What AI assistive technology means for people with autism
Every year on June 4, the world marks World Day for Assistive Technology. The idea at the centre of it is simple but radical: support tools — whether for communication, mobility, learning, or daily living — should be available to everyone who needs them, not just people in wealthy countries or well-funded institutions. The day draws attention to a global gap. Need is widespread. Access is not.
This year, I want to talk about something specific: the growing role of artificial intelligence in assistive technology and what it actually means for people with autism. Not in a “look how far we’ve come” way. More in a “here’s what’s real, here’s what’s promising, and here’s what we need to get right” way.
Why? Because I’m autistic, and this matters to me personally.
What AI in assistive technology actually is
Let’s get clear on terms. AI in assistive technology is not magic, and it is not a single product. It is a layer of functionality, built into different tools, that helps a system interpret information, respond to users, and do certain things automatically based on patterns it has learned.
In practical terms, this might mean a tool that converts speech to text in real time, one that adjusts language complexity based on how you interact with it, or one that helps you structure a message when the words feel stuck. It can also mean something that filters sound in a noisy environment or breaks a complex set of instructions into smaller steps you can follow one at a time.
These are not hypothetical. Apps like Proloquo2Go use machine learning to predict vocabulary based on context and user history so if someone frequently selects “apple” at snack time, the app begins to prioritise that word. Voiceitt, developed with Microsoft, translates atypical speech patterns into clear speech, and in a 2022 trial, 83% of users said it improved their ability to express their needs. Google’s Project Euphonia has been working on speech recognition that actually works for people with non-standard speech, reducing error rates by around 80% compared to generic systems.
These are meaningful developments. They represent a genuine shift from assistive tools that were rigid and required users to adapt to them towards tools that try to meet users where they are.
Where it makes a real difference
For autistic people, barriers tend to cluster around three areas: communication (expressing or interpreting language in real time), sensory processing (sensitivity to sound, light, or other input), and social interpretation (reading tone, subtext, or unspoken cues). None of these is universal — autism is not one experience — but they are common enough to shape how many of us move through the world.
AI tools are increasingly being designed with these specifically in mind.
In communication, AI-assisted augmentative and alternative communication (AAC) systems can help generate spoken output, predict what someone is trying to say, and suggest sentence structures that make intent clearer. AAC tools help people with autism participate more actively in conversations and social activities, allowing them to communicate their needs and make autonomous decisions, supporting a greater level of independence in daily life.
For sensory processing, adaptive systems can filter noise or modulate input. For the classroom or workplace, real-time transcription means you do not have to hold rapid speech in your working memory whilst also trying to process what was said. You can read it back. You can revisit it. That matters.
For social navigation, some tools now try to provide contextual cues — clarifying tone in messages, explaining implied meanings, or offering structured prompts for conversations that feel uncertain. For autistic individuals, AI interactions can carry lower demand: they wait as long as needed for a response and do not feel hurt if you step away for a while. For someone who experiences anxiety around real-time conversation, that is not a trivial thing.
AI could also help personalise assistive technologies in real time, automatically adjusting settings based on how someone interacts or detecting signs of stress or frustration and offering targeted support at the right moment.
What assistive technology does not do well
I want to be clear about the limitations because they matter and are often glossed over in enthusiasm for what is new.
Most AI systems are trained on neurotypical communication patterns. That means they can misread autistic communication, labelling tone as flat, inferring incorrect intent from literal phrasing, or nudging output towards standard expressions that do not reflect how the user actually thinks or speaks.
Autistic people’s engagement with AI technology is rarely purely beneficial, and the research backs this up. AI-based monitoring systems used in classrooms and clinics sometimes track students’ facial expressions or gaze and send reports to instructors, rather than empowering the students themselves, creating agency by proxy, where the technology becomes another layer through which others interpret and manage a disabled person, rather than a tool of self-expression.
This is a real risk. When a tool consistently “corrects” how you communicate, the message it sends, however unintentionally, is that your way of communicating is wrong. That is not assistive. That is normalising. Technical and ethical concerns persist around AI in autism care, even as the tools grow more capable.
There are also privacy concerns. Many AI assistive tools rely on continuous data collection, such as voice recordings, behavioural tracking, and inferred emotional states. For people who depend on these tools daily, opting out is not always realistic. That makes transparency and strong privacy protections non-negotiable, not optional extras.
And then there is the access gap. Most assistive technologies for autism are either extremely costly or geographically limited, making them inaccessible for lower-income families. An AI tool that is genuinely useful but locked behind a subscription, requires a stable broadband connection, or is only available in English, is not solving an equity problem. It is replicating one.
What autistic communities actually think
Here is the thing: there is no single autistic view on assistive technology. I cannot speak for everyone, and I would not try to.
Some people value tools that reduce communication effort, support sensory regulation, or make unpredictable environments feel more manageable. Others are concerned that assistive systems can push us toward conforming to neurotypical norms or override the ways we naturally express ourselves. Many of us hold both views at once, depending on the context and how much control we have over the tool.
What matters is that individuals can embrace an autonomous trajectory for themselves and that AI is designed to support that, not to determine it. The question is not whether a tool is technically impressive. It is whether it gives control to the person using it.
Governance structures must support direct participation where possible, and even for higher-support autistic users, systems should be adaptable so that as communicative capacity develops, control transitions back to the person, not away from them.
What responsible AI assistive tech should look like
If AI is going to play a meaningful role in supporting autistic people, it needs to be built differently. Researchers and advocates have been increasingly clear about what that requires:
Users should be able to adjust, limit, or turn off predictive and interpretive features, not have them on by default. Systems should make it clear when they are guessing or inferring, especially around emotional and social contexts. Assistance should be modular, so users choose when they want support and when they do not.
Training data needs to include neurodivergent communication styles, not just neurotypical patterns of speech, tone, and sentence structure. And data collection should be minimised, clearly explained, and not required for basic functionality.
Assistive technologies can both promote and undermine autonomy, which means their development requires ethical oversight. The excitement around what AI can do must be matched by seriousness about who it is designed for, and who gets to decide what “helpful” actually means.
The bigger picture
The rise of AI in assistive technology is not just a technological shift. It is a social one.
On World Day for Assistive Technology, the central question is not what these tools are capable of. It is who they serve, who shaped them, and who gets to use them at all. Access to effective support should not be determined by geography, income, or whether your way of communicating happens to match what a model was trained on.
AI has real potential here. But potential is not the same as progress. Progress looks like tools designed with autistic people, not just for them. It looks like systems that expand what someone can express, rather than narrowing it to what fits a statistical average. And it looks like the same quality of support is available in Lagos as in London.
That is the standard. We are not there yet. But it is worth naming clearly, so we know what we are working towards.
