Computer vision systems are increasingly used by blind individuals to navigate their lives, helping, for example, locate objects such as doors or chairs. Yet these recognition systems do not work for many personal objects a blind user might want to find, such as keys or a special notebook. In response, efforts created personalized recognition systems, where individuals train their phones to identify and locate things, like a coffee mug or white cane, using example images/videos. However, these tools are trained on data from high-resource contexts, not necessarily reflecting India’s material culture. This paper discusses the contribution of the ORBIT-India dataset, which extends these tools to the Indian context, home of the world’s largest blind population. The ORBIT-India dataset comprises 105,243 images from 587 videos, representing 76 unique objects. We use this experience to examine dataset collection practices translated from high- to low-resource settings, providing recommendations to support cross-geography dataset collection.

Media representations of people with disabilities profoundly influence societal perceptions, yet have historically been absent, stereotyped, or inaccurate. As AI-generated visual media becomes increasingly prevalent, there is a critical opportunity to address these misrepresentations. Responding to the lack of collectively negotiated representation standards, this paper presents our human-centric approach to engaging disability communities meaningfully in AI data practices. Over three months, we worked closely with three disability organizations across the Global North and South to develop the Community Library Creator that introduces design scaffolds to support communities in defining ‘good’ representation and curating community-centric AI datasets; laying the foundations for community-specific evaluation metrics and future model adaptations. We contribute qualitative insights into the complexities of community-led data curation; discuss the value and practical challenges of intersecting human insights with AI requirements; and reflect on human-centered AI approaches that empower communities to share their perspectives and actively shape AI data practices.

Current AI sound awareness systems can provide deaf and hard of hearing people with information about sounds, including discrete sound sources and transcriptions. However, synthesizing AI outputs based on DHH people's ever-changing intents in complex auditory environments remains a challenge. In this paper, we describe the co-design process of SoundWeaver, a sound awareness system prototype that dynamically weaves AI outputs from different AI models based on users’ intents and presents synthesized information through a heads-up display. Adopting a Research through Design perspective, we created SoundWeaver with one DHH co-designer, adapting it to his personal contexts and goals (e.g., cooking at home and chatting in a game store). Through this process, we present design implications for the future of “intent-driven” AI systems for sound accessibility.

Assistive technologies, such as smartphone-based object-recognition (OR) apps, provide visual assistance to people who are blind or lowvision to enable increased independent participation in society.While previous research has explored the functional accessibility of object-recognition technologies, little attention has been given to their social accessibility, particularly in interdependent sociocultural contexts of the Global South.Through a mixed-methods approach, employing a seven-day diary study followed by one-onone interviews with seven OR app users in India, we explore their experiences in depth.Our findings highlight the nuances of what interdependence looks like in a multicultural, Indian society, as people navigate public and private spheres with a camera-based assistive technology designed for independent, western contexts.We argue for the necessity to design assistive technologies following the interdependence framework that accommodates the social and cultural context of the Global South.Additionally, we propose design guidelines for assistive technologies in community-oriented societies, emphasizing community-centered approaches, cultural alignment, and locally adaptable designs.

Initiating joint attention (JA) is a fundamental first step in social interactions. In sighted individuals, it relies predominantly on visual cues, such as gaze and hand gestures. These features can reduce opportunities for blind and visually impaired (BVI) and sighted people to interact. Understanding the strategies to navigate these challenges is necessary to develop technology that can facilitate case study of five children with mixed visual abilities engaging in activities rich with JA opportunities. In a teacher-led classroom, the children experimented with the use of an AI-powered headset designed to support BVI people in social situations. Interaction analysis established that situational complexity affects the children’s responses to initiation attempts. Furthermore, the headset adds to this complexity, affecting the frequency and reactions to attempts to initiate JA. The findings informed the creation of a JA initiation framework and suggestions for future design.

Nasogastric tubes (NGTs) are feeding tubes that are inserted through the nose into the stomach to deliver nutrition or medication. If not placed correctly, they can cause serious harm, even death to patients. Recent AI developments demonstrate the feasibility of robustly detecting NGT placement from Chest X-ray images to reduce risks of sub-optimally or critically placed NGTs being missed or delayed in their detection, but gaps remain in clinical practice integration. In this study, we present a human-centered approach to the problem and describe insights derived following contextual inquiry and in-depth interviews with 15 clinical stakeholders. The interviews helped understand challenges in existing workflows, and how best to align technical capabilities with user needs and expectations. We discovered the tradeoffs and complexities that need consideration when choosing suitable workflow stages, target users, and design configurations for different AI proposals. We explored how to balance AI benefits and risks for healthcare staff and patients within broader organizational, technical, and medical-legal constraints. We also identified data issues related to edge cases and data biases that affect model training and evaluation; how data documentation practices influence data preparation and labeling; and how to measure relevant AI outcomes reliably in future evaluations. We discuss how our work informs design and development of AI applications that are clinically useful, ethical, and acceptable in real-world healthcare services.

Large multi-modal models (LMMs) hold the potential to usher in a new era of automated visual assistance for people who are blind or low vision (BLV). Yet, these models have not been systematically evaluated on data captured by BLV users. We address this by empirically assessing CLIP, a widely-used LMM likely to underpin many assistive technologies. Testing 25 CLIP variants in a zero-shot classification task, we find that their accuracy is 15 percentage points lower on average for images captured by BLV users than web-crawled images. This disparity stems from CLIP's sensitivities to 1) image content (e.g. not recognizing disability objects as well as other objects); 2) image quality (e.g. not being robust to lighting variation); and 3) text content (e.g. not recognizing objects described by tactile adjectives as well as visual ones). We delve deeper with a textual analysis of three common pre-training datasets: LAION-400M, LAION-2B and DataComp-1B, showing that disability con-tent is rarely mentioned. We then provide three examples that illustrate how the performance disparities extend to three downstream models underpinned by CLIP: OWL-ViT, CLIPSeg and DALL-E2. We find that few-shot learning with as few as 5 images can mitigate CLIP's quality-of-service disparities for BLV users in some scenarios, which we discuss alongside a set of other possible mitigations.

Understanding the mechanisms of information storage and transfer in Transformer-based models is important for driving model understanding progress. Recent work has studied these mechanisms for Large Language Models (LLMs), revealing insights on how information is stored in a model's parameters and how information flows to and from these parameters in response to specific prompts. However, these studies have not yet been extended to Multi-modal Large Language Models (MLLMs). Given their expanding capabilities and real-world use, we start by studying one aspect of these models -- how MLLMs process information in a factual visual question answering task. We use a constraint-based formulation which views a visual question as having a set of visual or textual constraints that the model's generated answer must satisfy to be correct (e.g. What movie directed by \emph{the director in this photo} has won a \emph{Golden Globe}?). Under this setting, we contribute i) a method that extends causal information tracing from pure language to the multi-modal setting, and ii) \emph{VQA-Constraints}, a test-bed of 9.7K visual questions annotated with constraints. We use these tools to study two open-source MLLMs, LLaVa and multi-modal Phi-2. Our key findings show that these MLLMs rely on MLP and self-attention blocks in much earlier layers for information storage, compared to LLMs whose mid-layer MLPs are more important. We also show that a consistent small subset of visual tokens output by the vision encoder are responsible for transferring information from the image to these causal blocks. We validate these mechanisms by introducing MultEdit a model-editing algorithm that can correct errors and insert new long-tailed information into MLLMs by targeting these causal blocks. We will publicly release our dataset and code.

The opportunity for artificial intelligence, or AI, to enable accessibility is rapidly growing, but widely impactful applications can be challenging to build given the diversity of user need within and across disability communities. Teachable AI systems give users with disabilities a way to leverage the power of AI to personalize applications for their own specific needs, as long as the effort of providing examples is balanced with the benefit of the personalization received. As an example, this paper presents the design and evaluation of Find My Things, an end-to-end application that can be taught by people who are blind or low vision to find their personal things. Through synthesis of the design process, this paper offers design considerations for the teaching loop that is so critical to realizing the power of teachable AI for accessibility.

Recent advances in AI and machine learning (ML) promise significant transformations in the future delivery of healthcare. Despite a surge in research and development, few works have moved beyond demonstrations of technical feasibility and algorithmic performance. However, to realize many of the ambitious visions for how AI can contribute to clinical impact requires the closer design and study of AI tools or interventions within specific health and care contexts. This article outlines our collaborative, human-centered approach to developing an AI application that predicts treatment outcomes for patients who are receiving human-supported, internet-delivered Cognitive Behavioral Therapy (iCBT) for symptoms of depression and anxiety. Intersecting the fields of HCI, AI, and healthcare, we describe how we addressed the specific challenges of (1) identifying clinically relevant AI applications ; and (2) designing AI applications for sensitive use contexts like mental health. Aiming to better assist the work practices of iCBT supporters, we share how learnings from an interview study with 15 iCBT supporters surfaced their practices and information needs and revealed new opportunities for the use of AI. Combined with insights from the clinical literature and technical feasibility constraints, this led to the development of two clinical outcome prediction models. To clarify their potential utility for use in practice, we conducted 13 design sessions with iCBT supporters that utilized interface mock-ups to concretize the AI output and derive additional design requirements. Our findings demonstrate how design choices can impact interpretations of the AI predictions as well as supporter motivation and sense of agency. We detail how this analysis and the design principles derived from it enabled the integration of the prediction models into a production interface. Reporting on identified risks of over-reliance on AI outputs and needs for balanced information assessment and preservation of a focus on individualized care, we discuss and reflect on what constitutes a responsible, human-centered approach to AI design in this healthcare context.

Novel AI-infused educational technologies can give children with blindness the opportunity to explore concepts learned incidentally through vision by using alternative perceptual modalities. However, more effort is needed to support the meaningful use of such technological innovations for evaluations at scale and later wide-spread adoption. This paper presents the development and pilot evaluation of a curriculum to enable educators to support blind learners’ self-exploration of social attention using the PeopleLens technology. We reflect on these learnings to present four design guidelines for creating curricula aimed to enable meaningful use. We then consider how formulations of “success” by our participants can help us think about ways of assessing efficacy in low-incidence disability groups. We conclude by arguing for our community to widen the scope of discourse around assistive technologies from design and engineering to include supporting their meaningful use.

Artificial Intelligence (AI) for accessibility is a rapidly growing area, requiring datasets that are inclusive of the disabled users that assistive technology aims to serve. We offer insights from a multi-disciplinary project that constructed a dataset for teachable object recognition with people who are blind or low vision. Teachable object recognition enables users to teach a model objects that are of interest to them, e.g., their white cane or own sunglasses, by providing example images or videos of objects. In this paper, we make the following contributions: 1) a disability-first procedure to support blind and low vision data collectors to produce good quality data, using video rather than images; 2) a validation and evolution of this procedure through a series of data collection phases and 3) a set of questions to orient researchers involved in creating datasets toward reflecting on the needs of their participant community.

AI technologies are often used to aid people in performing discrete tasks with well-defined goals (e.g., recognising faces in images). Emerging technologies that provide continuous, real-time information enable more open-ended AI experiences. In partnership with a blind child, we explore the challenges and opportunities of designing human-AI interaction for a system intended to support social sensemaking. Adopting a research-through-design perspective, we reflect upon working with the uncertain capabilities of AI systems in the design of this experience. We contribute: (i) a concrete example of an open-ended AI system that enabled a blind child to extend his own capabilities; (ii) an illustration of the delta between imagined and actual use, highlighting how capabilities derive from the human-AI interaction and not the AI system alone; and (iii) a discussion of design choices to craft an ongoing human-AI interaction that addresses the challenge of uncertain outputs of AI systems.

