Understanding How Autistic Young Adults Experience AI-Based Manipulation

Problem

Prior work has found that autistic young adults are more likely to experience harms on social media due to interpreting interface cues literally (Page et al., 2022). For example, participants in prior studies understood “Facebook Friends” as literal friends, which in turn led some to disclose private information or even send money to unsafe online connections. In the context of AI systems, these challenges may be amplified. Particularly when technologies such as targeted advertising adopt personable framing or use language like “this product is for you,” which can be interpreted as direct, individualized communication.

Current Research

We conducted an experimental survey where participants:

Ranked their preferences of various consumer items
Performed intervening tasks answering several questions
Viewed a chatbot interaction where the chatbot exhibited opposite consumer preferences
Lastly, re-ranked their preferences for those items.

We found that autistic young adults with substantial support needs were more likely to change their preferences than the sample from the general population. Moreover, they were also more likely to change their preferences than autistic young adults without substantial support needs. These findings suggest that autistic individuals with substantial support needs are more susceptible to celebrity chatbot persuasion.

Currently, we are in the process of designing and running an interview study where we examine participants' mental models regarding targeted advertising on social media.

Understanding How Autistic Young Adults Experience AI-Based Manipulation

Autistic young adults are more likely to experience harms on social media due to interpreting interface cues literally. In the context of AI systems, these challenges may be amplified.

Problem

Current Research

Publications