Sponsor content — CAPTCHAs: The internet’s distorted text avengers

This email was produced by Quartz Creative on behalf of Accenture, and not by the Quartz editorial staff. When cult streetwear brand [Supreme]( released $30 logo-stamped bricks in 2016—yes, [regular red bricks]( sold out almost immediately and went on to fetch $1,000 on eBay. While ludicrous to many, this sort of collecting is unremarkable for Supreme’s youthful obsessives, who’ll line up for blocks beside Supreme’s shops and overrun its web store with every “inventory drop.” The brand deliberately releases products in limited quantities, guaranteeing sellouts, and, in turn, seeding a [resale market]( where markups are magnitudes higher (you can buy plain T-shirts for $650). Consequently, Supreme’s online store is among the internet’s [most bot-abused sites](. There are hundreds of “shoebots” (with [dedicated]( [subreddits]( that buy out Supreme inventory at lightning speeds, dismaying real fans and garnering big bucks for botmakers. That was until March 2017 when Supreme’s site added a CAPTCHA. For years, bots had been bypassing item pages and checking out with impunity. But with CAPTCHA, bots must now convincingly pass themselves off as human. That’s hasn’t stopped scammers from building CAPTCHA-solving bots that can still siphon Supreme products, but it’s slowed them down. And as bots get better, so do CAPTCHAs. In 2017, the cost of [cybersecurity breaches was up 27%]( year over year, setting organizations back $11.7 million on average. Furthermore, [29% of business and IT executives]( report that their firm has been a target of adversarial bots at least once, and 77% believe automated systems create new risks—such as fake data, external data manipulation, and inherent bias. [Captcha_Atlas1] CAPTCHAs are one of the earliest cybersecurity tools adopted by organizations. Today, in combination with data analytics, aberration-detecting AI, and headline-stealing encryption services like blockchain, CAPTCHAs are keeping digital business free of fraudsters. [Sponsor Accenture]( [Quartz Obsession] CAPTCHAs Decemeber 8, 2018 Bots wear the street --------------------------------------------------------------- This email was produced by Quartz Creative on behalf of Accenture, and not by the Quartz editorial staff. When cult streetwear brand [Supreme]( released $30 logo-stamped bricks in 2016—yes, [regular red bricks]( sold out almost immediately and went on to fetch $1,000 on eBay. While ludicrous to many, this sort of collecting is unremarkable for Supreme’s youthful obsessives, who’ll line up for blocks beside Supreme’s shops and overrun its web store with every “inventory drop.” The brand deliberately releases products in limited quantities, guaranteeing sellouts, and, in turn, seeding a [resale market]( where markups are magnitudes higher (you can buy plain T-shirts for $650). Consequently, Supreme’s online store is among the internet’s [most bot-abused sites](. There are hundreds of “shoebots” (with [dedicated]( [subreddits]( that buy out Supreme inventory at lightning speeds, dismaying real fans and garnering big bucks for botmakers. That was until March 2017 when Supreme’s site added a CAPTCHA. For years, bots had been bypassing item pages and checking out with impunity. But with CAPTCHA, bots must now convincingly pass themselves off as human. That’s hasn’t stopped scammers from building CAPTCHA-solving bots that can still siphon Supreme products, but it’s slowed them down. And as bots get better, so do CAPTCHAs. In 2017, the cost of [cybersecurity breaches was up 27%]( year over year, setting organizations back $11.7 million on average. Furthermore, [29% of business and IT executives]( report that their firm has been a target of adversarial bots at least once, and 77% believe automated systems create new risks—such as fake data, external data manipulation, and inherent bias. [Captcha_Atlas1] CAPTCHAs are one of the earliest cybersecurity tools adopted by organizations. Today, in combination with data analytics, aberration-detecting AI, and headline-stealing encryption services like blockchain, CAPTCHAs are keeping digital business free of fraudsters. Origin story "Am I a robot?" --------------------------------------------------------------- [Businessman Being Scanned at Airport] The word “CAPTCHA” was coined in 2003 by Luis von Ahn and his researchers at Carnegie Mellon. CAPTCHA stands for Completely Automated Public Turing Test To Tell Computers & Humans Apart. Thus a CAPTCHA is an auto-generated Turing Test, or a way to determine whether AI is able to pass itself off as a human. The earliest CAPTCHAs were distorted word images: [Captcha_Image_TechVision] Von Ahn developed CAPTCHAs so that web companies could defend against bot-wielding spammers who’d terrorize forums and scalpers who’d buy out concerts in a fraction of a second. These bots get more sophisticated as automation gets more sophisticated. In a world where AI systems are predicting hurricane patterns and simulating protein folding, a simple shoe-buying bot employing basic programming is [fairly simple to build](. AI vision meets human vision Stop spam. Read books. --------------------------------------------------------------- Soon after being invented, CAPTCHAs became requisite for sites wanting basic cybersecurity: web stores, social media, email, etc. Millions of CAPTCHAs were being solved everyday by internet users. Eventually, von Ahn realized he could harness that CAPTCHA-solving manpower to do something useful, so his company began digitizing books for Project Gutenberg. They would scan physical books, using optical character-recognition software to translate words into digital script. And when words were too illegible for their computers to decipher, they’d be turned into CAPTCHAs. So, going forward, instead of seeing random distorted characters, CAPTCHAs started showing people words that computers couldn’t understand, and when enough users confirmed what those words were, they’d be uploaded to an ebooks database. Von Ahn called his project “reCAPTCHA” and provided a slogan: “Stop spam. Read books.” Fraud detection The CAPTCHA bot uprising --------------------------------------------------------------- In 2009, Google acquired reCAPTCHA and used its crowdsourced brainpower to digitize the entirety of The New York Times print archive from 1851, as well as all of Google Books. And when they finished that, they used reCAPTCHA to read all the signs on Google Maps street view that bots couldn’t. But there were two problems. First, countless CAPTCHA farm services started appearing in developing countries, offering low-wage workers around [$0.80 per 1,000 CAPTCHAs solved](. The bigger problem, however, was that AI computer vision was getting so good that bots were starting to solve text-based CAPTCHAs without human help. In response, Google distorted CAPTCHAs so they became harder to solve—warping the text and adding noise and random artifacts. This was actually counterproductive. Google found that humans were less successful at solving CAPTCHAS than Google’s own advanced AI systems. The bots were winning. On the plus side, the absurd word combinations gave way to a culture of [CAPTCHA art](. By the digits [33%]( The accuracy with which humans were able to solve distorted text CAPTCHAs in 2014 [99.8%]( The accuracy with which Google’s AI was able to solve distorted text CAPTCHAs in 2014 [200 million]( The number of reCAPTCHAs that are solved everyday [10]( The average number of seconds it takes for a human to solve a distorted text CAPTCHA [25 million]( The number of books digitized in Google Books [54%]( The annual growth rate of global corporate spending on AI projected between 2015 and 2020 Check this out A cybersecuring checkbox --------------------------------------------------------------- [Captcha_GIF_Check] When Alan Turing first conceived of the Turing Test in 1947, he suggested that a computer program’s resemblance to a human mind could be gauged by making it answer a series of questions written by an interrogator in another room. Seventy years later, Google designs a reCAPTCHA system that can spot bots by requiring something simple: click on a checkbox. By checking a box, Google’s 2013 reCAPTCHA update—called “No CAPTCHA”—sends an HTTP request with information users unwittingly provide: IP addresses, cookies, even mouse movements. These identifiers show you’re just a friendly human with human-like web surfing habits. In early 2017, Google began [rolling out]( version of No CAPTCHA where users don’t even need to check a box, making CAPTCHAs all but invisible henceforth. In an era when managers found that [97% of their business data was unreliable]( having innovative CAPTCHA solutions is critical to prevent bots from tampering—writing fake product or restaurant reviews, operating fake social media accounts, clicking on ads to artificially boost ad performance. Lessons in self-driving Bot-fighting bots are building better bots --------------------------------------------------------------- With No CAPTCHA, Google has made their bot-stopping algorithms invisible and thus harder to penetrate. However, when there’s any doubt of a user’s humanity, No CAPTCHA will serve users an additional image-recognition challenge—like picking an image of a storefront, or identifying all frames within an image that show a street sign. This CAPTCHA test harnesses crowdsourced brainpower yet again to do something remarkable: train artificial intelligence. As data improves in volume and quality, [AI systems are finding ever-wider enterprise applications]( including process automation, language interpretation, fraud detection, virtual assistants, and so much more. By labelling images, for example, CAPTCHA-solvers are helping build high-quality datasets for Waymo’s self-driving car computer vision software, or enhancing Google Photos’ facial recognition ability. CAPTCHA-driven AI is a classic case of supervised learning, one of [several advanced machine learning techniques]( (deep learning, reinforcement learning, etc.). With supervised learning, AIs learn to recognize patterns from thousands of labeled data points: one image dataset will teach AI to pick out dogs in a picture, for example, while another will help the AI distinguish individual dog breeds. [captcha_atlas2] Fun fact After seeing reCAPTCHAs so successfully harness crowdsourced brainpower while providing a service, von Ahn and one of his graduate students came up with another idea: Build a free foreign-language learning app where students were actually helping translate the internet. The project, Duolingo, launched in 2011, and now serves 200 million users learning 23 different languages. Please prove you are human An alternative CAPTCHA: How many tries does it take you to plug in a USB? OneThree or more Correct. Carry on, human Incorrect. A BOT! If your inbox doesn’t support this quiz, find the solution at bottom of email. The correct answer to the quiz is Three or more. If you choose to [unsubscribe]( you are unsubscribing from future sponsor content emails, not from the daily Quartz Obsession editorial email. Quartz | 675 Avenue of the Americas, 4th Fl | New York, NY 10011 | United States [Advertisement]