Start Up: AI’s indifferent control, inside the Indian scam call centres, Google breaks own Captcha, and more

Is your electricity meter telling the truth about your power use? Photo by benswing 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 9 links for you. Loooovely.. I’m @charlesarthur on Twitter. Observations and links welcome.

Why nothing works anymore • The Atlantic

Ian Bogost, in a tour de force:


No matter its ostensible function, precarious technology separates human actors from the accomplishment of their actions. They acclimate people to the idea that devices are not really there for them, but as means to accomplish those devices own, secret goals.

This truth has been obvious for some time. Facebook and Google, so the saying goes, make their users into their products—the real customer is the advertiser or data speculator preying on the information generated by the companies’ free services. But things are bound to get even weirder than that. When automobiles drive themselves, for example, their human passengers will not become masters of a new form of urban freedom, but rather a fuel to drive the expansion of connected cities, in order to spread further the gospel of computerized automation. If artificial intelligence ends up running the news, it will not do so in order to improve citizen’s access to information necessary to make choices in a democracy, but to further cement the supremacy of machine automation over human editorial in establishing what is relevant.

There is a dream of computer technology’s end, in which machines become powerful enough that human consciousness can be uploaded into them, facilitating immortality. And there is a corresponding nightmare in which the evil robot of a forthcoming, computerized mesh overpowers and destroys human civilization. But there is also a weirder, more ordinary, and more likely future—and it is the one most similar to the present.


The coda is remarkable, though you should take in the whole article from the start. In effect: “First we shape our tools, then our tools shape us, then our tools find more interesting things to do.”
link to this extract

Researcher breaks reCAPTCHA using Google’s speech recognition API • Bleeping Computer

Catalin Cimpanu:


A researcher has discovered what he calls a “logic vulnerability” that allowed him to create a Python script that is fully capable of bypassing Google’s reCAPTCHA fields using another Google service, the Speech Recognition API.

The researcher, who goes online only by the name of East-EE, released proof-of-concept code on GitHub.

East-EE has named this attack ReBreakCaptcha, and he says he discovered this vulnerability in 2016. Today, when he went public with his research, he said the vulnerability was still unpatched.

The researcher was not clear if he reported the bug to Google. Bleeping Computer has reached out to the researcher to inquire if Google was, at least, aware of the issue.

The proof-of-concept code the researcher released allows attackers to automate the process of bypassing reCAPTCHA fields, currently used on millions of sites to keep out spam bots.


Oops. But logical. Only works against the latest version of reCAPTCHA. But even so.
link to this extract

Inside the TalkTalk ‘Indian scam call centre’ • BBC News

Geoff White:


TalkTalk customers are being targeted by an industrial-scale fraud network in India, according to whistleblowers who say they were among hundreds of staff hired to scam customers of the British telecoms giant.

The scale of the criminal operation has been detailed by the three sources, who say they were employed by two front-companies set up by a gang of professional fraudsters.

The sources describe working in “call centres” in two Indian cities.

They say as many as 60 “employees” work in shifts in each office, phoning TalkTalk customers and duping them into giving access to their bank accounts.

The whistleblowers say they were given a script in which they were told to claim they were calling from TalkTalk. They say they then convinced victims to install a computer virus.


Everything’s exactly as I discovered in researching this stuff in 2012 – though White also has screenshots of the (perhaps intentionally incoherent) scripts that the scammers are given.
link to this extract

Snap tumbles for first time since IPO after analysts say sell • Bloomberg

Shelly Hagan:


After a euphoric public market debut, Snap Inc. shares dropped for the first time in three days after analysts began weighing in with their thoughts on the company’s true valuation.

The parent company of disappearing-photo app maker Snapchat priced shares in its initial public offering last Wednesday and they surged 44% on the first day of trading. On Friday the stock climbed a further 11%. By Monday, five of the seven analysts who cover the company had a sell rating on it while two said hold.

No analyst recommends buying the stock, according to data compiled by Bloomberg. Not all analysts are able to give their opinion on the stock yet, since those who work at banks involved in the IPO are prevented from doing so for a while.

Snap fell as much as 9%, to $24.61 and was trading at $25.41 at 11:47 a.m. in New York. That values the company at about $29bn.

“Academic literature suggests that the sexier and more glamorous a company’s IPO, the more likely it is to be overpriced at its IPO date and to suffer meaningful downwards earnings and valuation revisions in the first eight quarters after it goes public,” wrote Laura Martin, an analyst at Needham & Co., in a note to investors. She said Snap’s value is more like $19 to $23 a share.


There’s no sensible way to value Snap. And no point in owning its shares beyond speculation on a rise or fall.
link to this extract

Wearables as a platform, new 2017 and 2018 Apple Watch sales estimates, deep dive into 4q16 wearables market • Above Avalon members

From Neil Cybart’s paid-for daily news analysis of things Apple-y and related, in which he goes to town on IDC’s estimates for the wearables market (excluding AirPods) in 4Q 2016:


I have significant issues with IDC’s report and commentary. 

