Google’s “One True Answer” system is spouting racism and falsehoods; does Google want to fix it, and can it if so? Photo by kalleboo on Flickr.
You can now 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 12 links for you. None visited by the Russian ambassador. I’m @charlesarthur on Twitter. Observations and links welcome.
Should economists be more concerned about artificial intelligence? • Bank Underground
Mauricio Armellini and Tim Pike, writing on the Bank of England’s blog:
»
Hermann Hauser argues there were nine new General Purpose Technologies (GPTs) with mass applications in the first 19 centuries AD, including the printing press, the factory system, the steam engine, railways, the combustion engine and electricity. GPTs by definition disrupt existing business models and often result in mass job losses in the industries directly affected. For example, railways initiated the replacement of the horse and carriage, with resultant job losses for coachmen, stable lads, farriers and coach builders. Most of these GPTs took several decades to gain traction, partly because of the large amounts of investment required in plant, machinery and infrastructure. So there was sufficient time for the economy to adapt, thus avoiding periods of mass unemployment.
But the pace of technological progress sped up rapidly since the 19th century. Hermann identifies eight GPTs in the 20th century alone, including automobiles, aeroplanes, the computer, the internet, biotechnology and nanotechnology. Most recent innovations have been scalable much more quickly and cheaply. They have also been associated with the emergence of giant technology corporations — the combined market capitalisation of Apple, Google, Microsoft, Amazon and Facebook is currently about $2½ trillion. The faster these new waves of technology arise and the cheaper they are to implement, the quicker they are deployed, the broader their diffusion, the faster and deeper the rate of job loss and the less time the economy has to adapt by creating jobs in sectors not disrupted by GPTs.
And some technologies are evolving at lightning speed, such as the ongoing exponential increase in computing power. Computers have evolved in the past 40 years or so from initially being merely calculators to having applications that include smartphones and, in conjunction with the internet and big data, driverless cars, robots and the “Internet of Things”.
Looking to the future, how might these new GPTs affect the economy? The retail and distribution sector currently has over five million jobs. In the not too distant future, most consumer goods could be ordered online and delivered by either autonomous vehicles or drones. The warehouses in which the goods are stored could be almost entirely automated. Bricks and mortar stores might largely disappear.
«
TL:DR: robots and automation could overwhelm what we think of as work a lot more quickly than we expect.
link to this extract
How YouTube is changing our viewing habits • NPR
Zeynep Tufekci, talking on NPR:
»
A platform like YouTube has algorithms designed to recommend to you things that it thinks will be more engaging. And what I’ve found is that whatever I watched, it would push a more hardcore version of whatever it was I was watching across the political spectrum. So something that I found really striking was that if I watched Donald Trump rallies, I would get recommended white supremacist conspiracy theories. And you have examples from radicalized, you know, extremists when, some of their interviews, they talk about going down the rabbit hole of YouTube.
Scott Simon: I remember somebody once sent me a video purporting to show that man never landed on the moon. And there – for a couple of days thereafter, I had eight or 10 similar videos, each of them more convincing than the other, showing me why man never landed on the moon. Now, I don’t mind crawling out on a limb and saying, that is false, man landed on the moon. But at the same time, it makes you wonder, you know, people, who don’t consider themselves dumb will watch some of those videos and say, see, this proves it. I mean, that was the case of the person who sent it to me. Does YouTube or any other platform have some kind of responsibility not to put misinformation on their site?
TUFEKCI: So there’s two things going on. On the one hand, we do have freedom of speech. So if somebody wants to claim we never landed on the moon, I can see that as freedom of speech. But there’s no freedom to necessarily be recommended by YouTube, right? So what I see happening and what I see as troublesome is that if you watch something like that, YouTube could recommend to you something debunking falsehoods saying hey, check this out, right? Or…
SIMON: I mean, after seeing that video, I got a lot of stuff saying 9/11 never happened.
