the earlier studies. The solution to this is to design an experiment where inferences from mental content and inferences from phenomenal experience cause different judgments.

Schwarz et al. (1991, Experiment 1) asked subjects to list either 6 or 12 situations in which they behaved either assertively or unassertively. Pretests had shown that recalling 6 examples was experienced as easy, whereas recalling 12 examples was experienced as difficult. After listing examples, subjects had to evaluate their own assertiveness.

As one would expect, subjects rated themselves as more assertive when recalling 6 examples of assertive behavior than when recalling 6 examples of unassertive behavior.

But the difference in assertiveness ratings didn’t increase with the number of examples. Subjects who had to recall examples of assertive behavior rated themselves as less assertive after reporting 12 examples rather than 6 examples, and subjects who had to recall examples of unassertive behavior rated themselves as more assertive after reporting 12 examples rather than 6 examples.

If they were relying on the number of examples, then we should expect their ratings for the recalled quality to increase with the number of examples. Instead, they decreased.

It could be that it got harder to come up with good examples near the end of the task, and that later examples were lower quality than earlier examples, and the increased availability of the later examples biased the ratings in the way that we see. Schwarz acknowledged this, checked the written reports manually, and claimed that no such quality difference was evident.

V.

It would still be nice if we could do better than taking Schwarz’s word on that though. One thing you could try is seeing what happens when you combine the methods we used in the last two experiments: vary the number of examples generated and manipulate the perceived relevance of the experiences of ease and difficulty at the same time. (Last experiment, I promise.)

Schwarz et al. (1991, Experiment 3) manipulated the perceived value of the experienced ease or difficulty of recall by having subjects listen to ‘new-age music’ played at half-speed while they worked on the recall task. Some subjects were told that this music would make it easier to recall situations in which they behaved assertively and felt at ease, whereas others were told that it would make it easier to recall situations in which they behaved unassertively and felt insecure. These manipulations make subjects perceive recall experiences as uninformative whenever the experience matches the alleged impact of the music; after all, it may simply be easy or difficult because of the music. On the other hand, experiences that are opposite to the alleged impact of the music are considered very informative.

When the alleged effects of the music were the opposite of the phenomenal experience of generating examples, the previous experimental results were replicated.

When the alleged effects of the music match the phenomenal experience of generating examples, then the experience is called into question, since you can’t tell if it’s caused by the recall task or the music.

When this is done, the pattern that we expect from the first interpretation of the availability heuristic holds. Thinking of 12 examples of assertive behavior makes subjects rate themselves as more assertive than thinking of 6 examples of assertive behavior; mutatis mutandis for unassertive examples. When people can’t rely on their experience, they fall back to using mental content, and instead of relying on how hard or easy things feel, they count.

Under different circumstances, both interpretations are useful, but of course, it’s important to recognize that a distinction exists in the first place.


Medin, D. L., & Ross, B. H. (1996). Cognitive psychology (2nd ed.). Fort Worth: Harcourt Brace.

Schwarz, N., Bless, H., Strack, F., Klumpp, G., Rittenauer-Schatka, H., & Simons, A. (1991). Ease of retrieval as information: Another look at the availability heuristic. Journal of Personality and Social Psychology, 61, 195–202.

Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5, 207–232.

Wänke, M., Schwarz, N. & Bless, H. (1995). The availability heuristic revisited: Experienced ease of retrieval in mundane frequency estimates. Acta Psychologica, 89, 83-90.

Hedge drift and advanced motte-and-bailey

21 Stefan_Schubert 01 May 2016 02:45PM

Motte and bailey is a technique by which one protects an interesting but hard-to-defend view by making it similar to a less interesting but more defensible position. Whenever the more interesting position - the bailey - is attacked - one retreats to the more defensible one - the motte -, but when the attackers are gone, one expands again to the bailey. 

In that case, one and the same person switches between two interpretations of the original claim. Here, I rather want to focus on situations where different people make different interpretations of the original claim. The originator of the claim adds a number of caveats and hedges to their claim, which makes it more defensible, but less striking and sometimes also less interesting.* When others refer to the same claim, the caveats and hedges gradually disappear, however, making it more and more motte-like.

