gwern comments on Open thread, July 16-22, 2013 - Less Wrong

13 Post author: David_Gerard 15 July 2013 08:13PM

You are viewing a comment permalink. View the original post to see all comments and the full post content.

Article Navigation

Comments (297)
Tags: open open_thread

Comments (297)

You are viewing a single comment's thread. Show more comments above.

Comment author: gwern 20 July 2013 11:34:14PM *  25 points [-]

While reading a psychology paper, I ran into the following comment:

Unfamiliar things are distrusted and hard to process, overly familiar things are boring, and the perfect object of beauty lies somewhere in between (Sluckin, Hargreaves, & Colman, 1983). The familiar comes as standard equipment in every empirical paper: scientific report structure, well-known statistical techniques, established methods. In fact, the form of a research article is so standardized that it is in danger of becoming deathly dull. So the burden is on the author to provide content and ideas that will knock the reader’s socks off—at least if the reader is one of the dozen or so potential reviewers in that sub-subspecialty.

Besides the obvious connection to Schmidhuber's esthetics, it occurred to me that this has considerable relevance to LW/OB. Hanson in the past has counseled contrarians like us to pick our battles and conform in most ways while not conforming in a few carefully chosen ones (eg Dear Young Eccentric, Against Free Thinkers, Even When Contrarians Win, They Lose); this struck me as obviously correct, and that one could think of oneself as having a "budget" where non-conforming on both dress and language and ideas blows one's credit with people / discredits oneself.

This idea about familiarity suggests a different way to think of it is in terms of novelty and familiarity: ideas like existential risk are highly novel compared to regular politics or charities. But if these ideas are highly novel, then they are likely "distrusted and hard to process" (which certainly describes well many people's reaction to things on LW/OB), and any additional novelty like that of vocabulary or formatting or style, is more likely to damage reception or perhaps push readers past some critical limit than if applied to some standard familiar boring thing like evolution, where due to sufficient familiarity, idiosyncratic or novel aspects will not damage reception but instead improve reception. Consider the different reactions to Nick Bostrom and Eliezer Yudkowsky, who write about many of the same exact ideas and problems - but no one put on Broadway plays or YouTube videos mocking Bostrom or accusing him of being a sinister billionaire's tool in a plot against all that is good and just - while on the other hand, Hofstadter's GEB is dearly beloved for its diversity of novel forms and expressions, even if it's all directed toward exposition on pretty standard unshocking topics like Godel's theorems or GOFAI.

This line of reasoning suggests a simple strategy for writing: the novelty of a story or essay's content should be inverse to the novelty of its form.

If one has highly novel, perhaps even outright frightening ideas, about the true nature of the multiverse or the future of humanity, the format should be as standard and dry as possible. Conversely, if one is discussing settled science like genetics, one should spice it up with little parables, stories, unexpected topics and applications, etc.

Does this predict success of existing writings? Well, let's take Eliezer as an example, since he has a very particular style of writing. Three of his longest fictions so far are the Ultra Mega Crossover, "Three Worlds Collide", and MoR. Keeping in mind that the former were targeted at OB and the last at a general audience on FF.net, they seem to fit well: the Crossover was confusing in format, introduced many obscure characters or allusions, in service of a computationally-oriented multiverse that only really made sense if you had already read Permutation City, and so is highly novel in form & content, so naturally no one ever mentions it or recommends it to other people; "Three Worlds Collide" took a standard SF opera short-story style with stock archetypes like "the Captain", and saved its novelty for its meta-ethical content and world-building, and accordingly, I see it linked and discussed both on LW and off; MoR, as fanfiction, adapts a world wholesale, reducing its novelty considerably for millions of people, and inside this almost-"boring" framework introduces its audience to a panoply of cognitive biases, transhuman tropes like anti-deathism, existential risks, the scientific method, Bayesian-style reasoning, etc, and MoR has been tremendously successful on and off LW (I saw someone recommend it yesterday on HN).

Of course this is just 3 examples, but it does match the vibe I get reading why people dislike Eliezer or LW: they seem to have little trouble with his casual informal style when it's being applied to topics like cognitive biases or evolution where the topic is familiar to relatively large numbers of people, but then are horribly put off by the same style or novel forms when applied to obscurer topics like subjective Bayesianism (like the Bayesian Conspiracy short stories - actually, especially the Conspiracy-verse stories) or cryonics. Of course, I suppose this could just reflect that more popular topics tend to be less controversial and what I'm actually noticing is people disliking marginal minority theories, but things like global warming are quite controversial and I suspect Eliezer blogging about global warming would not trigger the same reaction as to, say, his "you're a bad parent if you don't sign kids up for cryonics" post that a lot of people hate.

Have I seen this "golden mean" effect in my own writing? I'm not sure. Unfortunately, my stuff seems to generally adopt a vaguely academic format or tone in proportion to how mainstream a topic is, and a great deal of traffic is driven by interest in the topic and not my work specifically; so for example, my Silk Road page is not in any particularly boring format but interest in the topic is too high for that to matter either way. It's certainly something for me to keep in mind, though, when I write about stranger topics.

Comment author: gwern 10 October 2013 04:42:00PM 2 points [-]

Though there’s probably no perfect way, the recent research mined keywords generated by users of the website the Internet Movie Database (IMDb), which contains descriptions of more than 2 million films. When summarizing plots, people on the site are prompted to use keywords that have been used to describe previous movies, yielding tags that characterize particular genres (cult-film), locations (manhattan-new-york), or story elements (tied-to-a-chair). Each keyword was given a score based on its rarity when compared to previous work. If some particular plot point – like, say, beautiful-woman – had appeared in many movies that preceded a particular film, it was given a low novelty value. But a new element – perhaps martial-arts, which appeared infrequently in films before the ’60s – was given a high novelty score when it first showed up. The scores ranged from zero to one, with the least novel being zero. Lining up the scores chronologically showed the evolution of film culture and plots over time. The results appeared Sept. 26 in Nature Scientific Reports.

