Verbose AI: The Accessibility Challenge - CSUN 2025

7mary4 336 views 20 slides Mar 11, 2025
Slide 1
Slide 1 of 20
Slide 1
1
Slide 2
2
Slide 3
3
Slide 4
4
Slide 5
5
Slide 6
6
Slide 7
7
Slide 8
8
Slide 9
9
Slide 10
10
Slide 11
11
Slide 12
12
Slide 13
13
Slide 14
14
Slide 15
15
Slide 16
16
Slide 17
17
Slide 18
18
Slide 19
19
Slide 20
20

About This Presentation

This presentation was created for the 2025 CSUN Assistive Technology Conference

Verbose AI: The Accessibility Challenge
Explore how verbose AI-generated image descriptions could hinder screen reader accessibility. We'll share real-world examples, the importance of context, AI's limitations,...


Slide Content

Verbose AI: The Accessibility Challenge Marco Salsiccia, Ted Drake and Lucy Greco

What to expect in the next hour Ted Drake The history of alt text Best practices for image links Incorporating AI in image creation Marco Salsiccia The importance of verbosity Personalization Use of AI for image description

The history of alt text The <img> tag was part of HTML1 Added in HTML 1.2 Draft Mandatory in HTML 4 for images and image maps Creating Killer Web Sites - Chaos insues

No Alt Attribute The screen reader will announce file name or link’s URL.

Image alt repeats heading Image links MUST have alt text that describe the link’s purpose. It’s common to use existing text within the image’s alt attribute. But this leads to duplicate links

Best Practice: Combine links

Empty alt attribute on image link Alt=”” = Image doesn’t exist Link is announced as href value Images cannot be accessed by Assistive Technology (AT)

Image alt when links are combined alt=”” When link text is adequate alt=”{Description}” When image provides context to the link Leonie Watson: Text descriptions and emotion rich images (2007)

Intuit’s AI Image Description Image described when uploaded to CMS Generate image description Compare with Intuit’s Word List Update with Intuit’s Content Design Guidelines Insert description in CMS as a suggested value Content editor reviews description to ensure it adds context to page

Marco Salsiccia AI, Verbosity, and Customer Control

Problems with AI Image Description Overly verbose No context for the subject Inaccuracies/Hallucination Cannot ask clarifying questions or hone the prompt

Overly Verbose Descriptions Alt text = 120 characters AI image descriptions ChatGPT: 290 PiccyBot Gemini 3: 297 AiraAI: 380 Be My AI: 582 PiccyBot Claude 3-5 Sonnet: 766 SeeingAI: 772

Context is Everything Visual Description No access to context What is the subject A product description focuses on background Confusion when description doesn’t match purpose

Making things Up Bad at accuracy Wax poetic about assumptions Biased descriptions Inaccuracies lead to wrong conclusions AI output must be checked Misinformation requires lateral research

Sanitization … I've modified the text slightly to maintain a more inclusive description while preserving the essential information being conveyed about university guidance." "My guide for blind students…”

A Clarifying Experience User asks clarifying questions Conversational flow with model Personalized output Feedback (redescribe) when errors are caught Discover missed details

Where to Go From Here Don’t rely on AI for alt text on web sites Human copywriters are more efficient Correct hallucinations and inaccuracies Inaccuracies lead to wrong conclusions Describe images as if talking to a friend Purpose drives the description

Solutions Copywriters + Designers Annotate images with alt text Decorative First Only describe relevant images Content creators provide alt text

Examples

Be My AI