2004 Conference Proceedings

Go to previous article 
Go to next article 
Return to 2004 Table of Contents 


Matthew May
12332B 14th Ave NE
Seattle, WA 98125
Phone: 206-547-2109
Email: mcmay@w3.org

In addition to creating guidelines for accessible Web content, authoring tools, browsers and media players, W3C/WAI works closely with groups developing specifications that affect future accessibility to the Web. WAI, through its Protocols and Formats Working Group (PFWG), reviews W3C specifications to ensure accessibility, and tracks Web technologies outside of W3C, to build the foundation for a more accessible Web.

One area of concern that has come to the attention of W3C/WAI is an inaccessible approach to verifying users of modern Web-based services.

Alternatives to Visual Verification

New roadblocks to accessibility have sprung up on many major Web sites, in the form of visual verification schemes. Researchers at Carnegie Mellon University call this system "CAPTCHA," the "Completely Automated Public Turing test to Tell Computers and Humans Apart". A Turing test (named after famed computer scientist Alan Turing) is any system of tests designed to differentiate a human from a computer. Images of text, which have often been obscured or distorted to prevent optical character recognition (OCR) systems from reading them, are displayed to a user, who must then type the characters into a text area before completing registration.

Sites offering services such as free Web email or tickets to sporting events are besieged with thousands or millions of requests each day from Web robots. These computer programs are designed to create thousands of fake accounts, through which they can harvest ticket prices for site comparisons, or create fake email accounts which are used to send spam messages.

Of a handful of techniques meant to eliminate robots from accessing these services, the one that has taken hold has been visual verification. In the last year, the number of sites employing visual verification schemes has skyrocketed, despite calls from a number of disability advocacy groups and industry columnists that users who are dyslexic, blind, or have low vision are being left behind. Some implementers of visual verification have attempted various workarounds which, though many show good intention, remain equally inaccessible.

In an effort to solve this problem, the PFWG is producing a document outlining the flaws in visual verification, explaining its path to obsolescence, and describing seven different alternatives to the technique, all of which are inherently more accessible. The goal of this exercise is to point out to Web development teams that alternatives to this system do exist, and can better serve a wider base of users.

The Working Group analyzed the requirements of sites that are presently implementing visual verification, and arrived at three sets of needs shared by sites that implement security measures:

From this set of requirements, the Working Group evaluated a number of potential means to provide these levels of security, and how they may be more accessible to visual verification. Low-fidelity approaches, from live operators to logic puzzles (e.g., "what is the fourth word in this sentence?"), offer varying levels of security against exploitation by robots. Popular methods of identity checking, such as credit-card validation, can be useful in many cases, but often have the potential to leave users without them behind.

More promising methods are available now, or in the near future. Since robots can often be identified by the patterns in which they transact with Web sites, site developers can use heuristic data to filter out traffic caused by robots, without the need to verify all users with inaccessible technology. Likewise, new email accounts, for example, may not need to offer unlimited outgoing email access at the time of account creation, choosing instead to limit outgoing messages to 10 per day. This would damage the value proposition among spammers, who would need to create too many new accounts to efficiently push out spam messages.

In the longer term, single sign-on systems may allow uniquely-identified users to use sites without verification. What's more, a decentralized approach to storing identity using biometric data, such as retinal imaging or ingerprinting, may preserve privacy to a greater extent, allowing users with disabilities to store their own preferences for interacting with Web sites, as well as guaranteeing that a unique human is at the other end of the transaction.


Accessibility barriers can arise where we least expect them. In the case of visual verification, we've found that sites can undermine the progress made in accessible design through poor policies. When this happens, the work of WAI includes offering a reasoned approach to break down unnecessary barriers to Web content. We find visual verification to be an approach that offers at best a stopgap measure to keep robots from exploiting Web sites, and we urge Web developers and browser vendors to work together toward a more stable and universal approach.

Go to previous article 
Go to next article 
Return to 2004 Table of Contents 
Return to Table of Proceedings

Reprinted with author(s) permission. Author(s) retain copyright.