An inquiry into CAPTCHA’ Past, Present and Future/s.
From Security Technology to Pattern Recognition Training Device
When CAPTCHA was invented in early 2000, it was, and still the idea behind its usage is, to ensure that humans use critical infrastructure and web services. But, in its basic principle, Shweta Ganjoo has briefly summarised the logic of CAPTCHAs, is similar to machine learning “all about identifying patterns”. While digitisation, machine learning technologies and AI can still be understood as young innovation, pattern recognition is not new at all.
Be it when people on the street recognise a person they already know, be it that we learn to assign shapes, colours and functions to a similar order in elementary school, or be it that we know where we are going, and should not go, as well as when we close our eyes to scenes when being in the cinema, etc. Patterns offer useful templates that help us structure and organise the world, in short: they reduce complex relations surrounding us.
Recently, the sociologist Armin Nassehi has argued in Muster that digitisation and the associated security issues (mentioned above) are not the problem that we face as a society, but the solution that we apply aiming to understand and secure even better the world we live in. Here, humans come into play, because we are much more intelligent machines. One can find this idea in the blockbuster Ex Machina (2015), when the android Ava asks a researcher what the artful patterns she draws each day do mean. As humans, we can move around very easily and manage all hurdles of our environment in big part thanks to our pattern recognition ability. Tech-corporations are using and implementing humans to train, and correct, and to improve pattern recognition, which for instance self-driving cars need in the future.
CAPTCHA can be understood as only one of the cultural techniques that tackles this challenge. In a technophile world, CAPTCHA is only one answer to a much bigger problem. As a very young innovation in human-machine-interaction, it allows us to improve the ways we understand the omnipresence and determination of patterns for the ways how we perceive things. CAPTCHA not only secures web services, but in our daily lives learns to perceive, sense, and know as we do by extracting patterns out of our CAPTCHA-bility.
This task entails that pattern recognition, which humans apply in their everyday life without efforts, is taken into service to train machines.
We need classifications to organise and manage what we see, and cannot see, what we fear, and cannot understand. Yet, at the same time patterns are historical, always in the making, can be undone. Patterns depend on contexts, they are not natural, but naturalised form of how we perceive.
Taking history seriously means to recognise that racism, sexism, ableism, classism, and all forms of discrimination are based on patterns that simplify too much our complex interrelations. CAPTCHA might help us understand that, just as CAPTCHA are programmed algorithms, use specific software, and rely on hardware and infrastructure, that enables but also disables access. This is why CAPTCHA prompts to question the basis of how patterns are programmed.
No CAPTCHA can survive an army of humans paid to crack it
A combination of cheap Internet access and the commodity nature of CAPTCHA itself has globalised the solving market. Today, you can find hundreds of "farms" solving large numbers of CAPTCHA, with retail prices as low as $1 per thousand.
So just how did CAPTCHA-busting turn into work real humans get paid to do?
CAPTCHAs are ubiquitous. We need to solve them to register email accounts, post comments, buy concert tickets. Although some AI systems have been developed to solve CAPTCHAs, the market discovered that it is far cheaper to farm out the problems to workers in developing countries.
Deciphering a line of distorted letters or sorting images is a full-time job for some people. For others, a way to make ends meet. There are a ton of jobs on offer. You can find reviews of the best ones (the platforms where you will make money faster).
"Yes, it's not big bucks and the work is a bit monotonous", one popular CAPTCHA-solving website admits. "But you can be sure that you will receive each cent you've earned and you will not be scammed".
We were curious so signed up to join the army of low-paid human (robots) cracking CAPTCHAs.
It takes the average person approximately 10 seconds to solve a typical CAPTCHA. The fastest human CAPTCHA-solvers are capable of doing 20-25 per minute or, on average, 1,000 per hour. Howeverf, in some of these platforms the worker may need to wait for 20-30 seconds for a new CATPCHA test to appear, which means the person needs to spend much more time to solve a given number of CAPTCHAs.
While the market for CAPTCHA-solving services has expanded, the wages of workers solving CAPTCHAs have been declining.
A research paper by the University of Calfornia (San Diego) looked at historical advertisements on getafreelancer.com and found that in 2007 CAPTCHA-solving routinely commanded wages as high as $10 / per 1,000 CAPTCHA.
By 2010, however, they had dropped to $0.75 / 1,000, with some workers earning as little as $0.5 / 1,000. In 2020, workers can expect $0.25-$0.60 per 1,000, according to one employer.
A leader in the market, Anti Captcha, reports their average worker makes about $100 per month, which (according to their website) is "a very good salary in such countries like India, Pakistan, Vietnam and others".
In general, their wages compare to paid to low-income textile workers in Asia and it’s likely CAPTCHA-solving is being outsourced to similar labour pools.
The work is repetitive, tedious, and demotivating. The platforms (one could describe them as factory floors) are onerous spaces where workers are timed and scolded — in red text — for being wrong.
Julian Posada is a PhD student at the University of Toronto (Canada) working on click workers. For his research, he established a list of 95 micro-work platforms and observed the traffic they generate. This is how he was able to see the importance of CAPTCHA platforms.
Most of them are established in Russia, he explained in an interview on 13 November 2020, even those with addresses in other countries such as Cyprus or elsewhere in Europe. Traffic, he said, comes mainly from Latin America, especially Venezuela, then from South Asia: India, Bangladesh, Thailand, Philippines, Indonesia. Finally, a portion of the traffic comes from Ukraine and Russia.
Among all the tasks that can be found on the web, CAPTCHA solving is one of the least paid, with 1,000 tasks to perform to get between 0.5 and 3 dollars, depending on the difficulty.
"This is a regression compared to Amazon's Mechanical Turk platform, which was able to feed people in the United States," said Posada.
