Start Up No.1,039: Facebook’s biased algorithm, Apple cuts HomePod price, US forces out Chinese investors, Trump whistleblowers abound, and more

You think you’re sending the army to the right place – but what if the reconnaissance photo is a deepfake? CC-licensed photo by Enough Project 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 9 links for you. Indicatively voted. I’m @charlesarthur on Twitter. Observations and links welcome.

Facebook ad algorithm is a race and gender stereotyping machine • The Intercept

Sam Biddle:


Rather than targeting a demographic niche, the researchers requested only that their ads reach Facebook users in the United States, leaving matters of ethnicity and gender entirely up to Facebook’s black box. As Facebook itself tells potential advertisers, “We try to show people the ads that are most pertinent to them.” What exactly does the company’s ad-targeting black box, left to its own devices, consider pertinent? Are Facebook’s ad-serving algorithms as prone to bias like so many others? The answer will not surprise you.

For one portion of the study, researchers ran ads for a wide variety of job listings in North Carolina, from janitors to nurses to lawyers, without any further demographic targeting options. With all other things being equal, the study found that “Facebook delivered our ads for jobs in the lumber industry to an audience that was 72% white and 90% men, supermarket cashier positions to an audience of 85% women, and jobs with taxi companies to a 75% black audience even though the target audience we specified was identical for all ads.” Ad displays for “artificial intelligence developer” listings also skewed white, while listings for secretarial work overwhelmingly found their way to female Facebook users.

Although Facebook doesn’t permit advertisers to view the racial composition of an ad’s viewers, the researchers said they were able to confidently infer these numbers by cross-referencing the indicators Facebook does provide, particularly regions where users live, which in some states can be cross-referenced with race data held in voter registration records.

In the case of housing ads — an area Facebook has already shown in the past has potential for discriminatory abuse — the results were also heavily skewed along racial lines.


Machine learning specialists have observed that ML picks out little differences and emphasises them. That’s what this does – and magnifies the existing differences. So it amplifies existing discrimination.
unique link to this extract

The newest AI-enabled weapon: ‘deep-faking’ photos of the Earth • Nextgov

Patrick Tucker:


Worries about deep fakes—machine-manipulated videos of celebrities and world leaders purportedly saying or doing things that they really didn’t—are quaint compared to a new threat: doctored images of the Earth itself.

China is the acknowledged leader in using an emerging technique called generative adversarial networks to trick computers into seeing objects in landscapes or in satellite images that aren’t there, says Todd Myers, automation lead and Chief Information Officer in the Office of the Director of Technology at the National Geospatial-Intelligence Agency.

“The Chinese are well ahead of us. This is not classified info,” Myers said Thursday at the second annual Genius Machines summit, hosted by Defense One and Nextgov. “The Chinese have already designed; they’re already doing it right now, using GANs—which are generative adversarial networks—to manipulate scenes and pixels to create things for nefarious reasons.”

For example, Myers said, an adversary might fool your computer-assisted imagery analysts into reporting that a bridge crosses an important river at a given point.  

“So from a tactical perspective or mission planning, you train your forces to go a certain route, toward a bridge, but it’s not there. Then there’s a big surprise waiting for you,” he said.


The concern seems a little overblown, but you have to worry about malicious actors, especially with open source.
unique link to this extract

Social media bosses could be liable for harmful content, leaked UK plan reveals • The Guardian

Heather Stewart and Alex Hern:


Under plans expected to be published on Monday, the government will legislate for a new statutory duty of care, to be policed by an independent regulator and likely to be funded through a levy on media companies.

The regulator – likely initially to be Ofcom, but in the longer term a new body – will have the power to impose substantial fines against companies that breach their duty of care and to hold individual executives personally liable.

The debate has been sharpened in recent months by the case of the British teenager Molly Russell and issues raised by the Christchurch shootings. Molly’s parents said she killed herself partly because of self-harm images viewed on social media.

The scope of the recommendations is broad. As well as social media platforms such as Facebook and search engines such as Google they take in online messaging services and file hosting sites.

