Source-linked AI summary
More Than Just Access: Generative AI as Communication Intermediary for Blind and Low-Vision Users
Protik Dey, Mohd Saifuzzaman, Taslima Akter
TL;DR
BLV users increasingly rely on GenAI to access information and replace some requests for interpersonal help, but they may be unable to verify its responses. Through interviews with 19 BLV participants, the paper examines this substitution and finds that its benefits depend on reliable, accessible, and safe exchanges, while users reserve consequential tasks for people.
Problem
Trust research often assumes users can visually inspect and independently verify automated outputs, leaving case-by-case GenAI substitution for BLV users and its effects on judgment and relationships underexamined.
Source: §1 · p. 1Method
Results
GenAI can extend independence and ease everyday requests when exchanges are reliable, reachable, and safe to disclose to, while participants reserved high-stakes tasks for people.
Source: §1 · p. 1, §4.2 · p. 3Takeaways & Limitations
The findings support designing GenAI explicitly as a communication partner and supporting BLV users’ independent judgment rather than treating the systems as general-purpose tools.
Source: §7 · p. 4Using GenAI for low-stakes tasks and returning to people for consequential ones puts the burden of identifying failures on BLV users.
Source: §5 · p. 4
Abstract
from arXiv · showhide
Generative AI (GenAI) tools are increasingly woven into how blind and low-vision (BLV) people communicate, not only with digital information, but with the physical world and with other people. Tools such as ChatGPT, Google Gemini, Be My AI, and Seeing AI translate visual and textual content into accessible form, and are beginning to substitute for interpersonal requests for help, such as asking a family member to read a label or describe a scene. Drawing on semi-structured interviews with 19 BLV participants, we examine GenAI as a communication intermediary and how it succeeds and fails as an alternative for reading, describing, and even asking another person for help. We also investigated what BLV users gain and risk when these tools take over that role. We conclude with design and policy implications for GenAI systems that communicate uncertainty honestly, protect information, and support BLV users' independence rather than substitute for it unsafely.
1 Introduction
For BLV users, GenAI increasingly mediates communication with information, surroundings, and other people, but its probabilistic answers can be difficult to verify. This paper examines what BLV users gain and risk when GenAI substitutes for interpersonal help.
1 Introduction
- GenAI increasingly mediates BLV communication, including tasks that previously required asking another person for help.Its outputs can be flexible and context-sensitive, but also unpredictable and sometimes incorrect.Source: §1 · p. 1
- Because BLV users often cannot independently verify responses, GenAI’s unpredictability becomes a communication problem, not just a usability flaw.Their information seeking and independent judgment can depend on non-visual cues that a system may or may not provide.Source: §1 · p. 1
- Replacing a person with GenAI changes whom users trust, what they disclose, and who is responsible if an exchange goes wrong.It can reduce the social cost of routine requests while removing some emotional connection and accountability.Source: §1 · p. 1
- The paper investigates when BLV users rely on GenAI instead of a person and what that substitution means for independent judgment and relationships.The authors conducted semi-structured interviews with 19 BLV users; they report that GenAI can extend independence and ease everyday requests when exchanges are reliable, reachable, and safe for disclosure.Source: §1 · p. 1
2 Related Work
Prior work examines trust in automated systems and BLV users’ accessibility needs, but often assumes users can visually inspect and verify outputs. Research has examined GenAI replacing interpersonal help, while leaving case-by-case reliance and its relational trade-offs underexamined.
2 Related Work
- Trust research identifies reliability, transparency, accountability, and privacy as central considerations, but commonly assumes users can visually inspect outputs and verify system behavior.That assumption does not hold for BLV users communicating through GenAI.Source: §2 · pp. 1–2
- BLV users use GenAI for reading and image understanding, but inaccessible interfaces and fluent errors can obstruct use and erode trust.Prior work notes problems when users cannot ask a system to look again or learn what it examined.Source: §2 · p. 2
- Prior work has examined GenAI replacing requests to people and users’ disclosure decisions, but case-by-case reliance and losses in accountability, sincerity, and disclosure remain underexamined.Existing findings describe users reserving high-stakes tasks for trusted people and using GenAI for smaller, frequent requests.Source: §2 · p. 2
3 Methodology
The study uses semi-structured interviews with 19 US-based BLV adults who use screen readers and have GenAI experience. The researchers explored communicative use, trust, and recommendations, then analyzed interview data through reflexive thematic analysis.
3 Methodology
- The researchers selected 19 participants from 33 respondents based on GenAI experience and frequency of use.Eligibility required US residence, age 18 or older, identifying as blind or low-vision, primarily using a screen reader, speaking English, and experience with GenAI; recruitment was mainly through the National Federation of the Blind.Source: §3 · p. 2
- Participants had experience with general-purpose GenAI and assistive tools, and all accessed them using screen readers.All used Be My AI or Seeing AI; several also used Aira, and participants included moderate-to-frequent users as well as one comparatively less experienced user.Source: §3 · p. 2
- Interviews asked about communicative tasks, usage patterns, trust, and how accessibility, privacy, and reliability shaped reliance on GenAI over people.Participants also discussed recommendations for making systems more accessible and communicating more confidently with BLV users.Source: §3 · p. 2
- Two authors used reflexive thematic analysis, jointly refining codes and interpretations through weekly meetings rather than calculating inter-rater reliability.After thematic saturation, they divided the remaining interviews and continued reviewing new codes together.Source: §3 · p. 2
4 Findings
Participants treated GenAI as a communication channel whose trustworthiness depended on how it worked, whether they could access it, and what they had to give up to use it. Their choices varied with task stakes, verification options, accessibility, and disclosure concerns.
