Re: Is it time for AI with Accessibility?

Hi Wayne,
when you say: "... Maybe what we need is an AI approach. What if we analyze
the image of web pages as blocks of content ..."
who is the "we" that is analyzing the image?  the AI?  that is the approach
that IBM AI Accessibility research has taken, and filed a few pattens.

and when you say:
"... While AI might be too slow for live pages it would certainly be useful
...This would eliminate the need for assistive technology to have internal
knowledge of the page structure, scripts and style..."
Although currently true that AI might be too slow at browser-render-to-user
time, its not too slow for designer & developers using AI infused tools.
AT might not need as much knowledge of the page structure, but AI and the
authoring and repair tools would still need this information to prevent,
auto correct, or prompt for designer & developer confirmation.

I think a combination of AI infused browser designer/developer tools with
coordination by AT is the best approach.  Someone always seem to be missing
from the table, usually the browser, but often the AT.  And until recently,
508 and similar regulations have provisions for the authoring tools, but
still none to very little on the browser developers themselves.

Remember how long it took before the zoom magnification was built in?

___________
Regards,
Phill Jenkins
See the new IBM Equal Access toolkit and accessibility checker at
www.ibm.com/able
pjenkins@us.ibm.com
Accessibility Executive
IBM Accessibility
linkedin.com/in/philljenkins/
www.ibm.com/able
twitter.com/IBMAccess
ageandability.com




From:	Wayne Dick <wayneedick@gmail.com>
To:	W3C WAI ig <w3c-wai-ig@w3.org>
Date:	12/07/2020 02:27 PM
Subject:	[EXTERNAL] Is it time for AI with Accessibility?



A new model for personalization I have been an advocate for visual...
                                                                            
                                                                            
                                                                            
                                                                            
                                                                            
                                                                            
                                                                            
                                                                            
                                                                            
                                  This Message Is From an External Sender   
                                  This message came from outside your       
                                  organization.                             
                                                                            
                                                                            
                                                                            
                                                                            
                                                                            
                                                                            

   A new model for personalization
   I have been an advocate for visual personalization since 2008. I have
   fought for WCAG criteria that enforce separation of content from
   presentation in an effort to impose users visual needs on extant web
   pages. However, so much content is supplied at runtime or in
   stylesheets, that this effort seems profoundly difficult.  Also, the
   approach opens security holes by requiring internal access to web
   content.
   Maybe what we need is an AI approach. What if we analyze the image of
   web pages as blocks of content that need classification into something
   like HTML elements. This process may similar to OCR or voice
   recognition, only the data should be more regular.
   Consider the application to low vision. This group needs page
   restructuring (like linerization), color control, spacing, reflow of
   text, and enlargement that exceeds most rendering algorithms.
   The difference between OCR and this structure recognition would be that
   we always have an image of the intended page from the runtime HTML.
   While AI might be too slow for live pages it would certainly be useful
   for books and publications.
   However, if large browsers would get in the game live delivery might be
   possible. Google, for example, is a leader in deep learning. This would
   eliminate the need for assistive technology to have internal knowledge
   of the page structure, scripts and style.
   We may be at the point where this is possible. I am interested in any
   research in this direction. Anybody know about anything like this in
   progress?



   Best, Wayne Dick