Imagine, it is well past midnight, you just had your nth cup of caffeinated beverage of the night because you have a very urgent deadline encroaching and you need to look something up on the internet pronto.
Then it happens, the webpage you are browsing asks you to 'select all images with bridges' to prove your humanity. Blurry images featuring bridges on the cusp of ruin, bridges tipping over to the next image by the barest of margins, bridges sneaking into bushes, avant-garde bridges that look unlike any bridge you have ever seen; make the existential task of proving your humanity difficult and frustrating.
However, most of us do not even have to imagine this since such experiences are becoming worryingly common as CAPTCHAs are becoming increasingly elaborate.
What are CAPTCHAs and why do they exist?
Anyone who has spent any significant time on the internet is familiar with CAPTCHAs. But what are they and more importantly why do these internet equivalent of speed breakers exist in the first place?
The Turing test is a test proposed by the mathematician Alan Turing to determine whether a computer can "think." CAPTCHA is the world's most frequently administered Turning test. CAPTCHA stands for the Completely Automated Public Turing test to tell Computers and Humans Apart.
CAPTCHAs are used to differentiate between real users and automated users, such as bots. At their most widely used form, CAPTCHAs present the user with a simple test like reading letters/digits or listening to speech and then ask the user to type in what they read or heard.
The characters featured in the image or sound clip are usually distorted in several ways to make it difficult for a machine to complete the test successfully. CAPTCHAs can prevent a wide variety of abuses, such as invalid account creation, spam comments on blogs and forums, maintaining poll accuracy, etc.
The fault in our CAPTCHAs
As good as CAPTCHAs are at providing security, they are just as good if not better at doing something else, like training AI.
In the golden days of the early 2000s relatively simple images of text sufficed to baffle most spambots. But a decade later, Google, while trying to digitise books and magazines for their Google Books project, ran into the problem of high scanning error rate.
Guess how they fixed that problem? They bought the reCAPTCHA program from Carnegie Mellon researchers and started presenting a digital image of the text from scanned books in the CAPTCHA tests. So, users typing the text as they saw meant converting the scanned image to text data which can be stored as a digital document, further improving Google's optical character recognition algorithms.
With so many people solving millions of CAPTCHAs, unwittingly training AIs to solve problems, AIs surpassing humans in solving certain problems was always an eventuality.
In 2014, Google, who now owned the world's most prominent CAPTCHA system, reCAPTCHA, found out that when it came to solving the most distorted text CAPTCHAs, one of its machine learning algorithms got the test right 99.8 percent of the time, while the humans could only get it right a lowly 33 percent of the time. To combat this, Google has moved on to a system called noCaptcha reCaptcha. It observes user data and behaviour patterns to let some users pass through with a tick on a checkbox that reads "I am not a robot" and presents others with the image labelling we see today.
However, bots keep getting better and keep on beating CAPTCHA technology so Google keeps on pushing out new iterations of reCAPTCHA. We have come to a point where Captchas have become much harder for humans but easier for bots.
What are the alternatives?
One of the biggest challenges researchers are facing in regards to coming up with CAPTCHA alternatives is the fact that humanity is a colourful tapestry of widely different cultures, languages, and religions resulting in wildly different life experiences. The challenge of coming up with methods that any human regardless of their background can use to prove their humanity has proven to be a very difficult one so far.
'Gamification' is one alternative to CAPTCHA that has been proposed. Yes, it still slows users down but it also introduces some fun into the process that ranges from mildly annoying to infuriating, depending on whom you ask. This approach typically asks users to drag and drop items into their designated slots, i.e.: planting plants in a pot. However, I have a hard time imagining having to 'plant a tree' to access a financial report being a financial analysts' idea of fun.
Another solution, known as the 'Honeypot' works by placing a hidden field in a form that is invisible to human users but visible to bots. The idea is that spambots will identify it as a normal field to complete, and any forms with that particular field completed will be marked as spam.
Some experts suggest rather than tests, implementing something called 'Continuous authentication', essentially observing the behaviour pattern of a user and looking for signs of automation. A real human being lacks complete control over their motor functions and so while a bot will interact with a page without moving a mouse, or by moving a mouse with extreme precision, human movements have an entropy that is hard to fake, at least for now.
CAPTCHAs may live on in a different form as well. Amazon received a patent in 2017 for a method involving providing a CAPTCHA challenge from a library or set of challenges that are designed in a manner that causes a human user to trivially get the answer to the challenge wrong.
Called Turing Test via failure, the only way to pass is to get the answer wrong. However, it remains a possibility that a bot could be programmed to answer the question or challenge incorrectly in the same manner as a human user. In response, the inventors claim that the type of wrong answers provided by a bot will be different from the type of wrong answers provided by a human.
Eugene Goostman is a 13-year-old boy from Odessa, Ukraine, he has a pet guinea pig and his father is a gynaecologist.
At least according to 33 percent of the judges of a Turing test contest, arranged to mark the 60th anniversary of his death, Eugene is a genuine boy. In reality 'he' is just 'it,' a chatbot developed in 2001 by three programmers. The bot uses personality quirks, humour and feigned a poor grasp of the English language to misdirect judges from its non-human tendencies and lack of real 'intelligence.'
So, what happens when bots can also start making errors like humans? Is fumbling through webpages, making typos and switching between browser tabs the endgame for AI?
Does it no longer mean 'To err is human'? Then what does it really mean to be human? Is there one definitive quality shared all across humanity that no machine can ever mimic but can be demonstrated to other machines at the same time?