Start Up No.1,100: the Huawei un-ban puzzle, 5G’s infrastructure problem, how Uber worsens congestion and pollution, ‘kayfabe’ and Trump, and more

A new research technique uses pixel differences to detect or prevent deepfakes. CC-licensed photo by Dorian 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 10 links for you. Maybe it’s your eyesight? I’m @charlesarthur on Twitter. Observations and links welcome.

Detecting deepfakes by looking closely reveals a way to protect against them • The Conversation

Siwei Lyu is Professor of Computer Science and the director of the Computer Vision and Machine Learning Lab at the University at Albany, State University of New York:


Some of my research group’s earlier work allowed us to detect deepfake videos that did not include a person’s normal amount of eye blinking – but the latest generation of deepfakes has adapted, so our research has continued to advance.

Now, our research can identify the manipulation of a video by looking closely at the pixels of specific frames. Taking one step further, we also developed an active measure to protect individuals from becoming victims of deepfakes.

In two recent (1) research papers (2), we described ways to detect deepfakes with flaws that can’t be fixed easily by the fakers.

When a deepfake video synthesis algorithm generates new facial expressions, the new images don’t always match the exact positioning of the person’s head, or the lighting conditions, or the distance to the camera. To make the fake faces blend into the surroundings, they have to be geometrically transformed – rotated, resized or otherwise distorted. This process leaves digital artifacts in the resulting image.

You may have noticed some artifacts from particularly severe transformations. These can make a photo look obviously doctored, like blurry borders and artificially smooth skin. More subtle transformations still leave evidence, and we have taught an algorithm to detect it, even when people can’t see the differences…

…As we develop this algorithm, we hope to be able to apply it to any images that someone is uploading to social media or another online site. During the upload process, perhaps, they might be asked, “Do you want to protect the faces in this video or image against being used in deepfakes?” If the user chooses yes, then the algorithm could add the digital noise, letting people online see the faces but effectively hiding them from algorithms that might seek to impersonate them.


Explained, of course, with videos.
unique link to this extract

Kudlow: US sales to Huawei won’t imperil national security • The New York Times

Associated Press:


[White House economics adviser Larry] Kudlow told “Fox News Sunday” and CBS’ “Face the Nation” that Huawei will remain on an American blacklist as a potential security threat. He stressed that additional US licensing “will be for what we call general merchandise, not national security sensitive,” such as chips and software generally available around the world.

“What’s happening now is simply a loosening up for general merchandise,” Kudlow said. “This is not a general amnesty.”

Trump made the announcement Saturday after meeting with China’s Xi Jinping at the Group of 20 summit in Japan. Trump said US companies could make the sales if the transactions don’t present a “great, national emergency problem.”

Several Republican senators immediately expressed concerns. In a tweet Saturday, Sen. Marco Rubio of Florida called the decision a “catastrophic mistake.” Sen. Lindsey Graham [Republican, South Carolina], told CBS that Trump’s agreement was “clearly a concession,” and also said it would be a mistake if sales to Huawei involved “major technology.”

Sen. John Barrasso, R-Wyo., described the Chinese company as a clear threat to US national security. “To me, Huawei in the United States would be like a Trojan horse ready to steal more information from us,” he told NBC’s “Meet the Press.”


The reversal on Huawei was predictable enough – Trump doesn’t do anything on principle, even when everyone around him knows that something should be done on principle – but this is just baffling. American companies were banned from selling to Huawei, and it looked like it would cripple the Chinese company. So is Google still on the banned list, given that its products aren’t generally available?
unique link to this extract

The downside of 5G: overwhelmed cities, torn-up streets, a decade until completion • WSJ

Christopher Mims:


5G networks don’t work like previous wireless cellular networks. Where 2G, 3G and even 4G rely on large towers with powerful antennas that can cover many square miles, the shorter-range, higher-frequency radio waves used by 5G networks—essential to their ability to deliver the 10- to 100-times faster speeds they promise—mean that 5G networks must have small cells placed much closer together.

Typically these small cells must be placed about 800 to 1,000 feet apart, says AT+T’s Ms. Knight. Small-cell antennas are typically the size of a pizza box, but can be much larger, and require both a fiber-optic connection to the internet and access to power. They go wherever there’s space: on buildings, new 5G-ready telephone poles and, often, retrofitted lampposts.

In 2018, the US had 349,344 cell sites, according to CTIA, a wireless industry trade organization. The organization estimates that—to achieve full 5G coverage—carriers will have to roll out an additional 769,000 small cells by 2026.

