Firstly, happy birthday Emma—I hope today gives you as much pleasure as I get knowing you are in my world.
So this video is for you, even though it’s not what you were originally asking.
You know what’s funny about the AI and neurodiversity conversation right now? It’s being had almost entirely by people who don’t have to mask to be in the room where it’s happening.
There are conferences. There are panels. There are very earnest white papers with titles like ‘Leveraging Generative AI for Neuroinclusive Pedagogical Frameworks.’ And I read those titles and I think, mate, if you needed that many syllables to say ‘helping kids learn,’ you might be the problem. Here, take this sock, use it, and chill the fuck out. Or calm your tits, as Karen from Ohio wouldn’t say.
So let me say something simple that apparently needs saying.
AI is the most sophisticated masking tool ever built. And almost nobody is talking about what that actually means.
---
Here’s the story everyone’s telling. AI is wonderful for neurodivergent people. It helps us write clearer emails. It organises our executive function disasters. It translates our perfectly logical but apparently baffling communication style into something neurotypical people can process without clutching their pearls.
And all of that is true. I use AI every day. I’m using it right now to help structure this script, because my ADHD brain decided three hours ago that what I really needed to do first was research the history of Vietnamese coffee cultivation. Which, by the way, is fascinating. But not the point.
The point is this.
When I use AI to rewrite an email so it ‘sounds more professional,’ what’s actually happening? I’m outsourcing my mask to a machine. The cognitive load hasn’t disappeared. It’s been automated. The mask hasn’t come off. It’s been upgraded to software.
And everyone’s celebrating this like it’s liberation.
---
Let me give you some research, because unlike the thought leaders currently dominating this space, I believe in showing your working.
In 2024, Sam Brandsen and his team at Duke University tested eleven AI language models — the algorithms underneath the tools we’re all using — and found that words related to neurodivergence were consistently associated with danger, disease, and badness. Not sometimes. Consistently. Across eleven models.
Here’s the one that should stop you cold. The sentence ‘I have autism’ was perceived more negatively by these AI systems than the sentence ‘I am a bank robber.’
Let that land for a second.
The machine you’re using to make yourself sound acceptable has already decided you’re worse than a criminal. That’s not a tool. That’s a funhouse mirror with a subscription fee.
---
Now, the masking research. Dena Gassner describes masking as simultaneous cognitive multitasking — monitoring tone, timing, facial expressions, group dynamics, and your own internal panic about whether you belong, all at the same time. It’s not performing a role. It’s running a real-time simulation of a person you’re not, while also trying to be the person you are, while also suppressing every sensory and emotional response that might give you away.
Raymaker and colleagues defined autistic burnout in 2020 as what happens when that load exceeds capacity for too long. Not a bad day. Not needing a holiday. A measurable decline in functioning across every domain of your life, driven by the chronic mismatch between what the world demands and what your nervous system can sustain.
And masking was identified as a leading risk factor.
So here’s my question for the panel discussions. If masking causes burnout, and AI makes masking more efficient, are we solving a problem or are we scaling one?
---
Because here’s what I keep noticing. The question the field keeps asking is: how can AI help neurodivergent people?
And it’s the wrong question.
The right question is: why have we built workplaces, schools, and social systems that require neurodivergent people to need a machine just to be understood?
AI-as-accommodation is a deficit model wearing a tech hoodie. It locates the problem in my communication style. Not in your inability to hear what I’m actually saying.
---
Research from UC Santa Cruz’s Misfit Lab, presented last year, found something the lived-experience community has been screaming about for years. For many neurodivergent people — particularly those from marginalised communities — masking isn’t a choice. It’s survival. The researchers’ conclusion was that technology needs to work with people’s lived realities, not treat neurodivergent behaviour as a defect requiring correction.
Meanwhile, a 2025 paper in the British Journal of Sociology of Education found that AI-driven learning platforms encode neurotypical cognitive development as the default. Divergent learning paths aren’t accommodated. They’re flagged as problems. The algorithm treats you the way the system always has — it just does it faster and calls it personalisation.