«
This is the other side of the Google result problem: YouTube’s “relevance” algorithm pushes people off towards mad stuff because it’s “more engaging” – that is, people (without critical faculties) watched it for longer than the stuff people know about, which is that 9/11 was perpetrated by Al Qa’ida and people have walked on the moon.
link to this extract
“And here’s what happens if you ask Google Home “is Obama planning a coup?” • Twitter
Rory Cellan-Jones is the BBC’s longstanding technology correspondent; he made a short video:
»
And here’s what happens if you ask Google Home “is Obama planning a coup?”
https://twitter.com/i/videos/tweet/838445095410106368?embed_source=clientlib&player_id=0&rpc_init=1
«
(I hope the video plays.)
There’s also this article at The Outline by Adrianne Jeffries:
»
Many of Google’s direct answers are correct. Ask Google if vaccines cause autism, and it will tell you they do not. Ask it if jet fuel melts steel beams, and it will pull an answer from a Popular Mechanics article debunking the famous 9/11 conspiracy theory. But it’s easy to find examples of Google grabbing quick answers from shady places.
Would you trust an answer pulled from the anti-vaccine alternative health content farm Mercola.com?
How about an answer from the Facebook page of a white nationalist group in Australia?
Would you trust a system that falls for a Monty Python joke?
What about a system that thinks Barack Obama is the current king of America — first because of an answer sourced from a Breitbart article, and now because of an article criticizing that answer?
«
It’s clear that Google’s “One True Answer” responses are woefully wrong in many important ways (The Outline offers another relating to whether any US Presidents have been members of the Ku Klux Klan). It really is time to kill it: it’s not working (because like YouTube’s recommendation algorithm, it relies on what people click, not what’s true), and it’s not fixable without a colossal human effort that Google won’t be willing to make.
link to this extract
Five AI startup predictions for 2017 • Bradford Cross
Cross is a founding partner at a venture capital company that has invested in machine learning startups. His prediction list is short:
»
1. Bots go bust
2. Deep learning goes commodity
3. AI is cleantech 2.0 for VCs [and hence ripe for a bust from disappointed overinvestment; it isn’t a separate feature but part of a stack]
4. MLaaS [machine learning as a service] dies a second death
5. Full stack vertical AI startups actually work
«
He goes on to explain them in more detail.
link to this extract
As Messenger’s bots lose steam, Facebook pushes menus over chat • TechCrunch
Sarah Perez:
»
Instead of forcing users to talk with a bot, developers can choose to create a persistent menu that allows for multiple, nested items as a better way of displaying all the bots’ capabilities in a simple interface.
The new persistent menus are limited to three items at the top-level, and its sub-menus are now limited to five. Before, if users wanted to engage with a menu like this, they often had to engage in conversations with the bot to discover the various sections and items.
Now, Facebook suggests to developers that they “consider stripping such exchanges down and cutting to the chase by putting the most important features in your menu.”
For example, a retailer’s bot might offer menu options that let you “go shopping,” “ask questions,” or “send messages.” If you clicked into the shopping section, the menu could update with a list of items to drill down into, like tops, bottoms, shoes and accessories.
«
This feels like it’s heading towards something that isn’t a bot.
link to this extract
Hidden backdoor found in Chinese-made equipment. Nothing new! Move along! • Bleeping Computer
Catalin Cimpanu:
»
DblTek stands for DBL Technology, a Hong Kong-based company that manufactures IP phones, SIM servers, various types of VoIP equipment and cross-network gateways. According to a report from cyber-security firm Trustware, GoIP GSM gateways allows hidden remote Telnet access via an account named “dbladm” that provides root-level access to the device.
Unlike the default “ctlcmd” and “limitsh” Telnet accounts, the “dbladm” account is not included in the product’s documentation.
While the first two use user-set passwords, the backdoor account uses a challenge-response authentication scheme. This scheme presents users with a string, on which they can perform various operations and deduce the password.
Backdoor password can be easily computed
Trustwave researchers said this scheme is very easy to reverse engineer. An attacker can create automated scripts that read the challenge, compute the response, and authenticate on the device.
Once they log in, because users have root privileges, they can take full control of the device, listen to ongoing traffic, or use the equipment for other actions, such as DDoS attacks or for relaying malicious traffic.
