Creation, Critique, and Consumption: Exploring Generative AI Descriptions for Supporting Blind and Low Vision Professionals with Visual Tasks

Basic Information

• Paper ID: 2510.08991
• Title: Creation, Critique, and Consumption: Exploring Generative AI Descriptions for Supporting Blind and Low Vision Professionals with Visual Tasks
• Authors: Lucy Jiang, Lotus Zhang, Leah Findlater (University of Washington)
• Classification: cs.HC (Human-Computer Interaction)
• Publication Time/Venue: ASSETS '25 Workshop: AT @ Work, Virtual 2025
• Paper Link: https://arxiv.org/abs/2510.08991

Abstract

Many blind and low vision (BLV) individuals are excluded from professional roles that may involve visual tasks due to accessibility barriers and persistent bias. Advanced generative AI systems can support BLV individuals by providing contextualized and personalized visual descriptions for creation, critique, and consumption. In this workshop paper, the authors provide design recommendations for better delivering contextualized visual descriptions across multiple professional tasks and discuss how these designs can improve autonomy, inclusivity, and skill development over time.

Research Background and Motivation

Problem Context

1. Severe Employment Gap: The employment rate of people with disabilities is approximately one-third that of non-disabled individuals, with BLV populations facing particularly acute employment barriers
2. Visual Tasks as Workplace Barriers: Modern workplaces involve numerous visually-communicative tasks (such as creating presentations, formatting documents, taking photographs, watching training videos, etc.) that pose significant obstacles for BLV professionals
3. Limitations of Traditional Assistive Technology: Existing accessibility solutions are primarily limited to providing basic visual information access rather than enabling full workplace participation

Research Motivation

• Rapid advances in generative AI technology create new opportunities for providing contextualized, personalized visual descriptions
• Need to move beyond basic information access to support comprehensive participation of BLV professionals in visual communication tasks
• Technological innovation can break down employment barriers and enhance workplace inclusivity for BLV populations

Core Contributions

1. Proposes a Design Framework for Specialized Visual Description Systems: Providing contextualized and personalized AI description services tailored to different professional scenarios
2. Constructs Two Concrete Application Scenarios: Video production for independent content creators and marketing material creation for large advertising agencies
3. Provides Systematic Design Recommendations: Covering visual task support across three dimensions—creation, critique, and consumption
4. Articulates Long-term Impact Mechanisms: Analyzing how these designs can improve autonomy, inclusivity, and skill development for BLV professionals

Methodology Details

Task Definition

This research focuses on designing generative AI visual description systems to support BLV professionals, encompassing three core task dimensions:

• Creation: Assisting BLV individuals in creating visual content
• Critique: Supporting evaluation and feedback on visual work
• Consumption: Facilitating understanding and processing of visual information

Design Framework

Scenario One: Video Production for Independent Content Creators

Core Needs Analysis:

• Difficulty identifying visual trends
• Challenges in shot composition and subject positioning
• Need for visual effect verification in post-production editing

AI Description System Design:

1. Trend Identification Support: Describing common visual accompaniments to popular audio tracks (gestures, on-screen text, etc.)
2. Shooting Process Assistance:
   • Ensuring optimal positioning of shooting subjects within the frame
   • Providing detailed content descriptions to assist artistic composition
3. Editing Process Enhancement:
   • Describing video color temperature
   • Assessing accuracy of filters and special effects
   • Providing artistic information beyond content editing

Scenario Two: Marketing Material Creation for Large Advertising Agencies

Core Challenges:

• Complexity of collaborative workflows
• Multi-format content production requirements
• Rapid iteration and real-time collaboration demands
• Strict brand guideline compliance

AI Description System Design:

1. Brand Consistency Support:
   • Precise brand guideline descriptions
   • Accurate color descriptions ensuring brand representation
2. Team Collaboration Enhancement:
   • High-level overview descriptions (overall visual appearance)
   • Object-level descriptions (e.g., sticky note groupings)
   • Collaborator cursor position tracking (as visual focus proxy)

Technical Innovation Points

1. Context-Aware Descriptions: Customizing description content and detail levels based on specific professional task requirements
2. Multi-Level Information Architecture: Providing hierarchical visual information from macro to micro perspectives
3. Real-Time Collaboration Support: Integrating dynamic visual feedback into team workflows
4. Personalized Adaptation: Adjusting description strategies based on user roles and task types

Experimental Setup

Note: This is a workshop paper that primarily provides design recommendations and conceptual frameworks, without traditional experimental setup and results.

