A US study has shown that some artificial intelligence applications designed to assist blind and visually impaired individuals struggle to provide direct answers to their questions. Instead of focusing on essential information, these apps tend to describe image details at length, which may limit their effectiveness in daily situations.
Research Methodology and Participation
The science website TechXplore reported in an article published the day before yesterday that the study was conducted by researchers from the University of Maryland in College Park. It was published as part of the CHI 2026 conference, which specializes in human-computer interaction, under the title "Say It My Way: Studying the Ability of Blind Users to Control Image-Related Conversational AI Responses."
The study involved 11 blind individuals who tested the "Be My AI" feature in the "Be My Eyes" application. This feature uses AI to describe images and answer questions about the user's surroundings. Participants tested the app in a laboratory setting and then used it in their daily lives for ten days. Researchers analyzed 418 interactions to evaluate the suitability of the responses to their needs.
Gap in Answer Accuracy
The results showed that participants' questions typically ranged between 10 and 20 words, while the app's answers exceeded the length of their questions by more than ten times. The average conversation lasted three rounds of questions and answers, while some conversations extended to 21 rounds. This sometimes forced users to rephrase their questions or provide additional information to obtain useful answers.
The study demonstrated that these difficulties become apparent when users need specific information. For instance, when one participant asked about the ripeness of food, the app focused on describing its appearance without providing clear practical guidance. In another instance, a user received a lengthy description of an image instead of having the location of their keys identified, which they were searching for.
Technical Challenges and Proposed Solutions
To overcome these issues, participants resorted to clarifying the intent behind their questions and breaking down complex tasks into short, specific questions. However, researchers noted that the need to exert this extra effort may reduce the ease of use of applications that are supposed to help people with visual impairments access information quickly and independently.
The study also identified difficulties in estimating distances and tracking changes over time, as well as limited guidance to help users position the camera and capture images in a way that allows the app to understand the scene accurately.
Researchers emphasized the importance of developing AI systems better capable of understanding user intent and providing concise, clear answers that align with their needs, rather than merely describing image contents. This would contribute to improving the performance of assistive technologies and enhancing the independence of blind and visually impaired individuals.
Source: Syrian Arab News Agency (SANA)
Comments (0)