Researchers say they tested GoIP 8-port GSM gateways, but they suspect that GoIP 1, 4, 16 and 32-port devices are affected as well since they use the same login binary in their firmware images.
«
Also linked in the story: 2012 report from a former Pentagon analyst saying China had backdoors in the equipment of 80% of the world’s telecoms.
link to this extract
An Android review for iOS users: conclusion (day 5) • BirchTree
Matt Birchler tried out Android, rather than iOS, for a month, and summed up his views of the differences over five days, culminating in this:
»
I think at the crux of my position can be best summed up by how fans of each platform talk about “power” features. If you asked me to give examples of the “power” of iOS, I would bring up:
• Extensive library of app extensions that let you share data nicely between apps.
• iMovie is a full consumer-grade video editing app.
• Ferrite is a shockingly powerful audio editor.
• Apps like Workflow and Launch Center Pro enable automations unlike anything we’ve seen on any computing platform before, and they make that power accessible to everyone.
• A rich third party ecosystem of apps built on powerful APIs are enabling people to slowly ditch their PCs for iPhones and iPads full time.
And when you ask me about the “power” of Android, this is what comes to mind:
• Ability to side-load apps not available on the Play Store.
• Custom launchers let you have a custom home screen
• Tasker allows me to make my phone do things based on the time of day, location, or other trigger.
• Ability to change default apps.
• Access to the file system.
• Ability to flash custom ROMs onto your device.
The notable difference in my two lists is that the iOS advantages have to do with you actually getting things done with your mobile device, while Android’s list is more about customizing the look of your device, as well as bringing over some more traditional PC features (file system and non-store software).
«
All four other posts are linked at the top of this piece, though the one on notifications differences is probably the most finely balanced in showing up the contrasts where both have strengths.
link to this extract
The night zombie smartphones took down 911 • WSJ
Ryan Knutson:
»
On a Tuesday night last October in Olympia, Wash., 911 operator Jennifer Rodgers stared at the list of incoming calls on her screen.
Normally, one or two calls at a time would trickle in at this hour. At 9:28 p.m., they began stacking up by the dozens like lines on an Excel spreadsheet.
An alarm alerting operators to unanswered 911 calls filled the room. It almost never sounds more than once. Tonight, it was going off constantly.
Ms. Rodgers had no idea what was happening. People in Olympia, a city of about 50,000 an hour’s drive south of Seattle, and the surrounding county were dialing 911 and hanging up before their calls were answered. Then they were dialing 911 again.
After about 15 minutes, a girl stayed on the phone long enough for Ms. Rodgers, a 911 operator for 15 years, to say through her headset: “Don’t hang up! Don’t hang up!”
“We didn’t mean to call 911!” the operator recalls the girl saying. “I’m not touching the phone! I’m not doing anything! I don’t know how to make it stop!”
For at least 12 hours on Oct. 25 and Oct. 26, 911 centers in at least a dozen U.S. states from California to Texas to Florida were overwhelmed by what investigators now believe was the largest-ever cyberattack on the country’s emergency-response system.
«
It transpires this attack works only against iPhones: it’s a shortcut leading to a link that is a “call” link, but rather than giving you a “cancel” option (as you get for a normal text string interpreted as a phone number) the iPhone assumes you must want to call the emergency number without delay.
Don’t try to copy it; the perpetrator may be facing a jail term. Meticulously reported.
link to this extract
How to keep messages secure • Teen Vogue
Nicole Kobie:
»
Heading to a protest, organizing with activists, or suddenly concerned about the politics of your parents? Don’t use SMS or Snapchat to chat about it – you need something safer.
To help you pick the right messaging app, Teen Vogue talked to a trio of security experts: Zeynep Tufekci, an associate sociology professor at the University of North Carolina, and the author of a book about networked protest; Alec Muffett, a software engineer who previously worked on security at Facebook; and Moxie Marlinspike, the security researcher who founded Open Whisper Systems, which developed the encryption used by WhatsApp and other messaging tools.