A salient example of this is that scientific claims (particularly in messy fields like psychology and economics) often come with a number of caveats and hedges, which tend to get lost when re-told. This is especially so when media writes about these claims, but even other scientists often fail to properly transmit all the hedges and caveats that come with them.

Since this happens over and over again, people probably do expect their hedges to drift to some extent. Indeed, it would not surprise me if some people actually want hedge drift to occur. Such a strategy effectively amounts to a more effective, because less observable, version of the motte-and-bailey-strategy. Rather than switching back and forth between the motte and the bailey - something which is at least moderately observable, and also usually relies on some amount of vagueness, which is undesirable - you let others spread the bailey version of your claim, whilst you sit safe in the motte. This way, you get what you want - the spread of the bailey version - in a much safer way.

Even when people don't use this strategy intentionally, you could argue that they should expect hedge drift, and that omitting to take action against it is, if not ouright intellectually dishonest, then at least approaching that. This argument would rest on the consequentialist notion that if you have strong reasons to believe that some negative event will occur, and you could prevent it from happening by fairly simple means, then you have an obligation to do so. I certainly do think that scientists should do more to prevent their views from being garbled via hedge drift. 

Another way of expressing all this is by saying that when including hedging or caveats, scientists often seem to seek plausible deniability ("I included these hedges; it's not my fault if they were misinterpreted"). They don't actually try to prevent their claims from being misunderstood. 

What concrete steps could one then take to prevent hedge-drift? Here are some suggestions. I am sure there are many more.

  1. Many authors use eye-catching, hedge-free titles and/or abstracts, and then only include hedges in the paper itself. This is a recipe for hedge-drift and should be avoided.
  2. Make abundantly clear, preferably in the abstract, just how dependent the conclusions are on keys and assumptions. Say this not in a way that enables you to claim plausible deniability in case someone misinterprets you, but in a way that actually reduces the risk of hedge-drift as much as possible. 
  3. Explicitly caution against hedge drift, using that term or a similar one, in the abstract of the paper.

* Edited 2/5 2016. By hedges and caveats I mean terms like "somewhat" ("x reduces y somewhat"), "slightly", etc, as well as modelling assumptions without which the conclusions don't follow and qualifications regarding domains in which the thesis don't hold.

The Web Browser is Not Your Client (But You Don't Need To Know That)

22 Error 22 April 2016 12:12AM

(Part of a sequence on discussion technology and NNTP. As last time, I should probably emphasize that I am a crank on this subject and do not actually expect anything I recommend to be implemented. Add whatever salt you feel is necessary)1


If there is one thing I hope readers get out of this sequence, it is this: The Web Browser is Not Your Client.

It looks like you have three or four viable clients -- IE, Firefox, Chrome, et al. You don't. You have one. It has a subforum listing with two items at the top of the display; some widgets on the right hand side for user details, RSS feed, meetups; the top-level post display; and below that, replies nested in the usual way.

Changing your browser has the exact same effect on your Less Wrong experience as changing your operating system, i.e. next to none.

For comparison, consider the Less Wrong IRC, where you can tune your experience with a wide range of different software. If you don't like your UX, there are other clients that give a different UX to the same content and community.

That is how the mechanism of discussion used to work, and does not now. Today, your user experience (UX) in a given community is dictated mostly by the admins of that community, and software development is often neither their forte nor something they have time for. I'll often find myself snarkily responding to feature requests with "you know, someone wrote something that does that 20 years ago, but no one uses it."

Semantic Collapse

What defines a client? More specifically, what defines a discussion client, a Less Wrong client?

The toolchain by which you read LW probably looks something like this; anyone who's read the source please correct me if I'm off:

Browser -> HTTP server -> LW UI application -> Reddit API -> Backend database.

The database stores all the information about users, posts, etc. The API presents subsets of that information in a way that's convenient for a web application to consume (probably JSON objects, though I haven't checked). The UI layer generates a web page layout and content using that information, which is then presented -- in the form of (mostly) HTML -- by the HTTP server layer to your browser. Your browser figures out what color pixels go where.

All of this is a gross oversimplification, obviously.

