RE: Accessibility of machine learning and generative AI draft - Comments from Josh

To Josh

Thanks so much for your detailed feedback and yes, if you can file them in GitHub that would be great.

To pick up on your point about repair and the innovation roadmap, the document has been broken up into two sections to try and address this – in terms of the innovation roadmap, the endgame from the user feedback that contributed to this is that AI offers the potential to seamlessly address accessibility issues without user intervention. One possible example is accessibility issues are addressed by the user agent as  AI assesses content on the fly before it is received by the user, but it’s hoped there’d be a lot more examples of this as the document is fleshed out. The second is more about the repair side and where automated tools can go and the current state of AI in addressing common accessibility aspects in WCAG such as the provision of alternative text and live captioning.

The implications, good and bad, are also something that needs to be fleshed out and you’re right about needing to put more of this in there. One example that often gets discussed here is the personalised voice feature on an iPhone where a non-verbal user can save their voice before losing it or use an AI generated voice from selection to make phone calls. This is a great AI accessibility innovation, but the downside is that automated systems that say ‘press 1 for <x> service’ and things like that often disconnect them believing the automated voice is a scam highlighting the privacy and security challenges, so some sort of identification mechanism would be needed. This use case isn’t highlighted in the document as it’s not strictly about content  but as you say there’s a lot that can be included as the scope is further defined.

Incidentally, do you have any capacity to be involved in the writing?  Your work across all the AUR documents was fantastic and would be great if you had some availability.

Scott.



, possibly being addressed in

Dr Scott Hollier
Chief Executive Officer
[Centre for Accessibility Australia logo]<https://www.accessibility.org.au/>
Centre For Accessibility Australia Ltd.
Phone: +61 (0)430 351 909
Email: scott.hollier@accessibility.org.au<mailto:scott.hollier@accessibility.org.au>
Address: Suite 5, Belmont Hub, 213 Wright Street, Cloverdale WA 6105
accessibility.org.au<https://www.accessibility.org.au/>
Subscribe to our newsletter<http://eepurl.com/drA-ib>

[X icon]<https://twitter.com/centrefora11y>[Instagram icon]<https://www.instagram.com/centreforaccessibility/> [Facebook icon] <https://www.facebook.com/centrefora11y/>  [LinkedIn icon] <https://www.linkedin.com/company/centreforaccessibility/>

CFA Australia respectfully acknowledges the Traditional Owners of Country across Australia and pay our respects to Elders past and present.

From: Joshue O'Connor - InterAccess <josh@interaccess.ie>
Sent: Friday, 11 October 2024 7:33 PM
To: public-rqtf@w3.org
Cc: Janina Sajka (janina@rednote.net) <janina@rednote.net>; Scott Hollier <scott@hollier.info>; Jason White <jason@jasonjgw.net>
Subject: Accessibility of machine learning and generative AI draft - Comments from Josh

Hi all,

I hope you are all well. Good to see this document progressing.

Some comments from me on the AI and Machine learning document (I can log these on GH also if helpful):

Over all some of the use cases are really about repair. What does the innovation roadmap look like? Where do we need it to be? It is better here to identify what AI does poorly, and what it does well? I think this document needs to be clearer about that and with an eye on a what the goal is. I’m not sure this doc outlines what AI/ML does poorly and therefore how we can make what it does better, rather than seeing AI/Machine learning as a panacea.

We are better off getting a clear view and presenting that - as well as by assessing the pros/cons maybe suggesting areas for further exploration, rather than a shopping list for current automated accessibility tools (which is what it currently looks like to me).

Some editorial comments below:

#Comment 1 on Scope:

I dont think the word ‘clearly' should be used here. This document could be better as a question, 'What is the current bearing of AI on accessibility?' or a gap analysis.

Remove ‘clearly'

"This is a draft collection of relevant information related to cross-disability accessibility guidance of how developments in machine learning and generative Artificial Intelligence (AI) clearly bears an impact on web accessibility standards and processes. Given the rapid changes in the consumption and development of AI design, this is intended to be a starting point to group the accessibility implications of machine learning and generative AI technologies.”

# Editorial:

"1.1.2 Machine learning
Machine learning represents a field of study within the AI domain that has a focus towards [on] statistical algorithm[s]- capable of learning from data and performing tasks without specific instructions. This form of AI tends to focus on determinations and predictions.”

# Suggestion about focus of the document:

"As online generative AI platforms such as ‘ChatGPT’ continue to offer consumers the unrestricted ability to create text and images, including video and audio from a variety of inputs, [it is important to as the question 'what is the benefit of these advances from an accessibility perspective. This leads to a follow on question 'Are their potential harms or other challenges?. This is because accessibility as a disciple in fundamtentially a quality issue and the principle of GIBO (Garbage in Garbage out) applies very much in the context of generative AI.’]”

#Suggestion ' it also is important to consider how accessible these outputs will be presented [and how this will effect the user experience of someone with a disability or who is a user of Assitive Technology], and if machine learning algorithm may address broader accessibility issues in everyday tasks. ‘

#Suggestion:
"This could be done through traditional means such as the addition of alternative text to images [which brings up the fundamental question of 'what is the quality of these alternate text desciptions' for images which do not have one. Or for recognising text that should [should have structural semantics identifying and fixing user interface componets that are not well formed, or are missing an accessible name or state information.”

# Follow on suggestion:

"Both of these use cases can to some degree today be assessed by automated accessibility checkers. There are quality issues here, as many checkers cannot make subjective assessments of if alternate text descriptions are actually useful. The question is to what degree can current AI or generative intelligence models bridge this gap and make qualitative repairs that are actually fit for purpose. This document aims to explore this and suggest where there are gaps and need for further research.”

I hope this helps!

Thanks

--
Joshue O’Connor
Director | InterAccess.ie