Object recognition has made great advances in the last decade, but predominately still relies on many high-quality training examples per object category. In contrast, learning new objects from only a few examples could enable many impactful applications from robotics to user personalization. Most few-shot learning research, however, has been driven by benchmark datasets that lack the high variation that these applications will face when deployed in the real-world. To close this gap, we present the ORBIT dataset and benchmark, grounded in the real-world application of teachable object recognizers for people who are blind/low-vision. The dataset contains 3,822 videos of 486 objects recorded by people who are blind/low-vision on their mobile phones. The benchmark reflects a realistic, highly challenging recognition problem, providing a rich playground to drive research in robustness to few-shot, high-variation conditions. We set the benchmark’s first state-of-the-art and show there is massive scope for further innovation, holding the potential to impact a broad range of real-world vision applications including tools for the blind/low-vision community. We release the dataset at https://doi.org/10.25383/city.14294597 and benchmark code at https://github.com/microsoft/ORBIT-Dataset.

There is a dearth of appropriate tools for young learners with mixed visual abilities to engage with computational learning. Addressing this gap, Torino is a physical programming language for teaching computational learning to children ages 7–11 regardless of level of vision. To create code, children connect physical instruction pods and tune their parameter dials to create music, audio stories, or poetry. Currently, the uptake of novel educational technologies to support inclusive education of children with disabilities continues to be limited at scale. We consider how the Torino Learning Environment supports non-specialist teachers to teach computational learning to children with mixed visual abilities in a UK-wide evaluation with 75 children and 30 teachers over a period of three months. We demonstrate how children can successfully learn with a novel physical programming language. We articulate how key design constructs such as persistent program overview and liveness supported non-specialist teachers to co-produce learning for children of different ages, visual and cognitive abilities. We conclude with reflective guidance on evaluating inclusive educational technologies at scale.CCS CONCEPTSHuman-centered computing → Accessibility → Empirical studies in accessibility

Computer vision systems to help blind users are becoming increasingly common, yet often these systems are not intelligible. Our work investigates the intelligibility of a wearable computer vision system to help blind users locate and identify people in their vicinity. Providing a continuous stream of information, this system allows us to explore intelligibility through interaction and instructions, going beyond studies of intelligibility that focus on explaining a decision a computer vision system might make. In a study with 13 blind users, we explored whether varying instructions (either basic or enhanced) about how the system worked would change blind users' experience of the system. We found offering a more detailed set of instructions did not affect how successful users were using the system nor their perceived workload. We did, however, find evidence of significant differences in what they knew about the system and they employed different, and potentially more effective, use strategies. Our findings have important implications for researchers and designers of computer vision systems for blind users, as well as more general implications for understanding what it means to make interactive computer vision systems intelligible.

Across the world, policy initiatives are being developed to engage children with computer programming and computational thinking. Diversity and inclusion has been a strong force in this agenda, but children with disabilities have largely been omitted from the conversation. Currently, there are no age appropriate tools for teaching programming concepts and computational thinking to primary school children with visual disabilities. We address this gap through presenting the design and implementation of Torino, a tangible programming language for teaching programming concepts to children age 7–11 regardless of level of vision. In this paper, we: (1) describe the design process done in conjunction with children with visual disabilities; (2) articulate the design decisions made; and (3) report insights generated from an evaluation with 10 children with mixed visual abilities that considers how children are able to trace (read) and create (write) programs with Torino. We discuss key design trade-offs: (1) readability versus extensibility; and (2) size versus liveness. We conclude by reflecting upon how an inclusive design approach shaped the final result.

Mobile technologies are valuable tools for the self-report of mental health and wellbeing. These systems pose many unique design challenges which have received considerable attention within HCI, including the engagement of users. However, less attention has been paid to the use of personal devices in public health. Integrating self-reported data within the context of clinical care suggests the need to design interfaces to support data management, sense-making, risk-assessment, feedback and patient-provider relationships. This paper reports on a qualitative design study for the clinical interface of a mobile application for the self-report of psychological wellbeing and depression during pregnancy. We examine the design tensions which arise in managing the expectations and informational needs of pregnant women, midwives, clinical psychologists, GPs and other health professionals with respect to a broad spectrum of wellbeing. We discuss strategies for managing these tensions in the design of technologies required to balance personal information with public health.

