What is a Turing Test?
Alan Turing, one of the founders and early theorists of what we now call computer science, posed the question, “Will it ever be possible to build a computer that can be mistaken for a human being?” Since then, computer scientists, experts in artificial intelligence and robotics have tried but have not succeeded in building a machine or crafting a piece of software that, under scrutiny, would be mistaken for a human. Every year, there is a competition where a panel of experts can ask questions of a hidden person or computer. They don’t know to which they are talking and, afterward, they vote on which contestants are human and which are machines. In the history of the competition no computer has fooled the panel.
What does this have to do with vision issues?
There are many web pages that contain inaccessible graphical word verification systems, also known as CAPTCHA. These intend to distinguish between actual humans legitimately trying to use the site and bots employed by spammers that will use the site for a commercial or malicious purpose. Thus, these graphics serve as Turing tests to separate the humans from the software.
The popularity of these Turing tests increases as spammers find more ways and web sites to use for their purposes. Some very popular sites, including BlogSpot, the host of this blog, Yahoo! And many other very useful web sites have implemented these tests.
The web version of the Turing test works by presenting a graphic that contains a word or sequence of characters drawn in such a manner that OCR software would probably not decipher it properly and then requiring the user to type it into a box on the web form. In its most basic form, no blind person can use it independently as they cannot see the graphic. Some web sites, Yahoo! For one, also provide an audio sample, distorted somewhat to prevent voice recognition software from understanding it. The audio samples, though, also make it difficult for people (like me) with perfect hearing to distinguish between the characters in the sample and virtually impossible for our friends with some hearing loss to use at all.
Who should be fixing this problem?
As a web accessibility advocate, I have long argued that such problems must be fixed by the content providers. Some advocates for people with vision impairments have suggested that Section 255 of the Telecommunications Act which requires all telecommunications equipment and services to be accessible to all would actually cover web sites as they can only be accessed via telecommunications products but this topic requires more attention and I’ll hold it off for a future posting. The web site hosts, though, would counter that they must do something to keep the spammers and other undesirables away.
The next group that could solve this problem are the Screen reader vendors. Historically, the AT companies have delivered many technologies that people believed too difficult to present in a manner that a blind person can understand. For many years, programmers at Henter-Joyce, Freedom Scientific, GW Micro, Dolphin Systems and others have proved the skeptics wrong by presenting an increasingly more complex collection of information to users of their products.
Why, then, do the AT companies claim that solving these relatively simple Turing tests evades the current generation of very smart people who work on screen readers? Clearly, this problem falls out of the scope of the work that some of these same programmers do today. But, those still involved in making screen readers, though, claim that this problem remains impossible.
Until about a week ago, I agreed that this problem is possibly too difficult to be solved in a screen reader. Then Blind Confidential received an anonymous tip from the world of research. Apparently, a graduate student has demonstrated software that can analyze these bitmaps and convert the contents to speakable text. My source said that this software runs pretty quickly and has shown a success rate between 95-98%. Although not perfect, I will take a probability in the high nineties over the near 0% we blinks have today.
This also begs the question, why does the multi million dollar screen reader industry claim this problem impossible while a graduate student, working on a meager university stipend, can solve it? Wouldn’t such an invention bring an AT company a terrific patent and a competitive advantage over the competition? Why do screen reader companies avoid doing research to advance their craft?
I believe that, until this technology reaches blind computer users, it should be removed from the web sites that currently employ such tests. I also believe strongly that screen reader companies should take a long look at this problem and, as Richard Stallman once said, “innovate, don’t litigate!”
Please join me in signing the petition below that specifically addresses Google’s use of this technique and, hopefully, we can start making headway in this problem in web accessibility.
The petition is entitled:
“Google Word Verification Accessibility”
It is hosted on the web by PetitionOnline.com, a free online petition
service, at the following URL: