AI & Neurodiversity

Artificial intelligence is trained on data reflecting the neurotypical majority, making it a normalising technology by default. This has specific, documented consequences for neurodivergent people. It also has potential to accommodate difference in ways previously impossible.

Umwelten is itself an AI-assisted project. Transparency about that matters. See the About.

Overview

AI and neurodivergent minds — the comprehensive topic overview. The normalisation problem, the surveillance problem, the conditional potentials, and the principles for getting it right.

The dangers

Algorithmic bias and neurodivergent people — hiring algorithms penalise autistic communication, proctoring software flags stimming, plagiarism detectors falsely accuse neurodivergent writers, content moderation suppresses disabled creators. The bias is structural, measurable, documented.

Emotion recognition AI and autism — systems trained on neurotypical faces misread autistic expression. Stillness gets coded as calm when it’s shutdown; intensity as aggression when it’s engagement. The EU banned these systems in workplaces and schools because the evidence supports it.

Autism screening, surveillance, and genetic testing — AI can detect autism from home videos and eye-tracking. Polygenic risk scores are being developed. Care facilities deploy wearable sensors. Each technology has potential benefits and surveillance risks. The Spectrum 10K controversy, the ASAN opposition to the NIH registry, and the Down syndrome parallel.

Social media algorithms and neurodivergent minds — how recommendation algorithms exploit autistic pattern-seeking and ADHD dopamine-seeking. TikTok’s documented suppression of disabled creators. The misinformation cascade. And the genuine community-building that happens on the same platforms.

The potentials

AI-assisted communication — LLM-enhanced AAC, SpeakFaster’s 57% motor efficiency gain, voice synthesis, LLMs as conversation partners. Real promise, conditional on who designs it and whose voice it amplifies.

Getting it right

Designing AI for neurodivergent people — the 23% statistic, participatory design principles, the AutSPACEs model, the accommodation-normalisation distinction, and what neurodiversity-affirming AI actually looks like.

This wiki

Umwelten is itself an AI-assisted project. The process is described in the About because intellectual honesty about knowledge production matters, especially in a project about minds that work differently.