AI: Building With, Not For, Disabled Individuals

Most existing technology for accessibility is developed *for* individuals with disabilities rather than *with* them. This distinction is crucial, as tools built solely for compliance often fail to function effectively in real-world scenarios. True innovation stems from a deep understanding of users' daily challenges, moving beyond mere technical standards to solve practical problems.

Compliance with minimum accessibility standards does not guarantee practical usability or meet the complex needs of diverse users. Without direct input from the target demographic, tools often miss critical nuances that impact their efficacy. Designing *with* individuals fosters solutions that genuinely integrate into their lives and address their specific requirements.

This collaborative approach transforms the development process, prioritizing user experience over abstract compliance metrics. It shifts the focus from a top-down model to one that is inclusive and responsive to the lived experiences of those who will use the technology. The goal is to create tools that are not only technically sound but also genuinely empowering and functional for their intended users.

When developers build with users, they gain invaluable insights into the actual daily challenges and strengths of individuals with disabilities. This approach leads to more robust, user-centered designs that enhance self-determination and real-world applicability.

Dr. Lisa Dieker's personal journey into inclusive design was deeply influenced by her son, who was born with Tourette's, dyslexia, and dysgraphia. Despite these conditions, he demonstrated remarkable abilities, becoming the 20th best gymnast in the United States.

His experience highlighted that the challenge was not his ability to succeed, but rather the availability of adequate support systems. This realization underscored the importance of providing tailored assistance to help individuals with unique abilities achieve their full potential, shifting the focus from inherent limitations to enabling environments.

Self-determination, especially in technology design for individuals with disabilities, fundamentally begins with giving them a voice. Dr. Dieker emphasizes the critical importance of actively asking people with disabilities for their opinions on new tools and engaging them in focus groups.

This direct user engagement is not only beneficial for the target demographic but also provides valuable insights that can improve tools for all users, including those who are not tech-savvy. Incorporating diverse perspectives from the outset ensures that tools are intuitive, functional, and user-friendly across a broad spectrum of abilities.

Overlooking this crucial step in the design process often leads to solutions that fail to address genuine needs, resulting in tools that are technically compliant but practically unusable. True self-determination is fostered when individuals have a direct say in the development of technologies intended to support them.

Referencing experts Jamie Basham and Dave Edy, Dr. Dieker conceptualizes AI as a "cognitive prosthetic." This analogy suggests that AI agents should function similarly to physical aids, like a wheelchair, by filling in functional gaps for students.

The primary goal of AI in this context is to support existing capabilities and compensate for areas where individuals may struggle, without supplanting the fundamental learning process itself. AI should enhance, not replace, core human functions, enabling greater independence and participation.

Current technology often satisfies minimum accessibility standards but still falls short for individuals with complex needs, such as those with multiple disabilities or visual impairments. Accessibility requirements are diverse, encompassing both visible and invisible conditions like Tourette's, which are often overlooked in standardized designs.

One-size-fits-all solutions rarely work effectively for a population with such varied challenges. The focus on minimum compliance often results in tools that are technically 'accessible' but practically unusable for those with unique or multiple disabilities.

Individuals with disabilities frequently use AI tools but are rarely involved in their actual design. Only 16% of global AI design representation includes people with disabilities, highlighting a significant disparity.

This lack of representation in design teams contributes to the high rates of unemployment and underemployment within the disabled population. The disconnect between users and creators results in tools that may not fully meet real-world needs.

Despite having various access tools, current AI agents often lack genuine disability-led design, trustworthy contexts, and reliable support for complex communication needs. AI outputs and literacy prompts are frequently too complicated for many users, creating new barriers rather than removing old ones.

Moreover, efforts to "beautify" digital interfaces often inadvertently break compatibility with essential screen readers, undermining accessibility for visually impaired users. This highlights a fundamental flaw in design approaches that prioritize aesthetics over essential functionality for diverse users.

These issues underscore the urgent need for a paradigm shift in AI development, moving towards design processes that inherently consider the diverse requirements of people with disabilities from conception to implementation. Without this shift, AI will continue to provide inadequate solutions, exacerbating existing inequalities.

