AI and Machine Learning - Scope unclear from title

[Chudesnov]

The scope of the document is defined as follows:

ai-accessibility/index.html

Lines 21 to 29 in 41232e4

	<ul>
	<li>Accessibility issues raised by the use of machine learning and generative AI in Web-based applications, including its interactive role in contributing to the user interface.</li>
	<li>The different considerations applicable to the application of machine learning in the authoring or content development environment, by contrast with the user's environment.</li>
	<li>Further discussion of the role of machine learning-based AI in evaluating the accessibility of Web content.</li>
	<li>The use of generative AI in software development, and in particular for writing or enhancing user interface code.</li>
	<li>Clarification of the ways in which machine learning and generative AI affect computational problems relevant to accessibility, both positively in offering new functionality and opportunities for personalized interfaces, and negatively in creating unique risks.</li>
	<li>Achieving appropriate combinations of AI and human expertise in meeting challenges of accessibility.</li>
	<li>Approaches to evaluating and refining machine learning systems during their development that take into account accessibility-related requirements.</li>
	</ul>

ai-accessibility/planning/scope.md

Lines 3 to 12 in 41232e4

	* Use of machine learning-based AI to enhance accessibility of Web content, including its application to
	+ Code generation
	+ Web content creation/authoring tools
	+ accessibility evaluation/content remediation
	+ Delivery of Web content to users
	* Application of machine learning-based AI by users themselves, including
	+ Assistive technologies
	+ Accessibility-related features of user agents
	+ Accessibility-related features of applications (e.g., speech recognition for caption generation in a meeting application)
	+ Applications that use machine learning to generate content, especially the accessibility of the content they create

The actual content of the document seems to be aiming to address both the current state of accessibility in content generation services based on large language models and diffusion models and the use of ML to create new technology or enhance existing technologies (whether ML-based or not) that help users overcome potential accessibility barriers.

However, the title "Accessibility of machine learning and generative AI" is ambiguous. It is unclear whether the title refers to the potential inaccessibility of new generative AI tools or content produced by them, as ML/GenAI is applied across a wide range of use cases and industries, or if it is primarily focused on using ML to improve existing technologies and inform relevant accessibility standards. I would argue that the title suggests the former, while its contents are mostly about the latter, with a bit of focus on the former.

I wonder if addressing everything ML/LLM-related in one document is a good approach and whether the title suits the current structure. It might be beneficial to do two things:

1. Clearly split the content according to the following distinctions either as part of the same document or in two different documents:

   • Evaluating the challenges posed by new generative AI tools (e.g., chatbots, media and code generation from text prompts) created without accessibility in mind, which means that as these are adopted by more people, more inaccessible content and products are produced (most likely without the person requesting it from the AI model even knowing that it’s inaccessible).
   • Focusing on the use of ML to improve current user agents, assistive technologies, evaluation/auditing tools, or development tools—and aligning these efforts with standards like UAAG, ATAG, etc.

2. Clarify the document's (or documents') objectives by choosing a title that accurately reflects its goal and focus, such as:

   • Accessibility Challenges Posed by Generative AI-Based Authoring Tools
   • Enhancing Accessibility through Machine-Learning-Enhanced Authoring Tools and User Agents
   • (For the combined document) Machine Learning and Accessibility: Challenges and Opportunities

[Chudesnov]

Possibly a duplicate of #12

[joshueoconnor]

Thank you @Chudesnov - discussed in RQTF, thanks for the suggestion. The final title/form of the document is a work in progress at this point. As we go, we expect this will be come clearer.

[joshueoconnor]

Discussed again in RQTF on 14th May - to be continued. Leaving open for reference.