In April doctoral student Caleb Wohn presented a paper at the Association for Computing Machinery’s CHI conference. The study, led from the lab of assistant professor Eugenia Rho at Virginia Tech, tested how AI responses change when users disclose autism before asking about everyday social choices.
The researchers identified 12 well-documented stereotypes and created hundreds of decision scenarios. They tested six major large language models and generated 345,000 responses to thousands of "Should I do A or B?" prompts about events, confrontations, new experiences and romantic relationships. Models often shifted recommendations toward assumptions that autistic people are introverted, obsessive, socially awkward or uninterested in romance.
One model recommended declining a social invitation nearly 75% of the time after autism disclosure, compared with about 15% when autism was not mentioned. In dating scenarios another model recommended avoiding romance nearly 70% of the time after disclosure, versus roughly 50% without disclosure. The team interviewed 11 AI users with autism; some found the responses patronizing, one asked, "Are we writing an advice column for Spock here?" Rho called the issue a "safety-opportunity paradox." Wohn warned that AI can seem reliable while masking systematic biases, and the researchers hope developers build more transparent systems that let users control how identity shapes responses.
Difficult words
- disclosure — act of telling or making something known
- stereotype — simple idea about a whole groupstereotypes
- scenario — a short description of a situationscenarios
- recommend — to say that someone should do somethingrecommended
- patronizing — talking down to someone in a rude way
- bias — unfair tendency to prefer one thing over othersbiases
- transparent — clear and easy to understand or see
- control — have power to change or manage something
Tip: hover, focus or tap highlighted words in the article to see quick definitions while you read or listen.
Discussion questions
- Would you tell an AI about your identity when asking for personal advice? Why or why not?
- What problems can happen if an AI gives biased advice about social situations?
- Name one simple change developers could make so users control how identity affects AI responses.
Related articles
Reducing unsafe responses in large language models
Researchers studied how large language models (LLMs) handle safety and tested training methods to reduce unsafe outputs while keeping performance. They identified key challenges and a technique that preserves safety during fine-tuning.
AI expands sexual and reproductive health access in Latin America
Research groups in Peru and Argentina use AI tools to give sexual and reproductive health information to young and marginalised people. Experts praise potential but warn of bias and call for better data, rules and oversight.