In some sense, the browser is self-evidently a client: It talks to an http server, receives hypertext, renders it, etc. It's a UI for an HTTP server.

But consider the following problem: Find and display all comments by me that are children of this post, and only those comments, using only browser UI elements, i.e. not the LW-specific page widgets. You cannot -- and I'd be pretty surprised if you could make a browser extension that could do it without resorting to the API, skipping the previous elements in the chain above. For that matter, if you can do it with the existing page widgets, I'd love to know how.

That isn't because the browser is poorly designed; it's because the browser lacks the semantic information to figure out what elements of the page constitute a comment, a post, an author. That information was lost in translation somewhere along the way.

Your browser isn't actually interacting with the discussion. Its role is more akin to an operating system than a client. It doesn't define a UX. It provides a shell, a set of system primitives, and a widget collection that can be used to build a UX. Similarly, HTTP is not the successor to NNTP; the successor is the plethora of APIs, for which HTTP is merely a substrate.

The Discussion Client is the point where semantic metadata is translated into display metadata; where you go from 'I have post A from user B with content C' to 'I have a text string H positioned above visual container P containing text string S.' Or, more concretely, when you go from this:

Author: somebody
Subject: I am right, you are mistaken, he is mindkilled.
Date: timestamp
Content: lorem ipsum nonsensical statement involving plankton....

to this:

<h1>I am right, you are mistaken, he is mindkilled.</h1>
<div><span align=left>somebody</span><span align=right>timestamp</span></div>
<div><p>lorem ipsum nonsensical statement involving plankton....</p></div>

That happens at the web application layer. That's the part that generates the subforum headings, the interface widgets, the display format of the comment tree. That's the part that defines your Less Wrong experience, as a reader, commenter, or writer.

That is your client, not your web browser. If it doesn't suit your needs, if it's missing features you'd like to have, well, you probably take for granted that you're stuck with it.

But it doesn't have to be that way.

Mechanism and Policy

One of the difficulties forming an argument about clients is that the proportion of people who have ever had a choice of clients available for any given service keeps shrinking. I have this mental image of the Average Internet User as having no real concept for this.

Then I think about email. Most people have probably used at least two different clients for email, even if it's just Gmail and their phone's built-in mail app. Or perhaps Outlook, if they're using a company system. And they (I think?) mostly take for granted that if they don't like Outlook they can use something else, or if they don't like their phone's mail app they can install a different one. They assume, correctly, that the content and function of their mail account is not tied to the client application they use to work with it.

(They may make the same assumption about web-based services, on the reasoning that if they don't like IE they can switch to Firefox, or if they don't like Firefox they can switch to Chrome. They are incorrect, because The Web Browser is Not Their Client)

Email does a good job of separating mechanism from policy. Its format is defined in RFC 2822 and its transmission protocol is defined in RFC 5321. Neither defines any conventions for user interfaces. There are good reasons for that from a software-design standpoint, but more relevant to our discussion is that interface conventions change more rapidly than the objects they interface with. Forum features change with the times; but the concepts of a Post, an Author, or a Reply are forever.

The benefit of this separation: If someone sends you mail from Outlook, you don't need to use Outlook to read it. You can use something else -- something that may look and behave entirely differently, in a manner more to your liking.

The comparison: If there is a discussion on Less Wrong, you do need to use the Less Wrong UI to read it. The same goes for, say, Facebook.

I object to this.

Standards as Schelling Points

One could argue that the lack of choice is for lack of interest. Less Wrong, and Reddit on which it is based, has an API. One could write a native client. Reddit does have them.

Let's take a tangent and talk about Reddit. Seems like they might have done something right. They have (I think?) the largest contiguous discussion community on the net today. And they have a published API for talking to it. It's even in use.

The problem with this method is that Reddit's API applies only to Reddit. I say problem, singular, but it's really problem, plural, because it hits users and developers in different ways.

On the user end, it means you can't have a unified user interface across different web forums; other forum servers have entirely different APIs, or none at all.2 It also makes life difficult when you want to move from one forum to another.

