By Stuart Kerr, Technology Correspondent, LiveAIWire

An AI-powered app called NeuroTranslator has grown from an experimental web tool into a paid product used by hundreds of people to decode the kind of everyday social exchange that once left them anxious or confused, all within about two years. For a Cape Town filmmaker named Kate D’hotman, who has autism and ADHD, the shift has been personal rather than technical. She now runs emails, arguments and awkward conversations past ChatGPT before responding to them.

D’hotman is one of a growing number of autistic and otherwise neurodivergent adults who have started treating large language models as informal communication aids, asking them to interpret tone, flag potential misunderstandings, or rehearse a difficult conversation in advance. None of this was designed specifically for autism. It is a repurposing of general-purpose chatbots, and increasingly of purpose-built tools, by people who found that predictable, non-judgemental text often works better for them than a live human reaction does.

How Neurodivergent Adults Are Actually Using These Tools

Reuters, in reporting carried openly by KSL.com, documented how D’hotman uses ChatGPT as an editor, translator and confidant, sometimes instructing it to adopt the tone of a psychologist while she works through a disagreement with a friend. She once uploaded months of messages between herself and that friend, asking the model to point out what she might have missed. “I know it’s a machine,” she said. “But sometimes, honestly, it’s the most empathetic voice in my life.”

Sarah Rickwood, a UK-based project manager with ADHD and autism, described a more practical use case to Reuters: her ideas tend to run ahead of her explanations, and ChatGPT helps her turn scattered thinking into emails and business cases that land the way she intends. Michael Daniel, an Australian engineer, built NeuroTranslator himself after his daughter’s autism diagnosis led to his own, discovering that a throwaway comment about his wife’s shirt had been read as criticism rather than the compliment he meant. The app, and its earlier web version called Autistic Translator, has since attracted several hundred paying subscribers.

What This Means for You

If you or someone close to you is neurodivergent, the practical takeaway is that these tools are already informal and unregulated, which cuts both ways. They are available now, at low cost, without a waiting list or a diagnosis requirement, which is precisely why people are using them. But nothing about a general-purpose chatbot has been clinically validated for this use, and the same feature that makes it comforting, its tendency to be agreeable and affirming, is also the feature that critics warn can quietly discourage someone from developing their own read on social situations rather than outsourcing it entirely.

Early Diagnosis Meets an Old Problem: Consent

A 2025 survey of the research literature, published on arXiv, catalogues a parallel and faster-moving use of AI in autism: early detection. Machine learning models are increasingly able to flag developmental concerns by analysing speech delay patterns, eye movement, or play behaviour in toddlers, often well before a traditional clinical screening would catch them. The same survey highlights social robots such as NAO and Kaspar being used in structured, repetitive interactions that appear to help some autistic children build social skills over time.

The appeal of earlier detection is obvious, but the paper is candid that it raises questions its authors do not resolve: how to obtain meaningful informed consent from a toddler’s guardians for biometric monitoring, how to prevent false positives from shaping a child’s educational path unfairly, and who is accountable when a diagnostic model trained on one population is applied to a child from a demographic it was never validated against.

Giving People Their Voice Back, Not Just a Translation of It

A separate strand of research is trying to solve a different problem entirely: not interpreting other people for autistic users, but helping non-speaking or minimally speaking users sound more like themselves. A prototype system called Speak Ease, described in a paper from Northeastern University researchers, combines speech recognition, a customised GPT model and personalised text-to-speech to let augmentative and alternative communication users convey humour, urgency or sadness, tones that older AAC devices tend to flatten into a single robotic register.

In feedback sessions with speech and language pathologists, the researchers ran into a genuinely difficult ethical question rather than a technical one. Several clinicians raised the spectre of facilitated communication, a discredited practice in which a human assistant was later found to be the true author of a disabled person’s supposed words. One pathologist in the study put the concern plainly, asking who is actually the author when software suggests what a user might want to say. The researchers concluded that AI-driven suggestions are acceptable for casual exchanges but become far riskier in serious conversations, where authorship and authenticity genuinely matter.

The Robot in the Retirement Home

Away from smartphones entirely, researchers at Saint Joseph’s University in Philadelphia have spent over two years piloting an expressive humanoid robot named Pepper inside Bancroft’s Flicker Residences, an assisted living community in New Jersey for aging adults with autism and other developmental disabilities. Pepper leads trivia games, plays music, and has, according to the nonprofit’s clinical staff, produced a level of genuine joy among residents that the pilot did not initially set out to measure.

The underlying research question is more serious than the games suggest. Almost nothing is known scientifically about how autism interacts with aging or dementia risk, largely because autistic adults have historically been excluded from long-term studies of cognitive decline. Bancroft’s clinical director has described the robot as a way to track behavioural changes consistently over time in residents who often communicate through gesture rather than speech, potentially catching early warning signs that current screening tools are not built to detect in this population.

The Real Risk Isn’t Replacement, It’s Overreliance

Not every observer of this trend is reassured by it. Critics interviewed alongside the Reuters reporting warned that constant access to an agreeable, on-demand interpreter could, for some users, reinforce avoidance of unpredictable human interaction rather than build tolerance for it. That concern echoes a broader pattern LiveAIWire has covered in our reporting on AI and emotional manipulation, where systems optimised to match a user’s emotional tone can end up validating a person’s read on a situation rather than genuinely testing it.

The deeper issue, as LiveAIWire’s wider look at emotional AI has detailed, is that the scientific basis for reading emotion from voice, text or facial expression at all remains genuinely contested, which matters more, not less, when the tool is being used by people who already communicate differently from the models these systems were built and tested on. Much like the debate captured in LiveAIWire’s coverage of AI standing in for therapists, the honest answer here is not that these tools are fake support. It is that they are unvalidated support, and unvalidated is not the same thing as harmless or the same thing as useless.

What ties the smartphone-based tools, the diagnostic research and the assisted-living robot together is a single unresolved question, one LiveAIWire has examined in the context of synthetic empathy more broadly: whether a system that convincingly mimics understanding, without any clinical oversight of what it actually gets right or wrong, is functioning as a genuine bridge to easier communication, or as a comfortable substitute for it. For now, the honest answer from the research is that nobody fully knows yet, and the people using these tools every day are not waiting for that answer before they decide for themselves.

About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and everyday life. LiveAIWire publishes original AI journalism every weekday at liveaiwire.com.