4 Findings
- Participants viewed GenAI as a channel to the visual world, digital content, and other people, whose trustworthiness depended on performance, reachability, and trade-offs.They used calibrated skepticism rather than full acceptance when they could not independently verify visual or textual translations.Source: §4 · pp. 2–3
4.1 Reading the World Aloud: GenAI as a Channel to Visual and Textual Information
- For reading and describing visual or textual content, participants could not usually verify GenAI’s translations independently and relied on their own judgment to assess possible errors.Verification was easiest when they already had relevant background knowledge; the channel was least trustworthy when they lacked another way to cross-check it.Source: §4.1 · p. 3
4.2 From Asking a Person to Asking a Machine
- GenAI reduced routine requests to friends, family, and bystanders, but participants preferred people for high-stakes financial, medical, and navigation-critical tasks.They weighed independence and reduced social burden against privacy concerns, sincerity, and human accountability; some treated GenAI output as a starting point rather than a final answer.Source: §4.2 · p. 3
4.3 No Accessibility, No Communication
- Inaccessible interfaces blocked communication before tool performance or privacy could matter, though participants said accessibility had improved and remaining barriers seemed temporary.Unlabeled fields, unclear buttons, or broken copying could prevent users from completing an exchange.Source: §4.3 · p. 3
4.4 What the Machine Knows to Help
- Participants consciously weighed sharing documents or surroundings against convenience and independence, while differing in how much corporate reputation and personalization shaped their privacy concerns.They valued tools recognizing their BLV needs but also worried about how much information systems retained over time.Source: §4.4 · p. 3
5 The Double-Edged Sword: Gains and Risk When GenAI Speaks for BLV Users
GenAI can reduce routine interpersonal requests and support privacy and independence, while introducing risks around confident errors and disclosure without full awareness. Participants balanced these risks by reserving consequential tasks for people, but that strategy leaves users responsible for anticipating when an exchange is high-stakes.
5 The Double-Edged Sword: Gains and Risk When GenAI Speaks for BLV Users
- GenAI absorbed smaller, frequent requests, extending existing relationships while supporting privacy and day-to-day independence.Participants still reserved requests that genuinely needed human assistance for people.Source: §5 · pp. 3–4
- Confident GenAI errors can be harder to trace and correct than a person's mistakes, especially in financial or medical exchanges.Participants managed this risk by using GenAI for low-stakes communication and returning to a person for consequential tasks.Source: §5 · p. 4
- Relying on users to identify failure assumes they can predict which exchanges will become high-stakes.This strategy places the burden of identifying failure on the BLV user rather than the system.Source: §5 · p. 4
- Participants sometimes disclosed images, documents, or speech without complete information about the consequences of sharing them.The cited concern about awareness spanned both younger and older users.Source: §5 · p. 4
6 Design Implications and Policy Recommendations
The recommendations call for GenAI systems to make failure, uncertainty, accessibility, and data practices legible during communication. They also call for task-sensitive modes and cross-platform protections so BLV users are not left to manage communication risks alone.
6 Design Implications and Policy Recommendations
- Systems should explicitly signal processing, failure, and low confidence through non-visual cues so users do not mistake unflagged errors for trustworthy responses.The passage frames uncertainty and failure as communicative events and calls for the cues a human would provide.Source: §6 · p. 4
- Accessibility should be built into GenAI system development because an inaccessible interface can shut down the communicative exchange.The recommendation treats accessibility as a persistent, system-wide state rather than an afterthought.Source: §6 · p. 4
- Systems should offer more conservative, uncertainty-forward modes for high-stakes tasks and more exploratory modes for casual use.Participants trusted tools differently when they replaced low-stakes convenience versus high-stakes tasks; trust earned in casual exchanges may be overgeneralized to consequential ones, with risks unevenly distributed.Source: §6 · p. 4
- At the point of disclosure, systems should explain accessibly what happens to images, documents, and recordings, including retention, reuse, and third-party sharing.The explanation should come before disclosure, be screen-reader accessible, and not assume baseline digital literacy.Source: §6 · p. 4
- Standardized, enforceable protections for data handling and accessibility should apply across GenAI products, shifting risk-management burdens away from individual BLV users.Ownership and corporate reputation function as imperfect, unevenly applied proxies for trust, and some users set concerns aside when a tool is their only accessible option.Source: §6 · p. 4
7 Conclusion
The conclusion presents GenAI as a communication intermediary that BLV users judge rather than adopt automatically. It calls for communication-partner design that supports independence without leaving users exposed to inaccessible, opaque, or silent failures.
7 Conclusion
- BLV users make independent judgments about relying on GenAI, weighing task stakes, reliability, accessibility, and disclosure in each exchange.They distinguish low-stakes tasks from high-stakes ones and decide when a human is still needed.Source: §7 · p. 4
- The findings underscore the limits of treating GenAI as a general-purpose tool as its role in daily communication grows.The conclusion links rising costs of poor accessibility, opaque disclosure, and silent failure to that expanding role.Source: §7 · p. 4
- The conclusion frames BLV communication support as a test of whether AI systems can earn a place in relationships previously served by people.It argues for designing GenAI explicitly as a communication partner.Source: §7 · p. 4