Other proposals in the online harm white paper include:
• Government powers to direct the regulator on specific issues such as terrorist activity or child sexual exploitation
• Annual “transparency reports” from social media companies, disclosing the prevalence of harmful content on their platforms and what they are doing to combat it
• Co-operation with police and other enforcement agencies on illegal harms, such as incitement of violence and the sale of illegal weapons.


unique link to this extract

Apple cuts price of HomePod worldwide, now $299 at the US Apple Store • 9to5Mac

Benjamin Mayo:


Apple has officially lowered the price of the HomePod worldwide. In the price has fallen from $349 to $299, breaking the $300 level. HomePod has seen promotional discounts at many third party retailers (eg: it’s $279 right now at Best Buy) over its lifespan, but now Apple has dropped the smart speaker’s standard list price.

This price cut represents a roughly 15% drop and seemingly applies to every region, not just the US. For example, the UK price has fallen from £319 to £279.

It is rare for Apple to reduce the price of any of its products mid-cycle, but not unprecedented. Apple dropped the price of the third-generation Apple TV from $99 to $69 in March 2015; the tvOS successor did not debut until six months later.


It’s not featured on the front of the site. The next thing it needs is a software update to play other services. What would be impressive – and sort of what you’d expect from a hardware company – would be if it got that before the annual iOS update in autumn.
unique link to this extract

CFIUS forces PatientsLikeMe into fire sale, booting Chinese investor • CNBC

Christina Farr and Ari Levy:


[US startup] PatientsLikeMe is being forced to find a buyer after the U.S. government has ordered its majority owner, a Chinese firm, to divest its stake.

PatientsLikeMe provides an online service that helps patients find people with similar health conditions. In 2017, the start-up raised $100m and sold a majority stake to Shenzhen-based iCarbonX, which was started by genomic scientist Jun Wang and is backed by Chinese giant Tencent.

That deal has recently drawn the attention of the Committee on Foreign Investment in the United States (CFIUS), which is aggressively cracking down on Chinese investments in American companies, particularly when national security and trade secrets are at risk.

CFIUS is now forcing a divestiture by iCarbonX, meaning PatientsLikeMe has to find a buyer, according to several people with knowledge of the matter. PatientsLikeMe started receiving notifications from CFIUS late last year, said the people, who asked not to be named because the details are confidential.

The move could have dire implications for the start-up community, as Chinese investors are scared away or forbidden from participating in deals that can help emerging businesses.


Also means CFIUS thinks that personal data is worth treating as a valuable national asset. That has big, big implications.
unique link to this extract

Prince Harry calls for ‘addictive’ Fortnite to be BANNED – ‘It shouldn’t be allowed’ •

Abbie Llewelyn:


The Duke of Sussex urged parents to protect their children from these “irresponsible” games during a visit to a YMCA in West London. He added that social media is “more addictive than drugs and alcohol”. Harry was speaking to mental health experts about the effects of social media and violent video games.

He said parents don’t know what to do about their children’s addiction to Fortnite.

The Duke said: “The game shouldn’t be allowed. Where is the benefit of having it in your household? It’s created to addict, an addiction to keep you in front of a computer for as long as possible. It’s so irresponsible.”

“Parents have got their hands up – they don’t know what to do about it. It’s like waiting for the damage to be done.”

Fortnite, which has millions of fans across the globe, is a game where players hunt for weapons to kill each other.


A brief note: the Daily Express, the newspaper which will surely run this story BIG in print, is notorious as a wildly right-wing paper catering to pensioners whose principal concern is whether it will be cold tonight, whether the value of their house is going to keep up, and the latest gossip about Princess Di, who died more than 20 years ago.

Anyhoooo, maybe Harry’s offspring will be explaining the attraction to him in 15 years or so. Or his father will explain that your young years are better spent playing strip billiards in Vegas. (Hard to argue with that, actually.)
unique link to this extract

‘Dozens’ of whistleblowers are secretly cooperating with House Democrats • The Atlantic

Russell Berman:


[Executive Oversight] Committee veterans told me, however, that the number of whistleblowers who’ve come forward since Trump became president is far higher than the number who cooperated with the panel during previous administrations. “The biggest difference wasn’t necessarily us switching to the majority; the biggest difference was Donald Trump being elected president,” said the Democratic aide, who spoke on the condition of anonymity because of the sensitivity of the committee’s investigative work. Democrats began hearing from whistle-blowers almost immediately after Trump was sworn in, the aide said, beginning with a report that then–National Security Adviser Michael Flynn had been exchanging text messages with his business partner during the inauguration.