Theoretical Foundation

• Analysis based on existing literature regarding challenges faced by BLV content creators
• Reference to existing research on visual editing assistance systems (e.g., Huh et al.'s text-video editing system)
• Integration of related work on digital graphic creation accessibility

Design Validation Methods

• Verifying problem prevalence through literature review
• Analyzing design requirements based on limitations of existing systems
• Drawing design inspiration from successful cases in related fields

Related Work

Visual Content Creation Assistance Technology

1. Chang et al.'s EditScribe: Using natural language verification loops to support non-visual image editing for BLV individuals
2. Huh et al.'s AVScript: Text-based video editing system integrating visual descriptions and speech
3. Zhang et al.'s A11yboard: Research on digital whiteboard accessibility

Digital Content Engagement by BLV Populations

1. Social Media Platform Engagement: BLV creators' daily life sharing and creative economy participation on video platforms
2. Accessibility Barrier Research: Difficulties in creating visually appealing content, filter function verification issues, trend-tracking challenges

Mixed-Ability Collaboration

1. Real-Time Collaboration Tools: Improvements in mixed-ability collaboration for text editors and presentation software
2. Collaborative Environment Accessibility: Accessibility of visually-oriented collaborative activities (wireframing, whiteboard discussions)

Conclusions and Discussion

Main Conclusions

1. Redefining Visual Literacy: BLV individuals possess deep visual understanding capabilities; technology should support and enhance rather than assume deficiency
2. Systematic Improvement of Workplace Inclusivity: Technological innovation can progressively reduce bias and improve autonomy, inclusivity, and skill development for BLV populations
3. Importance of Personalized Descriptions: Different professional scenarios require customized visual description strategies

Long-Term Impact Mechanisms

Drawing on Georgina Kleege's observation: "On average, a completely blind person born blind understands what vision means far better than the average sighted person understands what blindness means."

Expected Effects:

• Enhanced Autonomy: Reducing dependence on others' assistance
• Improved Inclusivity: Promoting more inclusive design practices and workplace culture
• Skill Development: Supporting BLV professionals in demonstrating creative capabilities

In-Depth Evaluation

Strengths

1. Strong Problem Orientation: Directly addresses core barriers to BLV workplace participation
2. Innovative Design Approach: Proposes the concept of contextualized, personalized AI description systems
3. High Practical Value: Provides concrete, actionable design recommendations
4. Solid Theoretical Foundation: Comprehensive literature references with thorough argumentation
5. Significant Social Value: Addresses workplace equality rights for marginalized populations

Limitations

1. Lack of Empirical Validation: As a conceptual paper, it lacks user research and system evaluation
2. Insufficient Technical Implementation Details: Limited description of specific technical architecture of AI systems
3. Missing Scalability Analysis: Insufficient discussion of design recommendation applicability to other professional scenarios
4. Absent Cost-Benefit Analysis: Does not consider practical costs of system development and deployment

Impact

1. Academic Contribution: Provides new design perspectives for accessibility technology research
2. Practical Guidance: Offers specific design guidance for relevant technology developers
3. Policy Implications: May influence workplace accessibility policy development
4. Social Value: Promotes societal reassessment of BLV individuals' professional capabilities

Applicable Scenarios

1. Content Creation Industries: Video production, graphic design, marketing creativity, and related fields
2. Collaborative Work Environments: Team work scenarios requiring real-time visual collaboration
3. Education and Training: Visual skills training and professional development support
4. Technology Development: AI-assisted tools and accessible technology product development

Future Research Directions

1. User Research: In-depth understanding of specific needs of BLV professionals across different occupations
2. Technical Implementation: Developing prototype systems and verifying technical feasibility
3. Effectiveness Evaluation: Designing evaluation metrics to verify system impact on user work efficiency and satisfaction
4. Cross-Domain Extension: Exploring applicability of design principles to other professional fields
5. Ethical Considerations: Investigating potential bias and privacy issues arising from AI description systems

Summary: This paper proposes an important and forward-looking research direction, providing better workplace support for BLV professionals through generative AI technology. While lacking empirical validation as a conceptual study, its design approach and social value merit further in-depth research and practical application exploration.