For example: 
• Who are these low-cost wearables competitors in the U.S. eating Fitbit’s market share in 2H16? IDC doesn’t name them. In reality, Fitbit’s troubles are increasingly found with consumers embracing higher-priced wearables containing additional utility. This is why Fitbit is running upmarket as fast as they can.

• Xiaomi is using a low-cost wearables strategy? The company is selling $15 plastic step and sleep trackers. This is like saying a phone company selling a $20 pay as you go phone with no apps is using a low-cost smartphone strategy. Xiaomi should not be included in the same discussion as Apple Watch or Fitbit. 

• While Apple Watch sales hit a record during 4Q16, unit sales were up 20% year-over-year. Calling this a “magnificent success” seems a bit hyperbolic, as if the Watch was a complete flop in 4Q15. 

• We discussed Garmin’s 4Q16 results a few weeks ago. I don’t know how IDC reached their estimate of Garmin selling 2.1M wearables at an ASP of $258 during 4Q16. Even if we assumed every dollar found in Garmin’s Fitness and Outdoors segments was related to wearables, which wasn’t the case, Garmin would have sold at most 1.4M to 1.5M wearables. In reality, Garmin likely sold less than 1M wearables. In addition, IDC says Garmin customers moved to higher-end devices that are able to do more than fitness tracking – this is the exact opposite of IDC’s main thesis for the wearables market. 

• IDC positions cellular smartwatches as a key to smartwatch sales success. Yet Samsung is the only company shipping cellular smartwatches at volume and they aren’t selling well compared to Apple Watch. 


In particular, IDC only quoting Xiaomi’s unit sales figures, and not revenue, seems unhelpful for understanding what’s going on. (And this is why you should subscribe to the Above Avalon newsletter, to get information like this.)
link to this extract

Electronic energy meters’ false readings almost six times higher than actual energy consumption • University of Twente


For quite some time now, rumours have been rife about electronic energy meters that give excessively high readings in practice. This prompted Prof. Leferink to investigate electronic meters, to see whether they can indeed give false readings. Together with co-workers Cees Keyer and Anton Melentjev from AUAS, he tested nine different electronic meters in this study. The meters in question were manufactured between 2004 and 2014. The meters were connected, via an electric switchboard, to a range of power-consuming appliances, such as energy saving light bulbs, heaters, LED bulbs and dimmers. The researchers then compared the actual consumption of the system with the electronic energy meter’s readings.  

In the experiments (which were entirely reproducible), five of the nine meters gave readings that were much higher than the actual amount of power consumed. Indeed, in some setups, these were up to 582% higher. Conversely, two of the meters gave readings that were 30% lower than the actual amount of power consumed.

The greatest inaccuracies were seen when dimmers combined with energy saving light bulbs and LED bulbs were connected to the system. According to Mr Keyer (lecturer Electrical Engineering at the AUAS and PhD student at the UT)  “OK, these were laboratory tests, but we deliberately avoided using exceptional conditions. For example, a dimmer and 50 bulbs, while an average household has 47 bulbs.” 

The inaccurate readings are attributed to the energy meter’s design, together with the increasing use of modern (often energy-efficient) switching devices. Here, the electricity being consumed no longer has a perfect waveform, instead it acquires an erratic pattern. The designers of modern energy meters have not made sufficient allowance for switching devices of this kind.

When they dismantled the energy meters tested, the researchers found that the ones associated with excessively high readings contained a ‘Rogowski Coil’ while those associated with excessively low readings contained a ‘Hall Sensor’. Frank Leferink (Professor of Electromagnetic Compatibility at the UT) points out that “The energy meters we tested meet all the legal requirements and are certified. These requirements, however, have not made sufficient allowance for modern switching devices”.


I think 582% is actually nearly seven times higher. (100% higher = 2x, 200% = 3x, and so on.) Moral: use percentages over 100 with great care, and never use those over 199; you’ll confuse yourself and everyone else.

Meanwhile, the findings are a concern. LEDs typically turn on and off incredibly fast. Standard meters maybe can’t deal with it.
link to this extract

BMW says ‘nein’ to Android Auto • TechCrunch

Frederic Lardinois:


I sat down with Dieter May, BMW’s senior vice president of Digital Services and Business Models (in an i3, of course).

During our conversation, we touched upon quite a few topics, ranging from self-driving cars, to the future of car ownership and the new business models that in-car technology enables. “We offer [Apple’s] CarPlay as an option but not Android Auto,” he said. “We believe the changes that are coming to the inside of the car and the user experience — like self-driving cars — you have to control the customer interface. That’s part of the brand experience and for that, I don’t want to have an Android screen and I especially want to be able to deeply integrate these systems.”

