AI is making CAPTCHA increasingly cruel for disabled users
Written by Robin Christopherson MBE, Head of Digital Inclusion at AbilityNet
A CAPTCHA, (an acronym for “completely automated public Turing test to tell computers and humans apart”), is a test used in computing to determine whether or not the user is human. You’ve all seen those distorted codes or image-selection challenges that you need to pass to sign up for a site or buy that bargain. Well, improvements in AI means that a crisis is coming … and disabled people are suffering the most.
CAPTCHAs are evil
Whatever the test — whether it’s a distorted code, having to pick the odd-one-out from a series of images, or listen to a garbled recording — CAPTCHAs have always been evil and they’re getting worse. The reason is explained in an excellent recent article from The Verge; Why CAPTCHAs have gotten so difficult. Increasingly smart artificial intelligence (AI) is the reason why these challenges are becoming tougher and tougher. As the ability of machine learning algorithms to recognise text, objects within images, the answers to random questions or a garbled spoken phrase improve month on month, the challenges must become ever-more difficult for humans to crack.
Jason Polakis, a computer science professor at the University of Illinois at Chicago, claims partial responsibility. In 2016 he published a paper showing that Google’s own image and speech recognition tools could be used to crack their own CAPTCHA challenges. “Machine learning is now about as good as humans at basic text, image, and voice recognition tasks,” Polakis says. In fact, algorithms are probably better at it: “We’re at a point where making it harder for software ends up making it too hard for many people. We need some alternative, but there’s not a concrete plan yet.”
We’ve all seen the ‘I am not a robot’ checkboxes that use clever algorythms to decide if the user’s behaviour navigating the website is random enough to be a human. These used to work well — letting us through with that simple checking of the box — but increasingly the bots are able to mimic a human’s mouse or keyboard use and we get the same old challenge of a selection of images popping up as an additional test of our humanity.
The Verge article quite rightly bemoans the place we’ve arrived at — highlighting how difficult these ever-more-obscure challenges are for people with normal levels of vision, hearing and cognitive abilities. We just can’t compete with the robots at this game.
Don’t forget the disabled — we’re people too
But what about all those people who don’t have ‘normal’ abilities? People with a vision or hearing impairment or a learning disability are well and truly thwarted when it comes to CAPTCHAs that test the vast majority of humans to the very limit and beyond. After reading the article, I came away feeling that this very significant group (a fifth of the population and rising) deserve a mention at the very least — after all, they’ve been suffering in the face of these challenges far, far longer than those who do not have a disability or dyslexia (and have been locked out of many an online service as a result).
At the very heart of inclusive design is the ability to translate content from one format into another. For example, if a blind person can’t see text on-screen, it should allow the ability to be converted into speech (that’s how I’m writing this article). If someone can’t easily read a certain text size or font style or in certain colours, then it should allow for resizing or the changing of fonts and colours — this is all basic stuff that most websites accommodate quite well. Images should be clear and their subject easy to understand — and they should include a text description for those who can’t see it at all. Audio should be clear. All aspects of ‘Web Accessibility 101’.
The whole point of CAPTCHA challenges is to allow for none of these. No part of the challenge can be machine-readable or the bots will get in. Text can’t be plain text that can be spoken out by a screenreader for the blind — it has to be pictures of characters so excruciatingly garbled that no text-recognition software can crack it. Ditto with an audio challenge. Pictorial challenges must be so obscure that object recognition software can’t spot the distant traffic lights amongst the foliage etc, etc. It has ever been thus.
Today the road signs need to be obscured by leaves because the bots are better than ever at recognising them — but five years ago the images were still chosen to be just complex enough so as to thwart the bots of the day. And because the bots are using the same machine-learning AI as the assistive software used by disabled people to convert content into a form that is understandable to them, they were locked out too.
Did I mention? — CAPTCHAs are evil
So long as websites want to keep the bots from registering spam accounts or posting bogus comments, there will need to be some way for developers to detect and deflect their attempts. The use of CAPTCHA challenges, however, is not and has never been a fair (or even legal) one. It discriminates and disenfranchises millions of users every day.
So, whilst the article neglects to mention the significant segment of users most egregiously affected by CAPTCHAs, I’m hopeful that its main message — namely that this arms-race is rapidly reaching a point where the bots consistently beat humans at their own game — is a herald of better times to come.
As CAPTCHAs actually begin to capture the humans and let the bots in, then they begin to serve the opposite objective to that intended. They should then disappear faster than a disillusioned disabled customer with money to spend but wholly unable to access your services.
So what’s the alternative?
Companies like Google, who have long provided commonly-used CAPTCHA services, have been working hard on a next-generation approach that combines a broader analysis of user behaviour on a website. Called reCAPTCHA v3, it is likely to use a mix of cookies, browser attributes, traffic patterns, and other factors to evaluate ‘normal’ human behaviour — although Google are understandably being cagey about the details.
So hopefully by now you get the bigger picture. Hopefully you’re saying to yourself, “Ah, but will the clever analysis cater for users who aren’t so average or will they once again be excluded by not being ‘normal’ enough?” Excellent question — I’m glad you’re on your game and on-board.
For example, will I, as a blind keyboard-only user of a website, be flagged as a bot and banished? Will a similar fate befall switch users (like the late and much missed Prof Stephen Hawking) who use certain software settings to methodically scan through a page. Dragon users issue voice commands that instantly move the mouse from one position to another in a very non-human way. I could go on.
I hope you get the picture. Moreover, I hope that Google and other clever types working on the issue elsewhere get the picture too. They certainly haven’t to date.
Originally posted here
Originally published at digileaders.com on February 22, 2019.