Re: [EXT] AI and the future of Web accessibility Guidelines

> On Apr 11, 2024, at 6:33 AM, Nat Tarnoff <nat.tarnoff@levelaccess.com> wrote:
> 
> I also think that while we have ML intelligent tools that can determine some of what is in a picture, giving alt text to an image that describes the image is often the wrong alt text. It won't deliver the meaning or emotion the image generates when in context. The difference might be "a bouquet of roses" versus "Our vibrant red and yellow roses are the perfect gift for your someone special". While technically accessible in the first count, the second is much more meaningful and effective. On the off chance AI can give more in depth information, I don't think it is going to get the feeling a human can provide until it is Turing complete.


hmm  interesting 

but I think that  "Our vibrant red and yellow roses are the perfect gift for your someone special”  is outside of the purpose/role of an alternative text for a picture..  

The purpose/role of the alternative is to provide the blind person (or other person with disability) with the same information as the person who can see gets from the picture.  

If the picture is a bouquet of roses — then a person with vision sees a bouquet of roses.   And the blind person should be able to get the same information.    

How AI can handle this
With AI they will be able to get not only the general description of the picture ( a bouquet of roses)
But will also be able to query it for more 
“How many roses”
What variety ?
What kind of vase?
What type of arrangement (ie. what else is with the roses?   Baby breath?  ) 
and much more information —  information that would never be in an alt text put in quickly to fulfill the alt text requirement 

AND if you did put all that in an alt text - blind users would complain you put in too much or the wrong information.
It is the ability to query and get what you want -  and not everything else -  that will be the power. 


If the author wants to convey "Our vibrant red and yellow roses are the perfect gift for your someone special”  
Then, the author should put that in the text on the page - because a sighted user would not get that from a picture.
and that is not a text description of what a sighted person sees - or gets from the picture.


Bottom line 
AI could not only ensure that there was an accessible alternative (in all the places where there is none today) 
  — but it also allows the user to get whatever level of information they want  (just a little — or a lot)  Something not possible with accessible alternatives provided by human users.

As to the “AI’s sometimes to nuts"
Note that AIs are now being trained with focused information - not “all information on the internet”.    And some are constraining answers to those found in specific datasets.

But training an AI off of twitter -  or any social media  - is just plain nuts in my opinion unless you are looking for ways to get wacky output from an AI.
Garbage in - garbage out is true for any data system - and AI is no different.   
But we don’t need to base our AIs on unreliable information.

Twitter may be great for getting samples of the ay people talk -but terrible as a source for reliable information…

That said - I still have my general concerns on AI.  And I still believe that we need to not depend on it til it is dependable. 
But I think that when it is — it will be far superior in its ability to  a) provide alternates everywhere   and  b) allowing the alternative to be shaped to better fit each individual and their needs.
And it think we will all be surprised as to how fast it will advance.

Will all alternatives always be great?   I expect there will be failures.  But the failures will be a small fraction of  the failures from human alternatives — and massively fewer than the completely missing alternatives we see today. 

Color me optimistic - but in the camp of - “great when it gets here” - lets keep an open mind (pro and con) as to when that will be

Best 

G