This rollout could mean three or four different carriers will be arriving at your street, each trying separately to dig to bury fiber. (And yes, fiber-optic cable almost always has to be buried.)


Terrific piece about the real-world implications of getting this done. The implication (to me at least) is that rural areas will be unlikely to see 5G: its range is too short and the cost disproportionate to the benefits it can provide compared to 4G, with its greater range.
unique link to this extract

GrubHub is buying up thousands of restaurant web addresses. That means Mom and Pop can’t own their slice of the internet • New Food Economy

H. Claire Brown:


Grubhub purchased three different domains containing versions of Shivane’s restaurant’s name—in 2012, 2013, and 2014. “I never gave them permission to do that,” she says. 

Shivane believes GrubHub purchased her restaurant’s web domain to prevent her from building her own online presence. She also believes the company may have had a special interest in owning her name because she processes a high volume of orders. She rattles off a list of names of local restaurants that she suspects may be in the same predicament. I find versions of about half those names on the list of GrubHub-owned domains. 

Additionally, it appears GrubHub has set up several generic, templated pages that look like real restaurant websites but in fact link only to GrubHub. These pages also display phone numbers that GrubHub controls. The calls are forwarded to the restaurant, but the platform records each one and charges the restaurant a commission fee for every order, according to testimony from GrubHub executives at a hearing at New York City Hall on Thursday. This happens on the GrubHub platform itself, too. The phone numbers you see displayed in the app typically aren’t a restaurant’s actual phone number, they’re the numbers that GrubHub uses to make sure it’s getting its commission. 


GrubHub says it’s doing it as a service to restaurants: “we have created microsites for them as another source of orders and to increase their online brand presence. Additionally, we have registered domains on their behalf, consistent with our restaurant contracts.” But now has stopped doing it. Odd.
unique link to this extract

‘Empty’ Uber cabs driving pollution and congestion • The Sunday Times

Nicholas Hellen:


Uber was launched in Britain with a promise that its smart technology, which matches passengers with the nearest vehicle for hire, would reduce traffic.

In 2014 Travis Kalanick, then its chief executive, told the Institute of Directors: “In our current model here in London there are 7½ cars taken off the road for every fully utilised Uber that is on the road.”

But James Farrar, a former Uber driver who obtained the figures after a two-year legal battle, said they provided hard evidence that the company’s approach added to congestion.

“They are competing on immediacy and availability and they do not carry any of the costs [of buying the cars]. That is going to lead to oversupply. You will cause congestion and these drivers will not have enough work.”

The figures, which tracked three drivers for a combined 7,500 hours, confirm that when they are looking for their next job they do not park, but typically spend 94% of their time cruising the streets, to maximise their chances of being offered another passenger.

David Dunn, 58, one of the three drivers, said he quit driving for Uber in Glasgow because he was having to work 80-hour weeks to recoup the £37,000 that he had spent on a car.


This doesn’t of course show how much of the time non-Uber taxis spend noodling around looking for trade, but it seems reasonable to think that if there are fewer taxis available, they spend less time not carrying passengers. Given that, maybe you’d want a licensing authority to mandate a maximum number of cars at some times, or that a certain proportion be electric (though that won’t help congestion), or similar. It’s the same story in the US.
unique link to this extract

Inside Apple’s long goodbye to design chief Jony Ive • Bloomberg

Mark Gurman:


He was in charge of a roughly two-dozen person design team that included artists whose passions extended to the development of surfboards, cars, and even DJing on weekends. Many of their spouses worked as designers, too…

…some people familiar with Apple are already worried about the new design leadership. Now that Ive is officially leaving, longtime studio manager Evans Hankey will run the hardware design group, Apple said. Hankey is a great team leader, but Apple now lacks a true design brain on its executive team, which is a concern, a person familiar with the design team said.

Hankey and Dye will report to Jeff Williams, Apple’s chief operating officer. While Williams is a talented executive, some people familiar with matter believe the shift is another sign of Apple becoming more of an operations company. Apple declined to comment.

“The design team is made up of the most creative people, but now there is an operations barrier that wasn’t there before,” one former Apple executive said. “People are scared to be innovative.”

…The design team is taking on this challenge without veteran members. Christopher Stringer and Daniele De Iuliis, a pair of key Ive lieutenants, kicked off the departures a few years ago, with Daniel Coster leaving to lead design at GoPro in 2016. The team lost three members in the past six months: Julian Hoenig, Rico Zorkendorfer and Miklu Silvanto.