Of the dozens of whistleblowers Democrats said they are working with, they have publicly confirmed that a handful work in the White House. All but Newbold, however, have come forward on the condition that they remain anonymous. Newbold spoke to the committee as part of its investigation of White House security clearances, and she’s not the only whistleblower involved in that matter, the panel confirmed in a memo describing her testimony. “Committee staff have spoken with other whistleblowers who corroborated Ms. Newbold’s account, but they were too afraid about the risk to their careers to come forward publicly,” the memo reads. The White House did not respond to a request for comment on this story.

Members from both parties interact privately with whistleblowers, but under a longstanding agreement within the committee, those who want to make on-the-record testimony must agree to be questioned by Democrats and Republicans alike.


The thing that puzzles me is that there are no sanctions when people are shown to have bent the law. The current outrage is that Jared Kushner was given a security clearance against the clear advice of the security clearance vetting agency. If you can’t properly sanction someone, all the “Oversight” and “Ethics Offices” are pointless. That, above all, is the lesson of Trump’s presidency: the checks and balances need to have some big sticks – such as jail time – they can wield.
unique link to this extract

Big Tech’s original sin • The New York Times

Charlie Warzel:


Facebook’s rapid rise to two billion-plus users, numerous privacy debacles and a steady stream of reported negative revelations suggest that, like its counterparts, the company’s quest for expansion trumped pressing concerns of privacy and transparency. A New York Times investigation last year reported that, “bent on growth,” Facebook executives “ignored warning signs” that Facebook could “disrupt elections, broadcast viral propaganda and inspire deadly campaigns of hate around the globe.”

Scale is also seductive at an engineering level, bottom line aside. Adding users and engagement, in one interpretation, might signal that you’re giving people what they want. In 2017, I asked a former senior Facebook employee if staff members had felt a sense of blame for Facebook’s inability to stop the spread of misinformation that plagued the platform during the 2016 election. Not exactly, the employee explained:

“They believe that to the extent that something flourishes or goes viral on Facebook — it’s not a reflection of the company’s role, but a reflection of what people want. And that deeply rational engineer’s view tends to absolve them of some of the responsibility, probably.”

We can see this sensibility today in the way the platforms tend to obfuscate and deflect responsibility. Just last week, a YouTube executive argued that its recommendation algorithms weren’t designed to nudge users toward more extreme videos. Similarly, Twitter has and will continue to argue it was not designed specifically to be disproportionately hostile to women and people of color. And Facebook will argue that it was certainly not designed to help foreign countries interfere in our elections.

But this defensive posture seems only concerned with intent. Even if we take the platforms at their word that they did not intend to profit from extremism or to become hubs for radicalization online, that doesn’t mean it doesn’t happen. Intent is far less important than the actual outcomes.


unique link to this extract

The whisper room: moderates on Twitter are losing their voice • Missouri University News Bureau


Michael Kearney, an assistant professor at the Missouri School of Journalism, found that while partisan users form highly partisan social networks on Twitter, moderate users – or those less politically engaged – continue to avoid politics, potentially creating an important void on social media.

“We are not necessarily getting farther and farther apart – it’s just the people in the middle are becoming more quiet and withdrawn,” Kearney said. “If you fail to consider all the people in the middle who do not care about politics as much, it seems like there is a more clear division when there is not, so social media might be artificially creating this sense that we are becoming more polarized.”

Kearney found that rather than increasing exposure to diverse viewpoints or sheltering users with self-reinforcing filter bubbles, social media simply amplifies and reflects the trends found in broader media environments. This was the first study of its kind to examine change in real-time behaviors of political polarization by looking at who Twitter users choose to follow during a general election.

Using software that he created himself, Kearney examined the user networks of 3,000 random followers of well-known partisan and entertainment-oriented accounts. Data was collected over six months leading up the 2016 general election, beginning shortly after Hillary Clinton and Donald Trump became the two major party nominees. Kearney found that as the election drew nearer, Democrats followed more Democrats, Republicans followed more Republicans, and moderates did not greatly expand who they followed on either political side.


Seriously, it’s taken him nearly three years to get this published? ArXiv is just over there, Prof Kearney, and it lets everyone take a gander at your possibly relevant work which would have been good to hear about a couple of years back. Not that it isn’t true now.
unique link to this extract

Errata, corrigenda and ai no corrida: yesterday’s calculation that today’s 40-year-olds would be tomorrow’s 65-year-olds in 15 years’ time attracted the notice of many people capable of maths. Yes, it should have said 25 years.

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 )

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.