Summary: The key difference between user interfaces for sighted users and blind users is not that between graphics and text; it's the difference between 2-D and 1-D. Optimal usability for users with disabilities requires new approaches and new user interfaces.
The typical advice for making websites accessible is to create a single design for all users, then ensure that it complies with additional guidelines for use by people with disabilities. This is also the approach we take in our own guidelines for accessibility: They aim at improving usability for users with disabilities by tweaking traditional websites and intranets to take special needs into account.
The main reason for this single-design-for-multiple-audiences approach is the assumption that most companies are unable to keep two different designs up-to-date. Thus, if they optimized a separate design for users with disabilities, they'd risk it rapidly becoming out of synch with the "main" website.
For most websites, this assumption is probably true. The average company allocates very limited resources to serving users with disabilities, so their best approach is to use these resources to improve the main design, rather than to design, implement, and maintain a separate site.
However, perfect usability for users with disabilities requires separate designs optimized for each of the main access modalities. An interface for blind users, for example, should be designed for auditory presentation. Such a design is inevitably better than simply reading out loud something designed for screen-based visual presentation, even if that presentation is modified to take blind users into account. Of course, in an ideal world, separate and equally targeted designs would also be available for low-vision users, users with motor skills challenges, and so on.
Optimize for Linear Access
The biggest potential gains reside in creating a special design optimized for auditory presentation . A good 1-D audio design not only serves blind and low-vision users, but will also help users in cars and other settings as auditory access to Internet content increases.
There's a fundamental difference between visual and auditory presentations in terms of dimensionality: screens are 2-D and depend on layout for presentation, and audio is 1-D and relies on sequence for presentation. Linearizing a 2-D layout is simply not as usable as having a good designer create a targeted 1-D layout.
In a 2-D layout, a good graphic designer organizes blocks of information to provide a visual of the website's structure and to prioritize the most important tasks by their relative size and 2-D location. For example, designers typically place the most important Web page elements in the center of the top screenfull, since that's where sighted users tend to look first. Although a targeted 1-D audio presentation should start with the most important information, most audio translations simply read a 2-D page aloud, starting at the top left, which mainly contains information that sighted users typically skip. Also, simply reading aloud eliminates size distinctions, which are key elements in 2-D designs.
The fundamental difference between user interfaces for sighted users and blind users is thus not the distinction between graphics and text, but that between 2-D and 1-D. Unfortunately, we don't know much about good 1-D layout for interaction design (we know how to produce good radio shows, but they are not interactive forms of audio content). However, it's quite likely that an optimal linearized presentation would make more use of hypertext than we find in 2-D layouts that benefit from visual scannability. Thus, a design for auditory use might end up being more N -dimensional than purely 1-dimensional.
Blind users might also benefit more from a 3-D user interface than sighted users. I imagine a gestural interface, where users are surrounded by different types of information located at different spots. They'd access this information by poking at points in space. Thus, designers could "park" search results and other key pieces of information in particular locations, where users could retrieve them with a gesture. For sighted users, such an interface would be useless: There would be no words floating in space -- unless they used a clumsy VR helmet. For blind users, however, gestures and unseen (but easily remembered) 3-D locations might beat linear read-outs.
Just as optimal 1-D design benefits sighted users needing hands-free content access, so, too can designs for users with other disabilities expand other users' options. Users with low vision, for example, can only see small amounts of information at any given time. Optimizing the design to suit their needs also benefits users of mobile devices or other small-screen devices, who essentially have the same limitations.
Regardless of the targeted user group, all designs must offer the same functionality and provide access to the same content. A good content management system will be necessary to ensure that all versions are kept in synch and no one misses out on updates.
Of course, the approach I advocate here is overly Utopian. I doubt at present that companies will spend enough money on users with disabilities to create sufficiently good alternative designs -- especially when such designs will require a completely new set of usability guidelines. However, the future is more promising: Once auditory access to Internet content becomes more mainstream, I expect that resources to create optimal audio designs will be more readily available.