Start Up No.949: AI finds missing voters, US cordcutters keep cutting, Creative Commons photos safe, White House goes deepfake, and more

YouTube’s algorithms can lead us down a rabbit hole – and they’re getting better at it. Photo by Kevin Dooley on Flickr.

You can sign up to receive each day’s Start Up post by email. You’ll need to click a confirmation link, so no spam.

A selection of 8 links for you. Use them wisely. I’m @charlesarthur on Twitter. Observations and links welcome.

How YouTube’s recommendation algorithm really works • The Atlantic

Alexis Madrigal:


YouTube wants to recommend things people will like, and the clearest signal of that is whether other people liked them. Pew found that 64% of recommendations went to videos with more than a million views. The 50 videos that YouTube recommended most often had been viewed an average of 456 million times each. Popularity begets popularity, at least in the case of users (or bots, as here) that YouTube doesn’t know much about.

On the other hand, YouTube has said in previous work describing its algorithm that users like fresher content, all else being equal. But it takes time for a post to build huge numbers of views and signal to the algorithm that it’s worth promoting. So, the challenge becomes how to recommend “new videos that users want to watch” when those videos are new to the system and low in views. (Finding fresh, potentially hot videos is important, YouTube researchers have written, for “propagating viral content.”)

Pew’s research reflects this: About 5% of the recommendations went to videos with fewer than 50,000 views. The system learns from a video’s early performance, and if it does well, views can grow rapidly. In one case, a highly recommended kids’ video went from 34,000 views when Pew first encountered it in July to 30 million in August.

The behavior of the system was explicable in a few other ways, too, especially as it adapted to making more clicks inside YouTube’s system. First, as Pew’s software made choices, the system selected longer videos. It’s as if the software recognizes that the user is going to be around for a while, and starts to serve up longer fare. Second, it also began to recommend more popular videos regardless of how popular the starting video was.

These conditions were almost certainly not hard coded into the algorithmic decision making. Like most of the Google sister companies, YouTube uses deep-learning neural networks, a kind of software that retunes its outputs based on the data fed into it. It’s not that a YouTube engineer said, “Show people kids’ videos that are progressively longer and more popular,” but rather that the system statistically deduced that this would optimize along all the dimensions YouTube desires.


The idea that YouTube’s algorithm is now going beyond simple understanding – why this video and not that? – and entering the point where it’s just trying to suck people in is quite unsettling when you consider that similar algorithms can beat the world’s best Go players.

At some point does it find a video sequence that nobody will be able to tear themselves away from?
link to this extract

Pay TV just lost one million subscribers in biggest quarterly loss ever • Exstreamist

Rob Toledo:


An executive at a major cable company said a few years ago that cutting the cord was a fad, and would not impact business in the long term.

This conversation was over two years ago, and almost every quarter since then, we have written the same article: that a record number of people are cutting the cord, ditching their expensive cable packages for more more flexible streaming services.

BTIG media analyst Rich Greenfield tweeted this week that cable and satellite companies lost over one million subscribers in the last quarter. This is the biggest loss of subscribers in one quarter seen by the pay TV industry ever.

Let that sink in. Over one million (now former) subscribers ditched their cable in a three month period.

This is not an anomaly, as each quarter for at least the past three years has seen quarterly falloff of cable and satellite customers.

In 2016, there were an estimated 99 million pay TV subscribers in the United States, with each year seeing a big decline, with estimates expecting this number to keep dropping.

While it used to be fairly simple in that a consumer several years ago would cancel their subscription and simply sign up for Netflix, the number of streaming services is on a rapid rise as well, which analysts believe has accelerated the cancellation of cable.


I wonder if Americans actively like the lack of adverts on services such as Netflix. This trend looks set to continue.
link to this extract

The Free Music Archive is closing this month • The Verge

Bijan Stephen:


The Free Music Archive was founded in 2009, the same year Barack Obama was inaugurated as this country’s first black president. As a project directed by the legendary Jersey City radio station WFMU, it was to be a “library of high-quality, legal audio downloads,” a place where artists could share their music and listeners could enjoy it for free. Now, following a funding shortage, the FMA plans to close sometime this month.

“The future is uncertain, has been my mantra lately,” says Cheyenne Hohman, who’s been the director of the Free Music Archive since 2014. The shutdown date was initially November 9th, but it has since been pushed back to November 16th because the FMA is in early talks with four different organizations that are interested in taking the project over. “The site may stay up a little bit longer to ensure, at the very least, that our collections are backed up on and the Wayback Machine.”