While each Apple designer specializes in specific product lines, they all contribute to each other’s products and plans. That means losing an individual designer is still a big deal, a former Apple executive said. “The design studio has no secrets,” this person said. “They all know what each other is working on.”


It’s definitely worth re-reading the New Yorker article from 2015 about Ive in the light of this announcement. It makes it feel a lot different. I didn’t think that Steve Jobs leaving Apple was the catastrophe some did. But Apple without Jobs and Ive isn’t the same beast.
unique link to this extract

Publishers says Apple is changing Apple News Plus, its subscription bundle • Business Insider

Lucia Moses:


publishers have had mixed views on Plus so far. Some saw it as a way to reap revenue from Apple’s massive customer base as many of them struggle to grow ad revenue. (Apple is sharing half of the revenue with publishers based on how much time users spend with the given publishers’ content, knowledgeable sources said.) The Wall Street Journal, New York magazine, Vox, and TheSkimm, opted in, as did Business Insider. Big magazine chains including Hearst, Meredith, and Condé Nast are also participating in the bundle, but are contractually obligated to do so as former owners of the app, according to sources.

Some publishers had concerns that the bundle would not produce meaningful revenue and that it would cannibalize their own subscription businesses, though. Major subscription publications The New York Times and Washington Post opted out of the bundle.

Apple gave away Plus for free for the first month, and in its first two days, it reportedly had about 200,000 subscribers, which is about what Texture had. But three months in, publishing execs who spoke for this article said the subscription revenue they’d gotten from the service was underwhelming based on two months of data after the trial ended.

One publishing exec said Apple projected publishers would get 10 times the revenue they made from Texture at the end of Apple News Plus’ first year. “It’s one twentieth of what they said,” the exec said. “It isn’t coming true.”


Got to admit, I don’t open Apple News (the app) from one month’s end to the next. The fact that it defines links using its own URL schema is almost worse than Google’s AMP. There are better news aggregators.
unique link to this extract

April 2017: How wrestling explains Alex Jones and Donald Trump • The New York Times

Nick Rogers, in April 2017:


Although the etymology of the word is a matter of debate, for at least 50 years “kayfabe” has referred to the unspoken contract between wrestlers and spectators: We’ll present you something clearly fake under the insistence that it’s real, and you will experience genuine emotion. Neither party acknowledges the bargain, or else the magic is ruined.

To a wrestling audience, the fake and the real coexist peacefully. If you ask a fan whether a match or backstage brawl was scripted, the question will seem irrelevant. You may as well ask a roller-coaster enthusiast whether he knows he’s not really on a runaway mine car. The artifice is not only understood but appreciated: The performer cares enough about the viewer’s emotions to want to influence them. Kayfabe isn’t about factual verifiability; it’s about emotional fidelity.

Although their athleticism is impressive, skilled wrestlers captivate because they do what sociologists call “emotional labor” — the professional management of other people’s feelings. Diners expect emotional labor from their servers, Hulkamaniacs demand it from their favorite performer, and a whole lot of voters desire it from their leaders.

The aesthetic of World Wrestling Entertainment seems to be spreading from the ring to the world stage. Ask an average Trump supporter whether he or she thinks the president actually plans to build a giant wall and have Mexico pay for it, and you might get an answer that boils down to, “I don’t think so, but I believe so.” That’s kayfabe. Chants of “Build the Wall” aren’t about erecting a structure; they’re about how cathartic it feels, in the moment, to yell with venom against a common enemy.


“Kayfabe” feels as though it describes quite a lot of politics right now. But definitely Trump.
unique link to this extract

Boeing’s 737 Max software outsourced to $9-an-hour engineers • Bloomberg

Peter Robison:


Increasingly, the iconic American planemaker and its subcontractors have relied on temporary workers making as little as $9 an hour to develop and test software, often from countries lacking a deep background in aerospace – notably India.

In offices across from Seattle’s Boeing Field, recent college graduates employed by the Indian software developer HCL Technologies Ltd. occupied several rows of desks, said Mark Rabin, a former Boeing software engineer who worked in a flight-test group that supported the Max.

The coders from HCL were typically designing to specifications set by Boeing. Still, “it was controversial because it was far less efficient than Boeing engineers just writing the code,” Rabin said. Frequently, he recalled, “it took many rounds going back and forth because the code was not done correctly.”

Boeing’s cultivation of Indian companies appeared to pay other dividends. In recent years, it has won several orders for Indian military and commercial aircraft, such as a $22bn one in January 2017 to supply SpiceJet Ltd. That order included 100 737-Max 8 jets and represented Boeing’s largest order ever from an Indian airline, a coup in a country dominated by Airbus.

