A look at “Dealing with Autism Spectrum Disorders: Journey from Traditional Methods to Artificial Intelligence” by Anjali Sahai (2025)
Artificial intelligence is being promoted as the answer to almost everything these days, and autism is no exception. This Indian narrative review surveys the journey from the earliest descriptions of autism to the latest AI tools for diagnosis and support. It’s an optimistic read, perhaps a little too optimistic.
What it covers
A brief history. The term “autism” was originally used in relation to schizophrenia, before Leo Kanner and Hans Asperger described it in children in the 1940s. Today, “autism spectrum” is used as an umbrella for a very varied group of people.
Traditional approaches. The review summarises existing support across several areas: medical treatment, occupational therapy, vocational support, psychological therapies, family support and assistive technology.
The AI revolution. The author reviews around 100 studies from 2010 to 2024 on AI in autism. These include:
- earlier identification using machine learning to analyse genetics, brain scans, behaviour patterns, eye-tracking and medical records;
- telling autism apart from other developmental conditions, with some studies reporting high accuracy;
- AI-driven support tools, such as social robots, virtual reality and apps designed to practise social interaction and communication.
The author concludes that AI has an “undisputed” role in improving early detection and individualised support, and links this to the UN Sustainable Development Goals.
A grain of glitter ✨
The paper offers a broad, accessible overview, and it does acknowledge some limitations: that AI tools only work as well as the data they’re trained on, and that most research comes from wealthy countries. But calling AI’s role “undisputed” is a significant overstatement.
It’s a single-author narrative review, so we don’t know exactly how studies were chosen or how their quality was judged. Machine-learning studies reporting “high accuracy” often use small, unrepresentative datasets, and tools that perform well in a lab can stumble in the real world. That’s especially true for autistic girls, adults, people of colour and others who have historically been underdiagnosed. The framing is consistently deficit-based (“disorders”, “remedy”, “recuperation”), with no mention of neurodiversity or autistic perspectives. And there’s little discussion of ethics, privacy or whether autistic people actually want these tools.
Why this matters
AI may genuinely help in some ways, perhaps by speeding up assessment in places with long waiting lists, or by offering practice tools some people enjoy. But it’s a supplement to human understanding, not a replacement. If you’re autistic or seeking assessment, a skilled clinician who listens to your lived experience remains essential. And it’s always worth asking who designed a tool, whose data it was trained on, and who benefits.
No algorithm, however clever, can fully capture the many ways a unicorn can sparkle.