On the developer end, something very ugly happens when a content provider defines its own provision mechanism. Yes, you can write a competing client. But your client exists only at the provider's sufferance, subject to their decision not to make incompatible API changes or just pull the plug on you and your users outright. That isn't paranoia; in at least one case, it actually happened. Using an agreed-upon standard limits this sort of misbehavior, although it can still happen in other ways.

NNTP is a standard for discussion, like SMTP is for email. It is defined in RFC 3977 and its data format is defined in RFC 5536. The point of a standard is to ensure lasting interoperability; because it is a standard, it serves as a deliberately-constructed Schelling point, a place where unrelated developers can converge without further coordination.

Expertise is a Bottleneck

If you're trying to build a high-quality community, you want a closed system. Well kept gardens die by pacifism, and it's impossible to fully moderate an open system. But if you're building a communication infrastructure, you want an open system.

In the early Usenet days, this was exactly what existed; NNTP was standardized and open, but Usenet was a de-facto closed community, accessible mostly to academics. Then AOL hooked its customers into the system. The closed community became open, and the Eternal September began.3 I suspect, but can't prove, that this was a partial cause of the flight of discussion from Usenet to closed web forums.

I don't think that was the appropriate response. I think the appropriate response was private NNTP networks or even single servers, not connected to Usenet at large.

Modern web forums throw the open-infrastructure baby out with the open-community bathwater. The result, in our specific case, is that if we want something not provided by the default Less Wrong interface, it must be implemented by Less Wrongers.

I don't think UI implementation is our comparative advantage. In fact I know it isn't, or the Less Wrong UI wouldn't suck so hard. We're pretty big by web-forum standards, but we still contain only a tiny fraction of the Internet's technical expertise.

The situation is even worse among the diaspora; for example, at SSC, if Scott's readers want something new out of the interface, it must be implemented either by Scott himself or his agents. That doesn't scale.

One of the major benefits of a standardized, open infrastructure is that your developer base is no longer limited to a single community. Any software written by any member of any community backed by the same communication standard is yours for the using. Additionally, the developers are competing for the attention of readers, not admins; you can expect the reader-facing feature set to improve accordingly. If readers want different UI functionality, the community admins don't need to be involved at all.

A Real Web Client

When I wrote the intro to this sequence, the most common thing people insisted on was this: Any system that actually gets used must allow links from the web, and those links must reach a web page.

I completely, if grudgingly, agree. No matter how insightful a post is, if people can't link to it, it will not spread. No matter how interesting a post is, if Google doesn't index it, it doesn't exist.

One way to achieve a common interface to an otherwise-nonstandard forum is to write a gateway program, something that answers NNTP requests and does magic to translate them to whatever the forum understands. This can work and is better than nothing, but I don't like it -- I'll explain why in another post.

Assuming I can suppress my gag reflex for the next few moments, allow me to propose: a web client.

(No, I don't mean write a new browser. The Browser Is Not Your Client.4)

Real NNTP clients use the OS's widget set to build their UI and talk to the discussion board using NNTP. There is no fundamental reason the same cannot be done using the browser's widget set. Google did it. Before them, Deja News did it. Both of them suck, but they suck on the UI level. They are still proof that the concept can work.

I imagine an NNTP-backed site where casual visitors never need to know that's what they're dealing with. They see something very similar to a web forum or a blog, but whatever software today talks to a database on the back end, instead talks to NNTP, which is the canonical source of posts and post metadata. For example, it gets the results of a link to http://lesswrong.com/posts/message_id.html by sending ARTICLE message_id to its upstream NNTP server (which may be hosted on the same system), just as a native client would.

To the drive-by reader, nothing has changed. Except, maybe, one thing. When a regular reader, someone who's been around long enough to care about such things, says "Hey, I want feature X," and our hypothetical web client doesn't have it, I can now answer:

Someone wrote something that does that twenty years ago.

Here is how to get it.



  1. Meta-meta: This post took about eight hours to research and write, plus two weeks procrastinating. If anyone wants to discuss it in realtime, you can find me on #lesswrong or, if you insist, the LW Slack.

  2. The possibility of "universal clients" that understand multiple APIs is an interesting case, as with Pidgin for IM services. I might talk about those later.