Even so, it’s not a perfect solution. “If it just goes into, it’s going to be there in perpetuity, but it’s not going to be changing at all,” Hohman says. “It’s not going to be the same thing, that sort of community and project that it was for … almost 10 years.”


link to this extract

Another use for AI: finding millions of unregistered voters • The New York Times

Steve Lohr:


For the last four years, Mr. Jonas has used his software for a multistate project known as Electronic Registration Information Center that identifies eligible voters and cleans up voter rolls. Since its founding in 2012, the nonprofit center has identified 26 million people who are eligible but unregistered to vote, as well as 10 million registered voters who have moved, appear on more than one list or have died.

“I have no doubt that more people are voting as a result of ERIC,” said John Lindback, a former senior election administrator in Oregon and Alaska who was the center’s first executive director.
Voter rolls, like nearly every aspect of elections, are a politically charged issue. ERIC, brought together by the Pew Charitable Trusts, is meant to play it down the middle. It was started largely with professional election administrators, from both red and blue states.

But the election officials recognized that their headaches often boiled down to a data-handling challenge. Then Mr. Jonas added his technology, which has been developed and refined for decades. It is artificial intelligence software fine-tuned for spotting and resolving identities, whether people or things.

“Every time you get two pieces of junk mail from the same place, that’s an entity resolution problem,” Mr. Jonas said. “They’re missed, but entity resolution problems are everywhere.”

Shortly after the election administrators tapped him, Mr. Jonas sketched out how his technology might be applied to their challenges. And they needed to take a very different path than another data-matching initiative, the Interstate Voter Registration Crosscheck System, which was already underway.

Crosscheck was begun in 2005, led by Ron Thornburgh, then the Republican secretary of state in Kansas, and later championed by Kris Kobach, the Republican secretary of state who is running for governor of Kansas.


I’m sure this will shock you, but Crosscheck produced lots of false positives which disenfranchised people wrongly, whereas ERIC is intended to both improve voter access and clean voter rolls so they’re more accurate.
link to this extract

2018 iPad Pro review: “What’s a computer?” • Ars Technica

Samuel Axon:


iOS is excellent software for phones, but it is not up to the task of driving creative professionals’ power user ambitions on a tablet—not even close. Copying, pasting, and editing text is an enormous hassle if you’re doing anything other than scribbling a couple of notes or shooting off an email. The multitasking features expanded upon in iOS 11 are still neat, and the iPhone X-like gesture for swiping quickly between apps like you’d swipe between Spaces on a Mac is powerful. But using this machine, you’ll be laboriously swiping between apps constantly to do the smallest things.

I already talked about the iPad Pro’s frustrating limitations of the USB-C connection and the lack of OS-wide support for external drives. This stuff is essential for power users, and iOS just doesn’t deliver. If you’ve ever used an iPad for productivity before, you know what I’m talking about. It’s infuriatingly close, and it gets marginally closer with each passing year, yet it never quite seems to arrive.

The problems here are surprising in part because they are very un-Apple. The company’s pitch to consumers and professionals alike has always been about the advantages of end-to-end integration, and that includes software and hardware built to work well together. But iOS feels like it is built for a completely different device, given that the new iPad Pro’s ambitions are much greater than those of prior iPads, or of the iPhone.

Then there’s app support. The OS’s limitations would be more tolerable if third-party (and first-party) apps picked up the slack, and the development tools are there to make it happen. Unfortunately, too many of the “pro” apps for the iPad Pro are deliberately stripped down for the tablet. And there are numerous tools that creatives and professionals would love to see on the iPad that just aren’t there.


I don’t agree. I’ve written and edited most of a book on an iPad Pro; I’ve produced and edited and given presentations from one. His criticism of the music element – that there’s no 3.5mm jack, and you need a wired connection for good audio editing – is strong on its face, but they you buy a $80 7-in-1 USB-C dongle from Hypershop which provides multiple USB-A, HDMI, SD, USB-C… and a 3.5mm jack.

Sure, dongles are an annoyance. But it’s there.
link to this extract

The Commons: the past is 100% part of our future • Flickr Blog

Don MacAskill is CEO of SmugMug (and now Flickr too):


The Big Three at Yalta

Photos from NASA, The Smithsonian, The National Archives UK, and The British Library, for example, have been shared in The Flickr Commons. As part of The Flickr Commons, all these organizations already were Pro or have received a free Pro account from us, so they have unlimited storage.

The Creative Commons (CC) organization has developed a suite of licenses that give individual photographers or groups great tools for licensing their photography for others to freely use. The photographer keeps their copyright and gives the public an easy way to use their images as long as the license terms are followed.

The Flickr Commons and Creative Commons are different, thus our storage changes affect each differently (or not at all).

Are Commons Photos Being Deleted?

No. And once more for good measure: no, Commons photos are not being deleted.