Who are the customers of these companies? "I don't have the answer", he said. "Traffic is very important, where does all the money come from to pay all these workers?"
The customers are mostly people who want to bypass CAPTCHAs. Some good, and others malicious. Developers use these services to test their apps before launching. Journalists may try to circumvent CAPTCHAs to scrap websites for research and investigation.
And then there are the e-mail spammers who need to solve CAPTCHA to create email accounts from which to spew advertisements, the blog spammers who want organic “clicks” and others trying to boost their search engine placement.
Mostly though, the humans who use CAPTCHA don't realize their labour is being tapped. More than 4.5 million websites use Google’s reCAPTCHA and the system collects hundreds of millions of daily solves. This equates to more than 100 person-years of labour every day.
A Google competitor, hCAPTCHA, estimates this free labour has likely earned Google billions since they acquired reCaptcha in 2009.
So-called machine learning and supposedly AI-driven technologies and practices, in reality, depend and rely on the data constantly provided by humans, knowingly or unknowingly.
We have created an environment, an ecosystem, in which it's hard to separate individual human tasks and machine individual tasks.
In this environment, work processes consist of intertwined tasks in which humans feed data into automated processes that then keep asking for more data.
The difference with CAPTCHA is friction: we are aware of our 'work' when solving CAPTCHAs because we are forced to stop, look and act; while in many other instances people have become accustomed to simply provide their inputs and their personal data automatically, with no friction at all.
Note: As you may have noticed, in the text above some words are cut at the end of the line. That's because text is "preformatted", which prevents the computer from recognising where it should move to the next line. And that's another instance of why we humans are much better than computers at recognising patterns and solving CAPTCHAs.
CAPTCHA Future(s): Utopia / Dystopia
In its cradle, “the Internet” was seen as a messianic technology, carrying the promise of freedom, participation and democracy. According to its pioneers, this technology had the potential to become a space in which we leave our bodily identities such as age, gender, sex and race behind, creating equal opportunities and access for all.
Today, in the face of the digital and omnipresent Frankenstein monster that we have collectively bred and inhabited since then, such imaginations of the Internet seem naïve.
Even though the internet has revolutionised information and communication infrastructures on a (nearly) global level, we are currently also finding ourselves trapped in a space structured by tech companies, platform algorithms and paywalls, filled with hate speech, surveillance and political manipulation, where the threat of phishing and identity thieves is always present.
Instead of giving us equal opportunities, the internet has become a free buffet for “data octopuses” who collect, analyse and capitalise the personal data we produce. In many senses, the Utopia turned out to be a Dystopia.
Guarding robots, working humans
In this present digital world, CAPTCHAs are presented as guardians. Their job is to tell humans and robots apart. By doing so, they are safeguarding our precious digital identities from ending up in the claws of sneaky bots, the internet villains. CAPTCHAs are the annoying gatekeepers of the platforms we visit on a daily basis.
However, as we have seen, CAPTCHAs turn the collective of everyday internet users into a silent work force.
While CAPTCHAs are supposedly telling humans and robots apart, they are simultaneously taking advantage of our unique human abilities, turning us into working human robots.
Thus, “we” work for “them” – the tech giants and the masters of AI, the saviour (or tyrant) of our time. Without us knowing, we collectively train this “new” super intelligence while solving CAPTCHAs. The current training program for the AI is unknown for most of us.
That is the state of the art today, which leads us to the question: utopia / dystopia? How do we imagine the future of CAPTCHA?
Dream(s) of CAPTCHA
Imagination is defined by the Oxford handbook of the Development of Imagination as “the capacity to mentally transcend time, place, and/or circumstance” – a capability (still) unique to humans. We wish to use this human and collective power – our imagination – to envision how a future of CAPTCHAs could look like.
Imaginations are not mere fictional stories or thoughts; they carry the potential of shaping our futures. According to Max Haiven and Alex Khasnabish, “Radical Imaginations” can be the spark of social movements: “The radical imagination is not just about dreaming of different futures. It’s about bringing those possibilities back from the future to workon the present, to inspire action and new forms of solidarity today.”
However, as they point out, the powerful also dream of the future; dreams that for us are “experienced […] as nightmares of insecurity, precarity, violence, and hopelessness.”
Thus, after looking at the past and present of CAPTCHA, we now use our imagination and turn to the future. How can we radically imagine a CAPTCHA which is contributing to a better world? These imaginations form a manifesto for CAPTCHA future(s) with 5 principles.
At the same time, we juxtapose these dreams with our nightmares, following the dreams of tech-giants and capitalist logics, going down the path of digital surveillance and oppression. We hereby invite you to come a long on an imaginary trip to the utopian and dystopian future(s) of CAPTCHA.
CREDITS: “CAPTCHA Past, Present and Future/s” was created between 12 and 14 November 2020, as part of the Silent Works Winter School, by Géraldine Delacroix, Monisha Caroline Martins, Julia Molin, Rebecca Puchta, André Rebentisch and Cagri Taskin; Sotiris Sideris and Jose Miguel Calatayud acted as facilitators and editors.
Sources of cover images: “I swear I’m not a robot!”, https://blog.cloudflare.com/moving-from-recaptcha-to-hcaptcha/ “CAPTCHA Past”, https://www.pandasecurity.com/en/mediacenter/panda-security/what-is-captcha/ “CAPTCHA Present”, https://www.change.org/p/it-s-time-to-finally-kill-captcha-0da5aea6-50d9-4b9a-b54d-f30d4f1e9478/ “CAPTCHA Future/s”, Screenshot from Blade Runner Final Cut, DVD, VLC Player Screenshot, taken on 13 November 2020.