  3. Ironically, despite my nostalgia for Usenet, I was a part of said September; or at least its aftermath.

  4. Okay, that was a little shoehorned in. The important thing is this: What I tell you three times is true.

Turning the Technical Crank

43 Error 05 April 2016 05:36AM

A few months ago, Vaniver wrote a really long post speculating about potential futures for Less Wrong, with a focus on the idea that the spread of the Less Wrong diaspora has left the site weak and fragmented. I wasn't here for our high water mark, so I don't really have an informed opinion on what has socially changed since then. But a number of complaints are technical, and as an IT person, I thought I had some useful things to say.

I argued at the time that many of the technical challenges of the diaspora were solved problems, and that the solution was NNTP -- an ancient, yet still extant, discussion protocol. I am something of a crank on the subject and didn't expect much of a reception. I was pleasantly surprised by the 18 karma it generated, and tried to write up a full post arguing the point.

I failed. I was trying to write a manifesto, didn't really know how to do it right, and kept running into a vast inferential distance I couldn't seem to cross. I'm a product of a prior age of the Internet, from before the http prefix assumed its imperial crown; I kept wanting to say things that I knew would make no sense to anyone who came of age this millennium. I got bogged down in irrelevant technical minutia about how to implement features X, Y, and Z. Eventually I decided I was attacking the wrong problem; I was thinking about 'how do I promote NNTP', when really I should have been going after 'what would an ideal discussion platform look like and how does NNTP get us there, if it does?'

So I'm going to go after that first, and work on the inferential distance problem, and then I'm going to talk about NNTP, and see where that goes and what could be done better. I still believe it's the closest thing to a good, available technological schelling point, but it's going to take a lot of words to get there from here, and I might change my mind under persuasive argument. We'll see.

Fortunately, this is Less Wrong, and sequences are a thing here. This is the first post in an intended sequence on mechanisms of discussion. I know it's a bit off the beaten track of Less Wrong subject matter. I posit that it's both relevant to our difficulties and probably more useful and/or interesting than most of what comes through these days. I just took the 2016 survey and it has a couple of sections on the effects of the diaspora, so I'm guessing it's on topic for meta purposes if not for site-subject purposes.

Less Than Ideal Discussion

To solve a problem you must first define it. Looking at the LessWrong 2.0 post, I see the following technical problems, at a minimum; I'll edit this with suggestions from comments.

  1. Aggregation of posts. Our best authors have formed their own fiefdoms and their work is not terribly visible here. We currently have limited support for this via the sidebar, but that's it.
  2. Aggregation of comments. You can see diaspora authors in the sidebar, but you can't comment from here.
  3. Aggregation of community. This sounds like a social problem but it isn't. You can start a new blog, but unless you plan on also going out of your way to market it then your chances of starting a discussion boil down to "hope it catches the attention of Yvain or someone else similarly prominent in the community." Non-prominent individuals can theoretically post here; yet this is the place we are decrying as moribund.
  4. Incomplete and poor curation. We currently do this via Promoted, badly, and via the diaspora sidebar, also badly.
  5. Pitiful interface feature set. This is not so much a Less Wrong-specific problem as a 2010s-internet problem; people who inhabit SSC have probably seen me respond to feature complaints with "they had something that did that in the 90s, but nobody uses it." (my own bugbear is searching for comments by author-plus-content).
  6. Changes are hamstrung by the existing architecture, which gets you volunteer reactions like this one.

I see these meta-technical problems:

  1. Expertise is scarce. Few people are in a position to technically improve the site, and those that are, have other demands on their time.
  2. The Trivial Inconvenience Problem limits the scope of proposed changes to those that are not inconvenient to commenters or authors.
  3. Getting cooperation from diaspora authors is a coordination problem. Are we better than average at handling those? I don't know.