Perinatal depression (PND) affects up to 15% of women within the United Kingdom and has a lasting impact on a woman's quality of life, birth outcomes and her child's development. Suicide is the leading cause of maternal mortality. However, it is estimated that at least 50% of PND cases go undiagnosed. This paper presents the results of the first feasibility study to examine the potential of mobile devices to engage women in antenatal mental health screening. Using a mobile application, 254 women attending 14 National Health Service midwifery clinics provided 2,280 momentary and retrospective reports of their wellbeing over a 9-month period. Women spoke positively of the experience, installing and engaging with this technology regardless of age, education, wellbeing, number of children, marital or employment status, or past diagnosis of depression. 39 women reported a risk of depression, self-harm or suicide; two-thirds of whom were not identified by screening in-clinic.

This research takes an orientation to visual impairment (VI) that does not regard it as fixed or determined alone in or through the body. Instead, we consider (dis)ability as produced through interactions with the environment and configured by the people and technology within it. Specifically, we explore how abilities become negotiated through video ethnography with six VI athletes and spectators during the Rio 2016 Paralympics. We use generated in-depth examples to identify how technology can be a meaningful part of ability negotiations, emphasizing how these embed into the social interactions and lives of people with VI. In contrast to treating technology as a solution to a 'sensory deficit', we understand it to support the triangulation process of sense-making through provision of appropriate additional information. Further, we suggest that technology should not try and replace human assistance, but instead enable people with VI to better identify and interact with other people in-situ.

Sophisticated ubiquitous sensing systems are being used to measure motor ability in clinical settings. Intended to augment clinical decision-making, the interpretability of the machine-learning measurements underneath becomes critical to their use. We explore how visualization can support the interpretability of machine-learning measures through the case of Assess MS, a system to support the clinical assessment of Multiple Sclerosis. A substantial design challenge is to make visible the algorithm's decision-making process in a way that allows clinicians to integrate the algorithm's result into their own decision process. To this end, we present a series of design iterations that probe the challenges in supporting interpretability in a real-world system. The key contribution of this article is to illustrate that simply making visible the algorithmic decision-making process is not helpful in supporting clinicians in their own decision-making process. It disregards that people and algorithms make decisions in different ways. Instead, we propose that visualisation can provide context to algorithmic decision-making, rendering observable a range of internal workings of the algorithm from data quality issues to the web of relationships generated in the machine-learning process.

We investigate how technology can support collaborative learning by children with mixed-visual abilities. Responding to a growing need for tools inclusive of children with vision impairments (VI) for the teaching of computer programing to novice learners, we explore Torino -- a physical programing language for teaching programing constructs and computational thinking to children age 7-11. We draw insights from 12 learning sessions with Torino that involved five pairs of children with vision ranging from blindness to full-sight. Our findings show how sense-making of the technology, collaboration, and learning were enabled through an interplay of system design, programing tasks and social interactions, and how this differed between the pairs. The paper contributes insights on the role of touch, audio and visual representations in designs inclusive of people with VI, and discusses the importance and opportunities provided through the 'social' in negotiations of accessibility, for learning, and for self-perceptions of ability and self-esteem.

There has been a surge in artificial intelligence (AI) technologies co-opted by or designed for people with visual disabilities. Researchers and engineers have pushed technical boundaries in areas such as computer vision, natural language processing, location inference, and wearable computing. But what do people with visual disabilities imagine as their own technological future? To explore this question, we developed and carried out tactile ideation workshops with participants in the UK and India. Our participants generated a large and diverse set of ideas, most focusing on ways to meet needs related to social interaction. In some cases, this was a matter of recognizing people. In other cases, they wanted to be able to participate in social situations without foregrounding their disability. It was striking that this finding was consistent across UK and India despite substantial cultural and infrastructural differences. In this paper, we describe a new technique for working with people with visual disabilities to imagine new technologies that are tuned to their needs and aspirations. Based on our experience with these workshops, we provide a set of social dimensions to consider in the design of new AI technologies: social participation, social navigation, social maintenance, and social independence. We offer these social dimensions as a starting point to forefront users' social needs and desires as a more deliberate consideration for assistive technology design.