Project RAISE, for example, discontinued facial tracking due to insufficient data on disabled users, revealing a common industry gap.

Developers often fail to disclose or verify whether disabled individuals are included in their AI training sets, leading to biased or incomplete models. This oversight can result in AI systems that do not accurately perceive or interact with a significant portion of the population.

To ensure universal usability and equitable AI, design must fundamentally pivot from being "built for" people with disabilities to being "built with" them. This inclusive approach extends beyond user interface to foundational data, ensuring that AI agents are truly representative and functional for all.

Many everyday technologies, such as microwaves and mobile devices, were initially designed without considering accessibility. Consequently, accessibility features are frequently added as afterthoughts, rather than being integral components of the initial design.

This reactive approach to accessibility often fails to meet the actual needs of users with physical or cognitive challenges, as retroactive fixes rarely account for the full spectrum of user requirements. Such an approach forces individuals with disabilities to wait for modified versions, creating delays and exacerbating inequalities.

Instead, Dr. Dieker advocates for a proactive "accessible-by-design" framework, where accessibility is embedded from the conceptual stage. This approach not only benefits people with disabilities but also enhances usability for everyone, creating more intuitive and versatile technologies. When accessibility is a core design principle, it leads to products that are inherently better for a broader user base.

This shift from reactive to proactive design is essential for fostering true inclusion and ensuring that technology genuinely empowers all individuals from the very beginning of its development cycle.

Project RAISE’s study, involving 150 students in its final year, revealed that true user agency in AI agents includes the right to reject the technology. While most students liked the AI agent ZB, one student chose to "fire" their agent within the first week of the study.

This incident underscores the importance of allowing users to decide if an AI agent is truly serving their needs. If an agent proves unhelpful, the ability to discontinue its use, or even opt for human assistance, is crucial for maintaining self-determination and ensuring that technology remains a tool for empowerment rather than imposition.

Developing AI for individuals with sensory and multiple disabilities necessitates distinct design modalities. Deaf users often struggle with current agents due to poor lip-reading capabilities and the absence of signing digits, highlighting a critical gap in visual communication.

Similarly, blind users may not benefit from or prioritize the physical visual features of an agent, requiring alternative interfaces. Over-designing or attempting to create 'do-it-all' agents can often fail to address specific deficits like reading or writing constraints, reinforcing the need for highly specialized and adaptable solutions tailored to individual needs.

Dr. Dieker strongly advocates for the use of closed-loop AI systems, especially in educational settings, and advises against autonomous AI for children under third grade. Project RAISE specifically utilized such systems, which were designed not to "learn" or act independently, ensuring a controlled and safe environment.

Concerns persist regarding the lack of adequate guardrails in fully autonomous AI agents, particularly when applied in sensitive areas like mental health. The potential for unintended consequences or harmful interactions necessitates stringent control mechanisms to protect vulnerable users.

Closed-loop systems prevent AI from evolving beyond its programmed parameters, offering a crucial layer of safety and predictability. This approach ensures that AI remains a supportive tool, controlled by human oversight, especially when interacting with children or individuals requiring structured support.

The primary ethical concern in the AI industry is the lack of diverse representation, particularly of women and people with disabilities, among its creators. This demographic imbalance means that AI is often developed by a population that does not reflect its user base, leading to potential biases and oversight.

AThere is a widespread lack of understanding concerning proper "guardrails" for AI, not only among parents and teachers but also among developers. This gap highlights the urgent need for clearer ethical guidelines and robust regulatory frameworks.

The industry must clearly define what "autonomous" means in practice, similar to the strict safety standards seen in sectors like autonomous vehicles (e.g., Waymo). Without inclusive creator teams, the ethical integrity and universal applicability of AI remain significantly compromised.

Dr. Dieker notes that the term 'disability' is used primarily in the United States to meet federal funding, support, and identification guidelines. While legally mandated, this term often fails to capture the nuances of individual conditions.

She proposes 'unique abilities' as a more descriptive and empowering term. Unlike transient challenges, a disability is characterized by its permanence, being an intrinsic part of an individual's life. This distinction emphasizes that while challenges may be overcome, a disability remains a constant, requiring ongoing support and inclusive design.