The Flickr Commons photos (those uploaded by the archival, governmental, etc. institutions we are working with) are safe. We are extremely proud of these partnerships. These photos won’t be deleted as a result of any of our announced changes. The only reason they’d disappear is if the organization that uploaded them decided to delete them.

Photos that were Creative Commons licensed before our announcement are also safe. We won’t be deleting anything that was uploaded with a CC license before November 1, 2018. Even if you had more than 1,000 photos or videos with a CC license. However, if you do have more than 1,000 photos or videos uploaded, you’ll be unable to upload additional photos after January 8, 2019, unless you upgrade to a Pro account.


Phew. (All the photos used to illustrate The Overspil are CC-licensed.)
link to this extract

New auto safety technologies push repair bills up • IEEE Spectrum

Robert Charette:


There is little debate over whether advanced driver assistance systems (ADAS) could reduce both the number and severity of vehicle crashes. A 2015 study [PDF] by the Motor & Equipment Manufacturers Association and Boston Consulting Group says equipping new vehicles with technologies including blind-spot warning, lane-departure warning, and collision-mitigation braking systems could eventually save 10,000 lives and eliminate or reduce the severity of millions of nonfatal injuries from motor vehicle accidents.

The additional cost of these advanced driver-assistance systems has slowed their adoption, however. A collision-mitigation system alone can increase the cost of a new vehicle by US $1,500 or more. Further, new research by the American Automobile Association (AAA) shows a significant increase in the cost of repairing these systems after even a minor accident. This finding could put off auto buyers even more.

According to AAA research, vehicles equipped with advanced safety features “can cost twice as much to repair following a collision due to expensive sensors and their calibration requirements.” For instance, a windshield repair for vehicles equipped with automatic emergency braking, adaptive cruise control, and lane departure warning systems could run as high as $1,650, the AAA found. This is in comparison to a typical windshield replacement cost which runs $210 to $230, although it is not uncommon to see it go as high as $500, according to Glass America.


Would it make you drive more carefully, perhaps?
link to this extract

White House shares doctored video to support punishment of journalist Jim Acosta • The Washington Post

Drew Harwell:


Critics said that video — which sped up the movement of Acosta’s arms in a way that dramatically changed the journalist’s response — was deceptively edited to score political points. That edited video was first shared by Paul Joseph Watson, known for his conspiracy-theory videos on the far-right website Infowars.

Watson said he did not change the speed of the video and that claims he had altered it were a “brazen lie.” Watson, who did not immediately respond to requests for comment, told BuzzFeed he created the video by downloading an animated image from conservative news site Daily Wire, zooming in and saving it as a video — a conversion he says could have made it “look a tiny bit different.”

Side-by-side comparisons support claims from fact-checkers and experts such as Jonathan Albright, research director of the Tow Center for Digital Journalism at Columbia University, who argued that crucial parts of the video appear to have been altered so as to distort the action.

A frame-by-frame breakdown by Storyful, a social-media intelligence firm that verifies media content, found that the edited video included repeated frames that did not appear in the original footage. The repeated frames were shown only at the moment of contact and made Acosta’s arm movement look more exaggerated, said Shane Raymond, a journalist at Storyful.

The video has quickly become a flashpoint in the battle over viral misinformation, turning a live interaction watched by thousands in real time into just another ideological tug-of-war. But it has also highlighted how video content — long seen as an unassailable verification tool for truth and confirmation — has become as vulnerable to political distortion as anything else.


First: how pathetic that the White House can’t use its own video. Second: utterly pathetic that it uses something from a conspiracy site; have they no pride? Third: didn’t expect that we’d be talking about doctored videos literally the day after I linked to a New Yorker article on it. Fourth: that the US can’t have any topic at all without it descending into partisan fury is a sad indictment of its political immaturity. It’s actually going backwards.
link to this extract

Errata, corrigenda and ai no corrida: none notified

2 thoughts on “Start Up No.949: AI finds missing voters, US cordcutters keep cutting, Creative Commons photos safe, White House goes deepfake, and more

  1. One weird thing about YouTube’s record algorithm: it seems to be unaware of the time of day, and of my personal patterns during the day.
    I work better in small bursts interspersed with distractions, and with background noise , so I’ve got a YouTube tab open most of the time. What I want on it during the day while working, during the day while on break, and in the evening varies wildly. I haven’t noticed recommendations adjusting to that which makes them mostly useless.

  2. Long ago, I thought it would do good for technical people to educate policy people about technical issues.
    Now, I’m coming to believe that any good I could do is in educating technical people that policy people don’t care about technical information – it’s just irrelevant at best, or even harmful at worst, to the way they (non-techies) achieve their goals.
    The Acosta Affair is layers and layers of tribal warfare.
    Nobody involved wants to hear about frame rates and compression artifacts.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.