“A DOLL POSSESSED BY A DEMON”
Recommender systems power YouTube’s controversial kids’ videos
Familiar cartoon characters are placed in bizarre scenarios, sometimes by human content creators, sometimes by automated systems, for the purpose of attracting views and ad money. First, from the New York Times:
“But the app [YouTube Kids] contains dark corners, too, as videos that are disturbing for children slip past its filters, either by mistake or because bad actors have found ways to fool the YouTube Kids algorithms.
“In recent months, parents like Ms. Burns have complained that their children have been shown videos with well-known characters in violent or lewd situations and other clips with disturbing imagery, sometimes set to nursery rhymes. Many have taken to Facebook to warn others, and share video screenshots showing moments ranging from a Claymation Spider-Man urinating on Elsa of ‘Frozen’ to Nick Jr. characters in a strip club.”
Full piece by SAPNA MAHESHWARI in the Times here.
On Medium, JAMES BRIDLE expands on the topic, and criticizes the structure of YouTube itself for incentivizing these kinds of videos, many of which have millions of views.
“These videos, wherever they are made, however they come to be made, and whatever their conscious intention (i.e. to accumulate ad revenue) are feeding upon a system which was consciously intended to show videos to children for profit. The unconsciously-generated, emergent outcomes of that are all over the place.
“While it is tempting to dismiss the wilder examples as trolling, of which a significant number certainly are, that fails to account for the sheer volume of content weighted in a particularly grotesque direction. It presents many and complexly entangled dangers, including that, just as with the increasing focus on alleged Russian interference in social media, such events will be used as justification for increased control over the internet, increasing censorship, and so on.”
Link to Bridle’s piece here.
Margarita comments: YouTube knows next to nothing about the content of its videos. This is despite the fact that research for video information retrieval has been progressing steadily in the last decade, with training and test sets being based on YouTube videos, for a lot of papers. The negligence on YouTube’s part is somewhat justified by the resource intensity of performing video signal information retrieval, considered in conjunction with the enormous rate with which gigabytes of new content emerge on the platform. However, YouTube has ignored other less sophisticated ways of learning about content too: through text analysis on comments or external links analysis, for example. (And YouTube has only recently expanded its algorithm to focus on time spent watching in addition to simply tracking clicks, likes, and search queries.)
- From the Verge, after the two pieces above were published: “YouTube says it will crack down on bizarre videos targeting children.” Link.
- An extremely long 2014 Slate Star Codex piece on similar themes: “Every single citizen hates the system, but for lack of a good coordination mechanism it endures. From a god’s-eye-view, we can optimize the system to ‘everyone agrees to stop doing this at once’, but no one within the system is able to effect the transition without great risk to themselves.” Link. Michael writes:Without meaning to, Scott Alexander gives a massive list of anti-social equilibria that could be pulled from a hypothetical reserach program founded on Greif.
- This video from May 2017 has more examples of the issue (it’s not fun to watch). Link.
- Mike Cook, an AI researcher who promotes generative techniques through the Procedural Generation Jam, published a quick response on his blog, which is understandably more optimistic about the possibilities of directing this technology. Link.
“EMOTIONS AS INFECTIOUS DISEASES”
Examining the transmission of emotions
“Human populations are arranged in social networks that determine interactions and influence the spread of diseases, behaviors and ideas. We evaluate the spread of long-term emotional states across a social network. We introduce a novel form of the classical susceptible–infected–susceptible disease model which includes the possibility for ‘spontaneous’ (or ‘automatic’) infection, in addition to disease transmission (the SISa model). Using this framework and data from the Framingham Heart Study, we provide formal evidence that positive and negative emotional states behave like infectious diseases spreading across social networks over long periods of time […] Determining to what extent particular emotions or behaviors are infectious is a promising direction for further research with important implications for social science, epidemiology and health policy. Our model provides a theoretical framework for studying the interpersonal spread of any state that may also arise spontaneously, such as emotions, behaviors, health states, ideas or diseases with reservoirs.”
Full paper by Alison L. Hill et al. here.
ht Lauren, who comments: Interesting to consider this social, interpersonal notion in comparison to popular interpretations of mental illness wherein the causes are largely considered to be at least in part due to some biological predisposition in a given individual. Here, the assumption is that one is “susceptible” simply by being a person (rather than, for example, a person with a specific genetic characteristic which makes them more or less likely to be discontent regardless of social contact).
“THE WEB CENTIPEDE”
How Reddit and 4chan influence Twitter and mainstream news
“Although previous work has examined information cascades, rumors, and hoaxes, to the best of our knowledge, very little work provides a holistic view of the modern information ecosystem. This knowledge, however, is crucial for understanding the alternative news world and for designing appropriate detection/mitigation strategies. Anecdotal evidence and press coverage suggest that alternative news dissemination might start on fringe sites, eventually reaching mainstream online social networks and news outlets. Nevertheless, this phenomenon has not been measured and no thorough analysis has focused on how news moves from one online service to another.
“In this paper, we address this gap by providing the first large scale measurement of how mainstream and alternative news flows through multiple social media platforms. We focus on the relationship between three fundamentally different social media platforms, Reddit, Twitter, and 4chan…”
Full paper by Zannettou et al. here.
- David Evans summarizes 147 papers in development economics, giving a sense of the scope of the field in 2017. Link. ht Sidhya
- “Now, in response to a Freedom of Information Act (FOIA) request from TCF, the Department of Education has provided information that sorts all 98,868 borrower defense claims received as of August 15, 2017, by school. The data represents the first-ever public record of the number of claims students have filed against each and every higher education institution in the country.” Link.
- On the complex infrastructure of Netflix. Link.
- On the “broke and broken” National Flood Insurance program. Link. ht Will
- “In a new paper published on the arXiv, researchers say they may have figured out a way to mitigate the problem for algorithms that are difficult for outsiders to examine—so-called “black box” systems.” Link. ht Margarita
- “Fooling an AI with a couple of pixels is called an adversarial example, and potential attackers can use them to trick or confuse an AI. For the first time, researchers have tricked an AI into thinking a real-world object—in this case, a 3D printed turtle—is a rifle.” Link. ht Margarita
- A paper on “the myths of data-driven campaigning.” Link.
- Mark Koyama: “Could Rome have had an industrial revolution?” Link.
- “The Florida—America’s biggest swing state—of early modern Europe appears to be Strasbourg in France: 30% of all witch trials on the continent occurred within 300 miles (500 km) of the city.” Link.
- For fun: the careful, careful process of making mirrors for the world’s largest telescope. Link.
Each week we highlight research from a graduate student, postdoc, or early-career professor. Send us recommendations: firstname.lastname@example.org.