Health and wellbeing applications increasingly raise ethical issues for design. User-centred and participatory design approaches, while grounded in everyday wisdom, cannot be expected to address ethical reflection consistently, as multiple value systems come into play. We explore the potential of phronesis, a concept from Aristotelian virtue ethics, for mHealth design. Phronesis describes wisdom and judgment garnered from practical experience of specific situations in context. Applied phronesis contributes everyday wisdom to challenging issues for vulnerable target users. Drawing on research into mHealth technologies for psychological wellbeing, we explore how phronesis can inform ethical design. Using a case study on an app for self-reporting symptoms of depression during pregnancy, we present a framework for incorporating a phronetic approach into design, involving: (a) a wide feedback net to capture phronetic input early in design; (b) observing the order of feedback, which directly affects value priorities in design; (c) ethical pluralism recognising different coexisting value systems; (d) acknowledging subjectivity in the disclosure and recognition of individual researcher and participant values. We offer insights into how a phronetic approach can contribute everyday wisdom to designing mHealth technologies to help designers foster the values that promote human flourishing.

A growing number of domains, including affect recognition and movement analysis, require a single, real number ground truth label capturing some property of a video clip. We term this the provision of continuum labels. Unfortunately, there is often an uncacceptable trade-off between label consistency and the efficiency of the labelling process with current tools. We present a novel interaction technique, setwise comparison, which leverages the intrinsic human capability for consistent relative judgements and the TrueSkill algorithm to solve this problem. We describe SorTable, a system demonstrating this technique. We conducted a real-world study where clinicians labelled videos of patients with multiple sclerosis for the ASSESS MS computer vision system. In assessing the efficiency-consistency trade-off of setwise versus pairwise comparison, we demonstrated that not only is setwise comparison more efficient, but it also elicits more consistent labels. We further consider how our findings relate to the interactive machine learning literature.

The use of depth-sensing computer vision to capture bodily movement is increasingly being exploited in healthcare. Yet, there are few descriptions of how real-world practices influence the design of such applications. To this end, we present the development and empirical evaluation of ASSESS MS, a system to support the clinical assessment of Multiple Sclerosis using Kinect. A key issue for developing machine-learning based systems is the need for standardized data on which statistical inferences can be made. We demonstrate that there are many aspects of clinical practice that are at odds with the need to capture standardized data for a computer vision system. We offer three design guidelines so address these: 1) Standardization is a multi-disciplinary issue and needs to be addressed early in the development process; 2) Tools that provide a view into what the camera “sees” can support the achievement of standardized data capture in real environments; 3) Tools to support standardized data capture should maintain the agency of human interaction. More broadly we show that when considering every day contexts, the traditional focus on measurement accuracy is only a small part of the effort needed to make a technology “work” in practice.

Computer visions (CV) systems are increasingly finding new roles in domains such as healthcare. These collaborative settings are a new challenge for CV systems, requiring the design of appropriate interaction paradigms. The provision of feedback, particularly of what the CV system can 'see,' is a key aspect, and may not always be possible to present visually. We explore the design space for audio feedback for a scenario of interest, the clinical assessment of Multiple Sclerosis using a CV system. We then present a mixed-methods experimental study aimed at providing some first insights into the challenges and opportunities of designing audio feedback of this kind. Specifically, we compare audio feedback that differentiates which body parts the CV system can see to audio feedback that is undifferentiated. The findings reveal that it is not enough to simply convey that something might be out of view of the camera as what the camera can 'see' depends on the specific configuration of participants and the peculiarities of the skeleton inference algorithms. The results highlight the importance of providing feedback which more naturally conveys spatial information in developing CV systems for collaborative use.