He expects that the car of the future (especially when we’re talking about autonomous cars) will offer far more personalization options, which in turn will enable new business models, too.

“If you have six screens in the car, you also get gesture control, voice control with a personal assistant, etc.,” May said. “You need to have control over that user experience — maybe you can get away with it if you’re a ‘mass producer,’ but not in the premium segment.”


Writing note: I don’t need to know that Lardinois sat down with May. It might be nice for Lardinois to say so, but the rest of us really don’t care. (“Told me” will do fine.) And I’d hope that he’d “touch upon quite a few topics” as a matter of course. It’s like a boring chef explaining how they cooked chips. Too much web “news” writing, never having had to cope with print’s strict wordcount tyranny, is flabby, slow and self-regarding.
link to this extract

Your understanding of the size of countries and continents is completely wrong • Relatively Interesting


The Mercator projection is a cylindrical map projection of a sphere to a two dimensional surface created by the Flemish geographer and cartographer Gerardus Mercator in 1569. It became the standard map projection for nautical purposes, and although the linear scale is equal in all directions around any point, the Mercator projection distorts the size of objects as the latitude increases from the Equator to the poles, where the scale becomes infinite.

As a result of these distortions…

• Greenland appears larger than Africa, but in reality Africa’s area is 14 times greater and Greenland’s is comparable to Algeria’s alone.
• Africa also appears to be roughly the same size as Europe, when in reality Africa is nearly 3 times larger.
• Alaska takes as much area on the map as Brazil, when Brazil’s area is nearly 5 times that of Alaska.
• Finland appears with a greater north-south extent than India, although India’s is greater.
• Antarctica appears as the biggest continent (and would be infinitely large on a complete map), although it is actually the fifth in area.


You already knew that Mercator was a convenient lie, but it’s nice to be reminded how much of a lie. (When it’s shown to children for the first time, is that fake news?) There’s a clever infographic accompanying the article, with questions and answers to: which is bigger, the US (inc. Alaska) or Russia? Is Colombia smaller or bigger than the UK? Is Tanzania the same size as Germany, smaller, or bigger?
link to this extract

Google’s “One True Answer” problem: when featured snippets go bad • Search Engine Land

Danny Sullivan says that he was happy when Google Home answered his question about whether guinea pigs can eat grapes:


I remember distinctly when this question first came to my mind. I had my refrigerator open. My guinea pig, hearing me in the kitchen, started squeaking for a treat. I saw the grapes in the fridge and wondering if he could eat them. Normally, that would mean shutting the fridge and finding my phone or computer to type a query. But I called out this question to the Google Home in my kitchen and got an immediate answer.

That is an incredible competitive advantage that Google has over Amazon, as well as Apple and Microsoft, when it comes to providing answers. The others are far more tightly curtailed in providing direct answers from databases and vetted resources. That makes them less prone to problematic results but also less helpful for a wide range of queries that people have.

Turning off featured snippets means Google will lose its competitive advantage with Google Home, as well as with spoken queries to smartphones. That’s why I think it’s unlikely this will happen. Google will likely tolerate the occasional bad attention for its problematic One True Answers for what it considers the greater good to its users and its competitive standing in keeping them.

Is there a way for Google to keep the good that featured snippets provide without causing problematic results? Not perfectly. Google processes over 5 billion queries per day, and even if featured snippets appear in only 15% of those at the moment (according to the Moz SERP features tracker), that’s nearly a billion One True Answers per day. Humans can’t vet all those.

But Google could consider not showing featured snippets in its web search results, when queries are typed. There’s no particular need for it to elevate one answer over the others in this way. By losing this display, it might force users to better use their own critical thinking skills in reviewing 10 possible answers that they are provided.

For spoken queries, having a One True Answer repeated — when it’s correct — is undoubtedly helpful. To better improve there, Google might revisit the sites it allows to appear as resources. This could include vetting them, as it does with Google News. Or, it could make use of some algorithm system to determine if a site is deemed to have enough authority to be featured.


I think “incredible competitive advantage” is wildly overstating things. Google Home and Amazon Echo are at the Californian Early Adopter stage. Giving the wrong (wildly, racistly, Nazist-y) answer is a competitive disadvantage, unless you’re trying to tap into the Californian Nazi Early Adopter market, which really *is* small.

As for “enough authority to be featured”, isn’t that meant to have already happened before it appears in the search results? Google “tolerating” things might seem like a good plan, until the bad publicity buries it. And it is really not going to be popular with any but the Californian Nazi Early Adopter market at this rate.

The mission statement of “organising the world’s information and making it accessible” might not include “showing you what’s true”. But people assume that. If they think Google isn’t doing that, bad things could follow for Google.
link to this extract

Errata, corrigenda and ai no corrida: none notified

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.