New web security system tests computers' emotions

Sorting the men from the replicants.

A new Captcha system seeks to separate humans from computers by testing empathy – and spreading awareness of human rights human rights abuses at the same time.

A Captcha – which stands for Completely Automated Public Turing test to tell Computers and Humans Apart – is the test used when logging into many sites to distinguish between real people and malicious programs, which may attempt to log into many thousands of accounts at the same time. You've all used one – signing up for a New Statesman commenting account, if nowhere else – and they are ripe for being put to good use.

ReCAPTCHA was the first socially-beneficial captcha, and still the most popular. It uses the combined might of all the human brain power wasted on Captchas to transcribe scanned books:

reCAPTCHA improves the process of digitizing books by sending words that cannot be read by computers to the Web in the form of CAPTCHAs for humans to decipher. More specifically, each word that cannot be read correctly by OCR is placed on an image and used as a CAPTCHA. This is possible because most OCR programs alert you when a word cannot be read correctly.

Since it took off, ReCAPTCHA has been used on innumerable sites, and is now displayed over 100 million times a day. But that success comes at a price. Now that the low hanging fruit has been plucked, fewer and fewer easily-transcribable words remain in its corpus, meaning that the system regularly throws up completely unintelligible words, words in other scripts, or things which just aren't language at all.

The civil rights captcha wants to be the replacement. Rather than using the captcha to perform useful work, like reCAPTCHA, it uses it to raise awareness about important issues:

Instead of visually decoding an image of distorted letters, the user has to take a stand regarding facts about human rights. Depending on whether the described situation is positively or negatively charged, the CAPTHA generates three random words from a database. These words describe positive and negative emotions. The user selects the word that best matches how they feel about the situation, and writes the word in the CAPTCHA. Only one answer is correct, the answer showing compassion and empathy.

As well as being important socially – example questions include "The parliament in St. Petersburg recently passed a law that forbids "homosexual propaganda". How does that make you feel?" – the Civil Rights Captcha is stronger against attack as well. It includes the same visual element as a reCAPTCHA, requiring potential attackers to decipher obfuscated words, but also requires any automated attack to parse a complex question, pick the right emotion, and only then work out which of the proffered words match that emotion.

The whole thing is rather reminiscent of Blade Runner:

We'll catch those pesky replicants yet.

Rutger Hauer, in the film Blade Runner.

Alex Hern is a technology reporter for the Guardian. He was formerly staff writer at the New Statesman. You should follow Alex on Twitter.

Getty
Show Hide image

A quote-by-quote analysis of how little Jeremy Hunt understands technology

Can social media giants really implement the health secretary’s sexting suggestions? 

In today’s “Did we do something wrong? No, it was social media” news, Health Secretary Jeremy Hunt has argued that technology companies need to do more to prevent sexting and cyber-bullying.

Hunt, whose job it is to help reduce the teenage suicide rate, argued that the onus for reducing the teenage suicide rate should fall on social media companies such as Facebook and Twitter.

Giving evidence to the Commons Health Committee on suicide prevention, Hunt said: “I think social media companies need to step up to the plate and show us how they can be the solution to the issue of mental ill health amongst teenagers, and not the cause of the problem.”

Pause for screaming and/or tearing out of hair.

Don’t worry though; Hunt wasn’t simply trying to pass the buck, despite the committee suggesting he direct more resources to suicide prevention, as he offered extremely well-thought out technological solutions that are in no way inferior to providing better sex education for children. Here’s a quote-by-quote analysis of just how technologically savvy Hunt is.

***

“I just ask myself the simple question as to why it is that you can’t prevent the texting of sexually explicit images by people under the age of 18…”

Here’s Hunt asking himself a question that he should be asking the actual experts, which is in no way a waste of anybody’s time at all.

“… If that’s a lock that parents choose to put on a mobile phone contract…”

A lock! But of course. But what should we lock, Jeremy? Should teenager’s phones come with a ban on all social media apps, and for good measure, a block on the use of the camera app itself? It’s hard to see how this would lead to the use of dubious applications that have significantly less security than giants such as Facebook and Snapchat. Well done.

“Because there is technology that can identify sexually explicit pictures and prevent it being transmitted.”

Erm, is there? Image recognition technology does exist, but it’s incredibly complex and expensive, and companies often rely on other information (such as URLs, tags, and hashes) to filter out and identify explicit images. In addition, social media sites like Facebook rely on their users to click the button that identifies an image as an abuse of their guidelines, and then have a human team that look through reported images. The technology is simply unable to identify individual and unique images that teenagers take of their own bodies, and the idea of a human team tackling the job is preposterous. 

But suppose the technology did exist that could flawlessly scan a picture for fleshy bits and bobs? As a tool to prevent sexting, this still is extremely flawed. What if two teens were trying to message one another Titian’s Venus for art or history class? In September, Facebook itself was forced to U-turn after removing the historical “napalm girl” photo from the site.

As for the second part of Jezza’s suggestion, if you can’t identify it, you can’t block it. Facebook Messenger already blocks you from sending pornographic links, but this again relies on analysis of the URLs rather than the content within them. Other messaging services, such as Whatsapp, offer end-to-end encryption (EE2E), meaning – most likely to Hunt’s chagrin – the messages sent on them are not stored nor easily accessed by the government.

“I ask myself why we can’t identify cyberbullying when it happens on social media platforms by word pattern recognition, and then prevent it happening.”

Jeremy, Jeremy, Jeremy, Jeremy, can’t you spot your problem yet? You’ve got to stop asking yourself!

There is simply no algorithm yet intelligent enough to identify bullying language. Why? Because we call our best mate “dickhead” and our worst enemy “pal”. Human language and meaning is infinitely complex, and scanning for certain words would almost definitely lead to false positives. As Labour MP Thangam Debbonaire famously learned this year, even humans can’t always identify whether language is offensive, so what chance does an algorithm stand?

(Side note: It is also amusing to imagine that Hunt could even begin to keep up with teenage slang in this scenario.)

Many also argue that because social media sites can remove copyrighted files efficiently, they should get better at removing abusive language. This is a flawed argument because it is easy to search for a specific file (copyright holders will often send social media giants hashed files which they can then search for on their databases) whereas (for the reasons outlined above) it is exceptionally difficult for algorithms to accurately identify the true meaning of language.

“I think there are a lot of things where social media companies could put options in their software that could reduce the risks associated with social media, and I do think that is something which they should actively pursue in a way that hasn’t happened to date.”

Leaving aside the fact that social media companies constantly come up with solutions for these problems, Hunt has left us with the burning question of whether any of this is even desirable at all.

Why should he prevent under-18s from sexting when the age of consent in the UK is 16? Where has this sudden moral panic about pornography come from? Are the government laying the ground for mass censorship? If two consenting teenagers want to send each other these aubergine emoji a couple of times a week, why should we stop them? Is it not up to parents, rather than the government, to survey and supervise their children’s online activities? Would education, with all of this in mind, not be the better option? Won't somebody please think of the children? 

“There is a lot of evidence that the technology industry, if they put their mind to it, can do really smart things.

Alas, if only we could say the same for you Mr Hunt.

Amelia Tait is a technology and digital culture writer at the New Statesman.