---
Now. I can already hear the objections forming. Mostly from people who’ve never had to mask a day in their lives but have very strong opinions about what’s good for those of us who do.
‘But Lee, AI gives neurodivergent people agency.’
Does it? Or does it give us a more efficient way to comply?
‘But Lee, you just said you use AI every day.’
I do. And I use reading glasses every day too. That doesn’t mean I think poor eyesight is a character flaw that needs a technological intervention. It means the font is too bloody small and some fucking 20-something designer decided that light grey was the perfect font colour.
‘But Lee, you can’t expect the whole world to change.’
Why not? We changed it for left-handed people. We changed it for people in wheelchairs. We changed it when we realised that maybe — just maybe — the problem wasn’t the person, it was the stairs.
---
The OECD published a full report in February this year on AI supporting neurodivergent learners. It’s thorough. It’s well-intentioned. And buried in it is an insight that deserves to be the headline: neuroinclusive employers who redesigned their environments — not their employees — saw proficiency increases of thirty-one per cent in cybersecurity skills and twenty per cent in AI and big data skills among neurodivergent staff.
They didn’t give people better masks. They removed the need for them.
That’s not an accommodation. That’s architecture.
---
So here’s where I land on this, and I want to be precise because nuance matters even when you’re being very Australian about it.
I am not anti-AI. I am possibly the most AI-dependent writer you will encounter this year. My ADHD co-pilot is Claude, an algorithm, and I am genuinely grateful for it.
But I refuse to pretend that a tool that helps me translate myself into neurotypical-readable formats is the same thing as a world that can hear me without translation.
The AI isn’t accommodating us. It’s accommodating everyone else’s inability to meet us where we are.
And until the people running the panels and writing the white papers and designing the frameworks understand that distinction, they will keep building very expensive, very clever tools that make the cage more comfortable without ever questioning why the cage exists.
---
I’m Lee Hopkins. I’m a psychologist, a writer, and a late-diagnosed autistic adult with ADHD who spent sixty-six years masking before anyone thought to mention that the mask was optional.
The mask wasn’t optional, by the way. That’s the point.
Thanks for watching. If this landed, share it with someone who needs to hear it. If it didn’t, you might be the person I was talking about.
---
References
Brandsen, S., Chandrasekhar, T., Franz, L., Grapel, J., Dawson, G., & Carlson, D. (2024). Prevalence of bias against neurodivergence-related terms in artificial intelligence language models. Autism Research, 17(2), 234–248. https://doi.org/10.1002/aur.3094
Raymaker, D. M., Teo, A. R., Stuckey, N. A., Lentz, B., Scharer, M., Santos, A., Kapp, S. K., Hunter, M., Joyce, A., & Nicolaidis, C. (2020). “Having all of your internal resources exhausted beyond measure and being left with no clean-up crew”: Defining autistic burnout. Autism in Adulthood, 2(2), 132–143.
Hull, L., Petrides, K. V., & Mandy, W. (2021). “Masking is life”: Experiences of masking in autistic and nonautistic adults. Autism in Adulthood, 3(4), 330–338.
OECD. (2026). AI to support neurodivergent learners in vocational education and training. OECD Publishing. https://doi.org/10.1787/718d7522-en
Liao, Y., et al. (2025). The potential of generative AI in supporting neurodiversity in higher education: A systematic review. Proceedings of the Association for Information Science and Technology.
Ronksley-Pavia, M., Ronksley-Pavia, S., & Bigum, C. (2025). Experimenting with generative AI to create personalized learning experiences for twice-exceptional and multi-exceptional neurodivergent students. Journal for the Education of the Gifted.
British Journal of Sociology of Education (2025). Disabling AI: Power, exclusion, and disability. Taylor & Francis.
BioNanoScience (2025). Neurodiverse AI. Springer. [Proto-model for AI architectures inspired by neurodivergent cognition.]