Based on resumes posted on social media, HCL engineers helped develop and test the Max’s flight-display software, while employees from another Indian company, Cyient Ltd., handled software for flight-test equipment.

In one post, an HCL employee summarized his duties with a reference to the now-infamous model, which started flight tests in January 2016: “Provided quick workaround to resolve production issue which resulted in not delaying flight test of 737-Max (delay in each flight test will cost very big amount for Boeing).”


Boeing says those programmers didn’t do the MCAS software that’s blamed for the crashes. There seems to be a deeper problem at Boeing, dumping its institutional memory (experienced staff) on the basis that its products are “mature”.
unique link to this extract

Amazon’s facial recognition creates dystopic future for trans and nonbinary people • Jezebel

Dhruv Mehrotra and Anna Merlan:


We reached out to all the companies using Rekognition for facial analysis as listed on Amazon’s information page for the product. Only two got back to us in a meaningful way. One was Limbik, a startup that uses machine learning to help companies understand whether their videos are being watched, and by who. They told us that Amazon’s binary gender settings posed a problem for them: “We have noticed this as an issue for us, as the better we can tag videos with proper tags the more accurate we can be with predictions and improvement recommendations. It would be best if we could get this type of information as it would help us categorize videos better and help with prediction.”

Without that information, Limbik added, they have to specify to customers what their analysis, using Rekognition, does and doesn’t do. “Since Rekognition only returns a binary value for gender, we have to make sure that, to customers, we specify that it is biological sex that is examined and not gender specifically and that it isn’t perfect. We have internal conversations about this issue and have discussed remedies but as we can have upwords of 1000 tags connected to a video coming from other Rekognition services, our internal tagging methods, manual human tagging and other methods, we haven’t found a good way to address this.”


Umm. The thing is, the recognition system is making determinations based on the shape of the face, which is sex-chromosome-determined, not gender-determined. To use a broad metaphor, it’s about where you were born, not what town you live in now. Nowhere in the story is this acknowledged, though.
unique link to this extract

Errata, corrigenda and ai no corrida: none notified

2 thoughts on “Start Up No.1,100: the Huawei un-ban puzzle, 5G’s infrastructure problem, how Uber worsens congestion and pollution, ‘kayfabe’ and Trump, and more

  1. re. Trump “even when everyone around him knows that something should be done on principle”. I’d laugh if it weren’t so sad. Really ? Who in there are you accusing of being principled ?

    Also, there’s still no beef in that burger: Huawei has buggy/hole-y software and there’s some level of gov. involvement in it, but that’s true of all tech and Chinese companies resp. (of all companies doing business in China actually). I understand letting China onto our infrastructure is iffy. I don’t understand how letting Trump/McConnell/Kushner isn’t, or what this has anything to do with clients (phones), and why in-depth audits can’t alleviate most fears (and find bugs in the process)

  2. re Kayfabe: from
    “If you watched the entire election cycle and concluded that Trump was nothing but a lucky clown, you missed one of the most important perceptual shifts in the history of humankind. I’ll fix that for you in this book.”

    Adams was one of the earliest public figures to predict Trump’s win, doing so a week after Nate Silver put Trump’s odds at 2 percent in his blog. The mainstream media regarded Trump as a novelty and a sideshow. But Adams recognized in Trump a level of persuasion you only see once in a generation.

    Trump triggered massive cognitive dissonance and confirmation bias on both the left and the right. We’re hardwired to respond to emotion, not reason. We might listen to 10 percent of a speech—a hand gesture here, a phrase there—and if the right buttons are pushed, we decide we agree with the speaker and invent reasons to justify that decision after the fact.

    The point isn’t whether Trump was right or wrong, good or bad. Win Bigly goes beyond politics to look at persuasion tools that can work in any setting—the same ones Adams saw in Steve Jobs when he invested in Apple decades ago. For instance:

    · If you need to convince people that something is important, make a claim that’s directionally accurate but has a big exaggeration in it. Everyone will spend endless hours talking about how wrong it is and will remember the issue as high priority.
    · Stop wasting time on elaborate presentation preparations. Inside, you’ll learn which components of your messaging matter, and where you can wing it.
    · Planting simple, sticky ideas (such as “Crooked Hillary”) is more powerful than stating facts. Just find a phrase without previous baggage that grabs your audience at an emotional level.

    Adams offers nothing less than “access to the admin passwords to human beings.” This is a must read if you care about persuading others in any field—or if you just want to resist the tactics of emotional persuasion when they’re used on you.

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.