When AI Functions as an Autism Support

When I was a child, I spent hours with action figures.

I was especially obsessed with Dragon Ball Z. I did not simply reenact episodes. The characters developed relationships, conflicts, histories, and entire stories in my head. Scenes continued long after I put the toys away.

For years, I thought this was just the child part of me refusing to disappear.

Eventually I understood it differently. The complicated internal stories had not gone away because they were part of how my imagination worked. As I grew older, they became fanfiction, original stories, and eventually novels. Other works shaped that imagination too. My Little Pony: Friendship Is Magic, for example, made me think deeply about friendship, mercy, restoration, and what gives a story its moral center.

The imagination was always there.

The difficulty was translating it.

As an autistic person, I can have a scene, argument, or idea alive in my head with far more detail than I can immediately put into orderly paragraphs. Ideas may arrive together rather than one after another. I can know what I am trying to say before I know how to make another person follow the path I took to get there.

That experience is mine, not a description of every autistic person. Autism is extraordinarily varied. But research on autistic adults likewise warns against reducing communication differences to a single deficit within the autistic person. A 2025 review of autistic adult communication found that barriers can arise from language and nonverbal differences, but also from environments and communication partners that do not accommodate autistic communication. The authors recommend a multifaceted approach to communication support rather than assuming one method will work for everyone.

For me, one of those supports has become conversational artificial intelligence (AI).

The Difference Between Having an Idea and Translating One

Speech-to-text was already useful to me because speaking an idea can be easier than forcing it through a keyboard one sentence at a time. Conversational AI added another layer.

I can dictate a large, tangled explanation and then work with the system to organize what I have said. I can ask it where my reasoning becomes difficult to follow. I can use it to compare possible structures, identify a missing transition, or point out that something I understand intuitively has never actually been explained to the reader.

That is very different, in my experience, from sitting down with nothing to say and asking a machine to invent something for me.

That distinction matters because if I already have the characters, conflict, argument, emotional direction, and intended meaning, the problem may be getting all of that through a bottleneck between my internal thought and the page. In those situations, I experience AI less as a replacement for imagination and more as an interface through which I can inspect and organize my own thinking.

There is precedent for thinking broadly about communication support. Researchers examining augmentative and alternative communication (AAC) for speaking autistic adults have emphasized that communication — not speech for its own sake — should be the priority, and that mainstream technologies can sometimes supplement communication in useful ways.

Generative AI is not the same thing as established AAC, and I do not think it should casually be placed in the same category. It can generate language I did not supply and introduce suggestions or directions of its own, which creates ethical and authorship questions that a simple transcription tool does not. For me, that history suggests a useful principle: we should pay attention to what a tool enables a person to do, not merely what category of technology it belongs to.

A Support Can Also Become a Substitute

My positive experience does not mean every use of AI is helpful.

Sometimes AI produces a sentence that is cleaner than mine and also much less like me.

That can be dangerous precisely because the sentence sounds good.

One of the most important things I have learned is that polish is not the same as voice. Sometimes the most polished sentence is the first one I reject. It may be too smooth, too generic, too certain, or simply something I would never say.

AI systems can also be wrong. They can invent information, misunderstand a source, flatten nuance, or confidently carry an incorrect assumption through an entire discussion. Anything factual still has to be checked.

I also worry about dependency. If the purpose of a support is to help me express my judgment, I do not want to gradually surrender that judgment to the support itself.

That is why I am uncomfortable with both extremes in the current AI debate.

I do not believe that because a tool can function as an accessibility support for me, every use of it is therefore justified. Accessibility does not erase rules about school assignments, employment, publication, contests, confidentiality, or authorship.

But I also do not think that the words “used AI” tell us enough to decide that a person’s contribution has disappeared.

Those positions can both be true.

Autism Does Not Make Us One Bloc

There is another reason for caution: autistic people do not have one collective opinion about AI.

Some autistic people may find conversational systems useful. Others may dislike them, distrust the companies behind them, object to how the technology was developed, worry about privacy or environmental costs, or simply find that interacting with AI makes their communication worse rather than better.

Those views deserve to be taken seriously.

Communication needs themselves vary across autistic adults and across situations. Research describes barriers arising not only from individual language or nonverbal differences, but also from the environments and communication partners around an autistic person.

So I would not prescribe my workflow to another autistic person.

What I would ask instead is whether we can make enough room for autistic people to discover which supports genuinely increase their ability to communicate without assuming that every unconventional method is evidence that the person behind it has less to say.

Questions I Ask Myself

Because conversational AI can cross the line from support into substitution, I have found it useful to ask myself a few questions while working:

Did this idea begin with me?

Can I explain why I believe what I am saying?

Am I making the important creative or argumentative decisions?

Does the final wording still sound like something I would actually say?

Have I checked factual claims rather than trusting the system because it sounds confident?

Am I following the rules of the place where I intend to submit the work?

And could I comfortably explain what the tool did if someone asked?

Those questions do not solve every ethical problem surrounding generative AI. They help me notice whether the technology is still supporting my work or quietly beginning to replace parts of it that I want to remain mine.

That boundary matters to me as a writer, but it also matters to me as an autistic person.

For much of my life, there was never a shortage of material inside my head. The challenge was finding dependable ways to bring more of it outside.

Conversational AI has become one of those ways.

I do not need it to imagine for me. I need to make sure that when I use it, the technology helps more of me become understandable rather than making me disappear behind it.

That, to me, is what an autism support should do.

Jonah Baird is an autistic author and founder of Clean Lantern Books, an independent publishing imprint. He writes fiction and nonfiction about disability, creativity, technology, moral imagination, and the practical realities of independent creative work. He lives in Fresno, California. For more information, email [email protected].

Sources

de Marchena, A., Cuneo, N., Gurbuz, E., Brown, M., Trujillo, J., & Bergstrom, J. “Communication in Autistic Adults: An Action-Focused Review.” Current Psychiatry Reports, 2025. https://pubmed.ncbi.nlm.nih.gov/40545515/

National Institute of Standards and Technology. “Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile.” 2024. https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence

Zisk, A. H., & Dalton, E. “Augmentative and Alternative Communication for Speaking Autistic Adults: Overview and Recommendations.” Autism in Adulthood. https://pubmed.ncbi.nlm.nih.gov/36601528/