To secure your messaging, they advise three steps. First, update your apps and Android or iOS to the latest version. Second, set a long PIN of at least eight characters to unlock your handset. And third, avoid SMS for texting, instead using a secure messaging app – whether it’s Facebook Messenger, WhatsApp, or a stronger tool such as Signal.
«
A sign of the times, surely, that it’s a feature in Teen Vogue which clearly explains why and how to use secure messaging. (The article is by a freelance rather than a staff writer; next – no joke! – she’s going to be writing about antivirus.)
link to this extract
How Uber deceives the authorities worldwide • The New York Times
Mike Isaac on Uber’s “Greyball” system:
»
When Uber moved into a new city, it appointed a general manager to lead the charge. This person, using various technologies and techniques, would try to spot enforcement officers [who might try to block UberX drivers, on the basis they were essentially unregulated cab drivers].
One technique involved drawing a digital perimeter, or “geofence,” around the government offices on a digital map of a city that Uber was monitoring. The company watched which people were frequently opening and closing the app — a process known internally as eyeballing — near such locations as evidence that the users might be associated with city agencies.
Other techniques included looking at a user’s credit card information and determining whether the card was tied directly to an institution like a police credit union.
Enforcement officials involved in large-scale sting operations meant to catch Uber drivers would sometimes buy dozens of cellphones to create different accounts. To circumvent that tactic, Uber employees would go to local electronics stores to look up device numbers of the cheapest mobile phones for sale, which were often the ones bought by city officials working with budgets that were not large.
In all, there were at least a dozen or so signifiers in the VTOS program that Uber employees could use to assess whether users were regular new riders or probably city officials.
If such clues did not confirm a user’s identity, Uber employees would search social media profiles and other information available online. If users were identified as being linked to law enforcement, Uber Greyballed them by tagging them with a small piece of code that read “Greyball” followed by a string of numbers.
When someone tagged this way called a car, Uber could scramble a set of ghost cars in a fake version of the app for that person to see, or show that no cars were available. Occasionally, if a driver accidentally picked up someone tagged as an officer, Uber called the driver with instructions to end the ride.
«
Isaac, it should be noted, has been doing especially amazing work in the past few months.
link to this extract
#UberLove • LinkedIn
Kelly Snodgrass worked at Uber for two and a half years, though she now works at Snap:
»
I found myself advocating for my male counterparts as objectively, the contributions put forth by the female individuals were not as valuable as the contributions of their male counterparts… but this wasn’t because they were female. The interesting context behind this is that as you may have heard, Uber is a company where the best ideas win…where you have to both come up with the best idea, AND execute on that idea in the best way. It’s a damn hard place to be successful! But the key here is that gender does not play a role, rather talent does. Uber tries really hard to fairly reward individuals…regardless of gender, and I was lucky enough to experience this first hand. I am a woman and had a great experience and wild success at Uber.
«
The best ideas win, eh? By definition, that means that identifying officials who might sic the regulations on Uber was a “best idea”, which tells you a great deal about Uber’s view of the outside world.
link to this extract
Uber’s VP of product and growth Ed Baker has resigned • Recode
Kara Swisher and Johana Bhuiyan:
»
Ed Baker, Uber’s VP of product and growth, has informed his team that he’s leaving the car-hailing company after more than three years.
In an email, the former Facebook exec wrote: “I have always wanted to apply my experience in technology and growth to the public sector. And now seems like the right moment to get involved.”
Before Uber — which he joined in 2013 along with former CFO Brent Callinicos and Uber SVP Emil Michael — Baker headed international growth at the social media giant for two years after the company acquired his dating app called Friend.ly in 2011.
With Baker’s leaving, marketplace head Daniel Graf will take over as acting head of product and growth. In addition, Uber has hired well-regarded Facebook product exec Peter Deng as head of its rider product. Also key in Graf’s organization is Aaron Schildkrout, who will be head of driver products.
But, because it is Uber, the Baker departure is complex: His resignation also comes at a time when Uber employees have complained about questionable behavior on his part.
«
link to this extract
Errata, corrigenda and ai no corrida: none notified
Filed under: links