Slightly Less Horrible Discussion

"Solving" community maintenance is a hard problem, but to the extent that pieces of it can be solved technologically, the solution might include these ultra-high-level elements:

  1. Centralized from the user perspective. A reader should be able to interact with the entire community in one place, and it should be recognizable as a community.
  2. Decentralized from the author perspective. Diaspora authors seem to like having their own fiefdoms, and the social problem of "all the best posters went elsewhere" can't be solved without their cooperation. Therefore any technical solution must allow for it.
  3. Proper division of labor. Scott Alexander probably should not have to concern himself with user feature requests; that's not his comparative advantage and I'd rather he spend his time inventing moral cosmologies. I suspect he would prefer the same. The same goes for Eliezer Yudkowski or any of our still-writing-elsewhere folks.
  4. Really good moderation tools.
  5. Easy entrance. New users should be able to join the discussion without a lot of hassle. Old authors that want to return should be able to do so and, preferably, bring their existing content with them.
  6. Easy exit. Authors who don't like the way the community is heading should be able to jump ship -- and, crucially, bring their content with them to their new ship. Conveniently. This is essentially what has happened, except old content is hostage here.
  7. Separate policy and mechanism within the site architecture. Let this one pass for now if you don't know what it means; it's the first big inferential hurdle I need to cross and I'll be starting soon enough.

As with the previous, I'll update this from the comments if necessary.

Getting There From Here

As I said at the start, I feel on firmer ground talking about technical issues than social ones. But I have to acknowledge one strong social opinion: I believe the greatest factor in Less Wrong's decline is the departure of our best authors for personal blogs. Any plan for revitalization has to provide an improved substitute for a personal blog, because that's where everyone seems to end up going. You need something that looks and behaves like a blog to the author or casual readers, but integrates seamlessly into a community discussion gateway.

I argue that this can be achieved. I argue that the technical challenges are solvable and the inherent coordination problem is also solvable, provided the people involved still have an interest in solving it.

And I argue that it can be done -- and done better than what we have now -- using technology that has existed since the '90s.

I don't argue that this actually will be achieved in anything like the way I think it ought to be. As mentioned up top, I am a crank, and I have no access whatsoever to anybody with any community pull. My odds of pushing through this agenda are basically nil. But we're all about crazy thought experiments, right?

This topic is something I've wanted to write about for a long time. Since it's not typical Less Wrong fare, I'll take the karma on this post as a referendum on whether the community would like to see it here.

Assuming there's interest, the sequence will look something like this (subject to reorganization as I go along, since I'm pulling this from some lengthy but horribly disorganized notes; in particular I might swap subsequences 2 and 3):

  1. Technical Architecture
    1. Your Web Browser Is Not Your Client
    2. Specialized Protocols: or, NNTP and its Bastard Children
    3. Moderation, Personal Gardens, and Public Parks
    4. Content, Presentation, and the Division of Labor
    5. The Proper Placement of User Features
    6. Hard Things that are Suddenly Easy: or, what does client control gain us?
    7. Your Web Browser Is Still Not Your Client (but you don't need to know that)
  2. Meta-Technical Conflicts (or, obstacles to adoption)
    1. Never Bet Against Convenience
    2. Conflicting Commenter, Author, and Admin Preferences
    3. Lipstick on the Configuration Pig
    4. Incremental Implementation and the Coordination Problem.
    5. Lowering Barriers to Entry and Exit
  3. Technical and Social Interoperability
    1. Benefits and Drawbacks of Standards
    2. Input Formats and Quoting Conventions
    3. Faking Functionality
    4. Why Reddit Makes Me Cry
    5. What NNTP Can't Do
  4. Implementation of Nonstandard Features
    1. Some desirable feature #1
    2. Some desirable feature #2
    3. ...etc. This subsequence is only necessary if someone actually wants to try and do what I'm arguing for, which I think unlikely.

(Meta-meta: This post was written in Markdown, converted to HTML for posting using Pandoc, and took around four hours to write. I can often be found lurking on #lesswrong or #slatestarcodex on workday afternoons if anyone wants to discuss it, but I don't promise to answer quickly because, well, workday)

[Edited to add: At +10/92% karma I figure continuing is probably worth it. After reading comments I'm going to try to slim it down a lot from the outline above, though. I still want to hit all those points but they probably don't all need a full post's space. Note that I'm not Scott or Eliezer, I write like I bleed, so what I do post will likely be spaced out]

View more: Next