...Unsurprisingly, the research also suggests that unfamiliar combinations of themes or plots that haven’t been encountered before (something like sci-fi-western) often have the highest novelty scores. “I think this reinforces this idea that novelty is often variations on a theme,” said Sreenivasan. “You use familiar elements broadly, and then combine them in novel ways.”

Sreenivasan’s analysis shows trends within particular genres as well. Action movies are essentially redefined in 1962 with the release of the first James Bond movie. Science-fiction films, on the other hand, show no similar creative uptick during the same period. According to the analysis, novelty in sci-fi has declined essentially since the genre first made it into movies. It’s possible that this has to do with early science-fiction films codifying the major tropes seen in these movies.

Another part of the analysis seem to correspond to theories put forth by social scientists about how much we enjoy novelty in creative works, said Sreenivasan. In general, humans enjoy new things. More specifically, there’s a tendency for people to look at and like things that are new but not too new. “If it’s way out there, it’s hard to palate,” said Sreenivasan. “And if it’s too familiar, then it seems boring.”

A model known as the Wundt-Berlyne curve illustrates this result. The amount of pleasure someone derives from a creative piece goes up as its novelty increases. But at a certain point, there is a maximum of enjoyment. After that, something becomes too unfamiliar to stomach anymore. Using the revenue generated by different films as a measure of its mass appeal, Sreenivasan found that more novel films sold more tickets until they reached a score of about 0.8. Afterwards, they appeared to decline in popularity and revenue.

(From the standard errors & shuffled results, the decline in revenue from 0.8 to 1.0 happens very fast, so one probably wants to undershoot novelty and avoid the catastrophic risk of overshoot.)

Comment author: gwern 21 November 2014 10:26:09PM 1 point [-]

"The Shazam Effect: Record companies are tracking download and search data to predict which new songs will be hits. This has been good for business—but is it bad for music?"

...But here’s the catch: if you give people too much say, they will ask for the same familiar sounds on an endless loop, entrenching music that is repetitive, derivative, and relentlessly played out. Now that the Billboard rankings are a more accurate reflection of what people buy and play, songs stay on the charts much longer. The 10 songs that have spent the most time on the Hot 100 were all released after 1991, when Billboard started using point-of-sale data—and seven were released after the Hot 100 began including digital sales, in 2005. “It turns out that we just want to listen to the same songs over and over again,” Pietroluongo told me. Because the most-popular songs now stay on the charts for months, the relative value of a hit has exploded. The top 1 percent of bands and solo artists now earn 77 percent of all revenue from recorded music, media researchers report. And even though the amount of digital music sold has surged, the 10 best-selling tracks command 82 percent more of the market than they did a decade ago. The advent of do-it-yourself artists in the digital age may have grown music’s long tail, but its fat head keeps getting fatter. Radio stations, meanwhile, are pushing the boundaries of repetitiveness to new levels. According to a subsidiary of iHeartMedia, Top 40 stations last year played the 10 biggest songs almost twice as much as they did a decade ago. Robin Thicke’s “Blurred Lines,” the most played song of 2013, aired 70 percent more than the most played song from 2003, “When I’m Gone,” by 3 Doors Down. Even the fifth-most-played song of 2013, “Ho Hey,” by the Lumineers, was on the radio 30 percent more than any song from 10 years prior.

...The problem is not our pop stars. Our brains are wired to prefer melodies we already know. (David Huron, a musicologist at Ohio State University, estimates that at least 90 percent of the time we spend listening to music, we seek out songs we’ve heard before.) That’s because familiar songs are easier to process, and the less effort needed to think through something—whether a song, a painting, or an idea—the more we tend to like it. In psychology, this idea is known as fluency: when a piece of information is consumed fluently, it neatly slides into our patterns of expectation, filling us with satisfaction and confidence. “Things that are familiar are comforting, particularly when you are feeling anxious,” Norbert Schwarz, a psychology professor at the University of Southern California, who studies fluency, told me. “When you’re in a bad mood, you want to see your old friends. You want to eat comfort food. I think this maps onto a lot of media consumption. When you’re stressed out, you don’t want to put on a new movie or a challenging piece of music. You want the old and familiar.”

... One of the popular songs of this past summer, “Problem,” combined a dizzy sax hook, ’90s-pop vocals, a whispered chorus, and a female rap verse. It was utterly strange and, for a while, ubiquitous. Greta Hsu, an associate professor at the University of California at Davis, who has done research on genre-blending in Hollywood, told me that although mixing categories is risky, hybrids can become standout successes, because they appeal to multiple audiences as being somehow both fresh and familiar.

Music fans can also find comfort in the fact that data have not taken over the songwriting process. Producers and artists pay close attention to trends, but they’re not swimming in spreadsheets quite like the suits at the labels are. Perhaps one reason machines haven’t yet invaded the recording room is that listeners prefer rhythms that are subtly flawed. A 2011 Harvard study found that music performed by robotic drummers and other machines often strikes our ears as being too precise. “There is something perfectly imperfect about how humans play rhythms,” says Holger Hennig, the Harvard physics researcher who led the study. Hennig discovered that when experienced musicians play together, they not only make mistakes, they also build off these small variations to keep a live song from sounding pat.

Powered by Reddit Powered by Reddit