Artificial intelligence sparks both excitement and caution, especially when it intersects with accessibility. While skepticism about AI’s current shortcomings is healthy and necessary, it’s equally important to explore its potential to create meaningful change. This article builds on Joe Dolson’s insightful critique by highlighting projects and opportunities where AI can genuinely improve lives for people with disabilities. Rather than dismissing the risks, we focus on what’s possible when AI is applied thoughtfully—with human oversight, context awareness, and ethical design. From smarter alternative text to personalized assistive tools, these eight areas offer a glimpse into a future where technology bridges gaps rather than widens them. Let’s dive into the opportunities that lie ahead, keeping a balanced perspective on both promise and pitfalls.
1. Smarter Alternative Text Generation
Computer-vision models for generating alt text have come a long way, but they still fall short. As Joe Dolson pointed out, current systems analyze images in isolation, missing context and failing to distinguish decorative from informative visuals. Yet the potential remains significant. Instead of replacing human judgment, AI can act as a collaborative tool—offering a rough starting point that authors can refine. For example, a model trained to recognize image usage within a page could flag which images likely need descriptions and which don’t, speeding up the accessibility workflow. Even for complex charts and graphs, AI can provide initial drafts that humans polish. The goal isn’t perfect alt text from a machine; it’s a partnership where AI reduces the grunt work and humans supply the nuance. With ongoing improvements in multimodal models, the path toward context-aware alt text is clearer than ever.
2. Human-in-the-Loop Authoring
The most effective AI applications in accessibility keep humans firmly in the loop. Rather than automating alt text entirely, we can design systems where AI suggests descriptions and users edit them. This approach acknowledges that current models are imperfect but still valuable. Imagine a tool that generates a caption for an image and then asks, “Does this match the intended meaning?” If the user says no, the AI can learn from the correction. This feedback loop improves accuracy over time while respecting the author’s expertise. For people with disabilities who create content, such tools can lower barriers by providing a scaffold. The key is transparency—users should know when AI is guessing and how to override it. By combining machine efficiency with human creativity, we can develop accessible content faster without sacrificing quality.
3. Context-Aware Image Analysis
Current AI models often treat images as standalone objects, but context matters enormously. A photo of a busy street might be decorative in a blog about city life or essential in a news article about traffic accidents. Training models to analyze how an image is used—its placement, surrounding text, and purpose—could revolutionize alt text generation. Such context-aware systems could automatically assign descriptions only to informative images and mark decorative ones as empty. This reduces the manual effort for content creators and ensures that assistive technologies don’t distract users with irrelevant details. Early experiments with multimodal learning show promise: models that combine text and image processing can infer whether a picture adds value or simply illustrates. As these techniques mature, we’ll see more precise accessibility tools that understand not just what is in an image, but why it’s there.
4. AI-Powered Document Structure Remediation
Accessible documents require proper headings, lists, and landmarks, yet many creators struggle with structure. AI can scan PDFs, Word files, or web pages and suggest correct semantic markup. For instance, a model could identify that a large bold sentence is actually a heading and propose an H2 tag. It can also highlight missing alt text on images or tables without summaries. This isn’t about replacing human editors—it’s about reducing the tedious work of manual remediation. Tools already exist that check for contrast ratios or heading order, but AI can go further by understanding the document’s narrative flow. By learning from examples of well-structured content, these systems can make intelligent suggestions that save time and improve compliance with WCAG. The result: more documents that work with screen readers and other assistive technologies, with less effort.
5. Personalized Assistive Interfaces
Accessibility isn’t one-size-fits-all. AI can tailor user interfaces to individual needs—adjusting font size, color contrast, or interaction modes based on a person’s disability. For example, a system might learn that a user with low vision prefers high-contrast, larger text, and automatically apply those settings across websites. Voice navigation could be enhanced with natural language understanding, allowing users to say “skip to the main content” and the AI interprets commands correctly. Machine learning can also detect user frustrations, such as repeated failed clicks on a small button, and offer alternative interactions. Personalization extends to learning preferences: some users may benefit from simplified language summaries of complex articles. By adapting in real time, AI makes digital experiences more inclusive without requiring manual configuration from the user.
6. Real-Time Captioning and Transcription
Live events, meetings, and videos benefit greatly from AI-generated captions. While automatic speech recognition has improved, it still struggles with accents, background noise, and domain-specific terminology. However, with continuous learning and speaker adaptation, AI can produce increasingly accurate captions. For deaf or hard-of-hearing individuals, real-time captions open up access to conversations that might otherwise be lost. Combine speech recognition with natural language processing, and you can also generate summaries or highlight key points. Some systems already allow users to customize caption appearance or choose between verbatim and summarized versions. The challenge remains accuracy and latency, but as models become more robust, we can envision captions that are nearly as reliable as human stenographers. This technology doesn’t just benefit people with hearing loss—it helps non-native speakers and anyone in noisy environments.
7. Sign Language Recognition and Translation
AI models that recognize sign language gestures can bridge communication gaps between deaf and hearing individuals. Computer vision systems can track hand shapes, movements, and facial expressions to interpret signs in real time, then convert them to text or speech. Conversely, AI can generate sign language avatars from spoken or written content. While far from perfect, these tools are evolving rapidly. The biggest hurdles are the diversity of sign languages (each with its own grammar) and the need for large, high-quality training datasets. Still, prototypes exist for applications like video call interpreters or educational tools that teach sign language. As these systems improve, they can facilitate smoother interactions in workplaces, schools, and public services. The key is to involve native signers in development to ensure cultural and linguistic accuracy.
8. Predictive Assistance for Cognitive Disabilities
For individuals with cognitive disabilities, AI can offer predictive support that reduces cognitive load. Example: an email client that suggests simple phrasing or reminds the user to include an attachment. Or a calendar app that predicts when a user might forget a recurring task and sends a gentle prompt. Natural language generation can present complex information in easier-to-understand formats, like bullet points or plain language summaries. AI can also detect patterns that indicate stress or confusion—such as repeated pauses in typing—and offer assistance. These tools are not about taking control but about providing just-in-time help. They must be designed with user consent and respect for privacy. When done right, they empower people to navigate daily tasks with greater independence and less frustration.
AI holds tremendous promise for accessibility, but only if we approach it with realistic expectations and ethical safeguards. The opportunities outlined here—from context-aware alt text to personalized interfaces—show that we are still in the early innings. Skepticism is healthy, but so is optimism. By investing in human-centered AI, we can create tools that amplify abilities rather than replace human judgment. The path forward requires collaboration between technologists, disability advocates, and end users. Together, we can ensure that AI becomes a force for inclusion, not exclusion.